Datasets:

kilian-group
/

arxiv-classifier

Dataset card Data Studio Files Files and versions Community

Subset (4)

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

Split (2)

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

paper_id stringlengths 9 16	version stringclasses 26 values	yymm stringclasses 311 values	created timestamp[s]	title stringlengths 6 335	secondary_subfield sequencelengths 1 8	abstract stringlengths 25 3.93k	primary_subfield stringclasses 124 values	field stringclasses 20 values	fulltext stringlengths 0 2.84M
1607.07633	3	1607	2018-04-27T10:53:22	Finite dual of a cocommutative Hopf algebroid. Application to linear differential matrix equations and Picard-Vessiot theory	[ "math.RA", "math.AC", "math.AG" ]	A fundamental tool of Differential Galois Theory is the assignment of an algebraic group to each finite-dimensional differential module over differential field in such a way that the category of differential modules it generates is equivalent, as a symmetric monoidal category, to the category of representations of the group. Its underlying set is then recognized as the group of differential automorphisms of the Picard-Vessiot field extension of the base field for this differential module. These results can be obtained by means of a Tannaka reconstruction process, applied to the abelian category of finite-dimensional differential modules. In this paper, we explore the possibility of extending this theory when the differential field is replaced by a more general differential ring $A$. In this case, it is reasonable to deal with differential modules which are finitely generated and projective over $A$. A major obstacle is that this category is not abelian, in contrast with the classical case when $A$ is a field. To overcome this difficulty, we develop some fundamental results concerning the \emph{finite dual}, hereby introduced, of a cocommutative Hopf algebroid, and a canonical monoidal functor sending (differential) modules to comodules. This functor is proved to be an equivalence of categories whenever a canonical ring homomorphism, which is introduced in this work, with domain in the aforementioned finite dual and with values in the convolutional ring, is injective. Module-theoretical sufficient conditions to get its injectivity are investigated. This machinery is applied to differential modules and their Picard-Vessiot theory.	math.RA	math	ON THE FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID. APPLICATION TO LINEAR DIFFERENTIAL MATRIX EQUATIONS AND PICARD-VESSIOT THEORY. LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS Abstract. A fundamental tool of Differential Galois Theory is the assignment of an algebraic group to each finite-dimensional differential module over differential field in such a way that the category of differential modules it generates is equivalent, as a symmetric monoidal category, to the category of representations of the group. Its underlying set is then recognized as the group of differential automorphisms of the Picard-Vessiot field extension of the base field for this differential module. These results can be obtained by means of a Tannaka reconstruction process, applied to the abelian category of finite-dimensional differential modules. In this paper, we explore the possibility of extending this theory when the differential field is replaced by a more general differential ring A. In this case, it is reasonable to deal with differential modules which are finitely generated and projective over A. A major obstacle is that this category is not abelian, in contrast with the classical case when A is a field. To overcome this difficulty, we develop some fundamental results concerning the finite dual, hereby introduced, of a cocommutative Hopf algebroid, and a canonical monoidal functor sending (differential) modules to comodules. This functor is proved to be an equivalence of categories whenever a canonical ring homomorphism, which is introduced in this work, with domain in the aforementioned finite dual and with values in the convolutional ring, is injective. Module-theoretical sufficient conditions to get its injectivity are investigated. This machinery is applied to differential modules and their Picard-Vessiot theory. Contents Infinite comatrix corings Introduction Motivation, background and overview Description of the main results 1. Preliminaries: Bialgebroids and Hopf algebroids 1.1. Notations and basic notions 1.2. Bialgebroids and Hopf algebroids 1.3. Examples of (right) Hopf algebroids. 2. The reconstruction process, Galois corings and principal bi-bundles. 2.1. 2.2. Review on Hopf algebroids constructed from fibred functors 2.3. Reconstruction and Galois corings 2.4. Principal bi-bundles attached to two different fibre functors 3. The finite dual of a ring extension. 3.1. The finite dual coring R◦ 3.2. The functors χ and L 3.3. The injectivity of the map ζ : R◦ → R∗ 3.4. Dual coalgebras 4. The finite dual of a right bialgebroid, and of a cocommutative Hopf algebroid. 4.1. Duality for bialgebroids 4.2. Duality for Hopf algebroids 5. Application: The finite dual of the first Weyl algebra, differential Galois groupoid and PV extensions 5.1. The Hopf algebroid structure of the first Weyl algebra 5.2. Differential modules Diff 5.3. The commutative Hopf algebroid attached to the first Weyl algebra 5.4. The differential Galois groupoid of a differential module. 5.5. Picard-Vessiot extensions for linear differential matrix equation, after Andr´e 5.6. Comparison with Malgrange's and Umemura's differential Galois groupoids References A as a Tannakian category 2 2 3 3 3 4 6 7 7 9 13 14 17 17 17 18 19 20 20 21 23 23 24 25 26 31 35 43 Date: September 17, 2018. 2010 Mathematics Subject Classification. Primary 18D10, 13N10, 16W25; Secondary 13B99, 16T10, 20G99, 34G10. Key words and phrases. Lie algebroids, Hopf algebroids, Differential modules, Linear differential equations, Rings of differential operators, Weyl algebra, Tannaka reconstruction, differential Galois groupoid, Picard-Vessiot extension of differential rings. Supported by grants MTM2016 -- 77033-P and MTM2013 -- 41992-P from the Spanish Ministerio de Econom´ıa y Competitividad and the European Union FEDER. 1 2 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS Introduction Motivation, background and overview. The classical Galois theory of linear differential matrix equations with coefficients in a differential field, has two essential interrelated parts. The first one deals with the representation theory of the differential Galois group attached to the system, while the second seeks for the space of solutions, namely, a simple differential algebra (the Picard-Vessiot extension) generated by a fundamental solution matrix and the inverse of its determinant. The group of differential automorphisms of this simple differential algebra coincides with the underlying set of the Galois group. In this paper we aim to explore the possibility of extending the first part of this theory when differential field is replaced by a differential ring, making use of the finite dual Hopf algebroid of the co-commutative Hopf algebroid built from a suitable Lie-Rinehart algebra (the module of global sections of the tangent bundle). Concerning the second part of the classical theory, we follow Yves Andr´e's approach given in [2]. Although, in [2], no use of groupoids or Hopf algebroids was made, we are able to identify the Picard-Vessiot extension of the differential polynomial algebra for any finite rank differential module, as the total isotropy Hopf algebra (over the base ring) of a commutative Hopf sub-algebroid of this finite dual Hopf algebroid. Given a Lie-Rinehart algebra L over a Dedekind domain A, we construct an affine groupoid over (the spectrum of) A whose category of A -- profinite1 representations is isomorphic, as a monoidal symmetric category, to the category of A -- profinite right modules over the universal enveloping algebroid of L, that is, the category of representations of L. In particular, this applies to the global sections of any Lie algebroid over an irreducible smooth curve over an algebraically closed field. For instance, if A is the coordinate ring of the complex affine line A1 C and L is the module of global sections of the transitive Lie algebroid of vector fields, then we have a monoidal equivalence between the category of all differential modules Diff A and the category of A-profinite comodules over a commutative Hopf algebroid U ◦, the finite dual of the first Weyl C -- algebra U viewed as the universal enveloping algebroid of L. Assume now we are given a differential A-module (or equivalently a linear differential matrix equation) (M, ∂) of rank m and consider the smallest category hMi⊗ generated by (M, ∂) and its dual, and closed under tensor products and sub-quotients. Then, in analogy with the classical differential Galois theory, we show that there exists an affine algebraic C -- M whose category of representations is equivalent as a monoidal symmetric category to hMi⊗. groupoid H We construct the representing algebra of (the objects of) H of U ◦. Furthermore, we show that the (fibre) groupoid H H linear group GLm(C) along the map A1 M(C) which is unique up to weak equivalences, is then termed the differential Galois groupoid attached to (M, ∂). Our approach is, in a sense, a naive one, since we do not construct a solution space (perhaps a differential algebra over A ⊗C A) whose 'groupoid of differential automorphisms' coincides with H M(C). We do, however, construct a Picard-Vessiot extension, in the sense of Andr´e [2], associated to a differential module M as a quotient Hopf A-algebra of U ◦ (M) by the Hopf ideal generated by the image of the subtraction of the source from the target. In other words, this is the total isotropy Hopf algebra of U ◦ (M) (Proposition 5.5.2). In this way, any of the isotropy groups of H M(C) is recognized as the group of differential automorphisms of this total isotropy Hopf algebra (Proposition 5.5.5). M as a finitely generated Hopf sub-algebroid U ◦ (M) M(C) is transitive and there is a monomorphism M(C) ֒→ G m of groupoids, where G m = (cid:0)A1 C(cid:1) is the pull-back groupoid of the general C × GLm(C) × A1 C; A1 C → {∗}. The algebraic groupoid H It is noteworthy to mention that in the literature there are other notions of Galois groupoid in the differ- ential context. For instance, Malgrange in [26], based on Umemura's theory [43], introduces a differential Galois groupoid (a Lie groupoid of a foliation). In the previous situation of the differential module (M, ∂) with rank m = 2, the attached Hopf algebroid of Malgrange's Galois groupoid is the quotient Hopf alge- j )−1]α=(α0,α1,α2) ∈ N3 over the algebra C[x, x1, x2], by broid of the Hopf algebroid C[x, x1, x2, y, y1, y2, yα some differential Hopf ideal, see the last part of Example 5.6.8. Thus, as a groupoid in the set theoretical sense, the objects set of Malgrange's Galois groupoid is the affine space A3 C, as it would be expected. Thus, Malgrange's and Umemura's approaches run in a totally different direction, see Subsection 5.6 for more details and explanations. C rather than A1 j , det(yǫi We already work in a more general context. Thus, let U be a co-commutative right Hopf algebroid over a commutative ring A, and let AU be the category of all right U -- modules which are finitely generated and projective as A -- modules (A -- profinite, for short). The category AU is monoidal symmetric and rigid, and 1i.e., finitely generated and projective A-modules. FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 3 obviously additive, but it is not abelian in general (even if A is a Dedekind domain). Thus, the Tannakian recontruction from [10, 5] will not produce a monoidal equivalence from AU to the category of A -- profinite representations of some affine groupoid. However, we will show that such a reconstruction process gives a commutative Hopf algebroid U ◦ over A and, what is crucial to get our results in the case where AU is not abelian, a homomorphism of rings ζ from U ◦ to the convolution ring U ∗. We prove that if ζ : U ◦ → U ∗ is injective, then AU is isomorphic, as a symmetric monoidal category, to the category AU◦ of A -- profinite right U ◦ -- comodules. In particular, this machinery can be applied when U = U(L) is the universal enveloping algebroid of a Lie-Rinehart algebra L over A, producing a an affine groupoid represented by U ◦ with the 'same' representation theory than L. We give some sufficient conditions for the injectivity of ζ. In particular, we show that ζ is injective for every co-commutative Hopf algebroid U whenever A is a Dedekind domain. Description of the main results. The paper is organized as follows. In Section 1, besides the statement of the general notations used along the paper, we collect some basic information on Hopf algebroids an their relationship to Lie-Rinehart algebras. Section 2 is devoted to describe those aspects of the reconstruction process related to corings (or cog´ebroıds, according to [5]) that will be useful in the sequel, including a review on the bialgebroids and Hopf algebroids constructed from fibre functors. Principal bi-bundles at- tached to pairs of fibre functors are also considered, with the aim of clarifying the Picard-Vessiot extension constructed in Section 5. The application of the reconstruction procces to a morphism of (possibly non commutative) rings A → R leads in Section 3 to the (right) finite dual A -- coring R◦, and a functor χ from the category AR of A -- profinite right R -- modules and the category AR◦ of A -- profinite right R◦ -- comodules. In contrast with the case of algebras over a field, it is not known if this functor is an equivalence of categories for a general A -- ring R. We construct a homomorphism of A -- bimodules ζ : R◦ → R∗, where R∗ denotes the right dual, as an A -- module, of R, and it is shown that, if ζ is injective, then χ : AR → AR◦ is an equivalence of categories (Proposition 3.3.2). A module-theoretical condition implying the injectivity of ζ are investigated in Proposition 3.3.4. As a consequence, χ is an equivalence of categories if A is a right hereditary right noetherian ring (e.g. if it is a Dedekind domain or a semi-simple Artinian ring). Section 4 starts by showing that, if (U, A) is a right bialgebroid, then the map ζ : U ◦ → U ∗ is a homomorphism of A⊗Aop -- rings. In fact, ζ becomes part of a canonical structure of left bialgebroid (U ◦, A). Theorem 4.2.2 contains our main contribution of Section 4, namely, the fact that if U is a cocommutative Hopf algebroid over a commutative ring A, then χ is a monoidal equivalence between AU and AU◦ . Section 5 developes the aforementioned application of our theory to differential modules, as described above. Besides, in Subsection 5.6, we explicitly describe the structure maps of the commutative Hopf algebroid attached to what is in the literature known as Malgrange's groupoid (or D-groupoid) [26]. We also discuss in this section, using some specific examples, Umemura's [42] approach to these groupoids. 1. Preliminaries: Bialgebroids and Hopf algebroids In this section, for the convenience of the reader unfamiliar with Hopf algebroids, we introduce some basic notions and results concerning this theory. We also fix some notations and terminology to be used along the paper. 1.1. Notations and basic notions. We work over a unital commutative ground ring K. All additive cate- gories will be assumed to be K -- additive categories, and additive functors between them will be K -- linear. For a category A, the notation X ∈ A means that X is an object of A. The identity morphism of an ob- ject X ∈ A is denoted simply by the object itself or by the symbol 1X. Algebra means associative and unital K-algebra and the notation Z(A) stands for the center of an algebra A. We denote by Ae := A ⊗K Ao the enveloping algebra of an algebra A, where Aop is the opposite algebra of A, whose elements are dis- tinguished from those of A by using the notation ao ∈ Aop, for a ∈ A. For simplicity, the same notation Ae := A ⊗L Aop will be used when K → L is commutative ring extension and A is an L-algebra. Modules are unital modules, and (A, B)-bimodules are assumed to be central K-bimodules. We denote by AModB the category of all (A, B)-bimodules; AMod and ModB denote, respectively, the category of left A-modules and the category of right B-modules. The corresponding hom-set functors will be denoted, respectively, by 4 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS HomA-(−, −) and Hom-B(−, −). For every (A, B)-bimodule AXB, we denote by X∗ = Hom-B(X, B) its right dual, while by ∗X = HomA-(X, A) its left dual, which are considered as (B, A)-bimodules with the actions: bϕa : x 7→ bϕ(ax), bψa : x 7→ ψ(xb)a, for all x ∈ X, ϕ ∈ X∗, ψ ∈ ∗X, a ∈ A, b ∈ B. (1) For every algebra A, let add(AA) denote the full sub-category of ModA whose objects are all finitely generated and projective (fgp for short) right A-modules; we also use the terminology right A-profinite modules for the objects in this sub-category. Given a morphism of algebras η : A → R, we denote by η∗ : ModR → ModA and ∗η : RMod → AMod the corresponding restriction of scalars functors. It is clear that, if η∗(R)A is fgp we then have a functor η∗ : add(RR) → add(AA). An algebra R is said to be an A-ring if there is a morphism of algebras η : A → R (also called a ring extension). By restriction of scalars, R becomes an A-bimodule and its multiplication factors through the tensor product over A, that is, its multiplication can be understood as a map µ : R ⊗A R → R. In this way the triple (R, η, µ) can be seen as a monoid in the monoidal category AModA. Saying that R is an Ae-ring is equivalent to say that there is a morphism of algebras s : A → R and an anti-morphism of algebras t : A → R such that s(a)t(a′) = t(a′)s(a), for every a, a′ ∈ A. Dually, an A-coring [40] is a comonoid object in the category AModA. That is, a triple (C, ε, ∆), where C is an A-bimodule together with two A-bimodule maps, the comultiplication and the counit ∆ : C −→ C⊗A C, ε : C −→ A, which satisfy the usual coassociativity and counitary properties. A right C-comodule is a right A-module M together with a right A-linear map M : M → M ⊗A C (called right coaction) such that (M ⊗A ∆C) ◦ M = (M ⊗A C) ◦ M, and (M ⊗A εC) ◦ M = M. A morphism of right C-modules f : M → N (right colinear map) is right A-linear map such that ( f ⊗A C) ◦ M = N ◦ f . The category of all right C-comodules and their morphisms will be denoted by ComodC. We recall from [36, 37, 39, 41] the definition and basic properties of (left, right) Hopf algebroids. A good survey on the subject is the monograph [4]. 1.2. Bialgebroids and Hopf algebroids. Fix A a ground K-algebra, and let V be an Ae-ring via a ring homomorphism η : Ae → V whose source and target maps are respectively denoted by s : A → V and t : Aop → V. The pair (A, V) is said to be a left biagebroid (or V is a left ×A-bialgebra) provided that its category of left V-modules is a monoidal category and the restriction of scalars functor Ol : V Mod → Ae Mod, Ol = ∗(s ⊗ t), is a strict monoidal functor. In particular, the underlying bimodule Ae V admits a structure of A-coring, and if we denote by ♦ the given tensor product of V Mod and consider two objects X, Y ∈ V Mod, then the underlying left Ae-module of X ♦Y is the A-bimodule X ⊗A Y whose left V-action is given by the formula v.(x ⊗A y) = v1 x ⊗A v2y, (2) where ∆(v) = v1 ⊗A v2 is the comultiplication of the underlying A-coring of V (we are using a simplified version of Sweedler's notation with sum understood). It is known from [37, Proposition 3.3] that the category V Mod is right closed with right inner hom- functors homV Mod(X, Y) := HomV-(V ♦X, Y), (3) where the left V-action of the right hand term is given by the V-bimodule V ♦X whose left V-action is (2) and its right V-action is induced by that of the factor V. This means that for any object X ∈ V Mod, the functor −♦X : V Mod → V Mod is left adjoint to the functor HomV-(V ♦X, −) : V Mod → V Mod called the right inner hom-functor and denoted by homV Mod(X, −). This adjunction is easily seen once observed that Y ♦X (cid:27) (V ♦X) ⊗V Y is a left V-linear isomorphism via the map x ⊗A y 7→ (1 ⊗A x) ⊗V y, and it uses the usual Hom-tensor adjunction isomorphism to get HomV-(Y ♦X, Z) (cid:27) HomV-((V ♦X) ⊗V Y, Z) (cid:27) HomV-(Y, HomV-(V ♦X, Z)). If we want to compute the right inner hom-functors in the category of modules V Mod via the forgetful functor Ol to the monoidal category Ae Mod, then it is better to resume the previous situation in form of a FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 5 diagram. So with the previous notations, we have a commutative diagram HomV-(Y ♦X, Z) Ol (cid:27) HomV-(Y, HomV-(V ♦X, Z)) Ol HomAe -(Y ⊗A X, Z) /❴❴❴❴❴❴❴❴❴❴❴❴❴❴❴ HomAe -(Y, HomV-(V ♦X, Z)) (cid:27) HomAe -(Y, TZ ) HomAe -(Y, HomAop -(X, Z)) (cid:27) / HomAe -(Y, HomV-(V ⊗Aop X, Z)), where the natural transformation TZ : HomV-(V ♦X, Z) → HomV-(V ⊗Aop X, Z) is defined by sending f 7→ (cid:2)v ⊗Aop x 7→ f (v1 ⊗A v2 x)(cid:3). The dashed arrow is then a natural isomorphism if and only if T− is a natural isomorphism, if and only if each map βX : V ⊗Aop X → V ♦X sending v ⊗Aop x 7→ v1 ⊗A v2 x is an isomorphism of left V-modules, where in the tensor product V ⊗Aop X we have used the bimodule V V1⊗Aop . Now, one can easily see that the maps βX are isomorphisms if and only if βV is an isomorphism. Following [37, Theorem 3.5], a left A-bialgebroid V is said to be a left Hopf algebroid provided that the functor Ol preserves right inner hom-functors. As we have seen this is equivalent to say that βV is an isomorphism. Its inverse induces then a well defined map β−1 V (−⊗A 1) : V → V ⊗Aop V sending v 7→ v+ ⊗Aop v− (the tensor product is defined using the bimodules V V1⊗Aop and 1⊗Aop VV ). This map is known as the anti- multiplication and its definition for bialgebras over fields extensions goes back to W. D. Nichols [31, Definition 4.]. Here the inverse map β−1 V (− ⊗A 1) will be referred to as the translation map. Analogously, an Ae-ring U with source and target s : A → U and t : Ao → U, is said to be a right A-bialgebroid whenever its category of right modules is a monoidal category and the restriction of scalars functor Or : ModU → ModAe (Or = (s ⊗ t)∗) is a strict monoidal functor. In this case the category ModU is left closed with left inner hom-functors homModU (X, Y) := Hom-U (X ♦ U, Y). (4) For each object X ∈ ModU , we have in this case that the functor X ♦− is left adjoint to Hom-U (X ♦ U, −), and the adjunction is given by the natural isomorphism Hom-U (X ♦Y, Z) (cid:27) Hom-U (Y ⊗U (X ♦ U), Z) (cid:27) Hom-U (Y, Hom-U (X ♦ U, Z)). (5) The corresponding β-maps in the category ModU are defined by βX : X ⊗Aop U → X ♦ U sending x ⊗Aop u 7→ xu1 ⊗A u2, where in the first tensor product we have used in its second factor the bimodule 1⊗Aop UU . One can check as before that Or preserves left inner hom-functors if and only if βU is an isomorphism. So the right A-bialgebroid U is said to be a right Hopf A-algebroid provided that βU is an isomorphism. In this case, the translation map β−1 U (1 ⊗A −) : U → U ⊗Aop U, where the first factor of the tensor product is the A-bimodule UAe , will be denoted by u 7→ u− ⊗Aop u+. As in [37, Proposition 3.7], the map β−1 U (1 ⊗A −) enjoy a list of properties. Here we mention few of them which we will need in the sequel: First note that β−1 U (u′ ⊗A u) = u′u− ⊗Aop u+, so we have u1, − ⊗Aop u1, + ⊗A u2 = u− ⊗Aop u+, 1 ⊗A u+, 2 ∈ (U1⊗Aop ⊗Aop 1⊗Aop U) ♦ U s(b) ⊗Aop s(a) = η(a ⊗ bo) − ⊗Aop η(a ⊗ bo) + ∈ U1⊗Aop ⊗Aop 1⊗Aop U. (6) (7) In case A = K which, by our conventions, automatically implies that s = t and that V (or U) is an ordinary bialgebra, it is well know that V is Hopf algebra if and only if βV is bijective. Remark 1.2.1. For a general left A-bialgebroid V, there is no hope of obtaining from the translation map β−1 V (− ⊗A 1) an endomorphism of V which could play the role of the antipode as in the case of Hopf algebras. Nevertheless, if we assume that A and V are commutative rings with s , t, which means that V is a commutative A-bialgebroid2, then the map S = (ε ⊗A V) ◦ β−1 V (− ⊗A 1) is well defined and gives the antipode for V. Conversely, the inverse of βV is given by β−1 S (v2)v′, whenever V having S as an antipode. Of course in this case the notions of bialgebroid and Hopf algebroid are obviously independent from the sides. V (v ⊗A v′) = v1 ⊗A 2We use the terminology: (A, V) is a commutative Hopf algebroid over K. / / _ _ / / 6 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS Let (A, U) be a right Hopf algebroid. Then the adjunction (5) and the analogue natural isomorphism to T−, give a right U-action on the right Ae-module Hom-Aop (X, Y), for every pair of right U-modules X and Y. This action is given by f . u : X −→ Y, ( x 7−→ f (xu−) u+ ), where, as before, β−1 U (1 ⊗A u) = u− ⊗Aop u+. (8) Now, fix a right U-module X and consider its left dual ∗X also as right U-module with the action (8) by taking YU = AU , the unit object of ModU , with right U-action a . u = ε(s(a)u) = ε(t(a)u). ev ∗X X ♦ x ⊗A ϕ ✤ / A, / ϕ(x) A 1 ✤ db′ / Hom-Aop (X, X) / 1X, (9) The following lemma is the version on the right of [4, Proposition 4.41]. Lemma 1.2.2. Let (A, U) be a right Hopf algebroid. Then every right U-module X which is finitely gener- ated and projective as a left A-module, admits ∗X as the right dual in the monoidal category ModU . Proof. Using the maps of Eq.(9), the right dual of X is given by the triple (∗X, ev, j−1 X ◦ db′), where jX : ∗X ♦ X −→ Hom-Aop (X, X), (ϕ ⊗A x 7−→ (cid:2)x′ 7→ ϕ(x′)x(cid:3)), is the canonical isomorphism. To show that jX is in fact a right U-linear map, one use the properties of the comultiplication of the underlying A-coring of U as well as Eq.(6). (cid:3) 1.3. Examples of (right) Hopf algebroids. Any Hopf algebra over a commutative ring is obviously a (left and right) Hopf algebroid. Below, we list some non trivial examples of Hopf algebroids, specially the ones with commutative base ring, which we will deal with in the forthcoming sections. Example 1.3.1. [32] Assume that the ground ring K is a field and let A be a commutative K-algebra which is a right H-comodule K-algebra with coation A : A → A ⊗K H, where H is a commutative Hopf algebra. Consider A ⊗K H as an (A ⊗K A)-ring via the following source and target: s : A → A ⊗K H, (a 7→ a ⊗K 1), t = A : A → A ⊗K H, (a 7→ a0 ⊗ a1). Then A ⊗K H is a commutative Hopf algebroid over K, with antipode S : A ⊗K H → A ⊗K H sending a ⊗ h 7→ a0 ⊗ a1S (h), where S is the antipode of H. Example 1.3.2. Let A be a commutative algebra over a field K of characteristic 0, and denote by DerK(A) the Lie algebra of all K-linear derivations of A. Consider an A -- module endowed with a structure of K -- Lie algebra L, and let ω : L → DerK(A) be a morphism of K-Lie algebras. The pair (A, L) is called Lie-Rinehart algebra with anchor map ω, provided (aX)(b) = a(X(b)), [X, aY] = a[X, Y] + X(a)Y, for all X, Y ∈ L and a, b ∈ A, where X(a) stands for ω(X)(a). Consider the (left) A-module direct sum A ⊕ L as a K-Lie algebra with the bracket: for any a, b ∈ A and X, Y ∈ L. Denote by τ : A ⊕ L → U(A ⊕ L) the canonical inclusion into its universal enveloping K-algebra. h(a, X), (b, Y)i = (cid:16)X(b) − Y(a), [X, Y](cid:17), As it was expounded in [33, 21, 28], associated to any Lie-Rinehart algebra (A, L) there is a universal object denoted by (A, U(L)). As an algebra, U(L) = U(A ⊕ L)/I, where I is the two-sided ideal generated by the set nτ(a′, 0).τ(a, X) − τ(a′a, a′X) a, a′ ∈ A, X ∈ Lo. There are canonical maps ιA : A → U(L), (cid:16)a 7→ a + I(cid:17) and ιL : L → U(L), (cid:16)X 7→ X + I(cid:17). The first one is an algebra map (whose image is not necessarily in the center), while the second is a K-Lie algebra map. Both maps are compatible in the sense that the following equations are fulfilled: for any a ∈ A and X ∈ L. Such equations determine in fact the universality of U(L). ιA(a)ιL(X) = ιL(aX), [ιL(X), ιA(a)] = ιA(X(a)), / / / / FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 7 As observed in [23, §4.2.1], the usual Hopf algebra structure of U(A ⊕ L) can be lifted to a structure of cocommutative (right) Hopf A-algebroid on U(L). The source and target are equal: t = s = ιA. The A- coring structure is an A-coalgebra structure whose underlying A-bimodule uses the right A-module structure derived from ιA, that is, the A-bimodule U(L)A with two-sided action a.u.a′ = u ιA(aa′), for every a, a′ ∈ A and u ∈ U(L) (recall that A is commutative). The comultiplication and the counit of U(L)A are given on generators by ∆(ιL(X)) = ιL(X) ⊗A 1U(L) + 1U(L) ⊗A ιL(X), ε(ιL(X)) = 0, ∆(s(a)) = s(a) ⊗A 1U(L) = 1U(L) ⊗A s(a), ε(s(a)) = a. for any a ∈ A and X ∈ L. The translation map βU(L) : U(L) → U(L)A ⊗A AU(L) is given on generators by s(a)− ⊗A s(a)+ := 1U(L) ⊗A s(a), ιL(X)− ⊗A ιL(X)+ := 1U(L) ⊗A ιL(X) − ιL(X) ⊗A 1U(L). Example 1.3.3. Here are some basic examples of Lie-Rinehart algebras. (1) The pair (A, DerK(A)) obviously admits the structure of a Lie-Rinehart algebra. (2) [24, Definition 3.3.1] A Lie algebroid is a vector bundle E → M over a smooth manifold, together with a map ω : E → T M of vector bundles and a Lie structure [−, −] on the vector space Γ(E) of global smooth sections of E, such that the induced map Γ(ω) : Γ(E) → Γ(T M) is a Lie algebra homomorphism, and for all X, Y ∈ Γ(E) and any f ∈ C∞(M) one has [X, f Y] = f [X, Y] + Γ(ω)(X)( f )Y. Then the pair (C∞(M), Γ(E)) becomes a Lie-Rinehart algebra. (3) Let (A, ∂i)i=1,··· ,n be a partial differential commutative K-algebra, that is, ∂i are K-algebra deriva- i A.∂i as a Lie tions of A such that ∂i ◦ ∂ j = ∂ j ◦ ∂i, for every i , j. Consider the A-module L := ⊕n K -- algebra with bracket [a.∂i, a′.∂ j] = a∂i(a′).∂ j − a′∂ j(a).∂i, for every i, j = 1, · · · , n. Then L admits canonically a structure of Lie-Rinehart algebra whose anchor map is given by: ω : L −→ DerK (A), (cid:2)a.∂i 7−→ [a′ 7→ a∂i(a′)](cid:3). Example 1.3.4. Under the assumptions of Example 1.3.2, it is clear that the anchor map ω : L → DerK (A) can be extended to an algebra map Ω : U(L) → EndK(A) which gives a left U(L)-action on the base algebra A. If we further assume that A, with this action, is a locally finite left module (i.e. each element generates a left U(L)-module which is a finite dimensional K-vector space), then A becomes a right comodule algebra over the usual finite dual commutative Hopf algebra U(L)′, see Example 3.4.1. In this way, as in Example 1.3.1 we obtain another (commutative) Hopf algebroid, namely, A ⊗K U(L)′. Example 1.3.5 (Base extension). Let (A, H ) be a commutative Hopf algebroid and φ : A → B be a mor- phism of commutative algebras. Then (B, B ⊗A H ⊗A B) admits a structure of commutative Hopf algebroid called the base ring extension of (A, H ) by φ. The structure maps of this Hopf algebroid are given as follows: s(b) = b ⊗A 1 ⊗A 1, t(b′) = 1 ⊗A 1 ⊗A b′, S (b ⊗A h ⊗A b′) = b′ ⊗A S (h) ⊗A b, ∆(b ⊗A h ⊗A b′) = X(h) (b ⊗A h1 ⊗A 1) ⊗B (1 ⊗A h2 ⊗A b′), ε(b ⊗A h ⊗A b′) = bb′φ(ε(h)). 2. The reconstruction process, Galois corings and principal bi-bundles. We recall in this section the construction of the universal coring from a given fibre functor [5]. We follow the presentation given in [14]. We also recall the notion of Galois coring. Roughly speaking, these are corings which can be reconstructed from their class of right comodules which are finitely generated and projective as right modules over the base algebra. 2.1. Infinite comatrix corings. Let ω : A → add(AA) be an additive faithful functor (referred to as a fibre functor), where A is an additive small category and A is an algebra, which is not assumed to be commutative. Here, add(AA) denotes the category of all finitely generated and projective right A -- modules. The image of an object P of A under ω will be denoted by ωP := ω(P), or even by P when no confusion may be expected. Given P, Q ∈ A, we denote by TPQ = HomA(P, Q) the K-module of all morphisms in A from P to Q. The symbol TP is reserved to the ring of endomorphisms of P. Clearly, S P = End(ωPA) 8 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS is a ring extension of TP via ω. In this way, every image ωP of an object P ∈ A, becomes canonically a (TP, A)-bimodule, and this bi-action can be extended to the following (TQ, A)-bimodule map TPQ ⊗TP P t ⊗TP p ✤ ωPQ / Q / tp = ω(t)(p), (10) since TPQ is already a (TQ, TP)-bimodule. The dual bi-action is given by the following (A, TP)-bimodule map Q∗ ⊗TQ TPQ q∗ ⊗TQ t ✤ ωQ∗ P∗ / q∗ / q∗t = q∗ ◦ ω(t). (11) For every object P ∈ A, one can define its associated finite comatrix A-coring P∗⊗TP P with the following comultiplication and counit P∗ ⊗TP P p∗ ⊗A p ✤ ∆ P∗ ⊗TP P / P∗ ⊗TP P ⊗A P∗ ⊗TP P, / PαP p∗ ⊗TP eαP ⊗A e∗ αP ⊗TP p P∗ ⊗TP P p∗ ⊗TP p ✤ εP∗ ⊗TP P / A / p∗(p) (12) where {(eαP , e∗ αP of the dual basis [13, Remark 2.2]. Now consider the following direct sum of A-corings )} ⊂ P × P∗ is any right dual basis for PA. The map ∆P∗⊗TP P does not depend on the choice B (A) = MP∈A P∗ ⊗TP P and its K-submodule Jω generated by the set (cid:26)q∗ ⊗TQ tp − q∗t ⊗TP p : q∗ ∈ Q∗, p ∈ P, t ∈ TPQ, P, Q ∈ A(cid:27) , (13) where the products are defined by the pairings of Eqs. (10) and (11). By [14, Lemma 4.2], Jω is a coideal of the A-coring B(A). Therefore, we can consider the quotient A-coring R (A) : = B(A)/Jω /Jω (14) =  MP ∈ A P∗ ⊗TP P and this is the infinite comatrix A-coring associated to the fibre functor ω : A → add(AA). Furthermore, it is clear that any object P ∈ A admits (via the functor ω) the structure of a right R(A)-comodule, which leads to a well defined functor χ : A → ComodR(A). Notation: We will denote by ϕ ⊗TP p := ϕ ⊗TP p + Jω, for ϕ ∈ ω(P)∗, p ∈ ω(P) and P ∈ A, a generator element in the infinite comatrix A-coring R(A). Under this notation the comultiplication of R(A) is given by ∆ : R(A) −→ R(A) ⊗A R(A), p∗ ⊗TP p 7−→ XαP p∗ ⊗TP eα,P ⊗A e∗ α,P ⊗TP p, where {eα,P, e∗ α,P} denotes a finite dual basis for PA = ω(P)A. The counit of R(A) is ε : R(A) −→ A, p∗ ⊗TP p 7−→ p∗(p). (15) (16) The above construction is in fact functorial, in the sense that if F : A → A′ is a K-linear functor and are fibre functors such that ω′ ◦ F = ω, then there is a morphism of A′ : ω′ ω : A A-corings R(F ) : R(A) → R(A′). / add(AA) There are two typical situations where the above process can be applied. Namely, starting with a ring extension η : A → R and consider the category AR of all right R-modules which by restriction of scalars are finitely generated and projective as right A-modules, the fibre functor is then the restriction of the forgetful functor η∗ : ModR → ModA, that is, η∗ : AR → add(AA). The associated A-coring R(AR) is simply denoted by R(R) or by R◦, and referred to as the (right) finite dual of the A-ring R. This in fact establishes a contravariant functor from the category of A-rings to the category of A-corings. The basic properties of the finite dual of a ring extension are presented in Section 3. The other situation which is somehow dual to the previous one, is that of an A-coring C, where the small category A is taken to be the category AC of all right C-comodules whose underlying right A-modules are / / / / / / / / / o o FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 9 finitely generated and projective. The fibre functor here is given by the restriction of the forgetful functor O : ComodC → ModA, that is, O : AC → add(AA). This situation was studied in [5, 14]. 2.2. Review on Hopf algebroids constructed from fibred functors. Starting with a K-linear monoidal (essentially small) category A (the tensor product ⊗ of A is implicitly assumed to be a K-linear functor in both factors) with a monoidal faithful K-linear functor ω : A −→ AModA with image in add(AA). Then, one can endow, using multiplication induced form the monoidal structure (see the formula (20) below), the associated infinite comatrix A-coring R(A) of equation (14) with a structure of (A ⊗TI Ao)-ring, where TI is the (commutative) endomorphism K-algebra of the identity object I, and TI → A is the injective K-algebra homomorphism induced by ω. It turns out that (A, R(A)) with this algebra structure is a (left) bialgebroid. If we further assume that A is a symmetric rigid monoidal category and that A is a commutative K-algebra, then (A, R(A)) has a structure of commutative Hopf algebroid over TI (compare with [5, Section 7]). For sake of completeness and for our needs, as well as for a non expertise reader convenience, we review in detail the structure maps of this construction. This detailed construction is to be used later. Consider the previous situation: ω : A −→ add(AA), where A is a not necessarily commutative algebra, and assume that the resulting functor, after composing ω with the embedding add(AA) ֒→ ModA, factors 3. To avoid some technical problems, throughout a (strict) monoidal faithful K-linear functor A → AModA we further assume that A is a Penrose category, in the sense that each of the K-modules TPQ is a central TI-bimodule over the commutative K-algebra TI (i.e., the left TI-action coincides with right one). This the case when A is for instance a braided or symmetric monoidal category. Let us first define an unitary and associative multiplication on R(A). As we have seen before, the infinite comatrix coring R(A) is the quotient A-coring B(A)/Jω, where and Jω is the K-submodule spanned by the set of elements as in Eq.(13). Given P, Q ∈ A, and ϕ ∈ P∗, ψ ∈ Q∗, we define (ϕ ⋆ ψ) ∈ (P ⊗A Q)∗ (remember that here we are denoting ω(P) := P), by B(A) = MP∈ A P∗ ⊗TP P (17) (ϕ ⋆ ψ) : P ⊗A Q −→ A, (cid:18)x ⊗A y 7−→ ψ(ϕ(x)y)(cid:19) . We use (18) to define componentwise an associative multiplication on B(A) by P∗ ⊗TP P ⊗A Q∗ ⊗TQ Q / (Q ⊗A P)∗ ⊗TQ⊗ P Q ⊗A P p∗ ⊗TP p ⊗A q∗ ⊗TQ q ✤ / q∗ ⋆ p∗ ⊗TQ⊗ P q ⊗A p (18) (19) To see that the map (19) is well defined, we used the fact that A is monoidal and that A → AModA is a (strict) monoidal functor. Since Jω is easily checked to be an ideal of B(A), we get that the following multiplication is well defined : (p∗ ⊗TP p) . (q∗ ⊗TQ q) = (q∗ ⋆ p∗) ⊗TQ⊗ P (q ⊗A p), (20) for every p∗ ⊗TP p and q∗ ⊗TQ q in R(A). A straightforward computation checks that this multiplication is associative. Let us show that there is a unit in R(A) for this multiplication. To this end, observe that there is an injective K-algebra map TI → A induced by ω. Therefore, one can consider the following well defined map η : Ae := A ⊗TI Ao −→ R(A), (21) where we denote by la′ : A → A the left multiplication map sending r 7→ a′r for any a′ ∈ A. The element idA ⊗TI 1A ∈ R(A) is clearly the unit for the multiplication (20). As before the source map s : A → R(A) sends a 7→ η(a ⊗TI 1o) and the target t sends a 7→ η(1 ⊗TI ao). h(a′ ⊗ ao) 7−→ (la′ ⊗TI a)i, The following result could be deduced from [5, Exemple, pp. 5849] and also from [20, Theorem 2.2.4]. However, a direct application of the former or the latter will not lead to an explicit description of the structure maps of the constructed Hopf algebroid. For our needs and for the reader convenience, we give here an elementary proof of these statements. 3This factorization condition is not needed when A is a commutative K -- algebra. / / 10 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS Proposition 2.2.1. Let A, A and ω be as above. Then the pair (A, R(A)) admits a structure of left bialgebroid. Assume furthermore that A is a commutative K-algebra, A is symmetric rigid monoidal K- linear and that ω is a symmetric monoidal faithful K-linear functor. Then (A, R(A)) is a commutative Hopf algebroid over TI. Proof. We need to endow the category of left R(A)-modules with a monoidal structure such that the restriction of scalars functor ∗η : R(A)Mod → Ae Mod becomes strict monoidal. Take two left R(A)- modules X, Y. The tensor product X ⊗A Y is then a left module over R(A) with the action given by λ : R(A) ⊗ X ⊗A Y (p∗ ⊗TP p) ⊗ (x ⊗A y) ✤ / X ⊗A Y (22) / PαP (p∗ ⊗TP eα,P)x ⊗A (e∗ α,P ⊗TP p)y, where we have used the comultiplication displayed in Eq. (15) and the fact that this map lands in an appropriate Sweedler-Takeuchi's product. The restriction of this action to scalars from Ae via η gives the canonical A -- bimodule structure of X ⊗A Y, that is, the action (a′ ⊗TI ao) . x ⊗A y = (a′x) ⊗A (ya), for every a, a′ ∈ A and x ∈ X, y ∈ Y. This in particular shows that the action is unital. The associativity property of the action given by Eq. (22) is derived as follows. Take two elements of the form p∗ ⊗TP p and q∗ ⊗TQ q in R(A), then λ(cid:18)(p∗ ⊗TP p) (q∗ ⊗TQ q) ⊗ (x ⊗A y)(cid:19) = λ(cid:18)(q∗ ⋆ p∗) ⊗TQ⊗ P (q ⊗A p) ⊗ (x ⊗A y)(cid:19) (q∗ ⋆ p∗) ⊗TQ⊗ P (eβ,Q ⊗A eα,P)x ⊗A (e∗ = XαP, βQ = XαP, βQ(cid:18)(p∗ ⊗TP eα,P) (q∗ ⊗TQ eβ,Q)x(cid:19) ⊗A (cid:18)(e∗ = λ(cid:18)(p∗ ⊗TP p) ⊗ λ(cid:18)(p∗ ⊗TP p) ⊗Ae (x ⊗A y)(cid:19)(cid:19) . β,Q ⋆ e∗ α,P) ⊗TQ⊗ P (q ⊗A p)y α,P ⊗TP p) (e∗ β,Q ⊗TQ q)y(cid:19) Given Z another left R(A) -- module, the natural isomorphism (X ⊗A Y) ⊗A Z (cid:27) X ⊗A (Y ⊗A Z) is clearly an isomorphism of left R(A)-modules. On the other hand, for any two arrows f , g in R(A)Mod, it is easily seen, using the above action, that f ⊗A g is a morphism in R(A)Mod. Now, consider the monoidal unit Ae A as a left R(A)-module via the action p∗ ⊗TP p a = p∗(ap) = p∗a(p) = ε(cid:18)p∗ ⊗TP p s(a)(cid:19) = ε(cid:18)p∗ ⊗TP p t(a)(cid:19) , where ε is the counit of R(A). For a left R(A)-module X, let ι : A ⊗A X (cid:27) X sending a ⊗A x 7→ s(a)x be the canonical isomorphism. Using the action of Eq.(22), we have ι(cid:18)p∗ ⊗TP p (a ⊗A x)(cid:19) = XαP = XαP ι(cid:18)p∗(aeα,P) ⊗A (e∗ α,P ⊗TP p)x(cid:19) p∗(aeα,P)eα,P ⊗TP p x = p∗a ⊗TP p x = p∗ ⊗TP p s◦(a)x = p∗ ⊗TP p ι(a ⊗A x), which shows that ι is left R(A)-linear. Similarly, one can show that the isomorphism X ⊗A A (cid:27) X is also left R(A) -- linear. Thus, R(A)A is the monoidal unit for the tensor product in R(A)Mod. Summarizing, we obtain that the category of left R(A)-modules is a monoidal category with unit the left module R(A)A, and the restriction of scalars functor ∗η : R(A)Mod → Ae Mod is a strict monoidal. Therefore, (A, R(A)) is left bialgebroid. Regarding the commutative case, we need first to introduce some notations: let us denote by τP,Q : P ⊗ Q → Q ⊗ P the symmetry of A and by (−)∨ : A → A the K-linear contravariant functor which sends any object P ∈ A to its dual P∨ and any morphism f : P → Q to it dual f ∨ : Q∨ → P∨. Under / / FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 11 assumption, there is a natural isomorphism γ− : (−)∗ ◦ ω → ω ◦ (−)∨, which leads to a natural isomorphism φP : ω(P) → ω(P∨)∗. Using the action of (10) and (11), the naturality of bothe γ and φ, reads φQ(ω( f ) p) = φP(p) ω( f ∨) γP(ϕ ω( f )) = ω( f ∨) γQ(ϕ), (23) (24) for every morphism f commutative the following diagrams : P → Q in A , p ∈ ω(P) and ϕ ∈ ω(Q)∗. On the other hand γ and φ render (ω(P∨)∗)∗ / ω(P)∗ φ∗ γ / ω(P∨) ❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ψ ω(P∨), (ω(P)∗)∗ ψ ω(P), γ∗ ω(P∨)∗ φ ❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ω(P) (25) where the vertical map is the canonical isomorphism of A-modules, sending q 7→ (cid:2)q∗ 7→ q∗(q)(cid:3). The natural isomorphism γ rends also the following diagrams ω(P)∗ ⊗A ω(P) evω(P) A A dbω(P) / ω(P) ⊗A ω(P)∗ ω(P∨) ⊗A ω(P) γ⊗Aω(P) u❧❧❧❧❧❧❧❧❧❧❧❧ )❘❘❘❘❘❘❘❘❘❘❘ (cid:27) (cid:27) (cid:27) ω(evP) / ω(I) ω(I) ω(dbP) ω(P)⊗Aγ )❘❘❘❘❘❘❘❘❘❘❘❘ 5❧❧❧❧❧❧❧❧❧❧❧ (cid:27) ω(P∨) ⊗A ω(P) ω(cid:0)P∨ ⊗ P(cid:1) / ω(cid:0)P∨ ⊗ P(cid:1) (26) commutative, where ev− and db− symbolize the evaluation and the dual basis morphisms attached to a dualizable object in a rigid monoidal category, respectively. Furthermore, φ is compatible with the tensor product, that is, for any pair of objects P, Q ∈ A, we have a commutative diagram: ω(P ⊗ Q) (cid:27) ω(P) ⊗A ω(Q) φP⊗AφQ / ω(P∨)∗ ⊗A ω(Q∨)∗ (cid:27) / (ω(Q∨) ⊗A ω(P∨))∗ φP⊗ Q (cid:27) (ω(Q∨ ⊗ P∨))∗ (27) ω(ιP, Q)∗ where ιQ, P : (Q ⊗ P)∨ (cid:27) P∨ ⊗ Q∨ is the canonical natural isomorphism in the category A. The natural transformation γ is also compatible with the tensor product, that is, we have another commutative diagram: ω(cid:0)(P ⊗ Q)∨(cid:1)∗, ω(P ⊗ Q) (cid:27) γP⊗ Q ω(cid:0)(P ⊗ Q)∨(cid:1) ω(ι−1 P, Q) (28) ω(P)∗ ⊗A ω(Q)∗ γP⊗AγQ / ω(P∨) ⊗A ω(Q∨) (cid:27) / ω(Q∨ ⊗ P∨) On elements, the commutativity of both diagrams of equations (72) and (28), are expressed by the following first two equalities: ιQ, P γQ⊗P(cid:0)q∗ ⋆ p∗(cid:1) = γP(p∗) ⊗A γQ(q∗), φQ⊗P(cid:0)q ⊗A p(cid:1) = (cid:16)φP(p) ⋆ φQ(q)(cid:17) ιQ, P, (q∗ ⋆ p∗) τP, Q = p∗ ⋆ q∗, (29) (30) (31) for every P, Q ∈ A, p ∈ ω(P), q ∈ ω(Q) and p∗ ∈ ω(P)∗, q∗ ∈ ω(Q)∗, where we have used the actions of equations (10) and (11). / / O O o o o o O O / / u ) / O O / ) O O / 5 / / . . / / / / / / O O 12 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS Take two generic elements (p∗ ⊗TP p), (q∗ ⊗TQ q) ∈ R(A) and using equality (31), we compute (p∗ ⊗TP p) (q∗ ⊗TQ q) = (q∗ ⋆ p∗) ⊗TQ⊗ P (q ⊗A p) = (q∗ ⋆ p∗) ⊗TQ⊗ P τP, Q(p ⊗A q), τP, Q is an arrow in A = (q∗ ⋆ p∗)τP, Q ⊗TP⊗ Q (p ⊗A q) = (p∗ ⋆ q∗) ⊗TP⊗ Q (p ⊗A q) = (q∗ ⊗TQ q) (p∗ ⊗TP p), which means that the multiplication (20) is commutative, and so R(A) is a commutative A ⊗TI A-algebra. The compatibility of the comultiplication of Eq.(15) and the counit of Eq.(16) with the multiplication of Eq.(20), are routine computation using the dual basis properties. The following map S : R(A) −→ R(A), (cid:16)p∗ ⊗TP p 7−→ φ(p) ⊗T P∨ γ(p∗)(cid:17), (32) which is well defined thanks to the naturality of γ− and φ−, is our candidate for the antipode map of the commutative bialgebroid (A, R(A)). Let us check that S transforms the source to the target and vice-versa, and that it is an algebra map as well. So, for two elements a, b ∈ A, we have S (η(a ⊗TI b)) = S (la ⊗TI b) = φ(b) ⊗TI∨ γ(la) (25) = lb ⊗TI a = η(b ⊗TI a). Hence S ◦ s = t and S ◦ t = s. The fact that S is a multiplicative map follows from the following computation: For any two generic elements (p∗ ⊗TP p), (q∗ ⊗TQ q) ∈ R(A), we have S(cid:16)(p∗ ⊗TP p) (q∗ ⊗TQ q)(cid:17) = S(cid:16)(q∗ ⋆ p∗) ⊗TQ⊗ P (q ⊗A p)(cid:17) (Q⊗ P)∨ γ(q∗ ⋆ p∗) = φ(q ⊗A p) ⊗T (29) = φ(q ⊗A p) ⊗T (Q⊗ P)∨ ι−1 Q, P γ(p∗) ⊗A γ(q∗) = φ(q ⊗A p) ι−1 Q, P ⊗T P∨ ⊗ Q∨ γ(p∗) ⊗A γ(q∗) (30) = φ(p) ⋆ φ(q) ⊗T P∨ ⊗ Q∨ γ(p∗) ⊗A γ(q∗) (20) = (φ(q) ⊗T Q∨ γ(q∗)) (φ(p) ⊗T = S (q∗ ⊗TQ q) S (p∗ ⊗TP p), P∨ γ(p∗)) which shows that S is a morphism of algebras. We still need to check that, for every element h ∈ R(A), we have h1S (h2) = η(ε(h) ⊗ 1), S (h1)h2 = η(1 ⊗ ε(h)). Take h ∈ R(A) of the form h = p∗ ⊗TP p, where p∗ ∈ ω(P)∗, p ∈ ω(P), for some object P ∈ A, then we have (summation understood) h1S (h2) = (p∗ ⊗TP eα,P)S (e∗ α,P ⊗TP p) P∨ γ(e∗ = (p∗ ⊗TP eα,P) (φ(p) ⊗T = (p∗ ⋆ φ(p)) ⊗T α,P)) P⊗ P∨ (eα,P ⊗A γ(e∗ α,P)) (26) = (p∗ ⋆ φ(p)) ⊗T P⊗ P∨ ω(dbP)(1A) = (p∗ ⋆ φ(p))ω(dbP) ⊗TI 1A (26) (25) ω(P) = (cid:0)ω(dbP)∗ ◦ (p∗ ⊗A φ(p))(cid:1) ⊗TI 1A = (cid:0)db∗ = (cid:0)db∗ = (cid:0)db∗ ◦ (ω(P)∗ ⊗A γ∗) ◦ (p∗ ⊗A φ(p))(cid:1) ⊗TI 1A ◦ (p∗ ⊗A γ∗φ(p))(cid:1) ⊗TI 1A ◦ (p∗ ⊗A ψ(p))(cid:1) ⊗TI 1A ω(P) ω(P) FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 13 = (p∗ ⊗A ψ(p))(cid:0)dbω(P)(1A)(cid:1) ⊗TI 1A = lp∗(p) ⊗TI 1A = η(ε(h) ⊗ 1A), which shows the first desired equality. The second one is similarly obtained (or can be obtained from the equality S 2 = id which is shown below). Lastly, since the natural isomorphisms dP : P → (P∨)∨ in A, satisfy the following equalities it is not difficult to show that S 2 = id, and this finishes the proof. γP∨ ◦ φP = ω(dP) ω(dP)∗ ◦ φP∨ ◦ γP = idω(P)∗ , (33) (cid:3) Remark 2.2.2. Given two symmetric rigid monoidal K-linear categories with symmetric monoidal faithful K-linear functors: ω : A → add(A) and ω′ : A′ → add(A), and F : A → A′ a monoidal K-linear functor such that A (◗◗◗◗◗◗◗◗◗◗ ω F add(A) A′ v♠♠♠♠♠♠♠♠♠ ω′ is a commutative diagram, then there exists a morphism φ =: R(F ) : (A, R(A)) → (A, R(A′)) of Hopf algebroids making commutative the following diagram of functors: AR(A) ω 9ttttttt A φ∗ A′ F O (A′)R(A′) ω′ 8qqqqqqqq ω ω O add(A) where O is the forgetful functor and φ∗ is the restriction of the induced functor φ∗ : ComodR(A) → ComodR(A′) sending any right R(A)-comodule (M, M) to the right R(A′)-comodule and acting obviously on morphisms. M M / M ⊗A R(A) M⊗Aφ / M ⊗A R(A′), 2.3. Reconstruction and Galois corings. It is well known that any coalgebra over a field can be recon- structed from its category of finite-dimensional right comodules. That is, it is isomorphic to an infinite comatrix coalgebra. Over a general base ring, one only obtains an A-coring morphism which is not always an isomorphism. Precisely, let R(AC) be the A-coring associated to the fibre functor O : AC → add(AA) for some A-coring C. Then there is a homomorphism of A-corings known as the canonical map: canAC : R(AC) −→ C, (cid:18)p∗ ⊗TP p + JA 7−→ (p∗ ⊗A C) ◦ P(p)(cid:19) . (34) Obviously any K-linear functor A → AC induces an analogue morphisms of corings canA. Following the terminology used in [14, 19]: Definition 2.3.1. An A-coring C is said to be an AC-Galois coring (or Σ = ⊕p∈AC P is a right Galois comodule) provided canAC is an isomorphism of A-corings. This is the case when ComodC is an abelian category having AC as a set of generators [14, Theorem 4.8], which is exactly the situation in the aforementioned coalgebra case (see [10, Proposition 4.13] for the case of corings over fields). The case of corings underlying certain commutative Hopf algebroids is of special interest as the following example shows. Example 2.3.2 (Geometrically transitive Hopf algebroids [5, 12]). Roughly speaking, a commutative flat Hopf algebroid (A, H ) over a base field K, is said to be geometrically transitive (GT for short), provided that the algebra map η = s ⊗K t : A ⊗K A → H is a faithfully flat extension. The fact that K should be a field is a crucial condition here. It turns out that GT Hopf algebroid (A, H ) is reconstructed from its A-profinite / / ( v / / $ $ s s / / 9 , , 8 v v / / 14 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS comodules, that is, from the category AH of the underlying A-coring H , as (the skeleton of) this category form a 'set' of generators, see [12, Corollary A (2)]. Indeed, every H -comodule is an inductive limit of comodules from AH . Besides, any comodule is faithfully flat as an A-module [12, Lemma 4.3 (b)]. In summary, by applying [14, Theorem 4.8], the canonical map of Eq. (34) is bijective and H is a Galois A-coring. 2.4. Principal bi-bundles attached to two different fibre functors. Apart from its own general interest, the material of this section will be used to clarify the construction of Picard-Vessiot extensions (over the affine line) in Section 5. In this subsection all algebras are commutative K-algebras over a base field K and are assumed to have K-points. The category of all commutative algebras over a commutative algebra A, is denoted by AlgA. We keep the notation of subsection 2.2 and fix a symmetric rigid monidal K-linear category A. For our needs it is convenient to assume further that the endomorphism algebra of the identity object of A is isomorphic to the base field, that is, TI (cid:27) K. For a given (non trivial) monoidal symmetric K-linear faithful functor ω : A → add(A) (i.e., a fibre functor), instead of denoting by R(A) the resulting Hopf algebroid A⊗ A(ω) := R(A) in order to specify which fibre functor from Proposition 2.2.1, we will use the notation R we are using and over which base algebra we are working. Given an algebra map ξ : A → C, we denote by ω ⊗A ξC : A → add(C) the extended fibre functor, which sends any object P ∈ A to the finitely generated and projective C-module ω(P) ⊗A ξC. Assume we are given two fibre functors ωi : A → add(Ai), i = 1, 2. If R is an A1 ⊗ A2-algebra, that is, we have an extension s ⊗ t : A1 ⊗ A2 → R, then we have two extended fibre functors, namely, ω1 ⊗A1 sR and ω2 ⊗A2 tR. In this way, we can define the following (A1, A2)-bimodule: R A1 ⊗ A2 (ω1, ω2) := LP ∈ A ω1(P)∗ ⊗TP ω2(P) Jω1 , ω2 where Jω1 , ω2 is the subbimodule generated by the set (cid:26)q∗ ⊗TQ ω2(t)(p) − (q∗ ◦ ω1(t)) ⊗TP p q∗ ∈ ω1(Q)∗, p ∈ ω2(P), t ∈ TPQ, P, Q ∈ A(cid:27) . Then there are two bimodule maps (35) (36) (37) α : A1 −→ R A1 ⊗ A2 (ω1, ω2),(cid:16)a1 7−→ la1 ⊗K 1A2(cid:17) and β : A2 −→ R A1 ⊗ A2 (ω1, ω2),(cid:16)a2 7−→ 1A1 ⊗K la2(cid:17). Our next aim is to show that the triple (R A2⊗ A2 (ω2))- bicomodule algebra which becomes a principal bibundle (see [16, Section 4] for the pertinent definition) when the involved Hopf algebroids are geometrically transitive (see Example 2.3.2 and [12, Theorem A] for more details). A1 ⊗ A2 (ω1, ω2), α, β) admits a structure of (R A1 ⊗ A1 (ω1), R The commutative algebra structure of R for a given two object P, Q ∈ A and elements p∗ 1 have that A1 ⊗ A2 (ω1, ω2) is defined in similar way as in Eq. (20). That is, ∈ ω1(Q)∗ and p2 ∈ ω2(P), q2 ∈ ω2(Q), we ∈ ω1(P)∗, q∗ 1 (p∗ 1 ⊗TP p2) . (q∗ 1 ⊗TQ q2) := (p∗ 1 ⋆ q∗ 1) ⊗TP⊗ Q (p2 ⊗A q2), (38) is a well defined multiplication, such that the map α ⊗ β : A1 ⊗K A2 → R Clearly, we have R assume that A1 = A2 = A, we denote by hα − βi the ideal of R A1 ⊗ A2 (ω1, ω2) is a K-algebra map. A⊗ A(ω), whenever A1 = A2 = A and ω1 = ω2 = ω. In case we only A⊗ A(ω1, ω2) generated by the set of elements A1⊗ A2 (ω1, ω2) = R (cid:8)α(a) − β(a) a ∈ A(cid:9), and by R A(ω1, ω2) := R A⊗ A(ω1, ω2) hα − βi (39) its quotient A-algebra with extension denoted by ι : A → R element p∗ ⊗TP p in this quotient algebra R A(ω, ω) := R R Specifically, R Hopf algebroid with source is equal to the target. A(ω1, ω2). The equivalence class of a generic A(ω1, ω2) is denoted by [p∗ ⊗TP p]. When ω1 = ω2, we denote A(ω) (here α and β are the source and the target maps, respectively, of this Hopf algebroid). A(ω) inherits canonically a structure of commutative Hopf A-algebra, that is, a commutative In the general situation we will denote by Isom⊗ A1 ⊗ A2(cid:0)ω2, ω1(cid:1) the functor Isom⊗ A1 ⊗ A2(cid:0)ω2, ω1(cid:1) : AlgA1 ⊗ A2 −→ Sets, (cid:16)R −→ Isom⊗(ω2 ⊗A2 tR, ω1 ⊗A1 sR)(cid:17) (40) FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 15 from the category of (A1 ⊗ A2)-algebras to the category of sets, which sends any (A1 ⊗ A2)-algebra R to the set of all tensorial natural transformations from the functor ω1 ⊗A1 sR to ω2 ⊗A2 tR (recall that the category A is assumed to be rigid, so that each of these natural transformations is already a natural isomorphism). We will employ the same notations for the case A1 = A2 = A, that is, we will denote by Isom⊗ AlgA → Sets the functor which sends any A-algebra C to the set Isom⊗(ω2 ⊗A C, ω1 ⊗A C). A(cid:0)ω2, ω1(cid:1) : A1 ⊗ A2(cid:0)ω2, ω1(cid:1) of equation (40) is Proposition 2.4.1. Keep the above notations. Then the functor Isom⊗ represented by the (A1 ⊗ A2)-algebra R isomorphisms, there is a commutative diagram A1 ⊗ A2 (ω1, ω2). Assume that A1 = A2 = A, then up to natural AlgA Isom⊗ %▲▲▲▲▲▲▲▲▲▲ A(ω2, ω1) Sets / AlgA⊗ A xqqqqqqqqqq Isom⊗ A⊗ A (ω2, ω1) where the horizontal functor is the canonical one. Furthermore the functor Isom⊗ by the quotient A-algebra R A(ω1, ω2) of equation (39). A(cid:0)ω2, ω1(cid:1) is represented Proof. The first statement of the proposition is [10, Proposition 6.6]. The rest of the proof is not difficult and left to the reader. (cid:3) Recall that a pair (P, α) is said to be a left comodule algebra over a commutative Hopf algebroid (A, H ), when α : A → P is an extension of algebras and P is endowed with a structure of left H -comodule whose underlying left A-module is αP such that the coaction λ : P → Hs ⊗A αP is an A-algebra map. Right comodule algebras are similarly defined, and bicomodule algebras are naturally introduced. Thus an (H , K )-bicomodule algebra over two commutative Hopf algebroids (A, H ) and (B, K ), is a triple (P, α, β) such that (P, α) is a left H -comodule algebra, (P, β) is a right K -comodule algebra and the left H -coaction is a morphism of right K -comodules (or equivalently the right K -coaction is left H -comodule morphism). Definition 2.4.2. ([16, Defintion 4.1]) A left principal (H , K )-bundle (P, α, β) for two Hopf algebroids (A, H ) and (B, K ) is an (H , K )-bicomodule algebra, that is, P is equipped with a left H -comodule algebra and a right K -comodule algebra structures with respect to the algebra maps α : A → P resp. β : B → P such that (1) β is a faithfully flat extension; (2) the canonical map canl : P ⊗B P → H ⊗A P, p ⊗B p′ 7→ p(−1) ⊗A p(0) p′ (41) is bijective. A triple (P, α, β) which only satisfies condition (2) is called left pseudo (H , K )-bundle. Right (pseudo) principal bundles are similarly defined and we use the notation canr for the corresponding canonical map. For instance, the opposite right principal bundle of a given left principal bundle (P, α, β), is the bundle (Pop, β, α) where the algebra Pop = P and the coactions are switched (interchanging the source and the target maps). A (pseudo) principal bibundle is a left and right (pseudo) principal bundle. That is, for pseudo bibundle, both canl and canr are required to be bijective. If, furthermore, both α and β are faithfully flat, then the bibundle is principal. For a given flat Hopf algebroid (A, H ) it is easily shown that the triple (H , s, t) is a principal (H , H )-bibundle, see [16, §4] for more examples and details. Now we come back to our situation. Consider as above the algebras R an object P ∈ A we fix a dual basis for ωi(P) ∈ add(Ai) using the notation {eαi, P , e∗ also make use of the notation pi ∈ ωi(P) and p∗ following well defined maps: Ai⊗ A j (ωi, ω j), for i = 1, 2. For }αi, P , i = 1, 2. We i ∈ ωi(P)∗, for any object P ∈ A. In this way we have the αi, P R A1 ⊗ A2 (ω1, ω2) λ / R A1⊗ A2 (ω1, ω2) ⊗A2 R A2 ⊗ A2 (ω2) (42) ⊗TP p2 ✤ (cid:16)p∗ 1 / Pα2, P p∗ 1 ⊗T p eα2, P ⊗A2 e∗ α2, P ⊗TP p2(cid:17), % / x / / 16 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS where the left A2-action on R A2 ⊗ A2 (ω2) is given by the algebra map α of equation (37), and R A1 ⊗ A2 (ω1, ω2) / R A1⊗ A1 (ω1) ⊗A1 R A1 ⊗ A2 (ω1, ω2) (43) ⊗TP p2 ✤ (cid:16)p∗ 1 / Pα1, P p∗ 1 ⊗T p eα1, P ⊗A2 e∗ α1, P ⊗TP p2(cid:17), where the right A1-action on R A1 ⊗ A1 (ω1) is given by the algebra map β of equation (37). The following Lemma and its subsequent Corollary are inspired from Proposition 2.4.1. Lemma 2.4.3. Consider (Ai, R Ai⊗ Ai (ωi)), for i = 1, 2, as commutative Hopf algebroids with structure maps given as in subsection 2.2. Then the triple (R A2⊗ A2 (ω2))- bicomodule algebra with left and right coactions given by equations (42) and (43), respectively. Fur- thermore, R A1⊗ A2 (ω1, ω2), α, β) is an (R A1⊗ A1 (ω1), R A1 ⊗ A2 (ω1, ω2) is a pseudo (R A2⊗ A2 (ω2))-bibundle. A1 ⊗ A1 (ω1), R Proof. The first statement is a straightforward verification. As for the last one, recall the natural transfor- mations γ and φ from equations (23) and (24). We only give the translation map which leads to the inverse of the canonical map can, the pertinent verification are routine computations which use the properties of γ and φ already mentioned in subsection 2.2. For the left one, that is, for canl the translation map is given by: τl R A(ω1) p∗ 1 ⊗TP p1 ✤ R / α A1⊗ A2 (ω1, ω2)β ⊗A2 R A1 ⊗ A2 (ω1, ω2)β ⊗T p eα2, P ⊗A2 φ(p1) ⊗T P∨ γ(e∗ α2, P ) / Pα2, P p∗ 1 While for the right canonical map canr the translation map is given by: R A(ω2) p∗ 2 ⊗TP p2 ✤ This finishes the proof. τr R / α A1 ⊗ A2 (ω1, ω2) ⊗A1 α P∨ φ(p∗ / Pα1, P γ(eα1, P ) ⊗T R A1 ⊗ A2 (ω1, ω2)β 2) ⊗A1 e∗ α1, P ⊗T p p2 (cid:3) Over the same algebra, the coaction of equations (42) and (43) are canonically lifted to the quotient algebras of Eq. (39) and the Lemma 2.4.3 still working in this case as well. Corollary 2.4.4. Assume that A1 = A2 = A. Then the bicomodule algebra structure of Lemma 2.4.3, induces a structure of (R A(ω1, ω2) is a pseudo (R A(ω2))-bicomdule algebra on R A(ω1, ω2). Furthermore, R A(ω1), R A(ω2))-bibundle. A(ω1), R The following is the main result of this section, it is the affine groupoid schemes version of the similar well known result for affine group schemes (over an affine scheme). As before, the ground ring K is assumed to be a field. For more notions and properties of geometrically transitive Hopf algebroids we refer to [5, 12]. Theorem 2.4.5. Let A be a symmetric rigid monoidal K-linear category with the endomorphism algebra of the identity object TI = K. Assume we have two fibre functors ωi : A → add(Ai), i = 1, 2 (i.e., faithful symmetric monoidal K-linear functors) and consider the associated (R A2⊗ A2 (ω2))-bicomodule algebra (R A1⊗ A1 (ω1), R A1⊗ A2 (ω1, ω2), α, β). Then, (i) if the Hopf algebroid R A1⊗ A1 (ω1) (resp., R A2 ⊗ A2 (ω2)) is geometrically transitive, then R A1 ⊗ A2 (ω1, ω2) is a left (resp., right) principal (R A1⊗ A1 (ω1), R A2⊗ A2 (ω2))-bundle. (ii) if A is an abelian locally finite K-linear category and both functors ωi are exact, then R A2 ⊗ A2 (ω2))-bibundle with opposite bibundle isomorphic to R A1 ⊗ A2 (ω1, ω2) A2 ⊗ A1 (ω2, ω1). (iii) if, in addition to the assumptions of item (ii), we also assume that ω1 = ω ⊗K A, for some fibre functor is a principal (R A1⊗ A1 (ω1), R ω : A → vectK, then R A(ω1, ω2) is a principal (R A(ω1), R A(ω2))-bibundle. A1 ⊗ A1 (ω1) is geometrically transitive, then by [12, Lemma 4.3(b)] every (left) R Proof. (i) If R comodule is faithfully flat as an A1-module. In particular R Combining this with part (i), we have that R Analogously, if we assume that R is a right principal (R A1 ⊗ A1 (ω1), R A1 ⊗ A2 (ω1, ω2) is a left principal (R A2 ⊗ A2 (ω2) is a geometrically transitive Hopf algebroid, then R A2⊗ A2 (ω2))-bundle. A1⊗ A1 (ω1)- A1 ⊗ A2 (ω1, ω2) is faithfully flat as an A1-module. A2⊗ A2 (ω2))-bundle. A1 ⊗ A2 (ω1, ω2) A1⊗ A1 (ω1), R / / / / / / FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 17 (ii) Under these assumptions the reconstructed Hopf algebroids R A2 ⊗ A2 (ω2) are, by [10, Th´eor`eme 1.12] (see also [5, Th´eor`emes 5.2 et 7.1]), both geometrically transitive. Therefore, part (ii) implies that R A1 ⊗ A2 (ω1, ω2) is a left and right principal bundle and so a principal bibundle. The opposite A1⊗ A2 (ω1, ω2) is clearly isomorphic to R bibundle of R A1 ⊗ A1 (ω1) and R A2 ⊗ A1 (ω2, ω1). A(ω1) and R (iii) We already know form Corollary 2.4.4 that R A(ω1, ω2) is a pseudo bibundle between the Hopf A-algebras R A(ω2). The only remaining condition is the faithfully flatness of the involved A-algebras. Up to the canonical symmetric monoidal equivalence of K-linear categories between A and finite-dimensional comodules over the Hopf K-algebra R A(ω1) is a flat A-module and so it is faithfully flat (as the unit splits by the counit). To show that R A(ω1, ω2) is a faithfully flat over A, we proceed as in the proof of [35, Th´eor`eme 4.2.2, page 155], by realising R A(ω1, ω2) as an inductive limit of an inductive system of finitely generated and projective A-modules whose transition morphisms are split monomorphisms. (cid:3) K⊗ K(ω), we know that R K(ω) = R 3. The finite dual of a ring extension. In this section we specialize the construction of Section 2 to the case of a ring homomorphism A → R, which leads to the (right) finite dual A -- coring R◦, and of a functor χ : AR −→ AR◦ . We will show that, in the case of coalgebras over a Dedekind domain, our R◦ is isomorphic to the usual finite dual defined as the subspace of R∗ of all linear forms whose kernel contains a cofinite ideal of R. When RA is finitely generated and projective, we show that R◦ is isomorphic to R∗. We also study when R◦ is Galois and, what is more important, when χ is an isomorphism of categories. All these results are related to the injectivity of a map R◦ → R∗ to be defined below. 3.1. The finite dual coring R◦. Given an A -- ring R, consider the category AR of all right R -- modules that are finitely generated and projective as right A -- modules. We define the right finite dual of the extension A → R as the A-coring R◦ = R(AR) described in Section 2. We know from equation (14) that R◦ is the factor A-coring B(AR)/J(AR), where B(AR) = MP∈AR P∗ ⊗TP P and J(AR) is the K-submodule spanned by the set of elements as in Eq.(13). The elements p∗ ⊗TP p = p∗ ⊗TP p + JAR , p∗ ∈ P∗, p ∈ P, P ∈ AR form a set of generators of R◦ as a K-module (and, of course, as an A-bimodule). Recall that the comulti- plication of R◦ is given by ∆◦ : R◦ −→ R◦ ⊗A R◦, p∗ ⊗TP p 7−→ p∗ ⊗TP eα,P ⊗A e∗ α,P ⊗TP p, where {eα,P, e∗ α,P} denotes a finite dual basis for PA. The counit of R◦ is p∗ ⊗TP p 7−→ p∗(p). ε◦ : R◦ −→ A, (44) (45) 3.2. The functors χ and L. Consider the homomorphism of A-bimodules b : R◦ ⊗R R → A defined as the composite and let ηR : R → ∗(R◦) be its image under the adjunction isomorphism HomA-(R◦ ⊗R R, A) (cid:27) HomR-(R, ∗(R◦)). b : R◦ ⊗R R (cid:27) R◦ ε◦ / A Explicitly, (46) Recall that, since R◦ is an A-coring, we know that ∗(R◦) is a ring with the convolution product. A straight- forward computation shows that ηR is an anti-homomorphism of rings. ηR(r)(p∗ ⊗TP p) = p∗(pr). From the well-known (see, e.g [6, 19.1]) functor l : ComodR◦ → ∗(R◦)Mod we get, after composing with the restriction of scalars functor associated to ηR, a functor L : AR◦ −→ AR. / 18 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS This functor is explicitly given on objects as follows: given a right R◦-comodule M : M → M ⊗A R◦, and using a Sweedler-type notation (summation understood), we define the following right R-action on M m · r = m0m∗ 1(m1r), r ∈ R, M(m) = m0 ⊗A m∗ 1 ⊗TP m1. Conversely, every object P ∈ AR is a right R◦-comodule with the coaction χP : P −→ P ⊗A R◦, p 7−→ eα,P ⊗A e∗ α,P ⊗TP p. (47) (48) This gives the object map of a functor χ : AR −→ AR◦ L ◦ χ = idAR. . A straightforward computation shows that 3.3. The injectivity of the map ζ : R◦ → R∗. Consider the image ζ : R◦ → R∗ of the A-bilinear map b : R◦ ⊗R R → A under the adjunction isomorphism This homomorphism of (A, R)-bimodules is explicitly given by Hom-A(R◦ ⊗R R, A) (cid:27) Hom-R(R◦, R∗) ζ : R◦ −→ R∗, p∗ ⊗TP p 7−→ (r 7→ p∗(pr)). (49) Clearly, we have ζ(p∗ ⊗TP p)(r) = ηR(r)(p∗ ⊗TP p), where ηR is the map defined in Eq.(46). For each module MA, define βM : M ⊗A R◦ → Hom-A(R, M) by βM(m ⊗A p∗ ⊗TP p)(r) = mp∗(pr). Lemma 3.3.1. βM is injective for every MA finitely generated and projective if and only if ζ : R◦ → R∗ is injective. Proof. Observe that, up to the canonical isomorphism A ⊗A R◦ (cid:27) R◦, we have that βA = ζ, which gives the direct implication. The converse is clearly deduced from the fact that the class of modules MA for which βM is injective is closed under finite direct sums and direct summands. (cid:3) Proposition 3.3.2. If ζ : R◦ → R∗ is injective, then the funtor χ : AR → AR◦ categories. is an isomorphism of Proof. Since we already know that L ◦ χ = idAR , we only need to prove that the composition χ ◦ L is also the identity functor. Let M be an object of AR◦ with coaction M : M → M ⊗A R◦. Then L sends M to a right R-module with the action (47). With this structure, M ∈ AR, and, therefore, by applying χ, we obtain a comodule χM : M → M ⊗A R◦. We need to check that M = χM and, since βM is injective by Lemma 3.3.1, it is enough if we prove that βM ◦ M = βM ◦ χM. This follows from the following computation: (βM ◦ M)(m)(r) = m0m∗ 1(m1r) = mr (βM ◦ χM)(m)(r) = eα,Me∗ α,M(mr) = mr. (cid:3) Proposition 3.3.3. If ζ : R◦ → R∗ is injective, then R◦ is a Galois A-coring. Proof. Recall that the canonical map canR◦ : R(R◦) → R◦ is defined by canR◦ (m∗ ⊗T M m) = m∗(m0)m∗ 1 ⊗T M1 m1. Therefore (canR◦ ◦ R(χ))(p∗ ⊗TP p) = canR◦ (p∗ ⊗Tχ(P) p) = p∗(eα,P)e∗ α,P ⊗TP p = p∗ ⊗TP p In this way, canR◦ ◦ R(χ) = idR◦ . By Proposition 3.3.2, χ is an isomorphism of categories and, henceforth, R(χ) is bijective. Therefore, canR◦ is bijective and R◦ is Galois. (cid:3) Proposition 3.3.4. Assume that for every homomorphism of right R-modules f : P → R∗, where P ∈ AR, there exists Q ⊆ AR and an exact sequence of right R-modules LQ∈Q Q is injective. g / P f / R∗. Then ζ : R◦ → R∗ / / FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 19 Proof. For every P ∈ AR we consider the adjunction isomorphism of (A, TP)-bimodules Hom-R(P, R∗) (cid:27) P∗, f 7→ (p 7→ f (p)(1)) with inverse Writing P⋆ = Hom-R(P, R∗), we obtain an isomorphism of A-bimodules P∗ −→ Hom-R(P, R∗), ϕ 7→ (p 7→ (r 7→ ϕ(pr))) P⋆ ⊗TP P LP∈AR R◦ (cid:27) Kn f ⊗TQ tp − f t ⊗TP p : f ∈ Q⋆, p ∈ P, t ∈ TPQ, P, Q ∈ ARo Up to this isomorphism, ζ : R◦ → R∗ is given by ζ( f ⊗TP p) = f (p), for f ∈ P⋆ and p ∈ P. Assume Pi fi ⊗TPi pi ∈ ker ζ, that is, Pi fi(pi) = 0. Write P = ⊕iPi and define f : P → R∗ and p ∈ P uniquely by the conditions f ιi = fi and πi(p) = pi for every i, where ιi : Pi → P is the i-th canonical injection. We use the notation πi : P → Pi for the canonical projections. Since p ∈ ker f , there exist a homomorphim of right R-modules g : Q → P and q ∈ Q such that p = g(q). The following computation Xi fi ⊗TPi pi = Xi fi ⊗Pi πi(g(q)) = Xi fiπig ⊗TQ q = f g ⊗TQ q = 0. finishes the proof. (cid:3) When RA is finitely generated and projective, it is well-known that R∗ is an A -- coring (see e.g. [4, Proposition 2.11]). Corollary 3.3.5. If RA is finitely generated and projective, then ζ : R◦ → R∗ is an isomorphism of A -- corings. Proof. Our map ζ : R◦ → R∗ is injective by Proposition 3.3.4, since R is a generator of AR. In this case, ζ is obviously surjective. (cid:3) Recall (see, e.g. [38, page 22]) that a ring is said to be right hereditary if every right ideal is projective as a right module. Corollary 3.3.6. If A is a right hereditary right noetherian ring, then ζ : R◦ → R∗ is injective. Proof. Recall that, over a right hereditary ring, submodules of projective right modules are projective modules (see, e.g., [38, Proposition I.9.5]). Thus, given a homomorphism of right R-modules f : P → R∗, where P ∈ AR, then Q := ker f is projective as a right A -- module, and, of course, it is finitely generated over A, since we are assuming that A is right noetherian. Now, apply Proposition 3.3.4. (cid:3) Example 3.3.7. Of course, every commutative Dedekind domain A fulfills the hypotheses of Corollary 3.3.6. Example 3.3.8. Obviously, if A is semi-simple Artinian, then A fulfills the hypotheses of Corollary 3.3.6. Problem 1. Corollaries 3.3.5 and 3.3.6 require "extreme" conditions to guarantee, by virtue of Proposition 3.3.4, that ζ is injective. More precisely, Corollary 3.3.5 imposes a strong condition on the ring extension A → R, while Corollary 3.3.6 restricts the kind of ground ring A we are allowed to work with. It would be interesting, in view of the consequences of the injectivity of ζ (see Theorem 4.2.2 and Corollary 4.2.4 below), to investigate more general hypotheses (presumably, module-theoretical conditions) that would imply it. 3.4. Dual coalgebras. Let R be an algebra over a commutative noetherian hereditary ring A. As a gen- eralization to the case of algebras over fields, it is possible to define a structure of A -- coalgebra over the A -- submodule R′ of R∗ consisting of those ϕ ∈ R∗ such that ker ϕ contains an A -- cofinite ideal of R (see [1, Theorem 2.8, Proposition 2.11]). Here, an ideal I of R is A -- cofinite if R/I is a finitely generated A -- module. The map ζ : R◦ → R∗ factorizes through R′. Indeed, for any generator p∗ ⊗TP p ∈ R◦, we need to prove that the cyclic right R -- module ϕR, where ϕ = ζ(p∗ ⊗TP p), is finitely generated as an A -- module (see [1, 2.3]). Let r, s ∈ R. Since ζ is right R -- linear, A is central in R, and by using a suitable dual basis, we get (ϕr)(s) = p∗(prs) = XαP p∗(eαP e∗ αP (pr)s) = XαP p∗(eαP se∗ αP (pr)) = XαP p∗(eαP s)e∗ αP (pr). 20 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS This implies that ϕr belongs to the A -- submodule of R∗ generated by the finite set of all p∗(eαP −)'s and, hence, ϕR is finitely generated as an A -- module. It follows from Example 3.3.7 that ζ : R◦ → R∗ is injective. Let τ : R◦ ⊗ R◦ → R◦ ⊗ R◦ denote the flip map. A straightforward computation shows that the following diagram is commutative. R◦ ζ R∗ τ∆◦ R◦ ⊗ R◦ ζ⊗ζ R∗ ⊗ R∗ m∗ $❏❏❏❏❏❏❏❏ xqqqqqqqqq (R ⊗ R)∗ In resume, we have that, if R is an algebra over a commutative noetherian hereditary ring A, then ζ : R◦ → R′ is an injective anti-homomorphism of A -- coalgebras. Proposition 3.4.1. Let R be an algebra over a commutative Dedekind domain A. Then ζ : R◦ → R′ is an anti-isomorphism of A -- coalgebras. Proof. We need just to show that ζ : R◦ → R′ is surjective. Let ϕ ∈ R′. We know that the cyclic right R -- module ϕR is a finitely generated A -- module. On the other hand, ϕR ⊆ R∗ and, therefore, ϕR is a torsion-free A -- module. Since A is a Dedekind domain, we get that ϕR is a fgp A -- module. Now, I := {r ∈ R : ϕr = 0} is a right ideal of R and, since ϕR (cid:27) R/I, we get that P := R/I is fgp as an A -- module. On the other hand, I ⊆ ker ϕ, which implies that there is ϕ ∈ P∗ such that ϕ(r + I) = ϕ(r) for all r ∈ R. Hence, ϕ = ζ(ϕ ⊗TP (1 + I)). (cid:3) Remark 3.4.2. The proof of Proposition 3.4.1 works to prove that if R is an algebra over a semisimple commutative ring A, then ζ : R◦ → R′ is an anti-isomorphism of coalgebras. 4. The finite dual of a right bialgebroid, and of a cocommutative Hopf algebroid. We show that the right finite dual U ◦ of a right bialgebroid (A, U) is a left bialgebroid. When A is commutative, this fact can be deduced from [5, Exemple, pp. 5849], although that construction cannot be directly extended to the setting of a non-commutive basis A because add(AA) is not a monoidal category. We also prove that the map ζ : U ◦ → U ∗ is a homomorphism of Ae -- rings, when U ∗ is endowed with the convolution product. Theorem 4.2.2 states that the finite dual of a cocommutative Hopf algebroid with commutative base ring is a commutative Hopf algebroid. This could also have been deduced from [5, Example, pp. 5849], we include an elementary proof. Our approach could be useful to treat specific examples. This will be illustrated in the last section. Theorem 4.2.2 also contains our main contribution in this section, namely, a sufficient condition to get a monoidal equivalence between the category of the A -- profinite modules over a cocommutative Hopf algebroid over A and the A -- profinite representations of its associated affine groupoid via the finite dual construction. This equivalence works in a non necessarily tannakian context. 4.1. Duality for bialgebroids. Let U be a right bialgebroid over a (possibly non commutative) K -- algebra A, and consider U as an A -- ring via its source map s : A → U. Consider the fibre functor s∗ : AU → add(AA) as in Section 2, and the corresponding right finite dual A -- coring U ◦ as in Subsection 3.1. Our aim is to endow U ◦ with the structure of a left bialgebroid over A in such a way that the map ζ : U ◦ → U ∗ defined in Eq.(49) becomes a morphism of Ae-rings. Here, the right convolution ring U ∗ is an Ae -- ring via the homomorphism of rings Lemma 4.1.1. Let (A, U) be a right bialgebroid. Then U ◦ admits a structure of Ae-ring such that ζ : U ◦ → U ∗ is a homomorphism of Ae-rings. ξ : Ae −→ U ∗, (cid:16)a ⊗ bo 7−→ (cid:2)u 7→ aε(t(b)u)(cid:3)(cid:17). (50) Proof. Observe first that A ∈ AU , where the right U -- module structure is given by the action a.u = ε(t(a)u) = ε(s(a)u). (51) / / $ x FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 21 With this right U-module structure, it is clear that A is the identity object of the monoidal category AU . Now, as in Eq. (21) of subsection 2.2, the element idA ⊗TA 1 ∈ U ◦ is the unit for the multiplication (20), and we have that is a homomorphism of rings (the associated source and target are denoted by s◦ and t◦, respectively) where the endomorphism ring TA of AU is the commutative subalgebra of A defined by η◦ : Ae −→ U ◦, (cid:16)(a ⊗ bo) 7−→ (la ⊗TA b)(cid:17), (52) (53) TA = na ∈ Z(A) aε(u) = ε(cid:0)t(a)u(cid:1), for every u ∈ Uo. Recall that the map ζ sends p∗ ⊗TP p 7→ (cid:2)u 7→ p∗(pu)(cid:3). In order to see that it is a homomorphism of Ae -- rings, let us check first that it is unital. So, ζ ◦ η(a ⊗ bo)(u) = ζ(la ⊗TA b)(u) = la(bu) = aε(t(b)u) = ξ(a ⊗ bo)(u), ∀u ∈ U. Hence ζ ◦ η = ξ, where ξ is the map defined in Eq.(50). Now, using the above multiplication Eq.(20), we have ζ(cid:16)(p∗ ⊗TP p) . (q∗ ⊗TQ q)(cid:17) (u) = ζ(cid:16)(q∗ ⋆ p∗) ⊗TQ⊗A P (q ⊗A p)(cid:17)(u) = (q∗ ⋆ p∗)((q ⊗A p) u) = (q∗ ⋆ p∗)(qu1 ⊗A pu2) = p∗(q∗(qu1)(pu2)) = p∗((pu2)(t(q∗(qu1)))). On the other hand, using the convolution multiplication defined by the coring UAe , we have ζ(p∗ ⊗TP p) ∗ ζ(q∗ ⊗TQ q) (u) = ζ(p∗ ⊗TP p)(cid:18)ζ(q∗ ⊗TP q)(u1)u2(cid:19) = ζ(p∗ ⊗TP p)(q∗(qu1)u2) = p∗(p(q∗(qu1)u2)) = p∗(p(u2t(q∗(qu1)))) = p∗((pu2)t(q∗(qu1))). Henceforth, ζ is multiplicative, and this finishes the proof. (cid:3) Proposition 4.1.2. If (A, U) is a right bialgebroid, then (A, U ◦) is a left bialgebroid. Proof. We know that U ◦ is constructed from the monoidal category AU and the forgetful functor ω = s∗ : AU → AModA, that is, we have that U ◦ = R(AU ) as in the notation of subsection 2.2. Therefore, we can apply the first statement of Proposition 2.2.1 to obtain the claim. (cid:3) Remark 4.1.3. There is a kind of symmetry in Proposition 4.1.2. This means that given a left bialgebroid (A, V), using it target map t : A → V and it associated category V A of left V-modules which are finitely generated and projective as A-modules via the fibre functor ∗t : V Mod → add(AA). The reconstruction process of Section 2 leads to a right bialgebroid(cid:0)A, R(V A)(cid:1) := (A, ◦V). 4.2. Duality for Hopf algebroids. In this subsection we assume that (A, U) is a left Hopf algebroid over a commutative K-algebra A, and the underlying A-coring of U is co-commutative. Bialgebras over field extensions studied in [31, 7] and those over a commutative algebra studied in [34], as well as the universal algebras of Lie algebroids or, in general, of Lie-Rinehart algebras (Examples 1.3.3 and 1.3.2), are all examples of this class of left bialgebroids. It is easily checked that, under the current assumptions, the source of U is equal to its target (i.e., s = t). This fact will be implicitly used in the sequel. Recall that for such a bialgebroid (A, U), the category AU consists of right U-modules whose underlying A-modules are finitely generated and projective, using either the functor s∗ or t∗. Proposition 4.2.1. Let (A, U) be a co-commutative right Hopf algebroid over a commutative algebra A. Consider the category AU with the fibre functor s∗ : AU → add(A). Then AU is a monoidal symmetric and rigid category with s∗ a strict monoidal fibre functor. 22 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS Proof. First observe that, since t = s, each object in AU is a central A-bimodule. On the other hand, for every pair of objects P, Q ∈ AU the flip map τP, Q : P ⊗A Q → Q ⊗A P, sending p ⊗A q 7→ q ⊗A p is actually an arrow in AU , since U is co-commutative. The rigidity of AU is immediate from Lemma 1.2.2, and the duals are described as follows: For every arrow f : P → Q in AU , its A-linear dual map f ∗ : Q∗ → P∗ is clearly an arrow in the same category AU , where P∗, Q∗ are objects of AU by the action of Eq.(8). Thus, P∗ is a dual object in AU for an object P ∈ AU , and the duality is given by the following arrows in AU ev : P∗ ♦ P = P∗ ⊗A P → A,(cid:0)p∗ ⊗A p 7→ p∗(p)(cid:1), (respectively called evaluation and dual-basis) where {eα,P, e∗ is now clear. db : A → P♦ P∗ = P ⊗A P∗,(cid:0)1 7→ eα, P ⊗A e∗ α, P(cid:1), α,P} is a dual basis for P. The rest of the proof (cid:3) (54) The main results of this section are the last two parts of the following theorem. Theorem 4.2.2. Let (A, U) be a co-commutative right Hopf algebroid over a commutative algebra A. Then (1) (A, U ◦) is a commutative Hopf algebroid over TA. (2) The functor χ : AU → AU◦ (3) If ζ : U ◦ → U ∗ is injective, then χ is an isomorphism of symmetric monoidal categories. is strict monoidal and preserves the symmetry. Proof. (1) This part follows from Propositions 4.2.1 and 2.2.1, and the Hopf algebroid structure maps are explicitly given as follows. The algebra structure is given by the multiplication of equation (20) and unit the algebra map η of equation (52). The the comultiplication is the algebra map ∆◦ : U ◦ −→ U ◦ ⊗A U ◦, p∗ ⊗TP p 7−→ XαP p∗ ⊗TP eα,P ⊗A e∗ α,P ⊗TP p, where {eα,P, e∗ α,P} denotes a finite dual basis for PA. The counit of U ◦ is p∗ ⊗TP p 7−→ p∗(p), ε◦ : U ◦ −→ A, and the antipode is the algebra map where ψ : P → (P∗)∗ is the canonical isomorphism of A-modules, as in subsection2.2. S ◦ : U ◦ −→ U ◦, (p∗ ⊗TP p 7−→ ψ(p) ⊗TP∗ p∗), (55) (56) (57) (48) by the coaction: (2) Let P, Q be two objects in AU . Then the right U ◦-comodule structure of χ(cid:0)P ⊗A Q(cid:1), is given as in β,Q) ⊗TP⊗A Q (p ⊗A q)(cid:17) (cid:16) p ⊗A q 7−→ X(eα,P ⊗A eβ,Q) ⊗A (e∗ χ(P⊗A Q) : P ⊗A Q −→ P ⊗A Q ⊗A U ◦, α,P} and {eβ,Q, e∗ where {eα,P, e∗ tensor product χ(P) ⊗A χ(Q), in the monoidal subcategory of right U ◦-comodules AU◦ comodule structure: β,Q} denote as above the dual basis of PA and QA. On the other hand, the , has the following α,P ⋆ e∗ χ(P)⊗A χ(Q) : P ⊗A Q −→ P ⊗A Q ⊗A U ◦, (cid:16) p ⊗A q 7−→ X p(1) ⊗A q(1) ⊗A p(2)q(2)(cid:17) Now, using equation (48) and the commutativity of the multiplication of U ◦, as given in (20), we get that χ(P)⊗A χ(Q) = χ(P⊗A Q). Therefore, χ(P) ⊗A χ(Q) = χ(P ⊗A Q), for any two objects P, Q in AU . The identity object of AU is the right U-module A with action a.u = ε(au), for any a ∈ A and u ∈ U. The image by χ of this object has the right coaction χ(A) : A → A ⊗A U ◦ sending a 7→ 1 ⊗A 1 ⊗TA a. Thus, by equation (52), we have A(a) = t◦(a), for every a ∈ A. Hence, χ(A) = (A, t◦) the identity object of the monoidal category AU◦ . We have then shown that χ is a strict monoidal functor. Lastly, since both monoidal categories AU and AU◦ have the flip as symmetry, one trivially obtains that χ is a symmetric monoidal functor. (3) Follows from Proposition 3.3.2. (cid:3) Corollary 4.2.3. Let (A, U) be a cocommutative right Hopf algebroid over a Dedekind domain A. Then the category AU is isomorphic, as a symmetric monoidal category, to AU◦ . Proof. This is deduced from Corollary 3.3.6 and Theorem 4.2.2. (cid:3) We also get the following consequence of Theorem 4.2.2, whose geometrical interpretation is that, under suitable conditions, given a Lie-Rinehart algebra there is a groupoid with the "same" representation theory (see Example 1.3.2 for details on the Hopf algebroid attached to a Lie-Rinehart algebra). FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 23 Corollary 4.2.4. Let L be a Lie-Rinehart algebra over a commutative ring A, and let U(L) denote its universal enveloping Hopf algebroid. If ζ is an injective map (e. g. if A is a Dedekind domain), then the category of A-profinte right U(L)-modules is isomorphic, as a symmetric rigid monoidal category, to the category of A-profinite right comodules over the commutative Hopf algebroid U(L)◦. Remark 4.2.5. Starting with a commutative Hopf algebroid (A, V), then, by Proposition 4.1.2, we know that (A, V ◦) is a right bialgebroid. The fact that (A, U ◦) is actually a right Hopf algebroid can be shown using the tecniques developed in [20]. In analogy with the classical situation of Hopf algebra over fields, it stills then to check that U ◦ is a co-commutative A-coring. It seems that in general, there is no direct way of proving this property. However, it can be derived under some assumptions, which are always fulfilled in the finite case (i.e., when VA or AV are finitely generated and projective). Precisely, if we assume that the map ζ : V ◦ → V ∗ of subsection 3.3 and the canonical map ΨV : V ∗ ×A V ∗ −→ (V ⊗A⊗ A V)∗, (cid:18)(p∗ ⊗TP p) ×A (q∗ ⊗TQ q) 7−→ (cid:20)u ⊗A⊗ A v 7→ p∗(pu) q∗(qv)(cid:21)(cid:19) (58) (here ×A is the Sweedler-Takeuchi's product [39, 41]), are injective and that VA is flat, then one can deduce, from the following commutative diagram, the co-commutativity of the comultiplication ∆◦ of the A-coring V ◦ ζ / V ∗ V ◦ ∆◦ V ◦ ×A V ◦ ζ×Aζ / V ∗ ×A V ∗ µ∗ ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ 4❥❥❥❥❥❥❥❥❥❥❥❥ ΨV (V ⊗A⊗ A V)∗ Following the observations of Example 3.3.5, the duality stated in [34, Propositions 3, 4] (see also [22, 8]), is now a particular instance of the one established hereby. 5. Application: The finite dual of the first Weyl algebra, differential Galois groupoid and PV extensions In this section, we illustrate our methods by treating, in an exhaustive way, the universal Hopf algebroid of the transitive Lie algebroid of vector fields over the complex affine line A1 C. In other words the first Weyl C-algebra. We first show that the category of differential modules is a Tannakian category (in the sense of [10]) which is identified with the category of comodules over the finite dual with underlying finitely generated free C[X] -- modules. Second we show that for a fixed differential module (M, ∂), this finite dual contains a commutative Hopf algebroid denoted by U ◦ with finitely generated underlying modules, is equivalent, as a symmetric monoidal C-linear category, to the full subcategory hMi⊗ of differential modules which are sub-quotients objects of (M, ∂). In analogy with the classical differential Galois theory over C(X), the associated affine algebraic groupoid of U ◦ (M), is then termed the differential Galois groupoid attached to (M, ∂) (or to the linear differential matrix equation defined by (M, ∂)). We also combine our result with those of [2], in order to give an explicit description of a Picard-Vessiot exetension of (C[X], ∂ ∂X ) for (M, ∂). In the last subsection, we compare our approach with that of Malgrange [26] and Umemura [42], and give several illustrating examples. (M) whose category of comodules comodU◦ (M) 5.1. The Hopf algebroid structure of the first Weyl algebra. Let A = C[X] be polynomial ring in the variable X over the field of complex numbers, and consider its (noncommutative) ring of differential operators U := C[X][Y, ∂ ∂X ], that is, the first Weyl algebra. We shall consider U as free right A-module with basis {Y n}n∈N, and with left action given by the rule aY = Ya + ∂a ∂X , for every a ∈ A. This algebra is clearly isomorphic to the universal right Hopf algebroid of the transitive Lie algebroid (A, DerC(A)), see Example 1.3.2. The structure maps are ∆(Y) = 1 ⊗A Y + Y ⊗A 1, ε(Y) = 0, and Y− ⊗A Y+ = 1 ⊗A Y − Y ⊗A 1. We are interested in describing the relationship between the category of comodules over the finite dual U ◦ and the category of differential modules over the differential ring A. To this end, we will apply Theorem 4.2.2 to the pair (A, U). But, first let us make the following general remark. / / 4 24 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS Remark 5.1.1. Theorem 4.2.2 applies for any ring of differential operators A[Y, δ], for δ any K -- linear derivative of A. In particular, Corollary 4.2.3 leads to a monoidal equivalence of categories between the category Diff A of A -- profinite differential modules and the A -- profinite representations of the affine groupoid represented by the finite dual of A[Y, δ], whenever A is a Dedekind domain (e.g. the coordinate ring of an irreducible smooth curve over C). If A is not a field, Diff A will probably fail to be abelian, as the example of C[X][Y, δ] shows by taking δ( f (X)) = X ∂ f (X) ∂X for all f (X) ∈ C[X]. What makes the Weyl algebra so special is that, in this case, Diff A is abelian. 5.2. Differential modules Diff A as a Tannakian category. Recall that a differential right module over the differential ring A is a finitely generated right A-module equipped with a C-linear map ∂ : M → M such that ∂(xa) = ∂x.a + ∂a.x, for every a ∈ A and x ∈ M (here ∂a stand for ∂a(X)/∂X). The linear map ∂ is called the differential of M. In all what follows the underlying A-module of a right differential module, will be considered as central (or symmetric) A-bimodule, that is, we have ax = xa, for every x ∈ M and a ∈ A. Every differential module is in fact free of finite rank as an A-module: if x ∈ M is a torsion element, then ax = 0 for some nonzero a ∈ A. It follows that (∂x)a + x∂a = 0, whence (∂x)a2 = 0. This shows that the (C-finite dimensional) torsion submodule t(M) of M is indeed a differential module. This is only possible when t(M) = 0 (see, e.g. [3, Lemma 4.2]). Using the notation of Section 2, the category AU is in this case the category of all differential modules over A, equivalently linear differential matrix equations. If we denote by {e1, . . . , em} any basis of M over A, the differential ∂ is then given by a matrix mat(M) = (ai j) ∈ Mm(A) such that ∂ei = − m Xj=1 e ja ji. (59) The minus sign in introduced, not only for historical reasons, but also for computational ones4. So if we identify an element y ∈ M with its coordinate column in An, we have Thus ker(∂) is the solution space of the following linear differential matrix equation ∂y1 ... ∂ym ∂y =   y′ i ... y′ m  y1 ... ym .   − mat(M) = mat(M) y1 ... yn  . (60) In analogy to [44, §2.2] we denote by Diff A the category AU . A morphism of differential modules f : (M, ∂) → (N, ∂) is an A-linear map f : M → N which commutes with differentials, that is, ∂ ◦ f = f ◦ ∂. A inherits a monoidal symmetric structure from the category of right U-modules. In this case, the tensor product of two objects (M, ∂), (N, ∂) in Diff A. First, we know from Section 4 that the category Diff A is again a differential module with differential map Next we list the properties of Diff ∂ : M ⊗A N −→ M ⊗A N, (cid:16)∂(x ⊗A y) = ∂(x) ⊗A y + x ⊗A ∂(y)(cid:17) (61) By Lemma 1.2.2, Diff A is then a monoidal symmetric and rigid C-linear category with identity object (A, ∂). The forgetful functor (which going to be a fibre functor in a Tannakian sense) is given as in subsection 2.1 A := AU → add(A), where η : A → U is the canonical by the restriction of scalars functor ω = η∗ : Diff ring extension. Observe that ω is a non trivial functor in the sense of [10] since we know that ω(A, ∂) = A and that Spec(A) , ∅. Moreover, Diff A is a full subcategory of the category ModU stable under finite limits and colimits. Hence, it is abelian. We summarize the properties of Diff A in the following lemma. Lemma 5.2.1. The category Diff A is C-linear abelian and the functor ω : Diff A → add(A) is strict monoidal C-linear faithful exact functor. Moreover, we have an isomorphism of rings EndDiffA(cid:16)(A, ∂)(cid:17) (cid:27) C. 4As we will notice in subsection 5.5, this depends on considering the principal bibundle structure of a Picard-Vessiot extension or its opposite bibundle, see Remark 5.5.7. FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 25 Remark 5.2.2. Observe furthermore, that Lemma 5.2.1 and [5, Proposition 2.5] imply that Diff A is a locally finite category over C, in the sense that each object is of finite length and the Hom-vector spaces are finite dimensional C-vector spaces. This in particular implies that the endomorphism ring of any differential module is a finite dimensional C-algebra. Therefore, the centre of the endomorphism ring of any simple object in Diff A of differential modules is a separable category in the sense of [5, D´efinitions page 5847]. A, coincides with the base field C. Thus, the category Diff Remark 5.2.3. Let (M, ∂) be a differential module with rank(M) = m. Denote by M∂ := ker(∂) the C-vector space of solutions in A of the equation (60). Then, there is an isomorphism of C-vector spaces HomDiffA ((A, ∂), (M, ∂)) (cid:27) M∂. Therefore, by Remark 5.2.2, we know that dimC(M∂) < ∞. In fact we have that dimC(M∂) ≤ m. Apart from the above structure, the usual linear algebra operations are also permitted in the category of differential modules. For instance, the exterior powers Vd M of a differential module M, are again differential modules with differential given by ∂ (x1 ∧ · · · ∧ xd) = d Xi=1 x1 ∧ · · · ∧ ∂xi ∧ · · · ∧ xd, such that the canonical linear map M⊗d in Diff given differential module (M, ∂). A. Similarly one can endows the symmetric dth-powers Symd(M) A-module with a differential, for a := M ⊗A · · · ⊗A M −→ Vd M is a differential map, i.e. a morphism On the other hand the internal hom-functors of Eq.(4), are explicitly given in this case by the hom- functors HomA(M, N) whose differential is defined by the formula Thus, up to isomorphisms, we have ∂l : M −→ N, (cid:16)x 7−→ (∂l)(x) = ∂l(x) − l(∂x)(cid:17). (62) homModU (M, N) = HomA(M, N). In particular, the vector space HomU(M, N) is identified with the linear maps whose differential is zero. The differential of the dual module is given then by for every ϕ ∈ M∗ and x ∈ M. ∂ : M∗ −→ M∗, (cid:16)(∂ϕ)(x) = ∂ϕ(x) − ϕ(∂(x))(cid:17), (63) 5.3. The commutative Hopf algebroid attached to the first Weyl algebra. Next we state some prop- erties of the commutative Hopf algebroid (A, U ◦) constructed from U by applying Theorem 4.2.2. First, we know from Example 3.3.7, that the canonical algebra map ζ : U ◦ → U ∗ defined by equation (49) is injective, where U ∗ is the right linear dual of U endowed with the convolution product. This fact will be implicitly used in the sequel. Proposition 5.3.1. Let A = C[X] and U = A[Y, ∂/∂X] its differential operator algebra. Then the commu- tative Hopf algebroid (A, U ◦) is a Galois A-coring. In particular the category Diff A is isomorphic to the full subcategory of right U ◦-comodules AU◦ and so EndComodU◦(cid:16)(A, t)(cid:17) (cid:27) C an isomorphism of rings. Proof. The first statement is a direct consequence of Proposition 3.3.3. The particular cases are immediate from Proposition 3.3.2 and Lemma 5.2.1. (cid:3) The subsequent contains further properties of the Hopf algebroid (A, U ◦). Corollary 5.3.2. Let A = C[X] and U = A[Y, ∂/∂X] its differential operator algebra. Then the commutative Hopf algebroid (A, U ◦) enjoys the following properties: (i) The algebra map η◦ : A ⊗C A → U ◦ induces on U ◦ a projective (A ⊗C A)-module structure. (ii) There is an isomorphism of symmetric monoidal C-linear categories Diff A (cid:27) comodU◦ , where comodU◦ denotes the full subcategory of right U ◦-comodules with finitely generated underlying A-modules. (iii) Every right U ◦-comodule is projective and faithfully flat as an A-module. In particular, the modules U ◦ A and AU ◦ are projective. 26 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS (iv) Any right U ◦-comodule is a filtered limit of subcomodules in comodU◦ . Proof. (i), (ii). We know from subsection 3.1, that (A, U ◦) was constructed from the pair (AU , s∗), or up to the isomorphism of Proposition 5.3.1, from the pair (Diff A, ω) of Lemma 5.2.1. By applying Deligne's Theorem [5, Th´eor`emes 5.2, 7.1], we have from one hand that the functor ω induces the equivalence sated in (ii) which proves this item. From another hand, we have that the Hopf algebroid (A, U ◦) is transitive and also separable by Remark 5.2.2. Therefore, it is geometrically transitive by [5, Corollaire 6.7], which by definition [5, D´efinition page 5845] means that U ◦ is a projective (A ⊗C A)-module, and this shows part (i). (iii). It follows directly from [5, Proposition 7.2], while item (iv) follows from [5, Proposition 6.2]. (cid:3) Remark 5.3.3. By [14, Theorem 5.7] and Corollary 5.3.2(iii), the category of right comodules over the Hopf algebroid (A, U ◦) admits Diff A as a generating set of small projectives if and only if the direct sum ⊕M∈ Diff A M is a faithfully flat right module over the ring ⊕M, N ∈ Diff A Hom(M, N) with enough orthogonal idempotents (i.e. Gabriel's ring of Diff A). Both equivalent conditions are fulfilled by Corollary 5.3.2(iv). 5.4. The differential Galois groupoid of a differential module. Fix a differential module M ∈ Diff a dual basis {ei, e∗ i }1≤i≤m. We set A with det(e1, . . . , em) := Xσ ∈ S m (−1)sg(σ)eσ(m) ⊗A . . . ⊗A eσ(1) ∈ M ⊗m , (64) where S m is the permutation group of m elements, sg(σ) is the signature of σ and M⊗m denotes the m-fold tensor product of the A-module M. We also denote := M ⊗A · · · ⊗A M detM := e∗ m ⋆ · · · ⋆ e∗ 1 ⊗T M⊗m det(e1, . . . , em) ∈ U ◦, (65) where, as in equation (18), the A-linear map e∗ e∗ m ⋆ · · · ⋆ e∗ 1 : M⊗m −→ A, 1 is defined by m ⋆ · · · ⋆ e∗ (cid:16)x1 ⊗A · · · ⊗A xm 7−→ e∗ m(xm) · · · e∗ 1(x1)(cid:17). Lemma 5.4.1. Keeping the previous notations, we then have (i) As an element in M ⊗m , the differential of det(e1, . . . , em) is ∂ det(e1, . . . , em) = tr(cid:0)mat(M)(cid:1) det(e1, . . . , em), (ii) (e∗ where tr(cid:0)mat(M)(cid:1) is the trace of mat(M); m ⋆ · · · ⋆ e∗ 1)(cid:16)det(e1, . . . , em)(cid:17) = 1. Proof. Both part (i) and (ii) are routine computations by using the formulae (61) and definitions. The details are left to the reader. (cid:3) A crucial consequence of this lemma is the subsequent. Lemma 5.4.2. The element detM of Eq.(65) is invertible in the algebra U ◦. one and basis e∗ 1 Proof. Let Vm M∗ be the m-exterior power of the dual module M∗ of M. This is a free A-module of rank tr(mat(M)) denotes the trace of the matrix mat(M). Thus, the module Vm M∗ is an object in the category m with differential ∂(e∗ 1 m) = −tr(mat(M))e∗ 1 m, where as before ∧ · · · ∧ e∗ ∧ · · · ∧ e∗ ∧ · · · ∧ e∗ Diff A. By Lemma 5.4.1(i), we define the following differential map 1 ∧ · · · ∧ e∗ f : A −→ (∧m M∗) ⊗A M⊗m , (cid:16) 1 7−→ (e∗ m) ⊗A det(e1, . . . , em)(cid:17), which we consider as a morphism in the category Diff A. Now let us check that det−1 M = (cid:16)e∗ m ⋆ · · · ⋆ e∗ 1 ⊗T So we compute their multiplication: M⊗m det(e1, . . . , em)(cid:17)−1 M⊗m det(e1, . . . , em) . (e∗ 1)(cid:17) ⊗T 1)(cid:17) ⊗T m ⋆ · · · ⋆ e∗ m ⋆ · · · ⋆ e∗ 1 m ⋆ · · · ⋆ e∗ e∗ 1 ⊗T 1 = (cid:16)(e∗ = (cid:16)(e∗ 1 ∧ · · · ∧ e∗ m)∗ ⋆ (e∗ ∧ · · · ∧ e∗ m)∗ ⋆ (e∗ = (e∗ 1 ∧ · · · ∧ e∗ m)∗ ⊗TVm M∗ (e∗ 1 ∧ · · · ∧ e∗ m). (66) ∧ · · · ∧ e∗ ∧ · · · ∧ e∗ m) m)∗ ⊗TVm M∗ (e∗ 1 ∧ · · · ∧ e∗ m) ⊗A det(e1, . . . , em)(cid:17) 1 (∧m M∗)⊗A M⊗m (cid:16)(e∗ (∧m M∗)⊗A M⊗m (cid:16) f (1)(cid:17) FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 27 = (cid:16)(e∗ The A-linear map (cid:16)(e∗ 1 1 ∧ · · · ∧ e∗ m)∗ ⋆ (e∗ m ⋆ · · · ⋆ e∗ ∧ · · · ∧ e∗ m)∗ ⋆ (e∗ m ⋆ · · · ⋆ e∗ 1)(cid:17) f ⊗TA 1. 1)(cid:17) f ∈ A∗ is defined by = = (cid:16)(e∗ 1 ∧ · · · ∧ e∗ (cid:16)(e∗ 1 ∧ · · · ∧ e∗ 1 ∧ · · · ∧ e∗ (e∗ m)∗ ⋆ (e∗ m)∗ ⋆ (e∗ m)∗(cid:16)e∗ m ⋆ · · · ⋆ e∗ m ⋆ · · · ⋆ e∗ m(cid:17) (e∗ 1 ∧ · · · ∧ e∗ 1)(cid:17)(cid:0) f (1)(cid:1) 1)(cid:17)(cid:16)(e∗ m ⋆ · · · ⋆ e∗ 1 ∧ · · · ∧ e∗ m) ⊗A det(e1, . . . , em)(cid:17) 1)(cid:16)det(e1, . . . , em)(cid:17) 5.4.1(ii) = 1, which shows that m ⋆ · · · ⋆ e∗ e∗ 1 ⊗T M⊗m det(e1, . . . , em) . (e∗ 1 and this finishes the proof. ∧ · · · ∧ e∗ m)∗ ⊗TVm M∗ (e∗ 1 ∧ · · · ∧ e∗ m) = 1 (cid:3) Remark 5.4.3. Similar to the context of commutative Hopf algebras, a grouplike element g in a commu- tative Hopf algebroid is always an invertible element with inverse S (g), its image by the antipode. In the particular situation of Lemma 5.4.2, it is easily seen from equation (66) that ∆(cid:0)det−1 M(cid:1) = (e∗ 1 = det−1 m)∗ ⊗TVm M∗ (e∗ ∧ · · · ∧ e∗ M ⊗A det−1 M . 1 ∧ · · · ∧ e∗ m) ⊗A (e∗ 1 ∧ · · · ∧ e∗ m)∗ ⊗TVm M∗ (e∗ 1 ∧ · · · ∧ e∗ m) which shows that det−1 M is a groulike element, and so is detM. Therefore, we have that On the other hand, the structure of right U ◦-comodule over AA which corresponds to this grouplike S (detM) = det−1 M . (67) element and, up to an isomorphism, is the right U ◦-comodule structure of Vm M∗ deduced by applying Proposition 5.3.1, is given by the coaction A → U ◦, a 7→ det−1 module (AA, ∂M) with differential M a. This corresponds to the differential For a given M, let ◦ ∂M : A −→ A, (cid:16)a 7−→ −tr(cid:0)mat(M)(cid:1) a + ∂a(cid:17). (68) U M be the (A ⊗C A) -- subalgebra of U ◦ generated by the image of M∗ ⊗T M M via the U M is generated as an (A ⊗C A) -- algebra by the ◦ canonical map M∗ ⊗T M M → U ◦. It is not hard to see that set ne∗ for any given a dual basis {ei, e∗ j ⊗T M eio1≤i, j≤m i }1≤i≤m as above. Proposition 5.4.4. The subalgebra U M is in fact a sub-bialgebroid of U ◦, which contains the element detM ◦ defined in Eq. (65). Furthermore, its localization algebra ◦ U M[det−1 M ] is a Hopf subalgebroid of U ◦. Proof. The first statement is immediate, since we already know that ∆(cid:0)e∗ j ⊗T M ei(cid:1) = Xk e∗ j ⊗T M ek ⊗A e∗ k ⊗T M ei, where ∆ denotes the comultiplication of U ◦. Since ∆ is already multiplicative, we get a coassociative and multiplicative A -- bimodule map ∆ : U M → U M ⊗A U M. Observe that we are using that A is a principal ideal ◦ ◦ ◦ domain and that U ◦ is torsionfree over A on both sides. The fact that detM ∈ formula of the multiplication in U ◦ displayed in (20). ◦ U M clearly follows from the Next we show that the antipode of U ◦ restricts to ◦ U M[det−1 ◦ the inclusion detu the determinant detM of Eq.(65). U M[det−1 M ] ⊆ U ◦. Denote by ui j the generating elements e∗ M ], since we already have by Lemma 5.4.2 i ⊗T M e j, for i, j = 1, . . . , m, and by Recall that the ( j, i)th cofactor of the matrix (ui j)1≤i, j≤m is then defined to be v ji = (−1)i+ j Xσ ∈ P( j,i) (−1)sg(σ)u j1σ( j1) . . . u jm−1σ( jm−1), 28 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS where P( j, i) is the set of all bijections from {1, . . . , m} \ { j} = { j1 < · · · < jm−1} to {1, . . . , m} \ {i}, and sg(σ) is the signature of σ viewed as a permutation in S m−1. Clearly, each of the v ji is an element in the (A ⊗ A)-subalgebra U M. Writing down this element using the notation of Eq.(64), we get ◦ v ji = (−1)i+ j e∗ jm−1 ⋆ · · · ⋆ e∗ j1 ⊗T ⊗m−1 M det(e1, . . . , g ei, . . . , em), where det(e1, . . . , ponent ei, see Eq.(64). g ei, . . . , em) is the resulting determinant of the vector (e1, . . . , em) after removing the com- Define now the following morphism in the category Diff A, by the composition hm : M∗ db⊗m−1 ⊗ M∗ / (M ⊗ M∗)⊗m−1 ⊗ M∗ / M⊗m−1 ⊗ (M∗)⊗m / M⊗m−1 ⊗ (cid:16)Vm M∗(cid:17), where db : A → M ⊗A M∗ is the coevaluation map of Eq.(54) up to the differential isomorphism M ⊗A M∗ (cid:27) HomA(M, M), the second map is the flip, and the third one is obvious. On elements, hm acts by hm(ϕ) = Xk1,..., km−1 (ekm−1 ⊗A . . . ⊗A ek1 ) ⊗A (e∗ k1 ∧ · · · ∧ e∗ km−1 ∧ ϕ), for every ϕ ∈ M∗, where each of the index kl runs the set {1, . . . , m}, for each of the l = 1, · · · , m − 1. Reordering the indices, we obtain hm(e∗ i ) = (−1)idet(e1, . . . , g ei, . . . , em) ⊗A (e∗ 1 ∧ · · · ∧ e∗ m). Using this equality and the above notation, we show that where as above γ : M → (M∗)∗ is the canonical isomorphism. Now, we compute γ(e j) = h(−1) je∗ ⋆ · · · ⋆ e∗ j1 ⋆ (e∗ 1 ∧ · · · ∧ e∗ jm−1 m)∗i ◦ hm, (69) (70) det−1 u . v ji (66) = (e∗ 1 ∧ · · · ∧ e∗ ∧ · · · ∧ e∗ m) (−1)i+ j e∗ ⋆ · · · ⋆ e∗ j1 ⊗T ⊗m−1 M det(e1, . . . , g ei, . . . , em) jm−1 (69) = (−1) j e∗ jm−1 = (−1) j e∗ jm−1 1 m)∗ ⊗TVm M∗ (e∗ ⋆ (e∗ ⋆ · · · ⋆ e∗ j1 1 ⋆ · · · ⋆ e∗ j1 ⋆ (e∗ 1 ∧ · · · ∧ e∗ m)∗ ⊗T ⊗m−1 ⊗∧m M∗ hm(e∗ i ) M ∧ · · · ∧ e∗ m)∗hm ⊗T M∗ e∗ i ) (70) = γ(e j) ⊗T M∗ e∗ i = SM(ui j). Thus, we have det−1 u . v ji = SM(ui j). (71) This with Eq.(67) show that S is restricted to the localization ◦ U M[det−1 M ], and this completes the proof. (cid:3) ◦ U M[det−1 (M) := A with a dual basis {ei, e∗ From now on, we denote by U ◦ Fix a differential module (M, ∂) ∈ Diff M ] the Hopf sub-algebroid of U ◦ stated in Proposition 5.4.4. i }1≤i≤m, and keep the above notations. Given two positive integers k, l, we denote by T (k, l)(M) := M⊗l ⊗A (M∗)⊗k which we consider as a differential module using the tensor product of Eq.(61). We denote by hMi⊗ the full sub-category of Diff A of finite sub- A belongs to hMi⊗ if it is a quotient of quotients differential modules of M. Thus, an object (X, ∂) of Diff the form X = X2/X1, where X1 ⊆ X2 ⊆ ⊕k, lT (k, l)(M) (finite direct sum). Since Diff A is by Lemma 5.2.1 an abelian category, a differential module (X, ∂) belongs to hMi⊗ if and only if it is a sub-object of an object finitely generated by those T (k, l)(M)'s. Let us denote by ωhMi⊗ : hMi⊗ −→ add(A) the restriction of the fibre functor ω : Diff A −→ add(A), and by (A, R(hMi⊗)) its associated commutative Hopf algebroid defined in Proposition 2.2.1, see also Eq. (14) and Remark 2.2.25. Since hMi⊗ is a symmetric rigid monoidal C-linear abelian, locally finite category and ωhMi⊗ is an exact faithful functor, we have that (A, R(hMi⊗)) enjoys similar properties (i)-(iv) of Corollary 5.3.2. Thus, (A, R(hMi⊗)) is a geometrically transitive Hopf algebroid, see Example 2.3.2 and [5, 12]. In particular we have that the fibre functor ωhMi⊗ induces a symmetric monoidal equivalence of categories hMi⊗ ≃ comodR(hMi⊗) to the category of R(hMi⊗)-comodules with finitely generated underlying 5In the notations of subsection 2.4, this is R(hMi⊗) = RA⊗ A(ω). / / / / / FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 29 A-modules. On the other hand, the embedding hMi⊗ ֒→ Diff functors, leads to the canonical map = AU which commutes with the forgetful A φ : (A, R(hMi⊗)) −→ (A, U ◦), (cid:18)p∗ ⊗TP p + JhMi⊗ 7−→ p∗ ⊗TP p + JAU(cid:19) . (72) The following is our main result of this subsection. Theorem 5.4.5. The morphism displayed in equation (72), induces an isomorphism R(hMi⊗) (cid:27) U ◦ (M) of Hopf algebroids, where U ◦ (M) is as in Proposition 5.4.4. Consequently, we get that the equivalence of cate- gories stated in Corollary 5.3.2(ii), rectricts to a symmetric monoidal C-linear equivalence of categories: χ : hMi⊗ −→ comodU◦ (M) , where comodU◦ (M) is the full subcategory of U ◦ (M)-comodules with finitely generated underlying A-modules. Proof. The map φ is clearly an injective morphism of Hopf algebroids. We then identify R(hMi⊗) with its image. Consider now the (A ⊗C A) -- subalgebra U M of U ◦ described in Proposition 5.4.4. Any generic ◦ ◦ ◦ U M ⊆ R(hMi⊗), element in as M is a differential module in the subcategory hMi⊗. The determinant detM of equation (65) and its inverse ⊗T M ei obviously belongs to the subalgebra R(hMi⊗), whence U M of the form e∗ j given by equation (66), are both elements in R(hMi⊗). Therefore, U ◦ (M) = ◦ U M[det−1 M ] ⊆ R(hMi⊗). Conversely, take an object N in hMi⊗ of the form N = M(k, l), for some positive integers, k, l. Consider (N, N) as a right U ◦-comodule, via the isomorphism χ (see subsection 3.2). Denote by C(N) := Xf ∈ HomU◦ (N,U◦ ) Im( f ) the A-subbimodule of coefficients of N. By Corollary 5.3.2(iii), U ◦ so by [15, Example 2.5], we can apply [15, Proposition 2.12]. Therefore, C(N) ⊆ U ◦ N (N) ⊆ (N ⊗A τ)(N ⊗A U ◦ true: C(N) ⊆ U ◦ differential modules, with N as before, we get by [15, Proposition 2.12] that A and AU ◦ are projective modules, and (M), since we have (M) ֒→ U ◦ is the canonical injection. The same inclusion holds (M), if we take N = ⊕(k, l)T (k, l)(M) a finite direct sum. Taking a sequence X1 ⊆ X2 ⊆ N of (M)), where τ : U ◦ C(X1) ⊆ C(X2) ⊆ C(N) ⊆ U ◦ (M), C(X2/X1) ⊆ C(X2) ⊆ C(N) ⊆ U ◦ (M). Henceforth, any object in hMi⊗ is a right U ◦ of the form p∗ ⊗TP p ∈ R(hMi⊗), thus P is an object in hMi⊗. We know that P is a right U ◦ and so p∗(p(0))p(1) ∈ U ◦ p∗ ⊗TP p = p∗(p(0))p(1), which shows that p∗ ⊗TP p ∈ U ◦ (M)-comodule (under the isomorphism χ). Now, take an element (M)-comodule, (M). Up to the canonical isomorphism canU◦ of equation (34), we have the equality (M). Therefore, R(hMi⊗) ⊆ U ◦ (M). In summary, we have an isomorphism (A, R(hMi⊗)) (cid:27) (A, U ◦ (M)) of Hopf algebroids. This leads to the following commutative diagram of symmetric monoidal categories: comodR(hMi⊗) hMi⊗ χ(cid:27) ≃ u❦❦❦❦❦❦❦❦❦❦❦ )❘❘❘❘❘❘❘❘❘ ≃ comodU◦ (M) Diff A = AU χ(cid:27) / comodU◦ which shows the stated equivalence of monoidal categories. (cid:3) M : AlgC → Grpds its associated presheaf of groupoids and let H Let (M, ∂) be a differential module over A and consider the attached Hopf algebroid (A, U ◦ (M)) of Theorem 5.4.5. Denote by H M(C) be its fibre at Spec(C), that is, the character groupoid of (A, U ◦ (M)) is a finitely generated Hopf algebroid (i.e., both A and U ◦ M is an affine algebraic C-groupoid scheme in the fpqc (fid`element plate quasi-compacte) topology. Moreover, the category of comodules comodU◦ is identified with the category of C-representations of the groupoid H (M) are finitely generated C-algebras), thus, H (M)). We know that (A, U ◦ M, see [10]. (M) In comparison with the classical case of differential Galois theory of fields (see [44, Theorem 2.33]), we have by Theorem 5.4.5 the following definition: / / u ✤ ✤ ✤ ✤ ✤ ) / 30 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS Definition 5.4.6. Let (A, U ◦ M(C) is referred to as a differential Galois algebraic groupoid of (A, ∂) for the differential module (M, ∂) (see Remark 5.4.8 below for a brief discussion on the uniqueness of this groupoid). M be as above. Then the character groupoid H (M)) and H M is a transitive groupoid scheme in the sense of [10], see also [5, 12]. Since H Next, we observe that, always for A = C[X], any differential Galois groupoid is transitive as an abstract groupoid. First we observe that, for any differential A-module (M, ∂), U ◦ (M) as a Hopf algebroid over C, satisfies similar conditions (i)-(iv) of Corollary 5.3.2. Thus, U ◦ (M) is a geometrically transitive Hopf algebroid M(C) , ∅, over C, so that H we have by [12, Corollary B] that H M(C) is a transitive groupoid in the set-theoretical sense (i.e., a groupoid in sets with only one connected component, or equivalently, the Cartesian product of the source and the target maps leads to a surjective map). In other words, H M(C) posses only one type of isotropy group. In the sequel we will show that the isotropy type group of H M(C) coincides with the so called differential Galois algebraic group of the differential module (M, ∂) following the terminology of Y. Andr´e [2, §3.2.1.2], (see Example 5.4.9 for more terminologies). Now, we proceed to show that H M(C) is in fact a sub-groupoid of the induced algebraic groupoid of the general linear group GLm(C) along the map A1 C → {∗} (the algebraic set with one point), where m is the rank of M. So let {ei, e∗ X ] of the complex general linear group of order m with indices i, j ∈ {1, · · · , m}. Recall that this a Hopf C-algebra with structure maps: i }1 ≤i≤ m be a dual basis for M. Consider the coordinate ring C[Xi j, det−1 ∆(Xi j) = m Xk=1 Xik ⊗A Xk j, ε(Xi j) = δi j, S (Xi j) = det−1 X Y ji, (73) where δi j is Kronecker's symbol, and where Y ji is the ( j, i)th cofactor of the matrix (Xi j). This is also a differential C-algebra, where the differential is given by ∂(Xi j)1≤,i, j≤m := (X′ i j)1≤,i, j≤m = mat(M) (Xi j)1≤,i, j≤m, (74) and the differential of det−1 tensor product algebra A ⊗C C[Xi j, det−1 X is by extension ∂(det−1 X ] via the map: X ) = −∂(detX)det−2 X . Extending this differential to the we will consider this tensor algebra as a differential algebra and obviously an extension of (A, ∂). ∂(cid:0)a ⊗C Xi j(cid:1) = ∂a ⊗C Xi j + a ⊗C ∂Xi j The base extension of the Hopf algebra C[Xi j, det−1 leads to the commutative Hopf algebroid A ⊗C C[Xi j, det−1 (A ⊗C A)-algebra (A ⊗C A)[Xi j, det−1 X ]. Furthermore, there is a surjective map: X ] via the algebra map C → A (see Example 1.3.5), X ] ⊗C A which is isomorphic to the polynomial φM : (A ⊗ A)[Xi j, det−1 X ] Xi j ✤ / U ◦ (M) / e∗ j ⊗T M ei, (75) of commutative Hopf algebroids. The affine algebraic groupoid attached to the Hopf algebroid (A ⊗ A)[Xi j, det−1 X ]) is geometrically transitive Hopf algebroid over C, see [12, Theorem A]. Its character groupoid is easily computed and it is given by: X ] is also transitive. Thus, (A, A ⊗ A)[Xi j, det−1 G m : A1 C × GLm(C) × A1 C pr3 ι pr1 A1 C, where the source and target are, respectively, pr3 and pr1 the third and first projections, the identity map sends x → ι(x) = (x, Im, x), where Im is the identity of GLm(C). The multiplication and the inverse maps of this groupoid are given by: (x, a, y) . (y, b, z) = (x, ab, z), (x, a, y)−1 = (y, a−1, x). This is clearly a transitive groupoid, and so there is only one type of isotropy group, namely, each of them is isomorphic to the general linear group GLm(C). The groupoid G m is in fact the induced groupoid of the groupoid (with only one object) GLm(C) along the map A1 C → {∗}, see [12, Example 2.4]. Now, using the morphism of equation (75) we claim the following corollary, where the particular claim can be compared with [2, Th´eor`eme 3.2.1.1(ii)]. / / / / / / o o FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 31 Corollary 5.4.7. There is a monomorphism of affine algebraic groupoids H lar, any isotropy group of H M(C) is identified with a closed sub-group of the algebraic group GLm(C). M(C) / G m . In particu- Next, we discuss the uniqueness of the differential Galois groupoid and give a motivating example. Remark 5.4.8. Keep the above notations and assume that there is a flat Hopf algebroid (A, H ) over C with H finitely generated as an algebra, such that there is a symmetric monoidal C-linear equivalence of categories hMi⊗ ≃ comodH to the category of right H -comodules with free of finite rank underly- ing A-modules. Assume further that (A, H ) is geometrically transitive (see Example 2.3.2 and [12], or equivalently that the associated affine C-groupoid scheme H is transitive). Then by Theorem 5.4.5 in con- junction with Deligne's Theorem [5, Theorem 5.2], there is a symmetric monoidal C-linear equivalence of categories ComodH ≃ ComodU◦ of all comodules (i.e., the extension of the previous equivalence to the ind-objects categories). Therefore by [16, Theorem A], there is a two-stage zig-zag of weak equivalences connecting (A, H ) and (A, U ◦ (M)), that is, they are weakly equivalent Hopf algebroids. Therefore their char- acter groupoids should be also weakly equivalent, meaning that H M(C) and H (C) are weakly equivalent affine algebraic groupoids. This means that, in contrast with the classical case of differential Galois theory where the Galois group is unique up to isomorphisms, in this framework the differential Galois groupoid is unique up to weak equivalences. (M) Example 5.4.9. Let (M, ∂) be a differential module whose underlying module M = A.m is a free A-module of rank one, endowed with the differential matrix mat(M) = a ∈ A, that is, ∂(m) = a(X)m. Then the Hopf algebroid U ◦ (M) is isomorphic to the Hopf algebroid (A ⊗C A)[T, T −1] (cid:27) A ⊗C C[T, T −1] ⊗C A, which is induced by the Hopf C-algebra C[T, T −1] (the coordinate algebra of the multiplicative group). (M) is generated as an (A ⊗C A)-algebra by the invertible element detM = m∗ ⊗T M m. Thus U ◦ Remark 5.4.10. As we can realize in Example 5.4.9, the differential ∂(m) = a(X)m does not influence the Hopf structure of U ◦ (M). In other words, taking a different differential (e.g., ∂(m) = b(X)m, for some b , a ∈ A) will leads, up to canonical isomorphism, to the same Hopf algebroid (A ⊗C A)[T, T −1]. Thus, the Hopf structure of U ◦ (M) does not take into account the differential of (M, ∂). This is perhaps why in the more (M)/hs − ti, is general context of [2, 3.2.2.2] the algebraic group attached to the (isotropy) Hopf A-algebra U ◦ referred to as the intrinsic differential Galois group of (A, ∂) for (M, ∂). As we will see in the next subsection, it turns out that in the case we are interested in, that is, the case of linear differential matrix equations over the complex affine line, the differential ∂ of the module M endows this Hopf algebra with a structure of simple differential algebra and convert it into a Picard-Vessiot extension of (A, ∂) for (M, ∂). 5.5. Picard-Vessiot extensions for linear differential matrix equation, after Andr´e. In this subsection we will perform the Picard-Vessiot theory for the particular case of polynomial algebra A = C[X]. In order to do so, we will use our results in combination with the general theory established in [2] for differential noetherian commutative rings with semisimple total ring of fractions. Precisely, we give a complete de- scription, using results from subsection 2.4, of the Picard-Vessiot algebra attached to a differential module (M, ∂) (or to a linear differential matrix equation) and observe that the outcome, in this particular situation, is not far from the classical situation of differential vector spaces over differential fields. We refer to [2, D´efinition 3.4.1.1] for the definition of the Picard-Vessiot extension of a commutative differential algebra which we are going to use in the sequel. Recall that our differential algebra A = C[X] is endowed with the usual differentiation d = ∂/∂X and the A-bimodule of differential forms Ω1 (the notation is that of [2]) is a central one which is a free A-module of rank one (i.e., Ω1 = DerC(A)∗ is the A-linear dual of the module of derivations DerC(A)). Thus, we are in the situation which is referred to in [2] as situation classique. Besides, we will implicitly use the symmetric monoidal equivalence between the category of free A-modules of finite rank with connections (for the previous Ω1) and the category of representations of the Lie-Rinehart algebra (A, DerC(A)), that is, the category AU in the notation of sub-section 5.2, where U is the first Weyl C-algebra of A. Let us fix a differential module (M, ∂) over A of rank m, with dual basis {ei, e∗ i }1≤i≤m and a differential matrix mat(M) = (ai j)1≤i, j≤m. Consider as before the attached C-linear abelian category hMi⊗ and its associated Hopf algebroid (A, U ◦ (M)) over C, constructed as in subsection 5.5. It is well known that this category admits a tensor generator (e.g., the differential module M ⊕ M∗) and have a fibre functor over / 32 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS Spec(A) , ∅, namely, the forgetful functor ω := ωhMi⊗ : hMi⊗ → add(A). Then by applying [10, Corollaire 6.20], the category hMi⊗ admits a fibre functor over the base field C, which we denote by ω′ : hMi⊗ → vectC. On the other hand, since hMi⊗ is a neutral tannakian category over C, we have that the dimension of the C-vector space ω′(M) is the rank of the underlying A-module of the differential module (M, ∂), that is, i }1≤i≤m a dual basis for the vector dimC(cid:0)ω′(M)(cid:1) = m, see [10, Th´eor`eme 7.1]. Let us then denote by {vi, v∗ space ω′(M). As we already mentioned, the Hopf algebroid (A, U ◦ Similar to the classical situation of linear differential matrix equations over a differential field, and as it was shown in [2, Th´eor`eme 3.4.3.1], the existence of ω′ is a fundamental step in building up the Picard- Vessiot extension of (A, ∂) for (M, ∂). Next, we will show, however, that the existence of ω′ comes for free with the information encoded in the Hopf algebroid structure of the pair (A, U ◦ (M)). This will provide us with a more conceptual way in overcoming that fundamental step in the Picard-Vissiot theory for our situation. (M)) is geometrically transitive over C and we know that H (C) , ∅, so we are in position to apply [12, Theorem A]. Therefore, for any point x ∈ A1 C, we can consider the associated isotropy Hopf C-algebra U ◦ (M), x which is by definition (see [12, Definition 5.1 and Lemma 5.2]) the base extension Hopf algebra (C, U ◦ (M), x := C x is C viewed as an A-algebra via the C-algebra map x : A → C. It turns out that [12, Theorem A] implies that the canonical Hopf algebroid extension x : (A, U ◦ (M), x) is a weak equivalence, which means that the induced functor x∗ : ComodU◦ establishes an equivalence of symmetric monoidal categories. Henceforth, the full sub-category of finite-dimensional comodules comodU◦ (i.e., the full sub-category of rigid objects) is equivalent, as symmetric monoidal category, to the category comodU◦ . Thus, by Theorem 5.4.5, we conclude that hMi⊗ is equivalent, as symmetric monoidal category, to comodU◦ . In this way we have a chain of symmetric monoidal C-linear faithful and exact functors: (M)) → (C, U ◦ → ComodU◦ x), where C x ⊗A U ◦ (M) ⊗A (M), x (M) (M), x (M) (M), x C ωx : hMi⊗ χ ⊗-≃ / comodU◦ (M) x∗ ⊗-≃ / comodU◦ (M), x O / vectC, where O is the forgetful functor. In summary, the previous observations assert the following, see also the proof of [10, Corollaire 6.20, pages 163-164]. Corollary 5.5.1. There is a point x ∈ A1 particular, the extended fibre functor ω′ ⊗C A : hMi⊗ → add(A) over A is naturally isomorphic to ω. C such that ω′ = ωx, up to a canonical natural isomorphism. In So far, we have two fibre functors the extension ℘ := ω′ ⊗C A and the restriction of the forgetful functor ω := ωhMi⊗ : hMi⊗ → add(A), we are then in the situation of subsection 2.4. Therefore, by Theorem 2.4.5, we can consider the principal (R A⊗ A(℘))-bibundle (RA⊗A(ω, ℘), α, β), which is lifted to a principal (R A(℘))-bibundle (RA(ω, ℘), ι), over the quotient Hopf A-algebras, see Eq. (39). Combining this with Theorem 5.4.5 and Corollary 5.5.1, we have that the first principal bibundle is isomorphic to a trivial bibundle (U ◦ (M)), namely, to the unit principal bibundle attached to U ◦ (M))-bibundle U (U ◦ (M), see [16]. A⊗ A(ω), R A(ω), R (M), U ◦ Now we consider the quotient algebra P := RA(ω, ℘) = RA⊗A(ω, ℘)/hα − βi with the canonical algebra extension ι : A → P and denote by [ f ] the equivalence class of an element f ∈ RA⊗A(ω, ℘). Define the map ∂ : P −→ P, (cid:16)hp∗ ⊗TP (p ⊗C a)i 7−→ h∂p∗ ⊗TP (p ⊗C a)i + hp∗ ⊗TP (p ⊗C ∂a)i(cid:17), where P ∈ hMi⊗, p∗ ∈ P∗, p ∈ ω′(X) and a ∈ A, and where we have used the differential of the duals as in given by Eq. (63). (76) Proposition 5.5.2. Keep the above notations. Then the pair (P, ∂) is a differential algebra which enjoys the following properties: (i) the map ι : (A, ∂) → (P, ∂) is a morphism of differential algebras; (ii) (P, ∂) is a simple differential algebra which is a Picard-Vessiot extension of (A, ∂) for (M, ∂); (iii) P is isomorphic to the quotient Hopf A-algebra U ◦ (iv) there is a surjective map (M)/hs − ti; ψM : A[Xi j, det−1 X ] Xi j ✤ / P (77) / he∗ i ⊗T M (v j ⊗C 1A)i := fi j, / / / / / FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 33 of differential algebras over (A, ∂). In particular, P (cid:27) A[Xi j, det−1 X ]/I as a differential algebra over A, where I is a maximal differential ideal. Furthermore, P is generated as an A-algebra by the entries of the matrix F = ( fi j)1≤i, j≤m and det−1 F with differential ∂F = mat(M) F (i.e., F is a fundamental matrix of solutions of the linear differential matrix equation (60)). Proof. The first claim and item (i) are easy verifications. Part (ii) follows from [2, Lemme 3.4.2.1(ii)] by combining Proposition 2.4.1 and Theorem 2.4.5(iii) with [2, §3.2.2.1]. As we have seen before, using (cid:27) RA⊗A(ω, ℘) of Hopf Theorem 5.4.5 in conjunction with Corollary 5.5.1, we get an isomorphism U ◦ algebroids. So part (iii) is obtained by going to the quotient Hopf A-algebras, that is, by considering the extended isomorphism of Hopf A-algebras U ◦ (M)/hs − ti (cid:27) RA⊗A(ω, ℘)/hs − ti = P. (M) The first claim of part (iv) is a direct consequence of part (iii) once taking into account the surjective morphism of Hopf algebroids given in Eq. (75). The particular statement is a direct implication of the first claim in this part and item (ii). The last claim of (iv) follows from the following computation: ∂(cid:16)he∗ i ⊗T M (v j ⊗C 1A)i(cid:17) = h∂e∗ i ⊗T M (v j ⊗C 1A)i m ⊗T M (v j ⊗C 1A)i Xk=1 he∗ m i (ek)) − e∗ i (ek))e∗ k k ⊗T M (v j ⊗C 1A)i i (∂ek)(cid:1)e∗ ⊗T M (v j ⊗C 1A)i − ⊗T M (v j ⊗C 1A)i ⊗T M (v j ⊗C 1A)i = k = (63) = = (59) = = i (ek)e∗ k m m Xk=1 h∂e∗ Xk=1 h(cid:0)∂(e∗ Xk=1 h∂(e∗ Xk=1 he∗ Xk, l halke∗ Xk=1 m m m ι(aik) fk j, i (el)e∗ k = 0 − i (∂ek)e∗ i (∂ek)e∗ k ⊗T M (v j ⊗C 1A)i m Xk=1 ι(aik)he∗ k ⊗T M (v j ⊗C 1A)i which shows that ∂F = mat(M) F and this finishes the proof. (cid:3) Now we focus on the structure of the group of differential algebra automorphisms of a Picard-Vessiot extension. As before we consider (P, ∂) a Picard-Vessiot extension of (A, ∂) for (M, ∂) and denote by Aut(A, ∂)(cid:0)(P, ∂)(cid:1) the group of differential A-algebra automorphisms, that is, an element in this group is an algebra automorphism σ : P → P such that σ ◦ ι = ι and ∂ ◦ σ = σ ◦ ∂. In this way, we obtain a functor valued in groups: Aut(A, ∂)(cid:0)(P, ∂)(cid:1) : AlgC −→ Grps, (cid:16)C −→ Aut(A⊗C, ∂⊗C)(cid:0)(P ⊗ C, ∂ ⊗ C)(cid:1)(cid:17), whose fibre at Spec(C) is the starting group Aut(A, ∂)(cid:0)(P, ∂)(cid:1). Our next task is to prove that this in fact is an affine algebraic group. Namely, we show that is isomorphic to each of the isotropy groups of the differential Galois groupoid H M(C). The proof will be done in several steps. First, we know from the definition of the Picard-Vessiot extension that the fibre functor ω′ : hMi⊗ → vectC is naturally isomorphic to the fibre functor κ : hMi⊗ → vectC, which sends any differential module N ∈ hMi⊗ to the finite-dimensional C-vector space ker(∂N⊗A P) the kernel of the differentiation of the differential module N ⊗A P, see [2, Lemme 3.4.2.1] for the proof of this fact. Let us denote by V = κ(M) and by {vi, v∗ i }1≤i≤m a dual basis of this vector space. Notice that, as we have seen before, dimC(V) = rank(M) = m. We will make the following choice for this basis vi = P j e j ⊗A f ji, for every i = 1, · · · , n. The following lemma will be implicitly used in the subsequent one. Lemma 5.5.3. Let k, l be a positive integer. Then, for any objet X ∈ hMi⊗, we have a monomorphism of C- vector spaces ⊕k,lT (k, l)(cid:0)κ(X)(cid:1) ֒→ ⊕k,lT (k, l)(X ⊗A P) (finite direct sums), which is extended to a monomorphism Lk,l T (k, l)(cid:16)κ(X) ⊗ C(cid:17) / Lk,l T (k, l)(cid:16)X ⊗A P ⊗ C(cid:17) / 34 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS of C-modules, for any C-algebra C. Proof. It is immediate by using the fact that κ is a fibre functor. (cid:3) We refer to [11, Chapitre II. §1. no 3] for the definition of the stabilizers sub-functors occurring in the following claim which can be compared with [2, Th´eor`eme 3.5.1.1]. the image of Aut Lemma 5.5.4. There is a monomorphism of functors Aut(A, ∂)(cid:0)(P, ∂)(cid:1) ֒→ GLC(V) (cid:27) GLC(ω′(M)). Moreover, (A, ∂)(cid:0)(P, ∂)(cid:1) is contained in Stab{κ(N)}, the stabilizer of every sub-object N of a finite direct sum ⊕k, lT (k, l)(M) in the category hMi⊗. In particular, the image of Aut⊗(ω′) containts that of Aut(A, ∂)(cid:0)(P, ∂)(cid:1). Proof. Let C be an object in AlgC and γ ∈ Aut(A⊗C, ∂⊗C)(cid:0)(P ⊗ C, ∂ ⊗ C)(cid:1). We set γ := (M ⊗A γ)V : V ⊗ C → V ⊗ C which by definition is a well defined C-linear automorphism. This gives the stated morphism of functors. The fact that this is a monomorphism can be obtained as follows: Take two automorphisms γ and γ′ and = γ′ . Using the basis {vi ⊗ 1C}1≤i≤m of the free C-module V ⊗ C, we get that γ( f ji ⊗ 1C) = assume that γ γ′( f ji ⊗ 1C), for every pair of indices i, j = 1, · · · , m. Thus γ = γ′, as they are (A ⊗ C)-algebra maps. As for the second claim, take C and γ as before and consider a monomorphism N ֒→ ⊕k,lT (k,l)(M) to a finite direct sum in the category hMi⊗. Denote by W = κ(N), so up to a canonical natural isomorphism, we have the following commutative diagram of C-modules: Lk,l T (k,l)(M ⊗A P ⊗ C) i❙❙❙❙❙❙❙❙❙❙❙ ⊕k,l T (k,l)(M⊗A γ) Lk,l T (k,l)(V ⊗ C) / Lk,l T (k,l)(M ⊗A P ⊗ C) ⊕k,l T (k,l)(γ) i❙❙❙❙❙❙❙❙❙❙❙ Lk,l T (k,l)(V ⊗ C) N⊗Aγ N ⊗A P ⊗ C i❚❚❚❚❚❚❚❚❚❚❚❚❚ W ⊗ C N ⊗A P ⊗ C i❚❚❚❚❚❚❚❚❚❚❚❚❚ /❴❴❴❴❴❴❴❴❴❴❴❴❴❴❴❴❴❴❴❴ W ⊗ C This implies that the front rectangle is commutative as well. Therefore, the action given γ stabilizes κ(N) ⊗ C and shows the claim. Lastly, the particular consequence is a direct application of [2, Th´eor`eme 3.2.1.1(iii)] combined with the previous statement. (cid:3) Proposition 5.5.5. Let (P, ∂) be the above Picard-Vessiot extension of (A, ∂) for (M, ∂). Then, we have an isomorphism Aut(A, ∂)(cid:0)(P, ∂)(cid:1) (cid:27) Aut⊗(ω′) of affine group schemes. Furthermore, the attached affine algebraic group Aut(A, ∂)(cid:0)(P, ∂)(cid:1) is isomorphic to each of the isotropy groups of the algebraic groupoid M(C) of Definition 5.4.6. H Proof. The last claim is a direct consequence of the first one in conjunction with Corollary 5.5.1. In order to prove the first statement, we only need to check that (the image of) Aut⊗(ω′) is contained in the image of Aut(A, ∂)(cid:0)(P, ∂)(cid:1) by the monomorphism of Lemma 5.5.4. So take an object C in AlgC and a natural isomorphism ξ ∈ Aut⊗(ω′)(C). Consider then the C-linear automorphism ξM and its associated invertible m × m-matrix (ci j)i, j with coefficients in C. Now, define the following (A ⊗ C)-algebra map: γξ : P ⊗ C −→ P ⊗ C, (cid:16)a fi j ⊗ c 7−→ Xl a fil ⊗ c jlc(cid:17). It is not difficult to check that γξ ∈ Aut(A⊗C, ∂⊗C)(cid:0)(P ⊗ C, ∂ ⊗ C)(cid:1). We still have to check that γξ this follows form the following computations: For each i = 1, · · · , m, we compute = ξM and γξ(cid:16)vi ⊗ 1C(cid:17) = γξ(cid:16)Xj e j ⊗A f ji ⊗ 1C(cid:17) = (M ⊗A γξ)(cid:16)Xj e j ⊗A f ji ⊗ 1C(cid:17) / / / 6 V i 6 V i / / ? O O ? O O / 7 W i ? O O 7 W i ? O O FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 35 e j ⊗A f jl ⊗ c jl = Xl = Xj, l = ξM(cid:16)vi ⊗ 1C(cid:17), which shows the desired equality and finishes the proof. vl ⊗ c jl = Xl (vl ⊗ 1C) c jl (cid:3) this group as the differential Galois isotropy group of (A, ∂) for the differential module (M, ∂). Proposition 5.5.5 suggests a terminology for the algebraic group Aut(A, ∂)(cid:0)(P, ∂)(cid:1), thus, we can refer to 0!, Example 5.5.6. Let (M, ∂) be a differential A-module of rank 2 with differential matrix mat(M) = 0 0 a for some non-zero polynomial a ∈ A. The associated Hopf algebroid U ◦ (M) is generated as an (A ⊗ A)-algebra . Now, by considering the following differential morphisms by the set ne∗ i ⊗T M e jo1≤i, j≤2 A −→ M, (cid:16)1 7−→ e1(cid:17); M −→ A, a2! 7−→ a2(cid:17) (cid:16) a1 (these are morphisms in the category hMi⊗), we show that f11 = f22 = 1P and f21 = 0. Therefore P is generated as an A-algebra by the element f12. Since, we already know that there is a non constant polynomial b ∈ A such that ∂b = a, we have another morphism in the category hMi⊗, namely, the one given by Using, this morphism, we have the following equalities in the algebra U ◦ t : A −→ M, (cid:16)1 7−→ (e2 − e1b)(cid:17). e∗ 1 ⊗T M e2 = e∗ 1 = e∗ 1 = e∗ ⊗T M (e2 − e1b) + e∗ 1 ⊗T M t(1A) + e∗ 1 1t ⊗TA 1A + e∗ ⊗T M e1b 1 ⊗T M e1b (M) ⊗T M e1b = −bl1A ⊗TA 1A + e∗ 1 ⊗T M e1b Passing to the quotient A-algebra P = U ◦ Galois isotropy group is a trivial group in this case. (M)/hs − ti, this implies that f12 = 0. Thus P = A and the differential Remark 5.5.7. Observe that the opposite principal bibundle (RA⊗A(℘, ω), β, α) of the principal bibundle (RA⊗A(ω, ℘), α, β) (see Theorem 2.4.5 and [16, Section 4]) provides us with another Picard-Vessiot exten- sion of (A, ∂), although, as an algebra it will be generated by a fundamental matrix of solutions F satisfying ∂F = −Fmat(M), which is obviously not a matrix of solutions for the system (60), but it is for the differ- ential module (M∗, ∂). This, explain the why of the minus sign in the differentiation of equation (59) and of the use of columns instead of rows in the formulation of the system (60). On the other hand, since P is a simple differential A-algebra and A its self is so, we can show by elementary arguments that P has no non zero divisor (see also [2, Proposition 3.4.4.4]), and so consider its total field of fractions Q(P). In this way, we will end up with a differential field extension C(X) → Q(P). If the field of constants of Q(P) coincides with C, then the extension Q(P)/C(X) can be referred to as the Picard-Vessiot field extension for (M, ∂). Lastly, as we have seen along the previous subsections, the results as well as the examples described therein, present a strong resemblance with the classical case of the differential field C(X). 5.6. Comparison with Malgrange's and Umemura's differential Galois groupoids. In this section we compute the Hopf algebroid structure of the coordinate ring of what is know in the literature as Malgrange's groupoid (or D-groupoid) for some special cases, and illustrate the differences between this groupoid and our approach. Before going on, the following notations are needed. For any positive integer n ∈ N \ {0}, we denote by ǫ0, ǫ1, · · · , ǫn the elements (1, 0, . . . , 0), (0, 1, 0, · · · , 0), · · · , (0, · · · , 0, 1) ∈ Nn+1, respectively. Given such an integer n, we define the following function Nn (k1, · · · , kn) ✤ kn / N / k1 + 2k2 + 3k3 + · · · + nkn := k(k1, ··· , kn). / / 36 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS We denote by k(n) := (kn)−1({n}) the inverse image of {n} by the function kn. Thus elements of k(n) are n-tuples (k1, · · · , kn) ∈ Nn of integers such that n = k1 + 2k2 + · · · + nkn. For instance, we have k(1) = (cid:8)1(cid:9), k(2) = (cid:8)(0, 1); (2, 0)(cid:9), k(3) = (cid:8)(0, 0, 1); (1, 1, 0); (3, 0, 0)(cid:9), Let A = C[X] be as before the one variable polynomial complex algebra and {x0, yn}n∈ N be a set of k(4) = (cid:8)(0, 0, 0, 1); (1, 0, 1, 0); (2, 1, 0, 0); (4, 0, 0, 0); (0, 2, 0, 0)(cid:9), · · · . independent variables over C. Consider the following commutative polynomial C-algebra which we also denote by HC. There are two algebra maps H := C[x0, y0, y1, · · · , yn, · · · , 1 y1 ], so that we can consider the A-bimodule sHt. Moreover, we have the following algebra maps: s : A → H , (cid:16)X 7→ x0 := x(cid:17) and t : A → H , (cid:16)X 7→ y0 := y(cid:17) sHt ∆ / sHt ⊗A sHt ∆(x) = x ⊗A 1, ∆(y) = 1 ⊗A y, ∆(yn) = Xk(k1 , ··· , kn ) ∈ k(n) n! k1! · · · kn!(cid:16)(cid:18) y1 1!(cid:19)k1 (cid:18) y2 2!(cid:19)k2 · · · (cid:18) yn n!(cid:19)kn (cid:17) ⊗A yk1 +k2 +···+kn , for n ≥ 1. Thus, for n = 1, 2, 3, 4, the image by ∆ of the variables yn's reads as follows: ∆(y1) = y1 ⊗A y1, ∆(y2) = y2 ⊗A y1 + y2 1 ⊗A y2, ∆(y3) = y3 ⊗A y1 + 3y1y2 ⊗A y2 + y3 1 ⊗A y3, ∆(y4) = y4 ⊗A y1 + 4y3y1 ⊗A y2 + 6y2y2 1 ⊗A y3 + 3y2 ⊗A y2 + y4 1 ⊗A y4, · · · . (78) (79) It is by construction that ∆ is actually a morphism of A-bimodules. There are other A-bimodule morphisms which are given as follows: sHt S tHs S (x) = y, S (y) = x, S (y1) = y−1 1 , and S (yn) = X (n,0,··· ,0) , k(k1 , ··· , kn ) ∈ k(n) − n! k1! · · · kn! S(cid:0)yk1+k2 +···+kn(cid:1)(cid:16)(cid:18) y1 1!(cid:19)k1−n (cid:18) y2 2!(cid:19)k2 · · · (cid:18) yn n!(cid:19)kn (cid:17), for n ≥ 2, for instance, S (y1) = y−1 1 , S (y2) = −y2y−3 1 , S (y3) = −y3y−4 1 + 3y2 2y−5 1 , S (y4) = −y4y−5 1 + 10y3y2y−6 1 − 15y3 2y−7 1 , · · · . And sHt ε / A ε(x) = X, ε(y) = X, ε(y1) = 1, and ε(yn) = 0, for every n ≥ 2. The C-algebra H is a differential algebra with differential given by: H δ / H δ(x) = 1, δ(y) = y1, δ(yn) = yn+1, for n ≥ 1. Thus, we have δ = ∂ ∂x + ∞ Xi=0 yi+1 ∂ ∂yi . (80) (81) (82) We will consider H as an (A ⊗ A)-algebra using the C-algebra map η := s ⊗ t : A ⊗ A → H given by equation (78). It is by construction that H is via η a faithfully flat (A ⊗ A)-bimodule. The following proposition can be seen as the algebraic counterpart of the "universal" geometric groupoid constructed in [26] and [42]. / / / / / FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 37 Proposition 5.6.1. The pair (A, H ) admits a structure of geometrically transitive commutative Hopf al- gebroid over C, whose comultiplication, counit and antipode, are given by equations (79), (81) and (80), respectively. Furthermore, (H , δ) is a differential extension of (A, ∂) via the source map. Proof. The last claim is clear. As for the proof of the first one, there are different ways to achieve it and the details are left to the reader. One way could be the use of direct computations, using a certain kind of induction, to show that these maps are compatible (i.e., they satisfy the pertinent commutative diagrams for the definition of a commutative Hopf algebroid, see for instance [16, 3.1] or [12, 3.2]), since we already know that the stated maps are obviously C-algebra maps. Another way, is to show that the associated presheaf H of the pair (A, H ) of C-algebras lands in fact in groupoids. To do so, one need to think of the variable y0 = y as if it was a function on x and the monomials yαi i as if they were formal derivative ∂αi y ∂xαi . In this way the compositions in the fibre groupoids are given by the chain rule of derivatives, and the inverse is by thinking this time that x is a function on y and using its derivative, see [29, 42]. Lastly, the fact that (A, H ) is geometrically transitive follows from the fact that η is a faithfully flat extension, as we have mentioned above, and from [12, Theorem A]. (cid:3) ] of H , for each r ∈ N \ {0} and set H0 Remark 5.6.2. Let (A, H ) be the Hopf algebroid of Proposition 5.6.1 and consider the sub C-algebra Hr := C[x, y0, y1, · · · , yr, 1 = C[x, y0] (cid:27) A ⊗ A. It is clear from the y1 definitions that the structure maps given in equations (79), (81) and (80), when restricted to Hr (denoted r), leads to a Hopf algebroid (A, Hr). Thus, the family {(A, Hr)}r ∈N is a family of by sr, tr, εr, ∆ Hopf sub-algebroids of (A, H ) and the canonical inclusions Hr → Hr+1 give an inductive system of Hopf A-algebroids such that r and S In order to clarify the connection with our approach, it is convenient to explain the idea which relates these Hopf algebroids with the Lie algebroids of jet vector bundles of the tangent bundle of A1 C. H = lim −−→ r ∈ N(cid:0)Hr(cid:1). Notice first that each of the algebras Hr is clearly the coordinate ring of a complex variety which we ∞. This means that we have a family of r and H is the coordinate ring of a variety denoted by J∗ denote by J∗ groupoids (Lie groupoids indeed) n(cid:0)J∗ r , A1 C(cid:1)or ∈ N is the Zariski open set of the infinite dimensional analytic space C × CN defined by the sequences (x, yn)n≥0 with y1 , 0 (referred to in the literature as the Lie groupoid of invertible jets). Following the general idea of [26, 3.2], if we consider the Lie algebroid Lie(J∗ r (i.e., the relative tangent via the source (or target) map), then Lie(J∗ r ) is identified with JrT , the jet vector bundle of order r of the tangent bundle T of A1 C. In this way, the module of global sections of the associated coherent sheaf is isomorphic as (right) A-module to the differential operators bimodule Diff r(A) of order r (see [30, Definition 9.67] for the definition of this A-bimodule). The general construction of the Lie algebroid structure of the jet bundles for a given Lie algebroid can be found in [9, 4.1], the explicit formulae for the bracket on the global sections of these jet bundles is detailed in [26, page 478]. It is noteworthy to mention also that these constructions can be performed in a purely algebraic way by using Lie-Rinehart algebras and their jets (or differential operators), in the sense of [30]. r ) of the groupoid J∗ 1 ] is isomorphic to U ◦ In relation with the constructions we have seen so far, we can easily check that he Hopf algebroid = C[x, y0, y±1 H1 (M) for a given differential module (M, ∂) of rank one, as was mentioned in Example 5.4.9. For a higher order, that is, for r ≥ 2, it is possible that the Hopf algebroid Hr can be related to the finite dual of the co-commutative Hopf algebroid VA(Lr) the universal enveloping algebroid of the Lie-Rinehart algebra Lr := Γ(JrT ) the global sections of the Lie algebroid of jets bundle of order r of the bundle T (see Example 1.3.3(2)). In summary it seems that with our approach (and also that of [2]) we only can treat the study of a system of linear differential equation associated to a representation of a given Lie algebroid (or Lie-Rinehart algebra, which in the situation of the present section is Γ(T A1 C) the global sections of the tangent bundle), and we are not able to analyse a system of partial differential equations attached to the jets of order higher than 2. This perhaps can be seen as a disadvantage of our approach, however, with this approach we are able also to study a system of linear differential equation attached to differential module with rank higher than 2 without being forced to change the base algebra (i.e., without being forced to increase the number of the coordinates system), which is the situation adopted in [26, page 493]. To be more precise, let us consider a with "prolongation" the groupoid J∗ ∞ = lim ←−−(cid:0)J∗ r(cid:1). The later 38 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS differential module (M, ∂) over (A, ∂) of rank m. In comparison with [26], the underlying module M can be considered as the global sections module of a trivial vector bundle of the rank m over A1 C, that is, we have M = Γ(E) (or E in the notation of [26]), where E = Cm × C. The differential structure map ∂ on M, is then interpreted as a flat connection ∇ on the vector bundle E (recall here, as in [26], were are in the case of smooth connected spaces). Following [26, 5.3, page 493], (here we take the discrete subset Z = ∅ of A1 C), the corresponding Hopf algebroid whose quotient leads to the description of the differential Galois group of ∇ (or to the associated one dimensional foliation without singularity, as Z = ∅), is the one with the base algebra OCm×C, since the Lie groupoid of invertible jets has the space Cm × C as the space of objects. Such a Hopf algebroid is given by the following differential polynomial algebra HCm+1 := Chx0, x1, x2, · · · , xm, y0, y1, · · · , ym, yiα, det(yiǫ j)−1ii, j=1,··· , m; α ∈ Nm+1\{0,ǫ0,··· ,ǫm} , (83) viewed as an Hopf algebroid with base algebra C[X0, X1, · · · , Xm], see Example 5.6.8 below where we illustrated the construction of this Hopf algebroid for m = 1. Therefore, if we want to employ the method of [26] for studying a differential module of rank m ≥ 2 over A = C[X0], then we are forced to use the Hopf algebroid of equation (83) as a Hopf algebroid over C[X0, X1, · · · , Xm]. This shows that our approach is somehow different from that adopted in [26], since all our Hopf algebras and Hopf algebroids have for the base algebra the polynomial algebra with no more than one variable, i.e., the algebra A. Next, we introduce the notion of Malgrange Hopf algebroid which is the coordinate ring of what is known in the literature as the D-groupoid, or Malgrange groupoid. We will treat only the case of Hopf algebroids of over the polynomial algebra A, that is, Hopf quotients of the above Hopf algebroid HC, since this is the case we are interested in this section. Recall first that a Hopf ideal of a given Hopf algebroid (A, H ) is an ideal I of H such that the pair of algebras (A, H /I) admits a unique structure of Hopf algebroid for which the morphism (idA, π) : (A, H ) → (A, H /I) becomes a universal morphism of Hopf algebroids. That is, any morphism (ida, φ) : (A, H ) → (A, K ) of Hopf algebroids such that I ⊆ ker(φ), factors through (idA, π) by a morphism of Hopf algebroids. This in fact is a naive definition which only consider subgroupoids of H with same set of objects H 0. The strict definition is a pair of ideals (I0, I1) of the pair of algebras (A, H ) satisfying pertinent conditions; since we will not use this general notion, we will not go on to the details. Here is the promised definition: Definition 5.6.3 ([26, 42]). Given (A, H ) as in Proposition 5.6.1 with H the associated presheaf of groupoids. A Malgrange Hopf A-algebroid over C is a quotient Hopf algebroid of the form (A, H /I), where I is a differential Hopf ideal, that is, a Hopf ideal I with δ(I) ⊆ I, where δ is the derivation of equation (82). Obviously the pair (A, H /I) defines a presheaf of groupoids which we denote by KI, and for every commutative C-algebra C we have that KI(C) is a sub-groupoid of H (C). Thus, KI is a sub- groupoid of H . An alternative terminology and the one which can be found in the literature, is to say that KI is a D-groupoid (or Malgrange groupoid over A1 C) defined by the Hopf ideal I (see also [18], where the notion of Hopf algebroid or that of presheaf of groupoids were not specified). Remark 5.6.4. As we have seen before, it is by construction that the Hopf algebroid (A, H ) of Proposition 5.6.1 is geometrically transitive in the sense of [12], see also Example 2.3.2. Therefore, the associated presheaf of groupoids H is a transitive groupoid scheme in the sense of [10]. It is noteworthy to mention that, taking a Malgrange groupoid KI associated to a differential Hopf ideal I, even if H defines a tran- sitive groupoid scheme, it is not necessarily that this the case for KI. That is, the (A ⊗ A)-module H /I does not necessarily have to be faithfully flat. Henceforth, Malgrange Hopf algebroids fail in general to be geometrically transitive. In the subsequent we give examples of Malgrange Hopf algebroids and describe, in some cases, Umemura's method [42] dealing with the study of certain partial differential equation. Example 5.6.5 ([42]). Let I be the ideal of H generated by the set {y1 − 1, yn}n≥2. Since δk(y1) = yk+1, for all k ∈ N∗, I is a differential ideal of H . Let us check that is also a Hopf ideal. We know that ∆(y1 − 1) = (y1 − 1) ⊗A 1 + 1 ⊗A (y1 − 1) ∈ H ⊗A I + I ⊗A H , and by using the formula (79) with n ≥ 2, we conclude that ∆(I) ∈ H ⊗A I + I ⊗A H . On the other hand, we have ε(y1 − 1) = 0 and ε(yn) = 0, whence ε(I) = 0. Concerning the image of I by the antipode, we have FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 39 1 − 1 = y−1 that S (y1 − 1) = y−1 1 (1 − y1) ∈ I, and by induction employing equation (80), we get S (I) ⊆ I. Therefore, (A, H /I) is a Malgrange Hopf A-algebroid which can be shown to be isomorphic to the Hopf algebroid (A, A ⊗ A). The associated groupoid is just the fibrewise groupoid of pairs ([12, Example 2.2]) and not the action groupoid ([12, Example 2.1]) of the additive on it self, as was claimed in [42, page 446], since obviously none of the images of x, y is a primitive element in H /I. Example 5.6.6. ([26, 3.5]) Consider A with the derivation X∂/∂X. As we have seen in Remark 5.1.1, in this case the category Diff A fails to be an abelian category. Thus, the classical Tannakian theory fails too, when considering the Lie-Rinehart algebra L = A.(X∂/∂X). Following [26], this is due perhaps to the fact that associated D-groupoid is not a reduced one. Let us check this by computing its coordinate ring, and compare Umemura's [42] method with this case. Consider then the Hopf algebroid (A, H ) of Proposition 5.6.1 with differential as in equation (82), and take the differential ideal I generated by(cid:8)δk(xy − y1)(cid:9)k≥0 (as we have mentioned before we thing of y1 as if it was the first derivative of y with respect to the variable x). Let us check first that I is a Hopf ideal. So, by the formula δn+1(xy1 − y) = nyn+1 + xyn+2, n ≥ 0 we have that whence ε(I) = 0. As for the image of I by the comuplitplication, we have that ε(xy1 − y) = 0, and ε(cid:0)δn(xy1 − y)(cid:1) = 0, for any n; ≥ 1 ∆(xy1 − y) = (xy1 − y) ⊗A y1 + 1 ⊗A (xy1 − y) ∈ I ⊗A H + H ⊗A I, ∆(δ(xy1 −y)) = ∆(xy2) = xy2 ⊗A y1 + xy2 1 ⊗A y2 = xy2 ⊗A y1 +(xy1 −y)y1 ⊗A y2 +y1 ⊗A xy2 ∈ I⊗A H +H ⊗A I, and ∆(cid:16)δ2(xy1 − y)(cid:17) = ∆(y2 + xy3) = y2 ⊗A y1 + y2 = (y2 + xy3) ⊗A y1 + (xy1 − y) ⊗A y3 + y2 1 ⊗A y2 + xy3 1 ⊗A y3 + 3xy1y2 ⊗A y2 + xy3 ⊗A y1 1 ⊗A (xy3 + y2) + 3xy1y2 ⊗A y2 ∈ I ⊗A H + H ⊗A I Seeking for a general formula of ∆(cid:16)δn(xy1 − y)(cid:17) and using induction shows that ∆(I) ⊆ I ⊗A H + H ⊗A I. Concerning the image by the antipode, we have 1 ∈ I, 1 ) = (xy1 − y)y−3 1 (cid:17)y−4 2y−1 1 1 y2 − xy1y2 ∈ I and S (xy1 − y) = yy−1 1 − x = −(xy1 − y)y−1 S(cid:16)δ(xy1 − y)(cid:17) = S(cid:16)xy2(cid:17) = −y(y2y−3 S(cid:16)δ2(xy1−y)(cid:17) = S(cid:16)y2 +xy3(cid:17) = −(y2+xy3)y−3 and one can also show, using a kind of induction, that S(cid:16)δn(xy1 − y)(cid:17) ∈ I, for any n ≥ 3. Thus I is a differential Hopf ideal and so (A, H /I) is a Malgrange Hopf algebroid. Since xy2 ∈ I and none of the elements x, y2 belong to I, the algebra H /I is not reduced. +(xy1−y)(cid:16)y3−y2 +3xy2(cid:0)y2y−4 1 (cid:1) ∈ I, 1 Now, we come back to Umemura's method as promised. Following [42], if we use the universal Taylor morphism, ι : C[x] −→ C[x][[Z]], a 7−→ Xn ≥ 0 n ∂ (a) n! Zn, where the differential is ∂ = x ∂ ∂x , then we can see that the image of x satisfies the equality ι(x) = x Exp(Z), and so x ∂ι(x) = ι(x). Moreover, ι can be extended to an algebra map ι : H → A[[Z]] by sending y 7→ ι(x), ∂x x 7→ x (i.e., the power series (x, 0, · · · , 0)) and ι(yn) = ∂nι(x) , for n ≥ 1. Thus, ι(y1) = Exp(Z) and ι(yn) = 0, ∂xn for n ≥ 2. Henceforth, we have a morphism κ : H /I → A[[Z]]6 of (A ⊗ A)-algebras, which is, of course, not injective nor a differential morphism. Thus there is no hope in obtaining an analogue result to the one claimed in [42, Proposition 6.1] for this case, and so Umemura's method only works for some specific 6 Note that the I-adic completion of this algebra map leads to a morphism of complete Hopf algebroids over A, where A is considered as a discrete topological ring, see [17]. 40 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS cases. This is due perhaps to the fact that the morphism κ does not take into account the whole system of equations (n − 1)y(n) − xy(n+1) = 0, for n ≥ 0, (84) in the sense that the images of yn by κ are zero up to degree 2. There are observations which should be highlighted here with respect to what we have seen so far. The previous power series algebra can in fact be seen as the convolution algebra of the co-commutative Hopf algebroid U, that is, we have U ∗ (cid:27) A[[Z]] where U is the universal enveloping algebroid of the Lie-Rinehart algebra A. ∂ and where the universal Taylor map ι is nothing but the target map for the (A ⊗ A)-algebra U ∗ (this is in fact a topological commutative Hopf algebroid, see [17] for more details). In this way, it is not clear, at least to us, if the commutative Hopf algebroid H /I can be identified with the finite dual of a certain co-commutative Hopf algebroid. According to what was explained in Remark 5.6.2, an answer to this question can be perhaps performed by using not only the finite dual of the universal enveloping algebroid of a given Lie-Rinehart algebra, but also the finite duals of the universal enveloping algebroids of the attached jets Lie-Rinehart algebras of any order. In this direction, it is perhaps possible that one could also establish a certain Picard-Vessiot theory for the system of equation (84). On the other hand, if we allow solutions of the system (84) to be with coefficients in the localized algebra C[X±1], then the system can be linearized and one can try to employ directly the theory we have developed hereby in order to solve this new linear system. In the following remark we illustrate the difference between the approach to the Galois groupoid of Definition 5.4.6 and the one introduced by B. Malgrange in [26] (see also [42]). Remark 5.6.7. Let place ourselves in the context of [26, 5.3] by taking the analytic smooth connected curve X to be the affine complex line A1 C and set as before A = C[X] its coordinate ring. This is indeed the case we are treating in this section. Now given a Malgrange groupoid KI, as in Definition 5.6.3, defined by a differential Hopf ideal I, then, following [25, 26], the attached Galois groupoid has X as the space of objects and an arrow is a germ of local diffeomorphism g : (X, p) → (X, g(p)) from an open neighbourhood of p in X to an open neighbourhood of g(p) in X, such that g is a solution of the differential equations generating the ideal I. The structure groupoid is the one induced from the groupoid Aut(X) of germs of local diffeomorphisms (´etale Lie groupoid in fact, see [27, Example 2.5(4)] for details). Therefore, for a differential module (M, ∂) of rank one over (A, ∂), the Galois groupoid H(M)(C) of Definition 5.4.6, have the same set of objects as the Galois groupoid attached to a Malgrange groupoid KI, for a given differential Hopf ideal I (notice that there are case when this two Galois groupoids coincide, for instance, when I is the differential Hopf ideal hynin≥2). For a differential module with rank m ≥ 2 over A, the situation is totally different. Precisely, let us consider such a differential module and consider as in Remark 5.6.2 the associated locally free vector bundle (E, π) of constant rank m over X. Following the procedure of [26, 5.3], in order to define the Galois groupoid Gal(F) (notation of [26, 5.2]) of the foliation F attached to (E, π) (and given by the connection ∇ corresponding to the differential ∂ of M), one has to consider a Malgrange Hopf algebroid extracted form a certain differential Hopf ideal IF of the Hopf algebroid HCm+1 of equation (83). The existence and the construction of this differential Hopf ideal, or equivalently that of the Malgrange groupoid KIF is guaranteed by [26, Th´eor`eme 4.5.1]. Indeed IF is the largest differential Hopf ideal for which the germs of sections of F are solutions. In this direction, the Galois groupoid H(M)(C) still having A1 C as set of objects, while the Galois groupoid Gal(F) has the space Am+1 (the analytic variety underlying the bundle (E, π), see [26, page 493]) as a set of objects. This at first glance shows that H(M)(C) is not isomorphic to Gal(F), it is not clear, however, at least to us, if they are weakly equivalent in the sense of Remark 5.4.8. C As we have seen in Remark 5.6.7, perfunctorily our differential Galois groupoid attached to a differential A-module is different from the one introduced by Malgrange in [26]. It seems that the two approaches are also far from begin similar. Specifically, let (M, ∂) be a differential A-module of rank 2. If we want to study the system of linear differential equations attached to (M, ∂) by using Malgrange's Hopf algebroids, then the 'universal' Hopf algebroid described in Proposition 5.6.1 is useless, since no isotropy group of the attached presheaf of groupoids will provide us with an algebraic closed subgroup of GL2(C). Besides, even for differential modules of rank one, like for instance the one considered in Example 5.4.9, it is not clear how to construct a differential Hopf ideal of this Hopf algebroid form this differential module (apriori the FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 41 defining ideal should be the differential ideal generated by y1 − xy which is clearly not a Hopf ideal, nor the one generated by z1 − xz, where z = y − x and z1 = z′). On the other hand, the 'universal' Hopf algebroid of Proposition 5.6.1 (or the one defined by any reduced algebraic variety as the coordinate ring of invertible jets [26, 42], like the one in Example 5.6.8), leads to partial differential equations defined by given differential Hopf ideals. This means that this approach and its methods of studying these equations goes somehow in the contrary direction of what is traditionally done in the differential Galois theory of differential fields (unless perhaps one is interested in founding partial differential equations attached to some dynamical system and providing its groupoid of "symmetries" [42, 29]). More precisely, usually one considers, in the first step, a system of (algebraic) differential equations as an initial data and look for a representation over a certain Lie algebroid (i.e., Lie-Rinehart algebra) which defines such system of equations. In the second step, one seek for a Picard-Vessiot extension of the differential base algebra as a formal solution space of the starting system of equations. Furthermore, if we want to use Malgrange's Hopf algebroids, then it is not clear from [26] or [43], how to construct a Picard-Vessiot extension, in the sense of [2], associated to the system of equations defining a differential Hopf ideal. In summary, our (algebraic) approach to differential Galois theory over differential rings seems to be different form the (geometric) one adopted in [26, 43], although, there are some similar aspects between the two approaches. In the following example, we give a map of bialgebroids involving the underlying bialgebroid of the Hopf algebroid of Proposition 5.6.1. Besides, we try to mimic the method employed in [42, Example 5.1], for the case affine complex plan (by taking a certain sub Lie-Rinehart algebra of the Lie algebra of derivations of the coordinate ring), and arrive to the conclusion that this method doesn't behave well in this case, which in some sense bears out the above explanations. Example 5.6.8. In this example we compute the structure maps of the Hopf algebroid describe in Proposi- tion 5.6.1, but for two variables instead of one. That is, a Hopf algebroid over C[x0, x1] the coordinate ring of the affine plan. The quotients by differential ideals of this 'universal' Hopf algebroid lead to Malgrange Hopf algebroids which allows the study of systems of certain algebraic partial differential equations with two variables. Let {x0, x1, yi α}i=0,1; α∈N2 be a set of independent variables over C. To simplify the notations we will adopt the following one: yi 0 := yi, for i = 0, 1; yi ǫ j := yi j, for i, j = 0, 1; and where ǫ0 = (1, 0), ǫ1 = (0, 1); yi α := yiα1α2 , where α = (α1, α2) ∈ N2 \ {0, ǫ0, ǫ1}. Then we have a family of variables: x0, x1, y0, y1; y00, y10, y01, y11; yi11, yi02, yi20, yi12, yi21, yi03, yi30, · · · · · · , i = 0, 1.  As before, we think of yi as functions of two variables x0, x1, and yi j their first partial derivatives ∂yi ∂x j , and yiα1α2 are their higher derivatives of the function (y0, y1) in two variables x0, x1. We consider the following commutative C-algebra . In this way the determinant det(yi j) can be viewed as the Jacobian ∂x ∂αyi α1 0 ∂x α2 1 K := HC2 = Chx0, x1, y0, y1, yi j, det(yi j)−1, yiα1α2ii, j=0,1, α ∈N2, α≥2 This is a (B ⊗ B)-algebra where B = C[x0, x1] is the polynomial algebra with two variables. Clearly it is a free (B ⊗ B)-module. The source and the target are s(xi) = xi and t(xi) = yi, for every i = 0, 1. The algebra K is a partial differential algebra extension of (B, ∂/∂x0, ∂/∂x1) via the source map. The partial derivations of K are given as in [26, page 470] by for any polynomial function P ∈ K . ∂iP = ∂P ∂xi +X ∂P ∂y jα y j(α+ǫi), i = 0, 1, 42 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS Next, we illustrate the Hopf algebroid structure of the pair (B, K ). The counit is the map which sends: ε(xi) = ε(yi) = xi, for all i = 0, 1; ε(yi j) = δi j (Kronecker symbol ), for all i, j = 0, 1; ε(yiα) = 0, for all i = 0, 1 and α ∈ N \ {0, ǫ0, ǫ1}. The comultiplication and the antipode are given as follows: ∆(yi j) = Xk yik ⊗A yk j, for i, j = 0, 1. While for any i = 0, 1, we have ∆(yi11) = yi20 ⊗B y00y01 + yi11 ⊗B y00y11 + yi11 ⊗B y10y01 + yi02 ⊗B y10y11 + yi0 ⊗B y011 + yi1 ⊗B y111, ∆(yi02) = yi20 ⊗B y2 01 ∆(yi20) = yi20 ⊗B y2 00 + yi11 ⊗B y11y01 + yi0 ⊗B y002 + yi11 ⊗B y01y11 + yi02 ⊗B y2 11 + yi11 ⊗B y10y00 + yi0 ⊗B y020 + yi11 ⊗B y00y10 + yi02 ⊗B y2 10 + yi1 ⊗B y102, + yi1 ⊗B y120, the antipode is given by: S (xi) = yi, for all i, j = 0, 1; S (yi j) = det(yi j)−1(−1)i+ jy ji, for all i, j = 0, 1, and for α ∈ N2 such that α = 2 and for any i = 0, 1, we have S (yi11) det(yi j)2 = (cid:0)y00y11 + y10y01(cid:1)(cid:16)S (yi0)y011 + S (yi1)y111(cid:17) − y10y00(cid:16)S (yi1)y102 + S (yi0)y002(cid:17) For higher degrees, that is, for α ∈ N with α ≥ 3, one can use the higher partial derivative Chain rules for both comultiplication and the antipode. We have that (B, K ) is a geometrically transitive Hopf algebroid. Assume we are given on B the following derivation δ(x0) = 1 and δ(x1) = x0 x1 (in other words we are considering the sub Lie-Rinehart algebra of DerC(B) generated by ∂ ). As we will observe ∂x0 below, this Lie-Rinehart algebra can be used to seek the solutions of the linear differential equation y′ = x0y over the polynomial algebra C[x0]. and x0 x1 ∂ ∂x1 From the definition of δ, we have that δn(x1) = pn(x0)x1, for any n ∈ N, where the sequence polynomials {pn(x0)}n≥0 in x0, satisfies the following non homogeneous recurrence: pn+1(x0) = x0 pn(x0) + pn(x0)′, p0(x0) = 1 n ≥ 1,  (85) : B → B[[Z]] sending b 7→ Following the method of [42] and by using the universal Taylor map i δn(b) n! Zn, we can conclude the following partial differential equations: Pn≥0 ∂i(x0) ∂x0 = 1, ∂i(x0) ∂x1 = 0, x1 ∂i(x1) ∂x1 = i(x1). (86) On the other hand, extending δ to the B[[Z]] by putting δ(Z) = 0 and considering the derivation ∂• = 1 ∂Z(cid:1) on B[[Z]], we have that i : (B, δ) → (B[[Z]], ∂•) is a morphism of differential algebras, so that the equation ∂• i(x1) = x0i(x1) holds true. Therefore, B[[Z]] contains a solution of the above linear differential equation (up to a certain scalar). 2(cid:0)δ + ∂ Now, if we want to use equations (86) with the purpose of guessing a differential Hopf ideal and con- struct a Malgrange Hopf algebroid according to Definition 5.6.3 (this is the method adopted, for instance in the example expounded in [42, §5]), then the adequate set of generators seems to be {y00 − 1, y01, x1y11 − y1} and their higher partial derivatives: ∂k i (x1y11 − y1), k, l, m ≥ 1, i = 0, 1. Unfortunately, i (y00 − 1), ∂l i(y01), ∂m − y11y01(cid:16)S (yi0)y020 + S (yi1)y120(cid:17), S (yi02) det(yi j)2 = 2y00y01(cid:16)S (yi0)y011 + S (yi1)y111(cid:17) − y2 01(cid:16)S (yi0)y020 + S (yi1)y120(cid:17), S (yi20) det(yi j)2 = 2y11y10(cid:16)S (yi0)y011 + S (yi1)y111(cid:17) − y2 − y2 00(cid:16)S (yi1)y102 + S (yi0)y002(cid:17) 10(cid:16)S (yi1)y102 + S (yi0)y002(cid:17) − y2 11(cid:16)S (yi0)y020 + S (yi1)y120(cid:17). FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 43 this is not the case, since the differential ideal J generated by this set, only leads to a structure of bial- gebroid on the quotient B-bimodule K /J. This is due to the fact that J is not stable under the antipode of K , as one can check from the equality S (y01) = det(yi j)−1y10. Nevertheless, we still have a morphism K /J → B[[Z]] of (B ⊗ B)-algebras, and one can also construct a morphism H → K /J of bialgebroids, sending y1 7→ y11 + J and yn 7→ y1(n−1)1 + J for n ≥ 2, where H is the Hopf algebroid of Example 5.6.5. It is noteworthy to mention that it is also possible that K /J becomes a Hopf algebroid after localizing on certain denominator set. In case of dimension two, that is, for a differential module (M, ∂) over A = C[x0] of rank 2 with matrix M := mat(M) = (ai j)1≤i, j≤2 ∈ M2(A), the previous method can be described as follows. Assume this time that δ is a derivation of the polynomial complex algebra C = C[x0, x1, x2], sending, x0 7→ 1, xi 7→ P1≤ j≤2 ai jx j, for i = 1, 2. In this case the Hopf algebroid to be considered is the pair (C, W), where W is the commutative polynomial algebra W := HC2 = C[x0, x1, x2, y0, y1, y2, yi j, yiα, det(yi j)−1]i, j=0,1,2, α ∈ N3\{ǫ0,ǫ1,ǫ2} whose structure maps are given as above, by thinking of the yi's as if they were certain functions on the variables x0, x1, x2. The equivalent recurrence of the recurrence given in equation (85), is the following recurrence on 2 × 2-matrices with entries in C[x0] + Mn M, n ≥ 1, (87) where the notation M′ ploying the universal Taylor map i : C → C[[Z]], this means that, we have n stand for the matrix whose entries are the derivations of the entries of Mn. Em- ∂i(xi) ∂x j = Xn≥0 a(n) i j n! Zn, for any i, j ∈ {1, 2}, where, for any n ≥ 0, the coefficients a(n) (87), we end up with the subsequent system of partial differential equations i j are the entries of the matrix Mn. Using the recurrence system of Mn+1 = M′ n M0 = I2,  The corresponding differential ideal J is the one generated by the set: ∂i(x0) ∂x1 ∂i(x2) ∂x0 ∂i(x0) ∂x2 = 0, = 0, = 0, (88) ∂i(x0) ∂x0 ∂i(x1) ∂x0 = 1, = 0, k=1 xk ∂i(xi) ∂xk for i = 1, 2. i(xi) = P2 ny00 − 1, y01, y02, y10, y20, yi − x1yi1 − x2yi2oi=1,2 . The quotient leads here also to a bialgebroid over the algebra C and it is not clear, up to now, how to relate this bialgebroid with the Hopf algebroid U ◦ (M) of subsection 5.5 attached to (M, ∂). Acknowledgement. The authors would like to thank the referee for bring our attention to the references [2] and [26] which leads us to improve considerably the earlier exposition of this paper. References [1] J. Y. Abuhlail, J. G ´omez-Torrecillas and R. Wisbauer, Dual coalgebras of algebras over commutative rings. J. Pure Appl. Alg. 153, 107 -- 120 (2000). [2] Y. Andr´e, Diff´erentielles non commutatives et th´eorie de Galois diff´erentielle ou aux diff´erences. Ann. Sci. ´Ecole Norm. Sup. 4e s´erie, 35 (5), 685 -- 739 (2001). [3] J. E. Bjork, Rings of differential operators. North-Holland Publishing Company Amsterdam, 1979. [4] G. Bohm, Hopf algebroids, Handbook of algebra, Vol. 6, North-Holland, Amsterdam, 2009, pp. 173 -- 236. [5] A. Brugui`eres, Th´eorie tannakienne non commutative. Commun. in Algebra 22, 5817 -- 5860 (1994). [6] T. Brzezi´nski and R. Wisbauer, Corings and Comodules. Cambridge University Press, Cambridge, 2003. [7] S. U. Chase, Group scheme actions by inner automorphisms. Commun. Algebra, 4(5), 403 -- 434 (1976). [8] S. Chemla, F. Gavarini and N. Kowalzig, Duality Features of Left Hopf Algebroids. Algebr. Represent. Theor. 19, 913 -- 941 (2016). [9] M. Crainic and R. L. Fernandes, Exotic Characteristic Classes of Lie Algebroids. http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.145.6167. 44 LAIACHI EL KAOUTIT AND JOS ´E G ´OMEZ-TORRECILLAS [10] P. Deligne, Cat´egories tannakiennes. In The Grothendieck Festschrift (P. Cartier et al., eds), Progr. math., 87, vol. II, Birkhauser, Boston, MA. 1990, pp. 111 -- 195. [11] M. Demazure and P. Gabriel, Groupes alg´ebriques. Tome I: G´eom´etrie alg´ebrique, g´en´eralit´es, groupes commutatifs, Masson & Cie, ´Editeur, Paris; North-Holland Publishing Co., Amsterdam, 1970, Avec un appendice Corps de classes local par Michiel Hazewinkel. [12] L. El Kaoutit, On geometrically transitive Hopf algebroids. J. Pure Appl. Algebra (2017), https://doi.org/10.1016/j.jpaa.2017.12.019 [13] L. El Kaoutit and J. G ´omez-Torrecillas, Comatrix corings: Galois corings, descent theory, and a structure theorem for cosemisimple corings. Math. Z., 244, 887 -- 906 (2003). [14] L. El Kaoutit and J. G ´omez-Torrecillas, Infinite comatrix corings. Int. Math. Res. Notices, 39, 2017 -- 2037 (2004). [15] L. El Kaoutit, J. G ´omez-Torrecillas and F.J. Lobillo, Semisimple corings. Alg. Collq., 11:4, 427 -- 442 (2004). [16] L. El Kaoutit and N. Kowalzig, Morita theory for Hopf algebroids, principal bibundles, and weak equivalences. Documenta Math., 22, 551 -- 609 (2017). [17] L. El Kaoutit, P. Saracco, Topological tensor product of bimodules, complete Hopf algebroids and convolution alge- bras. To appear in Commun. Contemp. Math. (2017), https://doi.org/10.1142/S0219199718500153. [18] G. Gasale, D-enveloppe d'un diff´eomorphisme de (C, 0), Annales de la Facult´e des Sciences Math´ematiques de Toulouse, Tome XIII, no. 4, 515 -- 538 (2004). [19] J. G ´omez-Torrecillas and J. Vercruysse, Comatrix corings and Galois comodules over firm rings, Algebras Repr. Th. 10 , 271-306 (2007). [20] Ph`ung H o Hai, Tannaka-Krein duality for Hopf algebroids. Israel J. Math. 167 , 193 -- 225 (2008). [21] J. Huebschmann, Poisson cohomology and quantization. J. Reine Angew. Math. 408, 57 -- 113 (1990). [22] L. Kadison and K. Szlach´ayi, Bialgebroid actions on depth two extensions and duality. Adv.Math.179 (1), 75 -- 121 (2003). [23] N. Kowalzig, Hopf algebroids and their cyclic theory, Ph. D. thesis, Universiteit Utrecht and Universiteit van Ams- terdam, 2009. [24] K. C. H. Mackenzie, General theory of Lie groupoids and Lie algebroids. London Math. Soc. Lecture Note Series 213. Cambridge University Press, 2005. [25] B. Malgrange, La vari´et´e caract´eristique d'un syst`eme diff´erentiel analytique. Ann. Inst. Fourier, tome 50, no. 2, 491 -- 518 (2000). [26] B. Malgrange, Le groupoıde de Galois d'un feuilletage. In Essay on geometry and related topics, Vol. 1,2. Monogr. Enseign. Math., vol. 38. Enseignement Math., Geneva, 2001, p. 465 -- 501. [27] I. Moerdijk and J. Mrcun, Lie groupoids, sheaves and cohomology, Poisson geometry, deformation quantisation and group representations, London Math. Soc. Lecture Note Ser., vol. 323, Cambridge Univ. Press, Cambridge, 2005, pp. 145 -- 272. [28] I. Moerdijk and J. Mrcun, On the universal enveloping algebra of a Lie algebroid. Proc. Amer. Math. Soc. 138, no. 9, 3135 -- 3145 (2010). [29] S. Morikawa and H. Umemura, On a general difference Galois theory II. Ann. Inst. Fourier, Grenoble, 59, no. 7, 2733 -- 2771, (2009). [30] J. Nestruev, Smooth manifolds and observables. Joint work of A. M. Astashov, A. B. Bocharov, S. V. Duzhin, A. B. Sossinsky, A. M. Vinogradov and M. M. Vinogradov. Translated from the 2000 Russian edition by Sossinsky, I. S. Krasil'schik and Duzhin. Graduate Texts in Mathematics, 220. Springer-Verlag, New York, 2003. [31] W. D. Nichols, The Konstant structure theorems for K/k-Hopf algebras. J. Algebra 97, 313 -- 328 (1985). [32] D. C. Ravenel, Complex Cobordism and Stable Homotopy Groups of Spheres. Pure and Applied Mathematics Series, Academic Press, San Diego, 1986. [33] G. Rinehart, Differential forms on general commutative algebras, Trans. Amer. Math. Soc. 108, 195 -- 222 (1963). [34] D. Rumynin, Duality for Hopf algebroids. J. Algebra, 223, 237 -- 255 (2000). [35] N. Saavedra Rivano, Cat´egories tannakiennes, Lect. Notes in Math., Vol. 256, Springer-Verlag,1972. [36] P. Schauenburg, Bialgebras over noncommutative rings and a structure theorem for Hopf bimodules, Appl. Categ. Structures 6 no. 2, 193 -- 222 (1998). [37] P. Schauenburg, Duals and doubles of quantum groupoids (×R-Hopf algebras), New trends in Hopf algebra theory (La Falda, 1999), Contemp. Math., vol. 267, Amer. Math. Soc., Providence, RI, 2000. pp. 273 -- 299. [38] B. Stenstrom, Rings of Quotients, Springer, Berlin, 1975. [39] M. Sweedler, Groups of simple algebras. Inst. Hautes ´Etudes Sci. Publ. Math. 44, 79 -- 189 (1974). [40] M. Sweedler, The predual theorem to the Jacobson-Bourbaki theorem. Trans. Amer. Math. Soc. 213, 391 -- 406 (1975). [41] M. Takeuchi, Groups of algebras over A ⊗ A, J. Math. Soc. Japan 29, no. 3, 459 -- 492 (1977). [42] H. Umemura, On the definition of the Galois groupoid. Ast´erisque, 323, 441 -- 452 (2009). [43] H. Umemura, Differential Galois theory of infinite dimention. Nagoya Math. J. 144, 59 -- 134 (1996). [44] M. van der Put and M. F. Singer, Galois theory of linear differential equations. A Series of Comprehensive Studies in Mathematics, Vol. 328. Springer-Verlag, Berlin, 2003. FINITE DUAL OF A COCOMMUTATIVE HOPF ALGEBROID, DIFFERENTIAL MATRIX EQUATIONS AND PV THEORY. 45 Universidad de Granada, Departamento de ´Algebra y IEMath. Facultad de Educaci´on, Econon´ıa y Tecnolog´ıa de Ceuta. Cor- tadura del Valle, s/n. E-51001 Ceuta, Spain E-mail address: [email protected] URL: http://www.ugr.es/kaoutit/ Department of Algebra and CITIC, Universidad de Granada, E18071 Granada, Spain E-mail address: [email protected] URL: http://www.ugr.es/gomezj/
1311.5475	1	1311	2013-11-21T16:49:55	p-filiform Leibniz algebras of maximum length	[ "math.RA" ]	The descriptions (up to isomorphism) of naturally graded $p$-filiform Leibniz algebras and $p$-filiform ($p\leq 3$) Leibniz algebras of maximum length are known. In this paper we study the gradation of maximum length for $p$-filiform Leibniz algebras. The present work aims at the classification of complex $p$-filiform ($p \geq 4$) Leibniz algebras of maximum length.	math.RA	math	P-FILIFORM LEIBNIZ ALGEBRAS OF MAXIMUM LENGTH. L.M. CAMACHO, E.M. CA NETE, J.R. G ´OMEZ, B.A. OMIROV Abstract. The descriptions (up to isomorphism) of naturally graded p-filiform Leibniz algebras and p-filiform (p ≤ 3) Leibniz algebras of maximum length are known. In this paper we study the gradation of maximum length for p-filiform Leibniz algebras. The present work aims at the classification of complex p-filiform (p ≥ 4) Leibniz algebras of maximum length. AMS Subject Classifications (2010): 17A32, 17A36, 17A60, 17B70. Key words: Lie algebra, Leibniz algebra, nilpotency, natural gradation, characteristic sequence, p-filiform, gradation of maximum length. 1. Introduction Leibniz algebras were introduced at the beginning of the 90s by J.-L. Loday in [14], which are a "non-commutative" generalization of Lie algebras. The right multiplication operator on an element of a Leibniz algebra is a derivation, which is a property inherited from Lie algebras. Active investigation on Leibniz algebra theory shows that many results of the theory of Lie algebras can be extended to Leibniz algebras. Distinctive properties of non-Lie Leibniz algebras have also been studied [2, 3]. For a Leibniz algebra there is a corresponding associated Lie algebra, which is the quotient algebra by the two-sided ideal generated by squares of elements of a Leibniz algebra (denoted by I). From the theory of Lie algebras it is well known that the study of finite dimensional Lie algebras was reduced to the nilpotent ones, due to Levi's theorem and Mal'cev's decomposition (see [13, 15]). The case of Leibniz algebras is analogous to Levi's theorem [3]. Namely, a Leibniz algebra is decomposed into a semidirect sum of its solvable radical and a semisimple Lie algebra. The structure of solvable Lie algebra can be obtained from the structure of its nilradical [16]. This approach has recently been extended to the case of Leibniz algebras [10]. Therefore, the main problem when describing of finite-dimensional Leibniz algebras is the nilpotent radical. Thus, the study of nilpotent Leibniz algebras is a crucial problem. Since the description of the set of n-dimensional nilpotent Leibniz algebras is an unsolvable task (even in the case of Lie algebras), we have to study nilpotent Leibniz algebras under certain conditions (conditions on index of nilpotency, various types of gradation, characteristic sequence etc.). The well-known gradations of nilpotent Lie and Leibniz algebras are very helpful when investigating of the properties of those algebras without restrictions on the gradation. Indeed, we can always choose an homogeneous basis and thus the gradation allows to obtain more explicit conditions for the structural constants. Moreover, such gradation is useful for the investigation of cohomologies for the considered algebras, because it induces the corresponding gradation of the group of cohomologies. The concept of length of a Lie algebra was introduced by G´omez, Jim´enez-Merch´an and Reyes in [11], [12]. Where, they distinguished an interesting family: algebras admitting a gradation with the greatest possible number of non-zero subspaces. Actually, the gradations with a large number of non-zero subspaces enable us to describe the multiplication on the algebra more exactly. They called such algebras algebras of maximum length. In fact, they only consider the connected gradation. There exist non connected algebras with the greatest possible number of non-zero subspaces. Nevertheless, according to G´omez et al. the notion of algebras of maximum length has already been used. In [1], [5] - [7] the classification of p-filiform Leibniz algebras of maximum length for 0 ≤ p ≤ 3 is already closed. The present paper aims at classification of n-dimensional p-filiform Leibniz algebras of maximum length with n and p generic. This work has been funded by Mathematics Institute and V Research Plan of Sevilla University, by the Grants of Junta de Andaluc´ıa, by the Grants (RGA) No:11-018 RG/Math/AS− I -- UNESCO FR: 3240262715 and IMU/CDC-program. 1 2 L.M. CAMACHO, E.M. CA NETE, J.R. G ´OMEZ, B.A. OMIROV Throughout the paper we consider finite-dimensional vector spaces and non-split algebras over the field of the complex numbers. Moreover, we have omitted null products in the multiplication table of an algebra. 2. Preliminares Recall [14] that an algebra L over a field F is called a Leibniz algebra if it satisfies the following Leibniz identity: where [−, −] denotes the multiplication of an algebra L. Let L be a Leibniz algebra, then L is naturally filtered by the descending central sequence [x, [y, z]] = [[x, y], z] − [[x, z], y], ∀x, y, z ∈ L L1 = L, Lk+1 = [Lk, L], k ≥ 1. An nilpotent Lebniz algebra L has nilindex equal to s if s is the minimum integer such that Ls 6= {0} and Ls+1 = {0}. We denote by Rx the operator of right multiplication on element x, i.e., Rx : L → L such that Rx(y) = [y, x] for any y ∈ L. Let x be an element of the set L\ L2. For the nilpotent operator Rx we define a descending sequence C(x) = (n1, n2, . . . , nk), which consists of the dimensions of the Jordan blocks of the operator Rx. In the set of such sequences we consider the lexicographic order, that is, C(x) = (n1, n2, . . . , nk) < C(y) = (m1, m2, . . . , ms) if and only if there exists i ∈ N such that nj = mj for some j < i and ni < mi. Definition 2.1. The sequence C(L) = max C(x)x∈L\L2 is called the characteristic sequence of the algebra L. Let L be an n-dimensional nilpotent Leibniz algebra and p be an non negative integer (p < n). Definition 2.2. The Leibniz algebra L is called p-filiform if C(L) = (n − p, 1, . . . , 1 ). A Leibniz algebra L is Z-graded if L = ⊕i∈ZVi, where [Vi, Vj] ⊆ Vi+j for some i, j ∈ Z with a finite number of non null spaces Vi. We will say that a Z-graded nilpotent Leibniz algebra L admits a connected gradation if L = Vk1 ⊕ Vk1+1 ⊕ · · · ⊕ Vk1+t and Vk1+i 6=< 0 > for some i (0 ≤ i ≤ t). Definition 2.3. The number l(⊕L) = l(Vk1 ⊕ Vk1+1 ⊕ · · · ⊕ Vk1+t) = t + 1 is called the length of the gradation, where ⊕L is a connected gradation. The gradation ⊕L has maximum length if l(⊕L) = dim(L). We define the length of an algebra L as follows l(L) = max{l(⊕L) such that ⊕ L = Vk1 ⊕ · · · ⊕ Vkt is a connected gradation}. An algebra L is called of maximum length if l(L) = dim(L). Thus, we resume the properties of gradation of maximum length: p {z }  (1) [Vi, Vj] ⊆ Vi+j , the subspaces of the gradation are not empty, the dimension of each subspace of the gradation equals 1, all subindices are different from each other. We define another type of gradation below. Given an n-dimensional Leibniz algebra L with nilindex s, put Li = Li/Li+1 with 1 ≤ i ≤ s and grL = L1 ⊕ L2 ⊕ · · · ⊕ Ls. Then [Li, Lj ] ⊆ Li+j and we obtain the graded algebra grL. If grL and L are isomorphic, grL ∼= L, we say that L is naturally graded. The classification of naturally graded p-filiform Lie algebras was done by Cabezas y Pastor in [4], which is given below. Theorem 2.1. Let L be a n-dimensional naturally graded p-filiform Lie algebra, with p > 1, n ≥ max{3p − 1, p + 8} and 3 ≤ r1 < r2 < · · · < rp−1 ≤ n − p odds. Therefore: • If rp−1 = n − p, then L is isomorphic to L(n, r1, r2, . . . , rp−2, n − p), • If rp−1 = n − p − 1, then L is isomorphic to L(n, r1, r2, . . . , rp−2, n − p − 1) or τ (n, r1, r2, . . . , rp−2, n − p − 1), P-FILIFORM LEIBNIZ ALGEBRAS OF MAXIMUM LENGTH 3 • If rp−1 = n−p−2, then L is isomorphic to L(n, r1, r2, . . . , rp−2, n−p−2), Q(n, r1, r2, . . . , rp−2, n−p−2) or τ (n, r1, r2, . . . , rp−2, n − p − 2), • If 2p − 1 ≤ rp−1 ≤ n − p − 3, then -- if n − p is odd, then L is isomorphic to L(n, r1, r2, . . . , rp−1) or Q(n, r1, r2, . . . , rp−1), -- if n − p is even, then L is isomorphic to L(n, r1, r2, . . . , rp−1). where L(n, r1, r2, . . . , rp−1) : [xi, xrj −i] = (−1)i−1yj, 1 ≤ i ≤ n − p − 1, 1 ≤ i ≤ rj −1 2 , 1 ≤ j ≤ p − 1. Q(n, r1, r2, . . . , rp−1) : [x0, xi] = xi+1, [xi, xrj −i] = (−1)i−1yj , [xi, xn−p−i] = (−1)i−1xn−p, 1 ≤ i ≤ n − p − 1, 1 ≤ i ≤ rj −1 1 ≤ i ≤ n−p−1 2 . 2 , 1 ≤ j ≤ p − 1, ([x0, xi] = xi+1,    τ (n, r1, r2, . . . , rp−2, n − p − 1) : [x0, xi] = xi+1, [xi, xrj −i] = (−1)i−1yj , [xi, xn−p−1−i] = (−1)i−1(xn−p−1 + yp−1), [xi, xn−p−i] = (−1)i−1 n−2i−p [x1, yp−1] = p+2−n xn−p. xn−p, 2 2 τ (n, r1, r2, . . . , rp−2, n − p − 2) : [x0, xi] = xi+1, [xi, xrj −i] = (−1)i−1yj , [xi, xn−p−2−i] = (−1)i−1(xn−p−2 + yp−1), [xi, xn−p−1−i] = (−1)i−1 n−p−1−2i xn−p−1, [xi, xn−p−i] = (−1)i(i − 1) n−p−1−i xn−p, [xi, yp−1] = p+3−n xn−p−2+i, 2 2 2 1 ≤ i ≤ n − p − 1, 1 ≤ i ≤ rj −1 1 ≤ i ≤ n−p−2 1 ≤ i ≤ n−p−2 2 2 , , 2 , 1 ≤ j ≤ p − 2, , 1 ≤ j ≤ p − 2, 2 1 ≤ i ≤ n − p − 1, 1 ≤ i ≤ rj −1 1 ≤ i ≤ n−p−3 1 ≤ i ≤ n−p−3 2 ≤ i ≤ n−p−1 1 ≤ i ≤ 2. 2 2 2 , , , with {x0, x1, . . . , xn−p, y1, . . . , yp−1} a basis. For naturally graded p-filiform non-Lie Leibniz algebras the result obtained is the following. Theorem 2.2. [8] Let L be a n-dimensional p-filiform non-Lie Leibniz algebra, with n − p ≥ 4. Then L is isomorphic to one of the following algebras If p is even: If p is odd: with {e1, . . . , en−p, f1, . . . , fp} a basis. M 1 :=([ei, e1] = ei+1, [e1, fj] = f p 2 +j, 1 ≤ i ≤ n − p − 1, 1 ≤ j ≤ p 2 . [ei, e1] = ei+1, [e1, f1] = e2 + f p [ei, f1] = ei+1, [e1, fj] = f p 2 +j, 1 ≤ i ≤ n − p − 1, 2 +1, 1 ≤ i ≤ n − p − 1, 2 ≤ j ≤ p 2 . [ei, e1] = ei+1, [e1, fj] = f⌊ p [ei, f⌊ p 2 ⌋+j , 2 ⌋ + 1] = ei+1, 1 ≤ i ≤ n − p − 1, 1 ≤ j ≤ ⌊ p 2 ⌋, 1 ≤ i ≤ n − p − 1. M 2 := M 3 := 4 L.M. CAMACHO, E.M. CA NETE, J.R. G ´OMEZ, B.A. OMIROV 3. p-filiform Leibniz algebras of maximum length In order to achieve our goal we use the following algorithm: 1. Firstly, we extend the naturally graded p-filiform Leibniz algebras by using the natural gradations. In this way, we can distinguish two cases: the natural graded p-filiform Lie algebras and the natural graded p-filiform non-Lie Leibniz algebras. 2. After that, we construct an homogeneous basis of a graded algebra of maximum length with respect to the basis of naturally gradation. 3. Finally, we classify the p-filiform Leibniz algebra in an homogeneous basis of maximum length. 3.1. Extension of Lie algebras. In this subsection we prove that there is no n-dimensional p-filiform Leibniz algebra of maximum length in the non split case, for p ≥ 4 and n ≥ max{3p − 1, p + 8}. Note that the study of the particular case for n < max{3p − 1, p + 8} can be found in [9]. multiplication whose naturally graded algebra is L. Further we will denote by eL the extension of L, that is, the family of algebras with the table of Theorem 3.1. Let L be a n-dimensional p-filiform Leibniz algebra, whose naturally graded associated algebra is isomorphic to L(n, r1, r2, . . . , rp−1) or Q(n, r1, r2, . . . , rp−1), with p ≥ 4, n ≥ max{3p − 1, p + 8} and 3 ≤ r1 < r2 < · · · < rp−1 ≤ n − p. Then the algebra L does not admit a gradation of maximum length. Proof: Note that, we have p ≥ 4, n ≥ 11 and 3 ≤ r1 < r2 · · · < rp−1 ≤ n − p, where all ri are odds. Let us suppose that L admits a gradation of maximum length. It is easy to see that natural gradation of the algebra L(n, r1, r2, . . . , rp−1 is L1 = hx0, x1i, Li = hxii with 2 ≤ i ≤ n − p and i 6= rj and Lrj = hxrj , yji with 1 ≤ j ≤ p − 1. Therefore the law of Study of the extension eL(n, r1, r2, . . . , rp−1). eL(n, r1, r2, . . . , rp−1) is defined by the following products, where the asterisks (*) denote the corre- sponding structural constants:  [x0, x0] = (∗)x3 + · · · + (∗)xn−p + (∗)y1 + · · · + (∗)yp−1, [x0, xi] = xi+1 + (∗)xi+2 + · · · + (∗)xn−p + (∗)y1 + · · · + (∗)yp−1, [x0, xi] = xi+1 + (∗)xi+2 + · · · + (∗)xn−p + (∗)y2 + · · · + (∗)yp−1, ... [x0, xi] = xi+1 + (∗)xi+2 + · · · + (∗)xn−p + (∗)yp−1, [x0, xi] = xi+1 + (∗)xi+2 + · · · + (∗)xn−p, [xi, x0] = −xi+1 + (∗)xi+2 + · · · + (∗)xn−p + (∗)y1 + · · · + (∗)yp−1, [xi, x0] = −xi+1 + (∗)xi+2 + · · · + (∗)xn−p + (∗)y2 + · · · + (∗)yp−1, ... [xi, x0] = −xi+1 + (∗)xi+2 + · · · + (∗)xn−p + (∗)yp−1, [xi, x0] = −xi+1 + (∗)xi+2 + · · · + (∗)xn−p, [xi, xrj −i] = (∗)xrj +1 + · · · + (∗)xn−p + (−1)i−1yj + (∗)yj+1 + · · · + (∗)yp−1, [xrj −i, xi] = (∗)xrj +1 + · · · + (∗)xn−p + (−1)iyj + (∗)yj+1 + · · · + (∗)yp−1, [xi, xj] = (∗)xi+j+1 + . . . (∗)xn−p + (∗)yrk + · · · + (∗)yp−1, [xi, yj] = (∗)xi+rj +1 + · · · + (∗)xn−p + (∗)yrk + · · · + (∗)yp−1, [yi, yj] = (∗)xri+rj +1 + · · · + (∗)xn−p + (∗)yrk + · · · + (∗)yp−1, 1 ≤ i ≤ r1 − 2, r1 − 1 ≤ i ≤ r2 − 2, ... rp−2 − 1 ≤ i ≤ rp−1 − 2, rp−1 − 1 ≤ i ≤ n − p − 1, 1 ≤ i ≤ r1 − 2, r1 − 1 ≤ i ≤ r2 − 2, ... , , 2 2 rp−2 − 1 ≤ i ≤ rp−1 − 2, rp−1 − 1 ≤ i ≤ n − p − 1, 1 ≤ i ≤ rj −1 1 ≤ j ≤ p − 1, rj + 1 ≥ 4, 1 ≤ i ≤ rj −1 1 ≤ j ≤ p − 1, rj + 1 ≥ 4, 1 ≤ i + j ≤ rk, 1 ≤ k ≤ p − 1, 0 ≤ i ≤ n − p − 1, 1 ≤ j ≤ p − 1 1 ≤ i, j ≤ p − 4, 7 ≤ rk ≤ p − 1, ri + rj + 1 ≤ rk ≤ p − 1. P-FILIFORM LEIBNIZ ALGEBRAS OF MAXIMUM LENGTH 5 Without loss of generality, one can assume that the general form of homogeneous generators of the gradation of maximum length are exs = x0 + n−pXi=1 1 (cid:19) 6= 0. p−1Xj=1 (2) aixi + bjyj and Aixi + Bjyj, a1 where det(cid:18) 1 Below we present the straightforward consequences of the generated elements ofeL(n, r1, r2, . . . , rp−1) via the above two vectors: A0 ext = A0x0 + x1 + n−pXi=2 p−1Xj=1 ] = (−1)i(1 − a1A0)xi+2 + (∗)xi+3 + · · · + (∗)xn−p + (∗)y2 + · · · + (∗)yp−1 ] = (−1)i(1 − a1A0)xi+2 + (∗)xi+3 + · · · + (∗)xn−p + (∗)y1 + · · · + (∗)yp−1 i−times i−times for 1 ≤ i ≤ r1 − 3, [fxs,fxs] = (∗)x3 + · · · + (∗)xn−p + (∗)y1 + · · · + (∗)yp−1, [ext, ext] = (∗)x3 + · · · + (∗)xn−p + (∗)y1 + · · · + (∗)yp−1, [fxs, ext] = (1 − a1A0)x2 + (∗)x3 + · · · + (∗)xn−p + (∗)y1 + · · · + (∗)yp−1, [[[fxs, ext],fxs], . . .fxs {z } [[[fxs, ext],fxs], . . .fxs } {z [[[fxs, ext],fxs], . . .fxs {z } [[[fxs, ext],fxs], . . .fxs {z } [[[fxs, ext], fxs], . . .fxs } {z ] = (−1)i(1 − a1A0)xi+2 + (∗)xi+3 + · · · + (∗)xn−p + (∗)yk+1 + · · · + (∗)yp−1 for 1 ≤ k ≤ p − 1. for rj−1 − 1 ≤ i ≤ rj − 3, 2 ≤ j ≤ p − 1, for r1 − 1 ≤ i ≤ r2 − 3, (rk −2)−times i−times i−times ] = (−1)i(1 − a1A0)xi+2 + (∗)xi+3 + · · · + (∗)xn−p + (∗)yj + · · · + (∗)yp−1 for rp−1 − 1 ≤ i ≤ n − p − 2. ] = (−1)rk−2(1 − a1A0)xrk + (∗)xi+3 + · · · + (∗)xn−p + (−1)rk−2(1 − a1A0)a1yk+ Let us take the new homogeneous basis constructed by the following vectors: z0 = exs, z2 = [z0, z1], zi = [zi−1, z0], 3 ≤ i ≤ n − p, pj = [z1, zrj −1], 1 ≤ j ≤ p − 1, z1 = ext, where pj = [z1, zrj −1] = (−1)rj −3(1 − a1A0)A0xrj + (∗)xrj +1 + · · · + (∗)xn−p+ + (−1)r1−3(1 − a1A0)yj + (∗)yj+1 + (∗)yp−1. The matrix of the change of basis is 1 a1 a2 a3 A0 1 A2 A3 0 0 ... 0 ... 0 ... 0 ... 0 0 0 ... 0 0 C2 (∗) 0 ... 0 ... 0 ... 0 ... 0 0 0 ... 0 0 C3 ... ... 0 0 ... ... 0 0 ... ... 0 0 ... ... 0 0 0 0 ... 0 0 0 ... 0  . . . . . . . . . . . . ... . . . ... . . . ... . . . ... . . . ar1 Ar1 (∗) (∗) ... Cr1 ... 0 ... 0 ... 0 A0 . . . . . . −Cr1 0 ... 0 . . . ... . . . . . . . . . . . . ... . . . ... . . . ... . . . ... . . . . . . ar2 Ar2 (∗) (∗) ... (∗) ... Cr2 ... 0 ... 0 (∗) . . . . . . . . . . . . ... . . . ... . . . ... . . . ... . . . . . . arp−1 Arp−1 (∗) (∗) ... (∗) ... (∗) ... Crp−1 ... 0 (∗) A0 . . . −Cr2 ... ... 0 . . . . . . (∗) ... ... . . . −Crp−1 . . . an−p b1 . . . An−p B1 . . . bp−1 . . . Bp−1 b2 B2 (∗) (∗) ... (∗) ... Cr2 ... 0 ... 0 a1 . . . . . . ... . . . ... . . . . . . . . . . . . a1 (∗) (∗) (∗) (∗) ... ... (∗) Cr1 ... ... 0 (∗) ... ... 0 (∗) ... ... 0 ... . . . Cn−p . . . . . . ... . . . ... . . . ... . . . . . . . . . ... 6 . L M . C A M A C H O , . E M . C A N E T E , J . R . G ´O M E Z , . B A . I O M R O V  (∗) (∗) ... (∗) ... (∗) ... ... 0 (∗) (∗) ... ... . . . Crp−1 ... (∗) −Cr1 (∗) ... (∗) 0 ... 0 (∗) −Cr2 ... 0 ... . . . −Crp−1 A0 . . . P-FILIFORM LEIBNIZ ALGEBRAS OF MAXIMUM LENGTH 7 where Ci = (−1)i(1 − a1A0), for 2 ≤ i ≤ rp − 1. Note that this matrix has rank equal to n (because of 1 − a1A0 6= 0). We put z0 ∈ Vks and z1 ∈ Vkt . Then, according to the definition of gradation of maximum length, we derive: L = Vks ⊕ Vkt ⊕ Vkt+ks ⊕ · · · ⊕ Vkt+(n−p−1)ks ⊕ V2kt+(r1−2)ks ⊕ · · · ⊕ V2kt+(rp−1−2)ks. This gradation is connected if and only if ks = ±1. Since the cases ks = 1 and ks = −1 are equivalent, one can assume ks = 1. To make the reasoning simple, we sometimes shall continue the notation ks, even though that ks = 1. Our next objective is to analyze the value of kt. We distinguish the following cases: • Let kt > 0. Then the properties (1) are satisfied if and only if kt = 2. In this case we conclude kt+ks < 2kt+(r1−2)ks < kt+(n−p−1)ks, that is, V2kt+(r1−2)ks = hp1, zmi with 2 ≤ m ≤ n−p, giving rise to a contradiction with the assumption of maximum length. • Let kt < 0. Then we consider the following subcases: -- If kt = −n + p + 1, then 2kt + (ri − 2)ks ≤ 2(−n + p + 1) + n − p − 2 = −n + p ≤ 0 for 1 ≤ i ≤ p − 1. Hence we can affirm that all subspaces V2kt+(ri−2)ks have negative subindices. Therefore, due to the connectedness of the gradation, we obtain: distance(pi, pi+1) = distance(2kt + ri − 2, 2kt + ri+1 − 2) = 1, ⇒ distance(ri, ri+1) = 1, which is impossible since the parameters are odd ri for 1 ≤ i ≤ p − 1. -- If kt > −n + p + 1, then there exists an zt, with 1 ≤ t ≤ n − p, such that zt ∈ V1 = hz0i. However, it is impossible because of zt is generator. Hence we get a contradiction. -- If kt < −n + p + 1, then it is easy to see that V0 = h0i, which contradicts the properties (1). Thus, it has been proved that there is no algebra of maximum length among the extension of the family L(n, r1, r2, . . . , rp−1). Furthermore, the above arguments can be used for some admissible value of rp−1, that is, this proof includes the particular cases eL(n, r1, r2, . . . , rp−2, n − p), eL(n, r1, r2, . . . , rp−2, n − p − 1) and eL(n, r1, r2, . . . , rp−2, n − p − 2). Study of the extension eQ(n, r1, r2, . . . , rp−1). This case is analogous to case eL(n, r1, r2, . . . , rp−1). The difference is only in the construction of the homogeneous basis of gradation of maximum length. Let us consider the following cases: Case 1: 1 + a1 6= 0. Then we take as a basis the following vectors: z0 = exs, z1 = ext, z2 = [z0, z1], zi = [zi−1, z0], for 3 ≤ i ≤ n − p, pj = [z1, zrj −1], for 1 ≤ j ≤ p − 1, where pj = A0(−1)rj −3(1 − a1A0)xrj + (∗)xrj +1 + · · · + (∗)xn−p+ (−1)r1−3(1 − a1A0)yj + (∗)yj+1 + + · · · + (∗)yp−1. same reasons we conclude that there is not any algebra of maximum length in this case. Since the chosen basis is the same as in the study of the extension eL(n, r1, r2, . . . , rp−1), for the 8 L.M. CAMACHO, E.M. CA NETE, J.R. G ´OMEZ, B.A. OMIROV Case 2: 1 + a1 = 0. It is clear that A0 6= −1 (because of 1 − a1A0 6= 0). Therefore, we can take the new basis defined by the following vectors: z0 = exs, z1 = ext, z2 = [z0, z1], zi = [zi−1, z0], for 3 ≤ i ≤ n − p − 1, zn−p = [zn−p−1, z1] = (−1)n−p−2(1 − a1A0)(1 + A0)xn−p, pj = [z1, zrj −1], for 1 ≤ j ≤ p − 1. It is not difficult to check that the matrix of basis transformation is non-singular. The associated gradation of maximum length is Vks ⊕ Vkt ⊕ Vkt+ks ⊕ · · · ⊕ Vkt+(n−p−2)ks ⊕ V2kt+(n−p−2)ks ⊕ V2kt+(r1−2)ks ⊕ · · · ⊕ V2kt+(rp−1−2)ks. Note that this gradation also satisfies the properties (1). As stated above, without loss of generality, we can assume ks = 1. We consider the following subcases: • Let kt > 0. Then, by considering the properties (1), we conclude kt = 2. So, we have ks = 1, kt = 2 and 3 ≤ r1 < n − p − 2, that is, 5 ≤ 2kt + (r1 − 2)ks < n − p. Thus, one can assert the existence of zt with 4 ≤ t ≤ n − p − 1, such that V2kt+(r1−2)ks = hp1, zti, which contradicts the assumption of maximum length (see the properties (1)). • Let kt < 0. Then from the properties of gradation of maximum length we get kt = −n + p + 2. Moreover, since r1 and r2 are odd we conclude distance(p1, p2) = distance(2kt + (r1 − 2)ks, 2kt + (r2 − 2)ks) = r2 − r1 > 1. Therefore, we obtain a contradiction with the assumption of maximum length again. (cid:3) The next theorem is proved by applying the same methods and arguments as in the proof of Theorem 3.1. Theorem 3.2. Let L be a n-dimensional p-filiform Leibniz algebra, whose naturally gradation leads to an algebra isomorphic to τ (n, r1, r2, . . . , n − p − 1) or τ (n, r1, r2, . . . , n − p − 2), with p ≥ 4 and n ≥ max{3p − 1, p + 8}. Then L does not admit a gradation of maximum length. 3.2. Extension of Leibniz algebras. In this subsection we study the description of p-filiform Leibniz algebras of maximum length from the extensions of naturally graded p-filiform non-Lie Leibniz algebras. The classification of Theorem 2.2 leads to considerate the extensions of the algebras M1 − M3. Firstly, we analyze the extension of the algebras M 1 and M 2. Theorem 3.3. Let L be a n-dimensional p-filiform Leibniz algebra, with n − p ≥ 4, p ≥ 4 and p even. Then, the algebra L is isomorphic to the one of the following pairwise non-isomorphic algebras: M 4(α) : 1 ≤ i ≤ n − p − 1, [xi, x1] = xi+1, 1 ≤ i ≤ p 2 , [x1, yi] = zi, [z1, y2] = [z2, y1] = αxn−p, α ∈ {0, 1}. n−p ∈ N in the algebra M 4(1). where n is even and n [xi, x1] = xi+1, [x1, yi] = zi, [y1, y2] = xn−p. 1 ≤ i ≤ n − p − 1, 1 ≤ i ≤ p 2 , M 5 : Proof: Similarly as above the generators of maximum length gradation of the algebra L have the form: aiei + exs = e1 + yj = fj + n−pXi=2 n−pXk=1 bjfj, pXi=1 2Xk=1,k6=j p ckj ek + dkj fk for 1 ≤ j ≤ p 2 . P-FILIFORM LEIBNIZ ALGEBRAS OF MAXIMUM LENGTH 9 Study of the extension gM 1. By considering the law of the algebra M 1 and the natural gradation we have: L1 = he1, f1, . . . , f p 2 i ⊕ L2 = he2, f p 2 +1, . . . , fpi ⊕ L3 = he3i ⊕ · · · ⊕ Ln−p = hen−pi. We construct the following new adapted basis: x1 = exs, whose associated gradation is xi = [xi−1, x1], 2 ≤ i ≤ n − p, yi, 1 ≤ i ≤ p 2 , zi = [x1, yi], 1 ≤ i ≤ p 2 , Vks ⊕ V2ks ⊕ · · · ⊕ V(n−p)ks ⊕ Vk1 ⊕ Vk2 ⊕ · · · ⊕ Vk p 2 ⊕ Vk1+ks ⊕ Vk2+ks ⊕ · · · ⊕ Vk p 2 +ks . Let us assume that this gradation has maximum length. We consider the products [zi, x1] with 1 ≤ i ≤ n − p − 1. Due to the law of the algebra M1 we obtain [zi, x1] = c1ie3 + (∗)e4 + · · · + (∗)en−p, i.e., we can assume that [zi, x1] = c1ix3, 1 ≤ i ≤ n − p − 1. On the other hand, by using the properties of the gradation we derive ([zi, x1] ∈ V2ks+ki, for 1 ≤ i ≤ n − p − 1, x3 ∈ V3ks , ks 6= ki, for 1 ≤ i ≤ n − p − 1. Therefor, c1i = 0, 1 ≤ i ≤ n − p − 1. By induction on a fixed i and any j and using the Leibniz identity, one can prove that [xi, xj ] = 0, 3 ≤ i, j ≤ n − p. Let us analyze the products [xi, yj], [yj, xi] for 1 ≤ i ≤ n − p, 1 ≤ j ≤ p According to the law of the algebra M1 we get [x2, yj] = (∗)e4 +· · ·+(∗)en−p, that is, [x2, yj] = Axm 2 and (i, j) 6= (1, j). with 1 ≤ j ≤ p 2 , 4 ≤ m ≤ n − p and some coefficient A. On the other hand, the properties of the gradation of maximum length deduce ([x2, yj] ∈ V2ks+kj , xm ∈ Vmks . Therefore, 2ks ≤ kj ≤ (n − p − 2)ks, which is only possible for A = 0, that is, [x2, yj] = 0, 1 ≤ i ≤ p 2 . By applying the similar argumentations it can be proved that [xi, yj] = 0 = [yj, xi], [xj , zi] = 0 = [zi, xj], [yi, zj] = 0, 1 ≤ j ≤ p 2 , 1 ≤ i ≤ n − p, 1 ≤ j ≤ n − p, 1 ≤ i ≤ p 2 , 1 ≤ i, j ≤ p 2 . (i, j) 6= (1, j), Thanks to the Leibniz identity we get [[zi, yj], x1] = 0. Since [xi, x1] = 0 if and only if xi = xn−p, then taking into account the product [zi, yj] = (∗)e4 + · · · + (∗)en−p, we obtain [zi, yj] = Aij xn−p for some coefficients Aij . Furthermore we obtain [zi, yi] = 0, 1 ≤ i ≤ p 2 by the properties of the gradation. Thanks to the Leibniz identity we derive Aij = Aji, 1 ≤ i, j ≤ p 2 . By following the same reasons we conclude [yi, yj] = Bijxn−p with Bii = 0. Finally, it is trivial to check that [zi, zj] = 0 for 1 ≤ i, j ≤ p Summarizing, the law of the algebra is determined by the following products: 2 , by using the Leibniz identity. 1 ≤ i ≤ n − p − 1, [xi, x1] = xi+1, 1 ≤ i ≤ p [x1, yi] = zi, 2 , [yi, yj] = Bijxn−p, 1 ≤ i, j ≤ p 1 ≤ i, j ≤ p [zi, yj] = Aijxn−p, 2 , i 6= j, 2 , i 6= j, Aij = Aji, L : with the conditions: (3) (if Bi0j0 6= 0 for some i0, j0, then Bi0k = 0 for all 6= j0 and Bsj0 = 0 for all s 6= i0, if Ai0j0 6= 0 for some i0, j0, then Ai0k = 0 for all k 6= j0 and Asj0 = 0 for all s 6= i0. 10 L.M. CAMACHO, E.M. CA NETE, J.R. G ´OMEZ, B.A. OMIROV Case 1: Bij = 0 for all i, j. Then the law of L has the following form: 1 ≤ i ≤ n − p − 1, [xi, x1] = xi+1, 1 ≤ i ≤ p [x1, yi] = zi, 2 , [zi, yj] = Aijxn−p, 1 ≤ i, j ≤ p 2 , i 6= j, Aij = Aji, L : where the parameters Aij satisfy the previous hypothesis. If all the parameters Aij are equal to zero, then it is easy to see that the algebra L has maximum length. Indeed, putting Vi = hxii, 1 ≤ i ≤ n − p, Vn−p+2j−1 = hyji, 1 ≤ j ≤ p 2 , Vn−p+2j = hzji, 1 ≤ j ≤ p 2 , we get L = V1 ⊕ V2 ⊕ · · · ⊕ Vn−p ⊕ Vn−p+1 ⊕ Vn−p+2 ⊕ Vn−p+3 ⊕ Vn−p+4 ⊕ · · · ⊕ Vn−1 ⊕ Vn. Thus we get the algebra M 4(0). Let us assume now that there exists Ai0j0 6= 0. Without loss of generality, we can suppose A12 = 0. Taking the following change of basis y′ i = y1 − yi, y′ i+1 = y2 + yi+1, z′ i = [x1, y′ i], z′ i+1 = [x1, y′ i+1], 3 ≤ i ≤ p 2 − 1, we obtain the algebra, which is defined by the following products 1 ≤ i ≤ n − p − 1, 1 ≤ i ≤ p 2 , [xi, x1] = xi+1, [x1, yi] = zi, [z1, y2] = xn−p, [z2, y1] = xn−p. L : Using the maximum length gradation properties and connectedness it is not difficult to check that n−p ∈ N and n even. this algebra admits the gradation of maximum length only for the cases of ks = n The associated gradation is decomposed into direct sum of the following spaces: y2 ∈ V(n−p−1)ks−1, xi ∈ Viks, y1 ∈ V1, z1 ∈ Vks+1, z2 ∈ V(n−p)ks−1, yi ∈ Vi−1, zi ∈ Vks+i−1, yq(ks−1)+i ∈ V2qks−1+i, zq(ks−1)+i ∈ V(2q+1)ks−1+i, y n−p−2 z n−p−2 (ks−1)+i ∈ V(n−p−2)ks−1+i, (ks−1)+i ∈ V(n−p−1)ks−1+i, 2 2 1 ≤ i ≤ n − p, 3 ≤ i ≤ ks, 3 ≤ i ≤ ks, 1 ≤ q ≤ n−p−4 , 1 ≤ q ≤ n−p−4 , 2 ≤ i ≤ ks − 1, 2 ≤ i ≤ ks − 1. 2 2 2 ≤ i ≤ ks, 2 ≤ i ≤ ks, Case 2: ∃i0, j0 such that Bi0j0 6= 0. Making the basis transformation y′ 2 = yj0 , without loss of generality, one can assume that B12 6= 0. Further, applying the properties of gradation of maximum length, the conditions (3) and the changes of basis, we arrive to the algebra of maximum length with the following table of multiplication: 1 = yi0 , y′ L : V−1 = hy1i, whose associated graded spaces are [xi, x1] = xi+1, [x1, yi] = zi, [y1, y2] = xn−p, 1 ≤ i ≤ n − p − 1, 1 ≤ i ≤ p 2 , V0 = hz1i, Vi = hxii, 1 ≤ i ≤ n − p, Vn−p+2k+1 = hyk+2i, 0 ≤ k ≤ p 2 − 2, Vn−p+2k = hzk+1i, 1 ≤ k ≤ p 2 − 1. In the following theorem we prove that there is no algebra of maximum length among algebras from The description of the extensiongM 2 is carried out in a similar way as for the extensiongM 1. extensiongM 3. (cid:3) P-FILIFORM LEIBNIZ ALGEBRAS OF MAXIMUM LENGTH 11 Theorem 3.4. Let L be a n-dimensional p-filiform Leibniz algebra with n − p ≥ 4, p ≥ 4 and p odd. Then L does not admit a gradation of maximum length. Proof: First of all, we shall denote q = ⌊ p 2 ⌋ for simplicity. Recall, the natural gradation of M 3 is L1 = he1, f1, f2, . . . , fq+1i ⊕ L2 = he2, fq+2, . . . , fpi ⊕ Li = heii, 3 ≤ i ≤ n − p. By considering the law of M 3 we denote by new generators the following: exs = e1 + yj = fj + n−pXi=2 n−pXk=1 Then we get aiei + bjfj, pXi=1 2Xk=1,k6=j p ckjek + dkj fk for 1 ≤ j ≤ q + 1. [[fxs,fxs] . . . ,fxs] } i−times 3 ≤ i ≤ n − p. = (1 + bq+1)i−1ei + (∗)ei+1 + · · · + (∗)en−p, [fxs,fxs] = (1 + bq+1)e2 + (∗)e3 + · · · + (∗)en−p + b1fq+2 + · · · + bqfp, {z [fxs, y1] = (c11 + dq+11)e2 + (∗)e3 + · · · + (∗)en−p + fq+2 + d21fq+3 + · · · + dq1fp, [y1,fxs] = c11[fxs,fxs], [fxs, yi] = (c1i + dq+1i)e2 + (∗)e3 + · · · + (∗)en−p + d1ifq+2 + · · · + fq+i+1 + · · · + dqifp, [yi,fxs] = c1i[fxs,fxs], [y1, y1] = c11[fxs, y1], [yi, yj] = c1i[fxs, yj ], 1 ≤ i, j ≤ q + 1, (i, j) 6= (1, 1). 2 ≤ i ≤ q + 1, the following vectors: Without loss of generality, one can assume 1 + bq+1 6= 0. Then the homogeneous basis of L is defined by 2 ≤ i ≤ q, x1 =fxs, xi = [xi−1,fxs], yi, zi = [x1, yi], 2 ≤ i ≤ n − p, 1 ≤ i ≤ q + 1, 1 ≤ i ≤ q. The matrix of the change of basis is as follows: bp bq 0 0 0 dp1 dp2 ... dpq+1 dq1 ... 1   1 0 0 0 0 c11 c12 ... a2 D1 0 0 0 c21 c22 ... a3 (∗) D2 0 0 c31 c32 ... a4 (∗) (∗) D3 0 c41 c42 ... c1q+1 c2q+1 0 ... 0 E1 ... Eq c3q+1 (∗) ... (∗) c4q+1 (∗) ... (∗) an−p (∗) (∗) (∗) . . . . . . . . . . . . . . . Dn−p−1 . . . . . . cn−p1 cn−p2 ... b1 0 0 0 0 1 d12 ... b2 0 0 0 0 d21 1 ... . . . . . . . . . . . . . . . cn−pq+1 d1q+1 d2q+1 (∗) ... (∗) 0 ... 0 0 ... 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . bq+1 bq+2 0 0 0 0 b1 0 0 0 dq+11 dq+12 dq+21 dq+22 ... 1 0 ... 0 ... dq+2q+1 1 ... d1q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . where Di = (1 + bq+1)i, 1 ≤ i ≤ n − p − 1 and Ei = c1i + dq+1i, 1 ≤ i ≤ q. The gradation associated to the above basis is Vks ⊕ V2ks ⊕ · · · ⊕ V(n−p)ks ⊕ Vk1 ⊕ · · · ⊕ Vkq+1 ⊕ Vk1+ks ⊕ · · · ⊕ Vkq+ks . Let us assume that this gradation has maximum length. Consider [x2, yi] = (1 + bq+1)(c1i + dq+1i)e3 + (∗)e4 + · · · + (∗)en−p, 1 ≤ i ≤ q. Therefore we conclude that [[x1, x1], yi] = Ax3, with A ∈ C. On the other hand, by considering the properties of the gradation we have ([x2, yi] ∈ V2ks+ki , for 1 ≤ i ≤ q, x3 ∈ V3ks , 12 L.M. CAMACHO, E.M. CA NETE, J.R. G ´OMEZ, B.A. OMIROV that is, we have either A = 0 or ks = ki. The last equality contradicts the assumption of maximum length, thus A = 0, i.e, we obtain (4) From the product we conclude c1q+1 = −1. c1i + dq+1i = 0, 1 ≤ i ≤ q. [x2, yq+1] = (1 + bq+1)(c1q+1 + 1)e3 + (∗)e4 + · · · + (∗)en−p, Finally, it contradicts the assumption of maximum length by comparison with the following equalities: [yq+1, x1] = −(1 + bq+1)e2 + (∗)e3 + +(∗)en−p − b1fq+2 − · · · − bqfp = −x2. By means of the gradation, it leads to Vks+kq+1 = V2ks , i.e, ks = kq+1, which contradicts the properties (1). (cid:3) References [1] Adashev J.Q., Canete E.M.,Derivations of the quasi-filiform Leibniz algebras of maximum length, Uzbek Math. J., 2, 2011, 3 -- 14. [2] Ayupov Sh.A., Omirov B.A., On some classes of nilpotent Leibniz algebras, (Russian) Sib. Mat. Zh., 42(1), 2001, 18 -- 29; translation in Sib. Math. J., 42(1), 2001, 15 -- 24. [3] Barnes D.W., On Levi's theorem for Leibniz algebras, Bull. Aust. Math. Soc., 86(2), 2012, 184 -- 185. [4] Cabezas J.M., Pastor E., Naturally graded p-filiform Lie algebras in arbitrary finite dimension, J. Lie Theory, 15, 2005, 379 -- 391. [5] Camacho L.M., Canete E.M., G´omez J.R., Omirov B.A., Quasi-filiform Leibniz algebras of maximum length, Sib. Math. J., 52(5), 2011, 840 -- 853. [6] Camacho L. M., Canete E.M, G´omez J.R., Omirov B.A., 3-filiform Leibniz algebras of maximum length, whose naturally graded algebras are Lie algebras, Linear Multilinear Alg., 59(9), 2011, 1039 -- 1058. [7] Camacho L. M., Canete E.M, G´omez J.R., Omirov B.A., 3-filiform Leibniz algebras of maximum length., Sumitted to Journal of Algebra (2013), arXiv:1310.6539v1. [8] Camacho L.M., G´omez J.R., Gonz´alez A.J., Omirov B.A. The classification of naturally graded p-filiform Leibniz algebras, Commun. Alg., 39(1), 2011, 153 -- 163. [9] Canete E.M. Algebras de Leibniz de longitud maxima, PhD Thesis, Universidad de Sevilla, 2012. (http://www.educacion.es/teseo). [10] Casas J.M., Ladra M., Omirov B.A., Karimjanov I.A. Classification of solvable Leibniz algebras with null-filiform nilradical, Linear Multilinear Alg., 61(6), 2012, 758 -- 774. [11] G´omez J.R., Jim´enez-Merch´an A., Reyes J. Maximum length filiform Lie algebras, Extracta Mathematicae, 16(3), 2001, 405 -- 421. [12] G´omez J.R., Jim´enez-Merch´an A., Reyes J., Quasi-filiform Lie algebras of maximum length, Linear Algebra and its Applications, 335, 2001, 119 -- 135. [13] Jacobson N., Lie algebras, Interscience Tracts in Pure and Applied Mathematics, No. 10, Interscience Publishers (a division of John Wiley and Sons), New York-London, 1962. [14] Loday J.L., Une version non commutative des alg´ebres de Lie: les alg´ebres de Leibniz. Ens. Math., 39, 1993, 269 -- 293. [15] Mal'cev A.I., Solvable Lie algebras, Amer. Math. Soc. Translation, 1950, p. 36. [16] Mubarakzjanov G.M. On solvable Lie algebras, (Russian), Izv. Vyss. Ucehn. Zaved. Matematika 1963, 114 -- 123. [L.M. Camacho -- E.M. Canete -- J.R. G´omez] Dpto. Matem´atica Aplicada I. Universidad de Sevilla. Avda. Reina Mercedes, s/n. 41012 Sevilla. (Spain) E-mail address: [email protected] --- [email protected] --- jrgomez [B.A. Omirov] Institute of Mathematics and Information Technologies of Academy of Uzbekistan, 29, Do'rmon yo'li street., 100125, Tashkent (Uzbekistan) E-mail address: [email protected]
1708.06143	1	1708	2017-08-21T10:37:48	On the catenarity of virtually nilpotent mod-$p$ Iwasawa algebras	[ "math.RA", "math.NT", "math.RT" ]	Let $p>2$ be a prime, $k$ a finite field of characteristic $p$, and $G$ a nilpotent-by-finite compact $p$-adic analytic group. Write $kG$ for the completed group ring of $G$ over $k$. We show that $kG$ is a catenary ring.	math.RA	math	On the catenarity of virtually nilpotent mod-p Iwasawa algebras William Woods October 21, 2018 Abstract Let p > 2 be a prime, k a finite field of characteristic p, and G a nilpotent-by-finite compact p-adic analytic group. Write kG for the com- pleted group ring of G over k. We show that kG is a catenary ring. 7 1 0 2 g u A 1 2 ] . A R h t a m [ 1 v 3 4 1 6 0 . 8 0 7 1 : v i X r a 1 Contents Introduction 1 Heights of primes and Krull dimension 1.1 Prime and G-prime ideals . . . . . . . . . . . . . . . . . . . . . . 1.2 Inducing ideals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Krull dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Control theorem 2.1 The abelian case . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Faithful primes are controlled by ∆ . . . . . . . . . . . . . . . . . 2.3 Primes adjacent to faithful primes . . . . . . . . . . . . . . . . . 3 Catenarity 3.1 The orbitally sound case: plinths and a height function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Vertices and sources 3.3 The general case: inducing from open subgroups . . . . . . . . . 3 6 6 7 7 11 11 13 16 19 19 24 29 2 Introduction Fix a prime p, a commutative pseudocompact ring k (e.g. Fp or Zp) and a compact p-adic analytic group G. (Such groups are perhaps most accessibly characterised as those groups G which are isomorphic to a closed subgroup of GLn(Zp) for some n.) The completed group ring kG (sometimes written k[[G]]) is defined by kG := lim←− N k[G/N ], where the inverse limit ranges over all open normal subgroups N of G, and k[G/N ] denotes the ordinary group algebra of the (finite) group G/N over k. This ring satisfies an obvious universal property [25, Lemma 2.2], and modules over it characterise continuous k-representations of G (which has the profinite topology). When k = Fp, Zp or related rings, this is often called the Iwasawa algebra of G. Iwasawa algebras (and related objects, such as locally analytic distribution alge- bras [20]) have recently become a very active research area due to their number- theoretic interest, for instance in the p-adic Langlands programme: see [19], for example. They are also interesting objects of study in their own right, as an interesting class of noetherian rings: see [3] for a 2006 survey of what is known about these rings. Our main result is the following. Theorem A. Take p > 2. Let G be a nilpotent-by-finite compact p-adic analytic group, and let k be a finite field of characteristic p. Then kG is a catenary ring. Recall that a ring R is said to be catenary if any two maximal chains of prime ideals with common endpoints have the same length, i.e. whenever P = P1 (cid:12) P2 (cid:12) · · · (cid:12) Pr = P ′ P = Q1 (cid:12) Q2 (cid:12) · · · (cid:12) Qs = P ′ are two chains of prime ideals of R which cannot be refined further (i.e. by adding an extra prime ideal Pi (cid:12) I (cid:12) Pi+1 or Qi (cid:12) I (cid:12) Qi+1), we have that r = s. This is a "well-behavedness" condition on the classical Krull dimension of kG: it says that, whenever P (cid:12) P ′ are adjacent prime ideals and the height h(P ) of P is finite, then we have h(P ′) = h(P ) + 1. This result goes some way towards redressing the long-standing gap between Iwasawa algebras and similar algebraic objects; for instance, similar catenarity results had already been established for classical group rings of virtually poly- cyclic groups (in a special case in [17], in full generality in [11]), for universal enveloping algebras of finite-dimensional soluble Lie algebras over C [6]; for quantised coordinate rings over C [27], and over [26]; for q-commutative power series rings [22]; and so on. 3 In proving this result, we crucially use the prime extension theorem, [24, Theo- rem A]. Before we can state this, we need to recall a few concepts. Let G be a nilpotent-by-finite compact p-adic analytic group. Then [23, The- orem C] there exists a unique maximal subgroup H of G with the property that H is an open normal subgroup of G, H contains a finite normal subgroup F , and H/F is nilpotent p-valuable. This H is called the finite-by-(nilpotent p-valuable) radical of G, and is written FNp(G). Given a prime ideal P of kG, as in [2, §1.3] and [24, Introduction], we will write P † for the kernel of the natural composite map G → (kG)× → (kG/P )×. We say that P is faithful if P † = 1, and P is almost faithful if P † is finite. We now state the prime extension theorem: Theorem. [24, Theorem A] Take p > 2. Let G be a nilpotent-by-finite compact p-adic analytic group, and let k be a finite field. Write H = FNp(G). If P is an almost faithful prime ideal of kH, then the ideal P kG is a prime ideal of kG. We will use this result to generalise Ardakov's analogue of Zalesskii's theorem [2, Theorem 8.6] to our current context. Recall [23, Definition 1.4, Lemma 1.10] that a group G is orbitally sound if, whenever H is a subgroup of G with finitely many G-conjugates, and H ◦ is the largest normal subgroup of G contained in H, we have [H : H ◦] < ∞; and recall that nilpotent p-valuable groups are indeed orbitally sound [2, Proposition 5.9]. As in [23], we will write throughout this paper ∆(G) =(cid:8)x ∈ G(cid:12)(cid:12) [G : CG(x)] < ∞(cid:9), ∆+(G) =(cid:8)x ∈ ∆(cid:12)(cid:12) o(x) < ∞(cid:9), where o(x) denotes the order of x. We will also often simply write ∆ and ∆+ to denote ∆(G) and ∆+(G). For the basic properties of these (closed, characteristic) subgroups, see [23, Lemma 1.3 and Theorem D]. Following Roseblade [17], we say that a prime ideal P of kG is controlled by the normal subgroup H of G if the right ideal (P ∩ kH)kG is equal to P . Theorem B. Let G be a nilpotent-by-finite, orbitally sound compact p-adic analytic group. Suppose P is an almost faithful prime ideal of kG. Then P is controlled by ∆. The analogous classical result, for group algebras of polycyclic-by-finite groups, was proved by Roseblade [17, Corollary H3]. Taken together with the results of [1] and [25], this also gives a precise partial answer to a question of Ardakov and Brown [3, Question G]: when G is as in Theorem B, this completely describes the prime ideals of kG in terms of closed normal subgroups, central elements and prime ideals of (classical) group algebras of finite groups over k. 4 This is all we need to deduce that kG is catenary when G is nilpotent-by-finite and orbitally sound - see Theorem 3.12. In order to pass from orbitally sound to general nilpotent-by-finite groups, we partly develop the theory of vertices and sources along the lines of [12] [13]. Theorem C. Let G be a nilpotent-by-finite compact p-adic analytic group, P a prime ideal of kG, H an orbitally sound open normal subgroup of G, and Q a minimal prime ideal above P ∩ kH. Write N for the G-isolator [23, Definition 1.6] of Q†, and write ∇ for the subgroup of G containing N defined by ∇/N = ∆(G/N ). Then P is induced from an ideal L of k∇. For the precise meaning of induced here, see §1.2. Theorem A then follows from Theorems B and C by adapting an argument from [11], as follows. Let G be a (not necessarily orbitally sound) nilpotent- by-finite compact p-adic analytic group, k a finite field of characteristic p > 2, and P a faithful prime ideal of kG. We already know [23, Theorem A] that G contains an open normal orbitally sound subgroup, which we denote nio(G). From Theorem C, we may deduce Corollary 3.20: that P is induced from some proper open subgroup H of G containing nio(G). If H = nio(G), then we can deduce from Theorem B (as above) that kH is catenary, and now by Lemma 3.22 we are done. In general, nio(G) ≤ H < G, and we may not have equality: but it is easy to see that [H : nio(H)] < [G : nio(G)] < ∞, and Corollary 3.23 establishes Theorem A by induction on the index [G : nio(G)]. 5 1 Heights of primes and Krull dimension 1.1 Prime and G-prime ideals Definition 1.1. Let G be a compact p-adic analytic group [5, Definition 8.14]. Suppose the group G acts (continuously) on the ring R, and that the ideal I ✁ R is G-stable. Then, following [15, §14], we will say that I is G-prime if, whenever A, B ✁ R are G-stable ideals and AB ⊆ I, then either A ⊆ I or B ⊆ I. Lemma 1.2. Let G be a compact p-adic analytic group and H a closed normal subgroup. (i) If P is a prime ideal of kG, then P ∩ kH is a G-prime ideal of kH. If H is open in G, then P is a minimal prime ideal above (P ∩ kH)kG. (ii) Let Q be a G-prime ideal of kH, and P any minimal prime of kH above Q. Then Q =Tx∈G P x. Furthermore, the set of minimal primes of kG above Q is {P xx ∈ G}. Proof. (i) The former statement follows from [15, Lemma 14.1(i)], and the latter from [15, Theorem 16.2(i)]. (ii) This follows from [15, Lemma 14.2(i)(ii)]. Definition 1.3. Let P be a prime ideal of a ring R. Then we define the height of P to be the greatest integer h(P ) := r for which there exists a (finite) chain P0 (cid:12) P1 (cid:12) · · · (cid:12) Pr = P (†) of prime ideals in R (or ∞ if no such longest finite chain exists). Suppose instead that the group G acts on R by automorphisms, and P is a G-prime ideal of R. Then the G-height of P is the greatest integer hG(P ) := r for which there exists a chain (†) of G-prime ideals in R (or ∞). Finally, suppose that the group G acts on R by automorphisms, and P is a G-orbital prime ideal of R (i.e. a prime ideal of R with finite orbit under the conjugation action of G). Then the G-orbital height of P is the greatest integer horb G (P ) := r for which there exists a chain (†) of G-orbital prime ideals in R (or ∞). We note the following immediate consequence of the correspondence of Lemma 1.2: Corollary 1.4. Let G be a compact p-adic analytic group and H an open normal subgroup. Take P a prime ideal of kG, and let Q be a minimal prime of kH above P ∩ kH. Then h(P ) = hG(P ∩ kH) = h(Q). 6 1.2 Inducing ideals Definition 1.5. Let H be an open (not necessarily normal) subgroup of G, and let L be an ideal of kH. We define the induced ideal LG ✁ kG to be the largest (two-sided) ideal contained in the right ideal LkG ✁ kG. In other r words, by [11, 2.1], LG is the annihilator of kG/LkG as a right kG-module, or by [15, Lemma 14.4(ii)], LG = \g∈G LgkG. Lemma 1.6. Induction of ideals is transitive: if H and K are open subgroups of G with H ≤ K ≤ G, and L ✁ kH, then LG = (LK)G. Proof. Let N be an open normal subgroup of G contained in H, and write (·) to denote the quotient by N , so that we have kG = kN ∗ G with H ≤ K ≤ G, and we may view L as an ideal of kN ∗ H. The result now follows from [13, Lemma 1.2(iii)]. 1.3 Krull dimension We recall some facts about Krull dimension, used here in the sense of Gabriel and Rentschler: see [7, §15]. Definition 1.7. Let 0 6= M be an R-module, and fix some ordinal α. We define the following notation inductively: • Kdim(M ) = 0 if M is an Artinian module, • Kdim(M ) ≤ α if, for every descending chain M0 ≥ M1 ≥ M2 ≥ . . . of submodules of M , we have Kdim(Mi/Mi+1) < α for all but finitely many i. Of course, if there exists some α such that Kdim(M ) ≤ α, but we do not have Kdim(M ) ≤ β for any β < α, then we write Kdim(M ) = α. Remark. Kdim(M ) is a measure of complexity of the poset of submodules of M . Kdim(M ) may not be defined for some modules M -- that is, we may not have Kdim(M ) ≤ α for any ordinal α. However, if M is a noetherian module, then Kdim(M ) is defined [7, Lemma 15.3]. Definition 1.8. Suppose that Kdim(M ) = α. We say that M is α-homogeneous if Kdim(N ) = α for all nonzero submodules N of M . Examples 1.9. 7 (i) Nonzero Artinian modules are 0-homogeneous. (ii) Prime rings R, as modules over themselves, are α-homogeneous (where we set α equal to Kdim(RR)) [7, Exercise 15E]. (iii) The property of being α-homogeneous is inherited by products [7, Corol- lary 15.2] and (nonzero) submodules (by definition). We now cite and adapt some standard results on Krull dimension. Lemma 1.10. (i) [11, 1.4(ii)] Let the ring R be α-homogeneous as a right R-module. If x ∈ R satisfies Kdim(R/xR) < Kdim(R), then x is a regular element of R. (ii) [10, Th´eor`eme 5.3] Suppose B is a finite normalising extension of A, and let M be a B-module. Then Kdim(MB) exists if and only if Kdim(MA) does, and if so, then they are equal. (iii) [7, Exercise 15R] If R is a right noetherian subring of a ring S such that S is finitely generated as an R-module, and M is a finitely generated S-module, then Kdim(MS) ≤ Kdim(MR). Corollary 1.11. Suppose A ⊆ C ⊆ B are right noetherian rings, and B is a finite normalising extension of A. Let M be a finitely generated B-module. Then, if Kdim(MB) exists, we have Kdim(MA) = Kdim(MC) = Kdim(MB). Proof. This follows immediately from Lemma 1.10(ii) and two applications of Lemma 1.10(iii). Lemma 1.12. Let G be a compact p-adic analytic group, H an open subgroup of G, and k a field of characteristic p. Let M be a finitely generated kG-module. (i) Kdim(MkG) = Kdim(MkH ). (ii) Suppose that M = W kG for some submodule W of MkH . Then we have Kdim(MkG) = Kdim(WkH ). (iii) MkG is α-homogeneous if and only if MkH is α-homogeneous. Proof. (Adapted from [11, 1.4(iii)-(v)].) (i) Let N be the (open) largest normal subgroup of G contained in H, so that kG is a finite normalising extension of kN . Now apply Corollary 1.11. (ii) Let N be as in (i). Then, by (i), it suffices to prove that Kdim(MkN ) = Kdim(WkN ). But, as a kN -module, M is a finite sum of modules (W g)kN for various g ∈ G, and these are all isomorphic, so in particular have isomorphic submodule lattices and therefore the same Kdim. 8 (iii) It is clear from the definition that, if MkH is α-homogeneous, then MkG is α-homogeneous. Conversely, suppose that MkG is α-homogeneous, and let W be a nonzero submodule of MkH . Then (W kG)kG is a nonzero submodule of MkG, so has Krull dimension α by assumption, and hence also Kdim(WkH ) = α by (ii). Lemma 1.13. Let G be a finite group, H a subgroup, and R ∗ G a fixed crossed product. Fix a semiprime ideal I of R ∗ G. If R ∗ G/I is α-homogeneous, then R ∗ H/(I ∩ R ∗ H) is α-homogeneous. Proof. (Adapted from [4, Lemma 4.2(i)].) Let M be a nonzero right ideal of the ring R ∗ H/(I ∩ R ∗ H), and write β = Kdim(MR∗H ). We wish to show that β = α. M is a right module over both R ∗ H and R; and R ∗ G/I is a right module over both R ∗ G and R. As R ∗ G and R ∗ H are both finite normalising extensions of R, we may apply Lemma 1.10(ii) to both of these situations to see that β = Kdim(MR∗H ) = Kdim(MR) and α = Kdim((R ∗ G/I)R∗G) = Kdim((R ∗ G/I)R). Now, as right R-modules, we have R ∗ H/(I ∩ R ∗ H) ∼= (R ∗ H + I)/I ≤ R ∗ G/I, and so M is isomorphic to some nonzero R-submodule of R ∗ G/I. In particular, this means that β = Kdim(MR) ≤ Kdim((R ∗ G/I)R) = α. But now (R ∗ G/I)R is α-homogeneous by Corollary 1.11, so we must have β = α. Corollary 1.14. Let G be a compact p-adic analytic group, H be an open subgroup of G, and N the largest open normal subgroup of G contained in H. Take k to be a field of characteristic p, and let Q be a prime ideal of kH, I = QG ∩ kN , and α = Kdim(kH/Q). Then kH/Q, kG/QG, kG/IkG are all α-homogeneous rings. Proof. As we observed in Example 1.9(ii), kH/Q is already α-homogeneous, as it is prime of Krull dimension α. We know from Definition 1.5 that the ideal QG can be written asTg∈G QgkG, and that this intersection can be taken to be finite. Hence, as a right kG- module, kG/QG is isomorphic to a (nonzero) submodule of the direct prod- uct of the various (finitely many) kG/QgkG; and each kG/QgkG is gener- ated as a kG-module by kH g/Qg, which is ring-isomorphic to kH/Q. Hence Kdim(kG/QG) = Kdim(kH/Q) by Lemma 1.10(ii). 9 Finally, as QG =Tg∈G(QkG)g, we see that I = \g∈G (QkG)g ∩ kN = \g∈G (QkG ∩ kN )g = \g∈G (Q ∩ kN )g, and so, as above, kN/I is a (nonzero) subdirect product of the various kN/(Q∩kN )g, which are all ring-isomorphic to kN/Q ∩ kN ; now Lemma 1.13 implies that kN/Q ∩ kN is α-homogeneous, so kN/I is also, and kG/IkG is generated as a kG-module by kN/I, so finally kG/IkG also inherits this property. We borrow a result from the standard proof of Goldie's theorem. Lemma 1.15. [21, Lemma 3.13] Suppose R is a semiprime ring, satisfying the ascending chain condition on right annihilators of elements, and which does not contain an infinite direct sum of nonzero right ideals. If I is an essential right ideal of R (i.e. a right ideal that has nonzero intersection I ∩J with each nonzero right ideal J of R), then I contains a regular element. These hypotheses are satisfied when R is G-prime and noetherian, for example. Proposition 1.16. With notation as in Corollary 1.14, suppose P is a prime ideal of kG containing QG. If P is minimal over QG, then h(P ) = h(Q). Proof. First, set I = QG ∩ kN . This is a G-prime ideal contained in P ∩ kN . Suppose for contradiction that the inclusion I ⊆ P ∩ kN is strict. First, we will show that P ∩ kN/I is essential as a right ideal inside kN/I. Indeed, the left annihilator L in kN/I of P ∩ kN/I is a G-invariant ideal which annihilates the nonzero G-invariant ideal P ∩ kN/I, so we must have L = 0; and so, given any right ideal T of kN/I having zero intersection with P ∩ kN/I, as we must have T ≤ L, we conclude that T = 0. Hence, by Lemma 1.15, we may find an element c ∈ P ∩ kN ⊆ kN which is regular modulo I. As kG/IkG is a free kN/I-module, c may also be considered as an element of P ⊆ kG which is regular modulo IkG. Hence Kdim(cid:0)kG/(QG + ckG)(cid:1)kG ≤ Kdim(cid:0)kG/(IkG + ckG)(cid:1)kG < Kdim(kG/IkG)kG = Kdim(kG/QG)kG as IkG + ckG ⊆ QG + ckG by Lemma 1.10(i) by Corollary 1.14, which, again by Lemma 1.10(i), shows that c ∈ P is regular modulo QG. However, we may now deduce from a reduced rank argument that P cannot be minimal over QG, as follows. Write ρ for the reduced rank [7, §11, Definition] of a right module over the semiprime noetherian (hence Goldie) ring R = kG/QG, and write (·) for images under the map kG → R. Now, c ∈ P implies cR ⊆ P , and so by [7, Lemma 11.3] we have ρ(R/cR) ≥ ρ(R/P ) ≥ 0. Further, if c is a 10 regular element of R, then cR ∼= R as right R-modules, so ρ(R/cR) = 0, again by [7, Lemma 11.3]. But now [7, Exercise 11C] implies that P cannot be a minimal prime of R. This contradicts the assumption we made at the start of the proof, and so we have shown that P ∩ kN = QG ∩ kN . We observed during the proof of Corollary 1.14 that QG ∩ kN = \g∈G (Q ∩ kN )g. But Q is a prime ideal of kH, so Q ∩ kN is an H-prime ideal of kN , so may be written as Q ∩ kN = \h∈H Qh 0 for some prime ideal Q0 of kN . Combining these two shows that P ∩ kN = QG ∩ kN = \g∈G Qg 0. Now, by applying [15, corollary 16.8] to both P ∩ kN and Q ∩ kN , we have that h(P ) = h(Q0) = h(Q) as required. 2 Control theorem 2.1 The abelian case We will require some facts about prime ideals in power series rings. Lemma 2.1. Let A be a free abelian pro-p group of finite rank and B a closed isolated (normal) subgroup. Take k to be a field of characteristic p. Write SpecB(kA) for the set of primes of kA that are controlled by B. Then the maps SpecB(kA) ↔ Spec(kB) P 7→ P ∩ kB QkA 7→ Q are well-defined and mutual inverses, and preserve faithfulness. Proof. If P is a prime ideal of kA, then P ∩ kB is an A-prime ideal (and hence a prime ideal) of kB by Lemma 1.2(i). 11 Conversely, note that, as B is isolated in A, the quotient A/B is again free abelian pro-p; so we may write A = B ⊕ C, where the natural quotient map A → A/B induces an isomorphism A/B ∼= C. Now, if Q is a prime ideal of kB, then kA/QkA = (kB/Q)[[C]] is a power series ring with coefficients in the commutative domain kB/Q, and is hence itself a domain. It follows from [1, Lemma 5.1] that QkA ∩ kB = Q, and by assumption, if P is controlled by B then we already have (P ∩ kB)kA = P . Now suppose the prime ideals P ✁ kA and Q ✁ kB correspond under these maps. Then, again viewing A as B ⊕ C, we may similarly consider kA/P as the completed tensor product [25, Definition 2.3] kB/Q ⊗ kC. Then the map k A → (kA/P )× can be written as B ⊕ C → (kB/Q)× ⊕ (kC)× . (kB/Q ⊗ k kC)× (b, c) 7→ ((b + Q), c), so it is clear that P is faithful if and only if Q is faithful. Lemma 2.2. Let A, B, k be as in Lemma 2.1. Take two neighbouring prime ideals P (cid:12) Q of kA, and suppose B controls P . Then (i) h(P ) + dim(A/P ) = r(A), (ii) h(Q) = h(P ) + 1, (iii) h(P ) = h(P ∩ kB). Proof. (i) This follows from [18, Ch. VII, §10, Corollary 1]. (ii) This follows from [18, Ch. VII, §10, Corollary 2]. (iii) Under the correspondence of Lemma 2.1, any saturated chain of prime ideals 0 = Q0 (cid:12) Q1 (cid:12) · · · (cid:12) Qn = P ∩ kB of kB extends to a chain of prime ideals 0 = P0 (cid:12) P1 (cid:12) · · · (cid:12) Pn = P of kA. As any two saturated chains of prime ideals in kA have the same length [18, Ch. VII, §10, Theorem 34 and Corollary 1], we need only check that this chain is saturated. Take two adjacent prime ideals I1 (cid:12) I2 of kB, so that h(I2) = h(I1)+1 [18, Ch. VII, §10, Corollary 2] and I1kA (cid:12) I2kA are prime. We will show that I1kA and I2kA are adjacent by showing that their heights also differ by 1. By performing induction on r(A/B), it will suffice to prove this for the case r(A/B) = 1, i.e. kA = kB[[X]]. It is clear that, when R is a commutative ring, dim(R[[X]]) ≥ 1 + dim(R) (where dim denotes the classical Krull dimension). But, giving R[[X]] the (X)-adic filtration, we see that gr(R[[X]]) ∼= R[x]. By [14, 6.5.6], we have 12 dim(R[[X]]) ≤ dim(gr(R[[X]])) = dim(R[x]) = 1 + dim(R), where this last equality follows from [14, 6.5.4(i)]. Hence, for any prime ideal I, we have dim(kA/IkA) − dim(kB/I) = dim((kB/I)[[X]]) − dim(kB/I) = 1. But, from (i), we see that dim(kA/IkA) = r(A) − h(IkA), dim(kB/I) = r(B) − h(I), and hence we conclude that h(I) = h(IkA). Setting I = I1, I2 now shows that h(I2kA) = h(I1kA) + 1 as required. 2.2 Faithful primes are controlled by ∆ As in [23], if P is a prime ideal of some completed group ring kG, we will write P † := ker(G → (kG/P )×), and say that P is faithful if P † = 1 and P is almost faithful if P † is finite. Fix a prime p, which will be arbitrary until otherwise stated. Recall the control theorem of Ardakov [2, 8.6]: Theorem 2.3. Let G be a nilpotent p-valued group of finite rank with centre Z. (i) If p is a prime ideal of kZ, then pkG is a prime ideal of kG. (ii) If P is a faithful prime ideal of kG, then P is controlled by Z. Proof. This is [2, 8.4, 8.6]. Lemma 2.4. Let G be finite-by-(nilpotent p-valuable), i.e. G = FNp(G). Then Z(G/∆+) = ∆/∆+. Proof. Given x ∈ G, the two conditions [G/∆+ : CG/∆+(x∆+)] < ∞ and [G : CG(x)] < ∞ are equivalent, as ∆+ is finite; this shows that we have ∆(G/∆+) = ∆/∆+. Take some x ∈ ∆, so that x satisfies this condition: then, given arbitrary g ∈ G, there exists some k such that gpk ∆+ ∈ CG/∆+ (x∆+), so that (x−1gx)pk ∆+, and it now follows from [9, III, 2.1.4] that x−1gx∆+ = g∆+. This shows that ∆/∆+ ≤ Z(G/∆+). Conversely, we must have Z(G/∆+) ≤ ∆(G/∆+) by definition. ∆+ = gpk We extend Theorem 2.3 to: Proposition 2.5. Let G be a finite-by-(nilpotent p-valuable) group and k a finite field of characteristic p. 13 (i) If p is a G-prime ideal of k∆, then pkG is a prime ideal of kG. (ii) If P is an almost faithful prime ideal of kG, then P is controlled by ∆. Proof. Adopt the notation of [25, Lemma 1.1 and Notation 1.2]. Let e ∈ cpik∆+ (p), and write f = eG. To prove (i), it suffices to prove that the ideal f · pkG ✁ f ·kG is prime. But, by the Matrix Units Lemma [25, Lemma 6.1], we have an iso- morphism f · kG ∼= Ms(e · kG1), where G1 is the stabiliser in G of e, and under which f · pkG 7→ Ms(e · p1kG1) for some G1-prime ideal p1 of k[[∆ ∩ G1]]. So, by Morita equivalence, it will suffice to show that the ideal e · p1kG1 ✁ e · kG1 is prime. Now recall from [25, Theorems A and C] that we have an isomorphism ψ : e · kG1 ∼= Mt(k′[[G1/∆+]]) under which e·p1kG1 7→ qk′[[G1/∆+]] for a (G1/∆+)-prime ideal q of k′[[∆∩G1/∆+]]. Hence we need now only show that qk′N ✁ k′N is prime, where N = G1/∆+. Note that, as G1 is open in G, we have ∆(G1) = ∆∩G1 [23, Lemma 1.3(ii)]; and from Lemma 2.4, ∆(G1)/∆+ = Z(G1/∆+). Hence, still writing N = G1/∆+, we see that q is an N -prime ideal of k′[[Z(N )]], and hence a prime ideal. But now qk′N is prime by Theorem 2.3(i). This establishes part (i) of the proposition. To show part (ii), take an almost faithful prime ideal P of kG. We would like to show that P is a minimal prime ideal above (P ∩ k∆)kG. But this is clearly true when ∆+ = 1 by Theorem 2.3; and in the general case, another application of the Matrix Units Lemma [25, Lemma 6.1] and [25, Theorems A and C], as above, reduces to the case ∆+ = 1. Hence, finally, we need only show that (P ∩ k∆)kG is prime; but P ∩ k∆ is a G-prime ideal of k∆ (again by Lemma 1.2(i)), so we are done by part (i) of the proposition. Until the end of this section, we will write (−)◦ to meanTg∈G(−)g. Corollary 2.6. Let G be a finite-by-(nilpotent p-valuable) group, and H an open normal subgroup of G containing ∆. Let k be a finite field of characteristic p. If P is an almost faithful G-prime ideal of kH, then P kG is a prime ideal of kG. Proof. Take a minimal prime Q of kH above P . Then we have Q◦ = P , so Q† is finite (as G is orbitally sound [23, Definition 1.4, Corollary 2.4]). Hence Q is controlled by ∆, by Proposition 2.5(ii), and by applying (−)◦ to both sides of the equality Q = (Q ∩ k∆)kH, we see that P is also: P = (P ∩ k∆)kH. In particular P kG = (P ∩ k∆)kG. But now Proposition 2.5(i) shows that (P ∩ k∆)kG is prime. 14 For the following results, we need to assume that p > 2 in order to be able to invoke [24, Theorem A]. Proposition 2.7. Let G be a nilpotent-by-finite, orbitally sound compact p- adic analytic group, and k a finite field of characteristic p > 2. Let H = FNp(G). If P is an almost faithful prime ideal of kG, then P is controlled by H. Proof. Let Q be a minimal prime ideal of kH above P ∩kH. Then (Q†)◦ = P †∩H is finite, so, as G is orbitally sound, Q† is also finite. By [15, Corollary 14.8], in order to prove that (P ∩ kH)kG is prime, it suffices to show that QkS is prime, where S is the stabiliser in G of Q. Let T = FNp(S). As H is a finite-by-(nilpotent p-valuable) open normal sub- group of S, we see that H must be an open normal subgroup of T . It is also clear that ∆(H) = ∆(T ) = ∆(S) = ∆(G) [23, Lemma 1.3(ii) and Theorem C]. Now, by Corollary 2.6, QkT must be prime; and we have that (QkT )† is finite. Now, by the prime extension theorem [24, Theorem A], (QkT )kS = QkS is prime. Lemma 2.8. Let G be a nilpotent-by-finite compact p-adic analytic group, and let H ≥ K be any two closed normal subgroups of G. Take P to be a prime ideal of kG. Let Q be a minimal prime ideal of kH above P ∩ kH. If P is controlled by H and Q is controlled by K, then P is controlled by K. Proof. By Lemma 1.2(ii), we have Q◦ = P ∩ kH, and so (P ∩ kK)kG = ((P ∩ kH) ∩ kK)kG = (Q◦ ∩ kK)kG = (Q ∩ kK)◦kG = ((Q ∩ kK)kH)◦kG = Q◦kG = (P ∩ kH)kG = P as K is normal in G as H is normal in G as Q is controlled by K as P is controlled by H. Now back to: Theorem 2.9. Let G be a nilpotent-by-finite, orbitally sound compact p-adic analytic group, k a finite field of characteristic p > 2, and P an almost faithful prime ideal of kG. Then P is controlled by ∆. Proof. Proposition 2.7 shows that P is controlled by H. Let Q be a minimal prime of kH above P ∩ kH: then Q◦ = P ∩ kH by Lemma 1.2(ii), so we see that (Q†)◦ = P † ∩ H is finite, so (as G is orbitally sound) Q† must also be finite. Hence, as Q is almost faithful, Proposition 2.5(ii) shows that it is controlled by ∆. Now Lemma 2.8 applies. 15 2.3 Primes adjacent to faithful primes We begin with a property of the "finite-by-(nilpotent p-valuable) radical" oper- ator. Lemma 2.10. Let G be a nilpotent-by-finite, orbitally sound compact p-adic analytic group, let N be a normal subgroup of G which is contained in ∆, and let F be a finite normal subgroup of G. Then the following three statements hold. (i) FNp(G/F ) = FNp(G)/F . (ii) Suppose that FNp(G/i∆(N )) = FNp(G)/i∆(N ). Then FNp(G/N ) = FNp(G)/N . (iii) Suppose N is ∆-isolated. Then we have either FNp(G/N ) = FNp(G)/N or N = ∆ = FNp(G). Proof. (i) This is clear from the construction of FNp(G) (see [23, Definition 5.3]). (ii) First, note that FNp(G)/N is a quotient of a finite-by-(nilpotent p-valuable) normal subgroup of G, and hence is still a finite-by-(nilpotent p-valuable) normal subgroup of G/N , i.e. FNp(G)/N ≤ FNp(G/N ). As both of these are of finite index in G, it will suffice to show that these indices are equal. Consider the natural surjection α : G/N → G/i∆(N ). We can see that ker α = i∆(N )/N = ∆+(∆/N ) ≤ ∆+(G/N ) ≤ FNp(G/N ) is a finite normal subgroup of G/N , and hence from (i) we see that FNp(G/N ) i∆(N )/N ∼= FNp(G/i∆(N )). That is, the restricted map αFNp(G/N ) : FNp(G/N ) → FNp(G/i∆(N )) is also surjective with kernel i∆(N )/N . Hence we have the following com- mutative diagram, in which the first two rows are exact, all three columns are exact, and C1 and C2 are the cokernels of the vertical maps. 16 1 1 1 1 1 1 / i∆(N )/N / FNp(G/N ) FNp(G/i∆(N )) / i∆(N )/N G/N G/i∆(N ) / 1 1 C1 1 C2 1 1 1 1 By the Nine Lemma [8, Chapter XII, Lemma 3.4], the third row is now also exact, so that C1 ∼= C2. But by assumption, C2 ∼= G/FNp(G), and hence [G/N : FNp(G/N )] = C1 = [G : FNp(G)] = [G/N, FNp(G)/N ], as required. (iii) Case 1. First, assume that ∆+ = 1. If G = FNp(G), then we clearly have FNp(G/N ) = FNp(G)/N for any Write H = FNp(G), and bH for the preimage of bH/N = FNp(G/N ). closed normal subgroup N . So suppose that H (cid:12) bH ≤ G, and take some z ∈ bH \ H. Now conjugation by z induces the automorphism x 7→ xζ on H/H ′ (where H ′ denotes the isolated derived subgroup), and hence also on H/H ′N , for some ζ ∈ t(Z× p ) [23, Lemma 4.2] satisfying ζ 6= 1 [24, Lemma 3.3]. If H/H ′N has nonzero rank, we may take an element x ∈ H whose image H = iH (H ′N ). in H/H ′N has infinite order; and now the image in bH/H ′N of hx, zi is not finite-by-nilpotent, contradicting the definition of bH. So we must have In particular, this implies that H = iH (H ′Z), where Z = Z(H) = ∆(G), and so, by [24, Lemma 3.5], we see that H is abelian, i.e. H = ∆. Further- more, this implies that H ′ = 1, and as N is already H-isolated (because ∆ is H-isolated), the equality H = iH (H ′N ) simplifies to give H = N . This is what we wanted to prove. Case 2. Now suppose instead that ∆+ 6= 1. As N is isolated in G, we see that -- ∆+ ≤ N , and N/∆+ is isolated normal inside G/∆+, contained in ∆/∆+; 17 / / / / / / / / / / / / / / / / / / / / -- ∆+(G/∆+) = 1; -- ∆(G/∆+) = ∆/∆+ = Z(FNp(G)/∆+); -- FNp(G/∆+) = FNp(G)/∆+; and so the result follows by applying Case 1 to G/∆+. Remark. If G is a compact p-adic analytic group, H is a closed normal subgroup, and Q is a G-stable ideal of kH, then Q† = (Q + 1) ∩ H is normal in G. Lemma 2.11. Let G be a nilpotent-by-finite, orbitally sound compact p-adic analytic group, and let k be a finite field of characteristic p > 2. If Q is a G- prime ideal of k∆, and FNp(G/Q†) = FNp(G)/Q†, then QkG is a prime ideal of kG. Remark. The hypothesis FNp(G/Q†) = FNp(G)/Q† (‡) has the following consequence. Let G be a nilpotent-by-finite, orbitally sound compact p-adic analytic group, k a finite field of characteristic p > 2, and let P (cid:12) P ′ be adjacent prime ideals of kG, with P almost faithful. Then P is controlled by ∆, by Theorem 2.9. Set Q := P ′ ∩ k∆. Consider i∆(Q†): if this is not equal to ∆, then by Lemma 2.10(ii), (iii), the hypothesis (‡) is satisfied. So suppose it is equal to ∆. Now, as Q contains the ideal ker(kG → k[[G/Q†]]) (the augmentation ideal of Q†), if we further have that FNp(G) = ∆, then kG/Q is a finite prime ring, which is therefore simple, and so Q must be a maximal ideal of kG of iG(∆) = G; otherwise, we again have (‡) by Lemma 2.10(ii), (iii). That is, under these conditions, we always have (‡) unless Q is a maximal ideal of kG and G is virtually abelian, in which case Q† is open in G. Proof. Write H = FNp(G). I above Q. Suppose the G-orbit of I splits into distinct H-orbits O1, . . . , Or, i=1 Pi = Q. In particular, since Pi is an H-prime of k∆, we have that PikH is prime by Proposition 2.5(i). As Q is a G-prime, we may write it asTg∈G I g for some minimal prime ideal and write Pi := TA∈Oi It remains to show that(cid:16)Tg∈G(PikH)g(cid:17) kG is prime. By [15, Corollary 14.8], A. Then Pi is an H-prime of k∆, and Tr it suffices to show that PikS is prime, where S = StabG(Pi). Write p = PikH, and note that p† = P † i ≤ ∆. Now, if FNp(G)/∆+ is non- abelian, we have FNp(S/p†) = FNp(S)/p†. If, on the other hand, FNp(G)/∆+ is abelian, then we must have Q† (cid:12) ∆, and as Q† is H-isolated orbital, we have i )g, we must have i , so that in particular [∆ : p†] = ∞. Hence again we have [∆ : Q†] = ∞. But as G is orbitally sound, and Q† =Tg∈G(P † that Q† is open in P † FNp(S/p†) = FNp(S)/p†. 18 Write (·) for the quotient map S → S/p†. Now, to show that PikS = pkS is prime, we need only show that pkS is prime. But p is a faithful prime ideal of kH, and H = FNp(S), so by [24, Theorem A], we are done. Lemma 2.12. Let k be a finite field of characteristic p > 2. Let G be a nilpotent-by-finite, orbitally sound compact p-adic analytic group, and let P (cid:12) Q be adjacent prime ideals of kG, with P almost faithful. Suppose that Q is not a maximal ideal of kG. Then Q is controlled by ∆. Proof. Q ∩ k∆ is a G-prime of k∆, and so (Q ∩ k∆)kG is prime by Lemma 2.11 and the accompanying remark. But P = (P ∩ k∆)kG ≤ (Q ∩ k∆)kG ≤ Q, (with the equality as a result of Theorem 2.9), and P and Q are adjacent, so (Q ∩ k∆)kG must equal either P or Q. Let us assume for contradiction that (Q ∩ k∆)kG = P . Then we must have P ∩ k∆ ≤ Q ∩ k∆ ≤ (Q ∩ k∆)kG = P, and by intersecting each of these with k∆, we see that P ∩ k∆ = Q ∩ k∆. In particular, by taking (·)† of both sides of this equality, we see that Q† ∩ ∆ is finite (as P is almost faithful). Let N be an open normal nilpotent p-valued subgroup of G, and let Z = Z(N ). By [23, Lemma 1.3(ii)], Z = ∆(N ) is a finite-index torsion-free subgroup of ∆, and so Q† ∩ Z = 1. Now, as N is nilpotent and the normal subgroup Q† ∩ N has trivial intersection with its centre, [16, 5.2.1] implies that Q† ∩ N = 1, and hence Q† must be a finite normal subgroup of G. So Q† ≤ ∆+, and in particular Q† = Q† ∩ ∆, which we earlier determined is finite. Hence Q is almost faithful, and must be controlled by ∆ by Theorem 2.9. In particular, we must have P ∩ k∆ 6= Q ∩ k∆. But this contradicts our assumption. 3 Catenarity 3.1 The orbitally sound case: plinths and a height func- tion Much of the material in this subsection is adapted from [17]. Unless stated otherwise, throughout this section, G is an arbitrary compact p-adic analytic group, and k is a finite field of characteristic p. We start by outlining our plan of attack: Lemma 3.1. Let R be a ring in which every prime ideal has finite height. Suppose we are given a function h : Spec(R) → N satisfying 19 • h(P ) = 0 whenever P is a minimal prime of R, • h(P ′) = h(P ) + 1 for each pair of adjacent primes P (cid:12) P ′ of R. Then R is a catenary ring. Proof. Obvious. Lemma 3.2. kG has finite classical Krull dimension, i.e. the maximal length of any chain of prime ideals is bounded. Proof. The classical Krull dimension of kG is bounded above by Kdim(kG) by [14, Lemma 6.4.5], which is equal to Kdim(FpG) by [14, Proposition 6.6.16(ii)], and this is bounded above by the dimension (in the sense of [5, Theorem 8.36]) of G, which is finite by definition (see the remarks after [5, Definition 3.12]). Definition 3.3. Let V be a QpG-module, and suppose it has finite dimension as a vector space over Qp. Take a chain 0 = V0 (cid:12) V1 (cid:12) · · · (cid:12) Vr = V of G -- orbital subspaces -- that is, Qp-vector subspaces of V with finitely many G-conjugates, or equivalently Qp-vector subspaces that are QpN -submodules for some open subgroup N of G. Assume further that this chain is saturated, in the sense that it cannot be made longer by the addition of some G -- orbital subspace Vi (cid:12) V ′ (cid:12) Vi+1. Such a chain is necessarily finite, as it is bounded above in length by dimQp (V ) + 1. We call the number r the G -- plinth length of V , written pG(V ). If pG(V ) = 1, we say that V is a plinth for G. Remark. The number r is independent of the Vi chosen. Indeed, fix a longest possible chain 0 = V0 (cid:12) V1 (cid:12) · · · (cid:12) Vr = V of G-orbital subspaces, and let G0 be the intersection of the normalisers NG(Vi), i.e. the largest subgroup of G such that each Vi is a QpG0-module. G0 is open in G. Now, given any chain 0 = W0 (cid:12) W1 (cid:12) · · · (cid:12) Ws = V of G-orbital subspaces, take H0 =Ts j=1 NG(Wj ), and note that G0 ∩ H0 is a finite-index open subgroup of G that normalises each Vi and Wj . Hence, by the Jordan-Holder theorem [7, Theorem 4.11], the chain Wj may be refined to a chain of length r; so if the chain Wj is saturated, then s = r. Definition 3.4. A G-group is a topological group H endowed with a continuous action of G. For example, closed subgroups of G, and quotients of G by closed normal subgroups of G, are G-groups under the action of conjugation. Let H be a nilpotent-by-finite compact p-adic analytic group with a continuous action of G. We aim to define pG(H). In fact, as plinths are insensitive to 20 finite factors, we may immediately replace H by the open subgroup formed by the intersection of the (finitely many) G-conjugates of any given open normal nilpotent uniform subgroup of H. Then there is a series 1 = H0 ✁ H1 ✁ · · · ✁ Hn = H (1) of G-subgroups such that Ai = Hi/Hi−1 is abelian for each i = 1, . . . , n. Let Qp for each i = 1, . . . , n, with G -- action given by conjugation. In this Vi = Ai ⊗ Zp case, we define pG(H) = pG(Vi). nXi=1 Lemma 3.5. pG(H) is well-defined, and does not depend on the series (1). Proof. Apply the Jordan-Holder theorem, as in the remark above. For our purposes, the most important property of pG is that it is additive on short exact sequences of G-groups, which also follows from the Jordan-Holder theorem. We record this as: Lemma 3.6. Suppose that 1 act sequence of G-groups. Then pG(A) + pG(C) = pG(B). / B / A / C / 1 is a short ex- We now define Roseblade's function λ. (Later, we will show that, in the case when G is nilpotent-by-finite and orbitally sound, λ is actually equal to the height function on Spec(kG).) Definition 3.7. λ(P ) =(cid:26) pG(P †) + λ(P π) P † 6= 1 hG(P ∩ k∆) P † = 1, where P π is the image of P under the map π : kG → kG/(P † − 1)kG ∼= k[[G/P †]]. This definition is recursive, in that if P is an unfaithful prime ideal, then λ(P ) is defined with reference to λ(P π); but P π is then a faithful prime ideal of kGπ, so this process terminates after at most two steps. We make the following remark on this definition immediately: Lemma 3.8. Let G be a nilpotent-by-finite, orbitally sound compact p-adic analytic group, k a finite field of characteristic p > 2, and P a faithful prime of k∆. Then λ(P ) = h(P ). Proof. λ(P ) is defined to be hG(P ∩ k∆). But, by Theorem 2.9 and Lemma 2.11, we see that there is a one-to-one, inclusion-preserving correspondence be- tween faithful prime ideals of kG and faithful G-prime ideals of k∆, so that hG(P ∩ k∆) = h(P ). 21 / / / / We return to the general case of G an arbitrary compact p-adic analytic group. Lemma 3.9. Let P (cid:12) Q be neighbouring prime ideals of kG, and write π : kG → kG/(P † − 1)kG ∼= k[[G/P †]]. Then λ(Q) − λ(P ) = λ(Qπ) − λ(P π). Proof. Firstly, as P ≤ Q, we have P † ≤ Q†, so the map ρ : kG → k[[G/Q†]] factors as kG π k[[G/P †]] σ / / k[[G/Q†]]. ρ We now compute λ(Q) − λ(P ) using Definition 3.7: λ(Q) − λ(P ) = pG(Q†) − pG(P †) + λ(Qρ) − λ(P π) = pG((Q†)π) + λ(Qρ) − λ(P π) = pG((Q†)π) + λ(Qπσ) − λ(P π) = pG((Qπ)†) + λ((Qπ)σ) − λ(P π) = λ(Qπ) − λ(P π) by Lemma 3.6 by definition of ρ as (Q†)π = (Qπ)† by Definition 3.7. Remark. Suppose G is a nilpotent-by-finite compact p-adic analytic group, and suppose we are given a subquotient A of G which is a plinth, with G- action induced from the conjugation action of G on itself. Then it is easy Qp) = 1. (Roseblade calls such plinths centric.) In- to see that dimQp (A ⊗ Zp deed, suppose A = H/K, where H and K are closed normal subgroups of G with K contained in H. Then we may replace G by an open normal nilpo- tent uniform subgroup G′, and A by A′ = H ′/K ′, where H ′ = H ∩ G′ and K ′ = iH ′ (K ∩ G′); after doing this, we still have that A′ is a plinth for G′, and Qp). But, as G′/K ′ is nilpotent, and A′ is a that dimQp (A ⊗ Zp Qp) = dimQp (A′ ⊗ Zp non-trivial normal subgroup, A′ must meet the centre Z(G′/K ′) non-trivially; and as A′ is torsion-free, we must have that A′ ∩ Z(G′/K ′) is a plinth for G′, and so must be equal to A′. Hence G′ centralises A′, and its plinth length is simply equal to its rank. Again, we will write (−)◦ to meanTg∈G(−)g. Lemma 3.10. Let G be a nilpotent-by-finite compact p-adic analytic group. Let U be a G-prime of k∆, and write ρ : k∆ → k[[∆/U †]]. Then h(U ) = hG(U ρ)+pG(U †). Proof. Let A = Z(∆), and let U1 be a minimal prime of kA above U ∩kA, so that U ∩ kA = U ◦ 1 ). 1 . Then hG(U ) = h(U1) by Corollary 1.4, and so hG(U ρ) = h(U ρ 22 / / 3 3 Now, from Lemma 2.2(i), we have h(U1)+dim(kA/U1) = r(A) and h(U ρ from which we may deduce that 1 )+dim(kA/U1) = r(Aρ), h(U1) = h(U ρ 1 ) + r(A) − r(Aρ). But r(A) − r(Aρ) = r(A ∩ ker ρ) = pG(U † ∩ A) by the above remark. Now this is just pG(U †), as A is open in ∆. Lemma 3.11. Let G be arbitrary compact p-adic analytic. Let H be a closed normal subgroup of G, and let K be an open subgroup of H which is normal in G. If P is a G-prime ideal of kH, then hG(P ) = hG(P ∩ kK). Proof. [Adapted from [17, Lemma 29].] We know that P = Q◦ for some prime Q of kH, and Q∩kK =Th∈H V h for some prime V of kK. Hence P ∩kK = V ◦. G for the height function on G-orbital primes, Then, writing horb hG(P ) = horb = horb = hG(P ∩ kK) G (Q) G (V ) by Lemma 1.2(ii) by Lemma 1.2(i) by Lemma 1.2(ii). Here, we deduce from Theorem 2.9 and [17, proof of theorem H2] the following corollary: Theorem 3.12. Let G be a nilpotent-by-finite, orbitally sound compact p-adic analytic group, and k a finite field of characteristic p > 2. Then kG is a catenary ring. Proof. Let P (cid:12) Q be neighbouring prime ideals of kG. We will first show that λ(Q) = λ(P ) + 1. By passing to k[[G/P †]], we may assume that P is a faithful prime ideal, by Lemma 3.9. Hence, by Theorem 2.9 (and as p > 2), we have that (P ∩k∆)kG = P . We also have either that (Q ∩ k∆)kG = Q, by Lemma 2.12, or Q† ≥ ∆ by the remark of Lemma 2.11; and so, in either case, we have P ∩ k∆ (cid:12) Q ∩ k∆. We will now show that P ∩ k∆ and Q ∩ k∆ are neighbouring G-primes of k∆. Suppose that they are not: then there must be a G-prime J strictly between them, i.e. P ∩ k∆ (cid:12) J (cid:12) Q ∩ k∆. Then, again by Lemma 2.11 and the remark made there, we see that JkG is a prime ideal of kG. Now it is clear that P ≤ JkG ≤ Q by the previous paragraph; and if JkG = Q, then intersecting both sides with k∆ shows that J = Q ∩ k∆, and likewise if JkG = P . Hence we must have P (cid:12) JkG (cid:12) Q, so that P and Q are not neighbouring primes. But this contradicts our initial assumptions. So we conclude that hG(Q ∩ k∆) = hG(P ∩ k∆) + 1. The right hand side is, by definition, just equal to λ(P ) + 1; and we have λ(Q) = λ(Qρ) + pG(Q†), where ρ : G → G/Q†. It remains to show that this is equal to hG(Q ∩ k∆). 23 Case 1. Q† is not open in G. Then Q is controlled by ∆ by Lemma 2.12 and the remark of Lemma 2.11, and so Qρ is controlled by ∆ρ, and in particular by iGρ(∆ρ) ≤ iGρ(∆(Gρ)). Write A = Z(∆(Gρ)) and B = A ∩ iGρ(∆ρ): as Qρ is controlled by iGρ(∆ρ), we have that Qρ ∩ kA is controlled by B. Furthermore, we can write Qρ ∩ kA = q ◦ for some prime q of kA, so that q is also controlled by B, and hence λ(Qρ) = hG(Qρ ∩ k[[∆(Gρ)]]) = hG(Qρ ∩ kA) = h(q) = h(q ∩ kB) = hG(Qρ ∩ kB) = hG(Qρ ∩ k[[iGρ(∆ρ)]]) = hG(Qρ ∩ k∆ρ) by definition by Lemma 3.11 by Corollary 1.4 by Lemma 2.2(iii) by Corollary 1.4 by Lemma 3.11 by Lemma 3.11. We also have pG(Q†) = pG((Q ∩ k∆)†). Hence λ(Q) = hG((Q ∩ k∆)ρ) + pG((Q ∩ k∆)†). Now we are done by Lemma 3.10. Case 2. Q† is open in G. We have already seen that this case only occurs when G = iG(∆), and so λ(Qρ) = λ(0) = 0, and pG(Q†) = pG(G), and hG(Q ∩ k∆) = hG(Q ∩ kA) = r(A). These are clearly equal, as A is open in G. In order to invoke Lemma 3.1, it remains only to show that λ(P ) = 0 when P is a minimal prime. But as all minimal primes are induced from ∆+, this follows immediately from the definition of λ: we will have P † ≤ ∆+ (and hence pG(P †) = 0), and P π ∩ k∆π will be a minimal G-prime of k∆π (and hence hG(P π ∩ k∆π) = 0). 3.2 Vertices and sources We now study a more general setting. Let G be an arbitrary compact p-adic analytic group, and P an arbitrary prime ideal of kG. Remark. Suppose G is orbitally sound and nilpotent-by-finite, N is a closed nor- mal subgroup of G, and I is a prime ideal of kG with N ≤ I † and [I † : N ] < ∞. Writing (·) for the natural map kG → k[[G/N ]], it is clear that the prime ideal I ✁k[[G/N ]] is almost faithful, and so, by Theorem 2.9, is controlled by ∆(G/N ), and that I is the complete preimage in kG of I, and is therefore controlled by the preimage in G of ∆(G/N ). This motivates the following definition: 24 Definition 3.13. Let I be an ideal of kG, and N a closed subgroup of G. We say that I is almost faithful mod N if I † contains N as a subgroup of finite index. We also write ∇G(N ) for the subgroup of NG(N ) defined by ∇G(N )/N = ∆(NG(N )/N ). Diagrammatically: G NG(N ) NG(N )/N ∇G(N ) ∆(NG(N )/N ) N/N N 1 We will extend this notion to ideals I with I † contained in N as a subgroup of finite index. Lemma 3.14. Let H be an open subgroup of N . Then there exists an open characteristic subgroup M of N contained in H. Proof. (Adapted from [15, 19.2].) Let [N : H] = n < ∞. Now, as N is topologi- cally finitely generated, there are only finitely many continuous homomorphisms N → Sn. where Sn is the symmetric group. Take M to be the intersection of the kernels of these homomorphisms. Lemma 3.15. Let N be a closed subgroup of G, and A = NG(N ). Suppose I is an ideal of kA, and I † ≤ N with [N : I †] < ∞. Then there is a closed normal subgroup M of A such that I is almost faithful mod M . Furthermore, this M can be chosen so that ∇G(N ) = ∇A(M ). Proof. Set H = I † in Lemma 3.14: then the subgroup M is characteristic in N , hence normal in A; M contains I †; and M is open in N , so we must have [I † : M ] < ∞. By definition, we have ∇G(N ) = ∇A(N ). Now, N/M is a finite normal subgroup of A/M , so is contained in ∆+(A/M ). Hence the preimage under the natural quotient map A/M → A/N of ∆(A/N ) is ∆(A/M ). But this is the same as saying that ∇A(N ) = ∇A(M ). 25 When G is a general compact p-adic analytic group, we will use the following lemma to translate between prime ideals of kG and prime ideals of kA for certain open subgroups A of G. Lemma 3.16. Let H be an open normal subgroup of G. Suppose P is a prime of kG, and write Q for a minimal prime of kH above P ∩ kH. Let B be the stabiliser in G of Q, and let A be any open subgroup of G containing B, so that H ≤ B ≤ A ≤ G. Proof. This follows from [15, 14.10(i)]. Then there is a prime ideal T of kA with P = T G, and furthermore this T Definition 3.17. A prime P ✁ kG is standard if it is controlled by ∆ and we satisfies T ∩ kH =Ta∈A Qa. have P ∩ k∆ =Tx∈G Lx for some almost faithful prime L ✁ k∆. prime of kH above P , so that P ∩ kH =Tx∈G Qx. If Q is a standard prime, Lemma 3.18. Let G be a nilpotent-by-finite compact p-adic analytic group and H an open normal subgroup. Let P be a prime ideal of kG, and Q a minimal then P is a standard prime. Proof. (Adapted from [15, 20.4(i)].) Write ∆ = ∆(G), ∆H = ∆(H), and P ∩ k∆ = \x∈G Sx and Q ∩ k∆H = \y∈H T y, for prime ideals S ✁ k∆ and T ✁ k∆H . On the one hand, P ∩ k∆H = (P ∩ k∆) ∩ k∆H Sx! ∩ k∆H (S ∩ k∆H )x , = \x∈G = \x∈G but on the other hand, P ∩ k∆H = (P ∩ kH) ∩ k∆H Qx! ∩ k∆H (Q ∩ k∆H )x = \x∈G = \x∈G = \x∈G T x. 26 Now, the conjugation action of G on ∆H has kernel CG(∆H ), which contains CG(∆) by [23, Lemma 1.3(ii)]. But CG(∆) =T CG(a), where the intersection runs over a set of topological generators a for ∆, and each CG(a) is open in G by definition of ∆. Now, as ∆ is topologically finally generated, we see that CG(∆) and hence CG(∆H ) are also open in G. That is, the conjugation action of G on ∆H factors through the finite group G/CG(∆H ), and hence the intersections above are finite, so that (by the pri- mality of T ), we have S ∩ k∆H ⊆ T x for some x ∈ G. Now, by assumption, Q is standard, so T is almost faithful. This means that S† ∩ ∆H ⊆ (T †)x is a finite group, and so, since [∆ : ∆H ] < ∞, we have that S† is also finite, so S is almost faithful. It remains to show that S◦kG = P . By Lemma 2.11, we see that S◦kG = P ′ is a prime ideal of kG contained in P . Now, (P ∩ kH)kG =\g∈G Qg kG =\g∈G \h∈H T hkH!g kG =\g∈G T g kG = (P ∩ k∆H )kG ⊆ (P ∩ k∆)kG = P ′ ⊆ P, by calculation above and as H is open and normal in G, we know from Lemma 1.2(i) that P is a minimal prime above (P ∩ kH)kG, so that P = P ′. Finally, the main theorem of this subsection: Theorem 3.19. Let G be a nilpotent-by-finite compact p-adic analytic group, P a prime ideal of kG, H an orbitally sound open normal subgroup of G, Q a minimal prime ideal above P ∩ kH, and N = iG(Q†). Then there exists an ideal L ✁ k[[∇G(N )]] with P = LG. Remark. The subgroup N is a vertex of the prime ideal P , and the ideal L is a source of P corresponding to the vertex N . 27 Proof. We follow the proof of [13, 2.3], as reproduced in [15, 20.5]. Trivially, H stabilises Q, i.e. H ≤ B := StabG(Q); and B normalises Q†. Set N := iG(Q†). Now we must have NG(Q†) ≤ A := NG(N ): indeed, if x ∈ G normalises Q†, then it permutes the (finitely many) closed orbital subgroups K of G containing Q† as an open subgroup, and hence it normalises N , which is generated by those K [23, Definition 1.6]. We are in the following situation: G A B NG(N ) StabG(Q) ⑦ ⑦ ⑦ ⑦ ⑦ ⑦ ⑦ ⑦ H orbitally sound ⑦ ⑦ ⑦ ⑦ ⑦ ⑦ ⑦ ⑦ ⑦ ⑦ ⑦ ∇G(N ) iG(Q†) N ❖❖❖❖❖❖❖❖❖❖❖❖❖❖❖ Q† Now, Lemma 3.16 shows that there is a prime ideal T of kA with P = T G and T ∩ kH =Ta∈A Qa. It will suffice to show the existence of a prime ideal L of k[[∇G(N )]] with T = LA, by Lemma 1.6. Let M be an open characteristic subgroup of N contained in Q†, whose existence is guaranteed by Lemma 3.14. Write ∇ = ∇G(N ), which we know is equal to ∇A(M ) by Lemma 3.15, and denote by (·) images under the natural map kA → k[[A/M ]]. Now Q is a prime ideal of kH with M ≤ Q† an open subgroup, so Q is an almost faithful prime ideal of kH; hence, as H is orbitally sound [23, Lemma 1.5(ii)], we see that Q is a standard prime of kH. But T ∩ kH = Ta∈A Qa clearly implies T ∩ kH = Ta∈A Q , by the modular law. Now Lemma 3.18 implies that T is also a standard prime ideal of kA: that . Lifting is, there is an almost faithful prime ideal L of k[[∆(A)]] with T = L this back to kA, we see that we have an almost faithful mod M prime ideal L of k∇ with T = LA as required. A a We end this subsection with an important application of this theorem. Recall 28 the definition of nio(G) from [23, Definition 2.5]. Corollary 3.20. Suppose G is a nilpotent-by-finite compact p-adic analytic group which is not orbitally sound. Let P be a faithful prime ideal of kG. Then P is induced from some proper open subgroup of G containing nio(G). Proof. Write H = nio(G). H is orbitally sound by [23, Theorem 2.6(ii)]. Let Q be a minimal prime ideal above P ∩ kH, so that N = iG(Q†) is a vertex for P by Theorem 3.19. Then P is induced from ∇G(N ), which is contained in NG(N ), and so P is induced from NG(N ) itself by Lemma 1.6. But, as nio(G) is orbitally sound, in particular it must normalise N [23, Theorem 2.6(i)]. Hence, if NG(N ) is a proper subgroup of G, we are done. Suppose instead that NG(N ) = G, i.e. that iG(Q†) is a normal subgroup of G. Then, for each g ∈ G, (Q†)g is a finite-index subgroup of iG(Q†) [23, Proposition 1.7]; and Q† is orbital in G, so there are only finitely many (Q†)g, and their intersection (Q†)◦ must also have finite index in iG(Q†). But (Q†)◦ = P † = 1, so in particular we have iG(Q†) = ∆+, and hence P is induced from ∇G(N ) = ∆, again by Theorem 3.19. Hence, as nio(G) contains ∆, P must be induced from nio(G) itself. 3.3 The general case: inducing from open subgroups Now we will proceed to show that kG is catenary. Lemma 3.21. Let H be an open subgroup of G, and P a prime ideal of kG. Suppose Q is an ideal of kH maximal amongst those ideals A of kH with AG ⊆ P . Then Q is prime, and P is a minimal prime ideal above QG. Proof. Suppose I and J are ideals strictly containing Q: then, by the maximality of Q, we see that I G and J G must strictly contain P . Hence I GJ G ⊆ (IJ)G [15, Lemma 14.5] strictly contains P , and so IJ strictly contains Q. Hence Q is prime. P is clearly a prime ideal containing QG, so to show it is minimal it suffices to find any ideal A of kH with P a minimal prime above AG. Let N be the normal core of H in G, and take A = (P ∩ kN )H : then by Lemma 1.6 we have AG = (P ∩ kN )G = (P ∩ kN )kG, and by Lemma 1.2(i), P is a minimal prime above this. Lemma 3.22. Let H be an open subgroup of G with kH catenary. If P (cid:12) P ′ are adjacent primes of kG, and P is induced from kH, then h(P ′) = h(P ) + 1. Proof. (Adapted from [11, 3.3].) Choose an ideal Q (resp. Q′) of kH which is maximal amongst those ideals A of kH with AG ⊆ P (resp. AG ⊆ P ′). Then Q and Q′ are prime, and P (resp. P ′) is a minimal prime ideal over QG (resp. 29 Q′G), by Lemma 3.21. Hence, by Proposition 1.16, we see that it suffices to show that h(Q′) = h(Q) + 1. Suppose not. Then there exists some prime ideal I of kH with Q (cid:12) I (cid:12) Q′; and we may choose a prime ideal J of kG which is minimal over I G. Then P ≤ J ≤ P ′. But h(Q) < h(I) < h(Q′) implies (by another application of Proposition 1.16) that h(P ) < h(J) < h(P ′), contradicting our assumption that P and P ′ were adjacent primes. Corollary 3.23. Let G be a nilpotent-by-finite compact p-adic analytic group, and k a finite field of characteristic p > 2. Then kG is a catenary ring. Proof. (Adapted from [11, 3.3].) Take two adjacent prime ideals P (cid:12) Q of kG, and assume without loss of generality that P is faithful. We proceed by induction on the index [G : nio(G)]. When this index equals 1, we are already done by Theorem 3.12, so suppose not. Then Corollary 3.20 implies that P is induced from some proper open subgroup H of G containing nio(G). As nio(G) is an orbitally sound open normal subgroup of H, it must be contained in nio(H) (by the maximality of nio(H)), and so we have [H : nio(H)] < [G : nio(G)]. By induction, kH is catenary, so we may now invoke Lemma 3.22 to show that h(Q) = h(P ) + 1. References [1] K. Ardakov. Localisation at augmentation ideals in Iwasawa algebras. Glas- gow Mathematical Journal, 48(2):251 -- 267, 2006. [2] K. Ardakov. Prime ideals in nilpotent Iwasawa algebras. Inventiones math- ematicae, 190(2):439 -- 503, 2012. [3] K. Ardakov, K.A. Brown. Primeness, semiprimeness and localisation in Iwasawa algebras. Trans. Amer. Math. Soc., 359:1499 -- 1515, 2007. [4] K.A. Brown. The structure of modules over polycyclic groups. Math. Proc. Cambridge Philos. Soc., 89:257 -- 283, 1981. [5] J.D. Dixon, M.P.F. du Sautoy, A. Mann and D. Segal. Analytic Pro-p Groups. Cambridge University Press, 1999. [6] O. Gabber. Equidimensionalit´e de la vari´et´e caract´eristique. Universit´e de Paris VI. [7] K.R. Goodearl, R.B. Warfield, Jr. An Introduction to Noncommutative Noetherian Rings. Cambridge University Press, 2004. [8] S. Mac Lane. Homology. Springer, 1963. [9] Michel Lazard. Groupes analytiques p-adiques. Publications Math´ematiques de l'IH ´ES, 26:5 -- 219, 1965. 30 [10] B. Lemonnier. Dimension de krull et cod´eviation. application au th´eor`eme d'eakin. Communications in Algebra, 6:16:1647 -- 1665, 1978. [11] M. Lorenz, E. Letzter. Polycyclic-by-finite group algebras are catenary. Math. Res. Lett., 6:183 -- 194, 1999. [12] M. Lorenz. Prime ideals in group algebras of polycyclic-by-finite groups: vertices and sources. S´eminaire d'alg`ebre P. Dubreil et M. P. Malliavin, Lect. Notes in Mathematics, 867:406 -- 420, 1981. [13] M. Lorenz and D. S. Passman. Prime ideals in group algebras of polycyclic- by-finite groups. Proc. London Math. Soc., 43:520 -- 543, 1981. [14] J.C. McConnell and J.C. Robson. Noncommutative Noetherian Rings. American Mathematical Society, 2001. [15] Donald S. Passman. Infinite Crossed Products. Academic Press Inc., 1989. [16] Derek J.S. Robinson. A course in the theory of groups. Springer-Verlag New York, 1996. [17] J.E. Roseblade. Prime ideals in group rings of polycyclic groups. Proc. London Math. Soc., 36:385 -- 447, 1978. [18] O. Zariski, P. Samuel. Commutative Algebra. D. Van Nostrand Company, 1965. [19] P. Schneider and J. Teitelbaum. Banach space representations and iwasawa theory. Israel J. Math., 127:359 -- 380, 2002. [20] P. Schneider and J. Teitelbaum. Locally analytic distributions and p-adic representation theory, with applications to gl2. J. AMS, 15:443 -- 468, 2002. [21] R.G. Swan. Goldie's theorem. http://math.uchicago.edu/~swan/expo/Goldie.pdf. [22] Edward S. Letzter, Linhong Wang. Prime ideals of q-commutative power series rings. Algebr. Represent. Theor., 14:1003 -- 1023, 2011. [23] W. Woods. On the structure of virtually nilpotent compact p-adic analytic groups. Preprint, arXiv:1608.03137 [math.GR], 2016. To appear in J. Group Theory. [24] W. Woods. Extensions of almost faithful prime ideals in virtually nilpotent mod-p Iwasawa algebras. Preprint, arXiv:1610.03740 [math.RA], 2016. [25] W. Woods. Maximal prime homomorphic images of mod-p Iwasawa alge- bras. Preprint, arXiv:1608.07755 [math.RT], 2016. [26] Milen Yakimov. On the spectra of quantum groups. In Memoirs of the American Mathematical Society, volume 229 (5). American Mathematical Society, 2014. [27] K.R. Goodearl, J.J. Zhang. Homological properties of quantized coordinate rings of semisimple groups. Proc. London Math. Soc., 94:647 -- 671, 2007. 31
1501.04697	2	1501	2016-08-30T21:43:59	Strong shift equivalence and the generalized spectral conjecture for nonnegative matrices	[ "math.RA", "math.DS" ]	We show that the weak and strong forms of the Generalized Spectral Conjecture (GSC) of Boyle and Handelman are equivalent. The GSC asserts that well understood necessary spectral conditions on a square matrix A over a subring S of the reals are sufficient for that matrix to be shift equivalent over S (in the weak form) or strong shift equivalent over S (in the strong form) to a primitive matrix over S. The foundation of this work is the recent result that the group NK_1(S) of algebraic K-theory exactly captures the refinement of shift equivalence over S by strong shift equivalence over S. The GSC remains open in general even in the case that S equals the real numbers.	math.RA	math	STRONG SHIFT EQUIVALENCE AND THE GENERALIZED SPECTRAL CONJECTURE FOR NONNEGATIVE MATRICES 6 1 0 2 MIKE BOYLE AND SCOTT SCHMIEDING Dedicated to Hans Schneider, in memoriam g u A 0 3 ] . A R h t a m [ 2 v 7 9 6 4 0 . 1 0 5 1 : v i X r a Abstract. Given matrices A and B shift equivalent over a dense subring R of R, with A primitive, we show that B is strong shift equivalent over R to a primitive matrix. This result shows that the weak form of the Generalized Spectral Conjecture for primitive matrices implies the strong form. The foundation of this work is the recent result that for any ring R, the group NK1(R) of algebraic K-theory classifies the refinement of shift equivalence by strong shift equivalence for matrices over R. Contents Introduction 1. 2. Shift equivalence and strong shift equivalence 3. Proof of Theorem 1.1 4. Reflections on the Generalized Spectral Conjecture References Appendix A. Correction 1 4 6 12 12 14 1. Introduction The purpose of this paper 1 is to prove the following theorem and explain its context. Theorem 1.1. Suppose R is a dense subring of R, A is a primitive matrix over R and B is a matrix over R which is shift equivalent over R to A. Then B is strong shift equivalent over R to a primitive matrix. 2010 Mathematics Subject Classification. Primary 15B48; Secondary 37B10. Key words and phrases. nonnegative matrix; spectra; shift equivalence; spectral conjecture. 1 This paper is an outgrowth of the paper [2], for which its authors were awarded (along with Robert Thompson) the second Hans Schneider Prize. 1 2 MIKE BOYLE AND SCOTT SCHMIEDING We begin with the context. By ring, we mean a ring with 1; by a semiring, we mean a semiring containing {0, 1}. A primitive matrix is a square matrix which is nonnegative (meaning entrywise nonnegative) such that for some k > 0 its kth power is a positive matrix. Definitions and more background for shift equivalence (SE) and strong shift equivalence (SSE) are given in Section 2. We recall the Spectral Conjecture for primitive matrices from [2]. In the statement, ∆ = (d1, . . . , dk) is a k-tuple of nonzero complex numbers. ∆ is the nonzero spectrum of a matrix A if A has characteristic polynomial of the form χA(t) = tmQ1≤i≤k(t − di). ∆ has a Perron value if there exists i such that di > dj when j 6= i. The trace of ∆ is tr(∆) = d1 + · · · + dk. ∆n denotes ((d1)n, . . . , (dk)n), the tuple of nth powers; and the nth net trace of ∆ is trn(∆) = Xdn µ(n/d)tr(∆d) in which µ is the Mobius function (µ(1) = 1; µ(n) = (−1)r if n is the product of r distinct primes; µ(n) = 0 if n is divisible by the square of a prime). Spectral Conjecture 1.2. [2] Let R be a subring of R. Then ∆ is the nonzero spectrum of some primitive matrix over R if and only if the following conditions hold: (1) ∆ has a Perron value. i=1(t − di) lie in R. (2) The coefficients of the polynomial Qk (3) If R = Z, then for all positive integers n, trn(∆) ≥ 0; if R 6= Z, then for all positive integers n and k, (i) tr(∆n) ≥ 0 and (ii) tr(∆n) > 0 implies tr(∆nk) > 0. It is not difficult to check that the nonzero spectrum of a primitive matrix satisfies the three conditions [2]. (We remark, following [8] it is known that the nonzero spectra of symmetric primitive matrices cannot possibly have such a simple characterization.) To understand the possible spectra of nonnegative matrices is a classical problem of linear algebra (for early background see e.g. [2]) on which interesting progress continues (see e.g. [7, 14, 15, 13] and their references). Understanding the nonzero spectra of primitive matrices is a variant of this problem and also an approach to it: to know the minimal size of a primitive matrix with a prescribed nonzero spectrum is to solve the classical problem (for details, see [2]); and it is in the primitive case that the Perron-Frobenius constraints manifest most simply. Finally, as the spectra of matrices over various subrings of R appear in applications, in which the nonzero part of the spectrum is sometimes the relevant part [1, 2], it is natural to consider the nonzero spectra of matrices over arbitrary subrings of R. The Spectral Conjecture has been proved in enough cases that it seems almost certain to be true in general. For example, it is true under any of the following conditions: • The Perron value of Λ is in R (this always holds when R = R) or is a quadratic SSE AND GSC 3 integer over R [2]. • tr(Λ) > 0 [2, Appendix 4] • R = Z or Q [11]. The general proofs in [2] do not give even remotely effective general bounds on the size of a primitive matrix realizing a given nonzero spectrum. The methods used in [11] for the case R = Z are much more tractable but still very complicated. However, there is now an elegant construction of Tom Laffey [14] which proves the conjecture for R = R in the central special case of positive trace, and in some other cases; where it applies, the construction provides meaningful bounds on the size of the realizing matrix in terms of the spectral gap. The nonzero spectrum of a matrix is a "stable" or "eventual" invariant of a matrix. For a matrix over a field, an obvious finer invariant is the isomorphism class of the non- nilpotent part of its action as a linear transformation. The classification of matrices over a field by this invariant is the same as the classification up to shift equivalence over the field; for matrices over general rings, from the module viewpoint (see Sec.2), shift equivalence is the natural generalization of the isomorphism class of this nonnilpotent linear transformation. For some rings, an even finer invariant is the strong shift equiv- alence class. The Generalized Spectral Conjecture of Boyle and Handelman (in both forms below) heuristically is saying that only the obvious necessary spectral conditions constrain the eventual algebra of a primitive matrix over a subring of R, regardless of the subring under consideration. Generalized Spectral Conjecture (weak form, 1991) 1.3. Suppose R is a sub- ring of R and A is a square matrix over R whose nonzero spectrum satisfies the three necessary conditions of the Spectral Conjecture. Then A is SE over R to a primitive matrix. Generalized Spectral Conjecture (strong form, 1993) 1.4. Suppose R is a subring of R and A is a square matrix over R whose nonzero spectrum satisfies the three necessary conditions of the Spectral Conjecture. Then A is SSE over R to a primitive matrix. The weak form was stated in [2, p.253] and [3, p.124]. The strong form was stated in [1, Sec. 8.4]), along with an explicit admission that the authors of the conjecture did not know if the conjectures were equivalent (not knowing if shift equivalence over a ring implies strong shift equivalence over it). Following [5] (see Theorem 2.1), we know now that the strong form of the Generalized Spectral Conjecture was not a vacuous generalization: there are subrings of R over which SE does not imply SSE (Example 4 MIKE BOYLE AND SCOTT SCHMIEDING Note! 3.5). The results of [5] also provide enough structure that we can prove Theorem 1.1, which shows that the two forms of the Generalized Spectral Conjecture are equivalent. In contrast to the statement of the Generalized Spectral Conjecture for primitive matrices, it is not the case that the existence of a strong shift equivalence over R from a matrix A over R to a nonnegative matrix can in general be characterized by a spectral condition on A. There are dense subrings of R over which there are nilpotent matrices which are not SSE to nonnegative matrices (Remark 3.6). There is some motivation from symbolic dynamics for pursuing the zero trace case of the GSC. The Kim-Roush and Wagoner primitive matrix counterexamples [10, 21] to Williams' conjecture SE-Z+ =⇒ SSE-Z+ rely absolutely on certain zero-positive patterns of traces of powers of the given matrix. We still do not know whether the refinement of SE-Z+ by SSE-Z+ is algorithmically undecidable or (at another extreme) if it allows some finite description involving such sign patterns. We are looking for any related insight. 2. Shift equivalence and strong shift equivalence Suppose R is a subset of a semiring and R contains {0, 1}. (For example, R could be Z, Z+, {0, 1}, R, R+, . . . ) Square matrices A, B over R (not necessarily of the same size) are elementary strong shift equivalent over R (ESSE-R) if there exist matrices U, V over R such that A = UV and B = V U. Matrices A, B are strong shift equivalent over R (SSE-R) if there are a positive integer ℓ (the lag of the given SSE) and matrices A = A0, A1, . . . , Aℓ = B such that Ai−1 and Ai are ESSE-R, for 1 ≤ i ≤ ℓ. For matrices over a subring of R, the relation ESSE-R is never transitive. For example, if matrices A, B are ESSE over R, j > 1 and Aj 6= 0, then Bj−1 6= 0; but if A is the n × n matrix such that A(i, i + 1) = 0 for 1 ≤ i < n and A = 0 otherwise, then A is SSE-R to (0). Over any ring R, the relation SSE-R on square matrices is generated by similarity over R (U −1AU ∼ A) and nilpotent extensions, ( A X 0 0 ) ∼ A ∼ ( 0 X 0 A ) [17]. Square matrices A, B over R are shift equivalent over R (SE-R) if there exist a positive integer ℓ and matrices U, V over R such that the following hold: Aℓ = UV AU = UB Bℓ = V U BV = V A . Herem ℓ is the lag of the given SE. It is always the case that SSE over R implies SE over R: from a given lag ℓ SSE one easily creates a lag ℓ SE [23]. For certain semirings R, including above all R = Z+, the relations of SSE and SE over R are significant for symbolic dynamics. The relations were introduced by Williams for the cases R = Z+ and R = {0, 1} to study the classification of shifts of finite type. Matrices over Z+ are SSE over Z+ if and only if they define topologically conjugate shifts of finite type. SSE AND GSC 5 However, the relation SSE-Z+ to this day remains mysterious and is not even know to be decidable. In contrast, SE-Z+ is a tractable, decidable, useful and very strong invariant of SSE-Z+. Suppose now R is a ring, and A is n×n over R. T o see the shift equivalence relation SE-R more conceptually, recall that the direct limit GA of Rn under the R-module homomorphism x 7→ Ax is the set of equivalence classes [x, k] for x ∈ Rn, k ∈ Z+ under the equivalence relation [x, k] ∼ [y, j] if there exists ℓ > 0 such that Aj+ℓx = Ak+ℓy. GA has a well defined group structure ([x, k] + [y, j] = [Akx + Ajy, j + k]) and is an R- module (r : [x, k] 7→ [xr, k]). A induces an R-module isomorphism A : [x, k] 7→ [Ax, k] with inverse [x, k] 7→ [x, k + 1]. GA becomes an R[t] module (also an R[t, t−1] module) with t : [x, k] 7→ [x, k + 1]. A and B are SE-R if and only if these R[t]-modules are isomorphic (equivalently, if and only if they are isomorphic as R[t, t−1] modules). If the square matrix A is n × n, then I − tA defines a homomorphism Rn → Rn by the usual multiplication v 7→ (I − tA)v, and cok(I − tA) is an R[t]-module which is isomorphic to the R[t]-module GA. For more detail and references on these relations (by no means original to us) see [5, 16]. Williams introduced SE and SSE in the 1973 paper [23]. For any principal ideal domain R, Effros showed SE-R implies SSE-R in the 1981 monograph [6] (see [24] for Williams' proof for the case R = Z). In the 1993 paper [3], Boyle and Handelman extended this result to the case that R is a Dedekind domain (or, a little more generally, a Prufer domain). Otherwise, the relationship of SE and SSE of matrices over a ring remained open until the recent paper [5], which explains the relationship in general as follows. Theorem 2.1. [5] Suppose A, B are SE over a ring R. (1) There is a nilpotent matrix N over R such that B is SSE over R to the matrix A ⊕ N = (cid:18)A 0 0 N(cid:19). (2) The map [I − tN] → [A ⊕ N]SSE induces a bijection from NK1(R) to the set of SSE classes of matrices over R which are in the SE-R class of A. We will say just a little now about NK1(R), a group of great importance in algebraic K-theory; for more background, we have found [18, 19, 22] very helpful. NK1(R) is the kernel of the map K1(R[t]) → K1(R) induced by the ring homomorphism R[t] → R which sends t to 0. The finite matrix I − tN corresponds to the matrix I − (tN)∞ in the group GL(R[t]) (with I denoting the N × N identity matrix and (tN)∞ the N × N matrix which agrees with tN in an upper left corner and is otherwise zero). Every class of NK1(R) contains a matrix of the form I − (tN)∞ with N nilpotent over R. NK1(R) is trivial for many rings (e.g., any field, or more generally any left regular Noetherian 6 MIKE BOYLE AND SCOTT SCHMIEDING ring) but not for all rings. If NK1(R) is not trivial, then it is not finitely generated as a group. From the established theory, it is easy to give an example of a subring R of R for which NK1(R) is not trivial (Example 3.5). 3. Proof of Theorem 1.1 Proof of Theorem 1.1. Given a square matrix M over R, let λM denote its spectral radius and define the matrix M by M(i, j) = M(i, j). By Theorem 2.1, let N be a nilpotent matrix such that B is SSE over R to the matrix (cid:18)A 0 0 N(cid:19) . Suppose M is a matrix SSE over R to N and M also satisfies the following conditions: (1) λ3M < λA (2) For all positive integers n, trace(3Mn) ≤ trace(An) . (3) For all positive integers n and k, if tr(3Mn) < tr(An), then tr(3Mnk) < tr(Ank). Then by the Submatrix Theorem (Theorem 3.1 of [2]), there is a primitive matrix C SSE over R to A such that 3M is a proper principal submatrix of C. Without loss of generality, let this submatrix occupy the upper left corner of C. Define M0 to be the matrix of size matching C which is M in its upper left corner and which is zero in other entries. Then B is SSE over R to the matrix (cid:18)C 0 0 M0(cid:19). Choose ǫ ∈ R such that 1/3 < ǫ < 2/3 and compute 0 (cid:18)I −ǫI I (cid:19)(cid:18)C 0 I(cid:19)(cid:18)C ǫ(C − M0) 0 M0(cid:19)(cid:18)I (cid:19)(cid:18) I M0 0 0 (cid:18)I 0 I ǫI 0 I (cid:19) = (cid:18)C ǫ(C − M0) (cid:19) I(cid:19) = (cid:18) (1 − ǫ)C + ǫM0 (1 − ǫ)(C − M0) M0 0 ǫ(C − M0) ǫC + (1 − ǫ)M0(cid:19) := G . −I The matrix G is SSE over R to B, and it is nonnegative. The diagonal blocks have positive entries wherever C does; because C is primitive, there is a j > 0 such that C j > 0, and therefore the diagonal blocks of Gj are also positive. Because neither offdiagonal block of G is the zero block, it follows that G is primitive. So, it suffices to find M SSE over R to N satisfying the conditions (1)-(3) above. Choose K such that tr(Ak) > 0 for all k > K. Let n be the integer such that N is n × n, and let J by the integer provided by Proposition 3.4 given n and K. Given this J, choose ǫ > 0 such that for any J × J matrix M with M∞ < ǫ, we have λ3M < λA and for k > K we also have tr(3Mk) < tr(Ak). Now let δ > 0 be as provided by Proposition 3.4 for this ǫ. SSE AND GSC 7 If we can now find an n × n nilpotent matrix N ′ which is SSE over R to N and satisfies N ′ < δ, then we can apply Proposition 3.4 to this N ′ to produce a matrix M SSE over R to N and with M < ǫ and with tr(M k) = 0 for 1 ≤ k ≤ K. This matrix M will satisfy the conditions (1)-(3). Pick γ > 0 such that γN∞ < δ. There is a matrix U in SL(n, R) such that U −1NU = γN. The matrix U is a product of basic elementary matrices over R, and these can be approximated arbitrarily closely by basic elementary matrices over R. Consequently there is a matrix V in SL(n, R) such that V −1NV ∞ < δ. Choose N ′ = V −1NV . (cid:3) To prove the Proposition 3.4 on which the proof of Theorem 1.1 depends, we use a correspondence proved in [5]. We need some definitions. Given a finite matrix A, let A∞ denote the N×N matrix which has A as its upper left corner and is otherwise zero. In any N×N matrix, I denotes the infinite identity matrix. Given a ring R, El(R) is the group of N × N matrices over R[t], equal to the infinite identity matrix except in finitely many entries, which are products of basic elementary matrices (these basic matrices are by definition equal to I except perhaps in a single offdiagonal entry). For finite matrices A, B, the matrices I −A∞ and I −B∞ are El(R[t]) equivalent if there are matrices U, V in El(R[t]) such that U(I − A∞)V = I − B∞. Definition 3.1. Given a finite matrix A over tR[t], choose n ∈ N and k ∈ N such that A1, . . . Ak are n × n matrices over R such that A∞ = k Xi=1 ti(Ai)∞ and define a finite matrix A♯ = A♯(k,n) over R by the following block form, in which every block is n × n: A♯ = A1 A2 A3 0 I 0 0 0 I . . . . . . 0 0 0 0 0 I 0 . . . 0 0   . . . Ak−2 Ak−1 Ak 0 . . . 0 . . . . . . 0 . . . . . . 0 . . . . . . 0 0 0 0 . . . I 0 0 0 0 . . . 0 I .   In the definition, there is some freedom in the choice of A♯: k can be increased by using zero matrices, and n can be increased by filling additional entries of the Ai with zero. These choices do not affect the SSE-R class of A♯. 8 MIKE BOYLE AND SCOTT SCHMIEDING Theorem 3.2. [5] Let R be a ring. Then there is a bijection between the following sets: • the set of El(R[t]) equivalence classes of N × N matrices I − A∞ such that A is a finite matrix over tR[t] • the set of SSE-R classes of square matrices over R. The bijection from El(R[t]) equivalence classes to SSE-R classes is induced by the map I − A∞ 7→ A♯. The inverse map (from the set of SSE-R classes) is induced by the map sending A over R to the N × N matrix I − tA. By the degree of a matrix with polynomial entries we mean the maximum degree of its entries. If M is a matrix over R[t], with entries M(i, j) = Pi,j,k mijktk, then we define M = maxk>0 maxi,j mijk. If M is a matrix over R, with M(i, j) = mij, then M∞ is the usual sup norm, M∞ = maxi,j mij. d over tkR[t], with entries aij = P1≤r≤d a(r) Lemma 3.3. Suppose R is a dense subring of R and A is an n × n matrix of degree 4n2 . Then there is an n × n matrix B over tk+1R[t] such that I − A∞ is El(R[t]) equivalent to I − B∞ and the following hold: ij tr. Suppose Pn ii = 0 and A ≤ 1 i=1 a(k) (1) degree(B) ≤ degree(A) + 3k. (2) B ≤ 4n3A . Proof. For finite square matrices I − C and I − D, we use I − C ∼ I − D to denote elementary equivalence over R[t] of I − C∞ and I − D∞. We have I − A = ∼     1 − a11 −a12 −a21 1 − a22 ... −an1 −an2 1 − a11 −a12 −a21 1 − a22 ... −an1 0 −an2 0 · · · −a1n · · · −a2n . . . · · · ... 1 − ann   · · · −a1n · · · −a2n . . . · · · · · · a(k) 11 tk a(k) 22 tk ... 1 − ann a(k) nn tk 1 0 ... := I − A1 .   In order, apply the following elementary operations: (1) For 1 ≤ j ≤ n, add column n + 1 to column j of I − A1, to produce a matrix I − A2. Then degree(A2) = degree(A); the diagonal entries of A2 lie in tk+1R[t]; and A2 ≤ 2A1 = 2A. Every entry in row n + 1 of I − A2 equals 1. (By definition these entries have no impact on A2.) SSE AND GSC 9 (2) For 1 ≤ i ≤ n, add (-1)(row i) of (I − A2) to row n + 1 to form I − A3. Then the entries of A3 lie in tkR[t], and the diagonal entries of A3 still lie in ii = 0 . We have A3 ≤ nA2 ≤ 2nA < 1 and tk+1R[t], since Pn i=1 a(k) degree(A3) ≤ degree(A) . (3) For 1 ≤ i ≤ n, add (−a(k) block form, ii tk)(row n + 1) of (I − A3) to row i to form I − A4. In I − A4 = (cid:18)I − A5 0 1(cid:19) x in which A5 is n × n and x = (x1 · · · xn). Adding multiples of column n + 1 to columns 1, . . . , n to clear out x, we see I − A5 ∼ I − A. We have degree(A5) ≤ degree(A) + k and A5 ≤ A3 + (A)(A3) ≤ 2A3 ≤ 4nA < 1 . In A5, the diagonal terms lie in tk+1R[t] and the offdiagonal terms lie in tkR[t]. In the next two steps, we apply elementary operations to clear the degree k terms outside the diagonal. We use part of a clearing algorithm from [9]. (4) Let bij be the coefficient of tk in A5(i, j). For 2 ≤ i ≤ n, add (−b1jtk)(row j) to row 1. Continuing in order for rows i = 2, . . . , n − 1: for i + 1 ≤ j ≤ n, add (−bijtk)(row j) to row i. Let (I − A6) be the resulting matrix. The entries of A6 on and above the diagonal lie in tk+1R[t]. We have degree(A6) ≤ degree(A5) + k ≤ degree(A) + 2k and A6 ≤ A5 + (n − 1)A52 ≤ nA5 ≤ 4n2A ≤ 1 . (5) Let cij denote the coefficient of tk in A6(i, j). For 2 ≤ j ≤ n, add (−cj1tk)(column j) of A6 to column 1. Continuing in order for columns i = 2, . . . , n − 1: for i + 1 ≤ j ≤ n, add (−cji)(column j) to column i. For the resulting matrix (I − B), the entries of B lie in tk+1R[t], with degree(B) ≤ degree(A6) + k ≤ degree(A) + 3k and B ≤ A6 + (n − 1)A62 ≤ nA6 ≤ 4n3A . (cid:3) 10 MIKE BOYLE AND SCOTT SCHMIEDING Proposition 3.4. Suppose R is a dense subring of R, n ∈ N and K ∈ N. Then there is a J in N such that for any ǫ > 0 there exists δ > 0 such that the following holds: if N is a nilpotent n × n matrix over R and N∞ < δ, then there is a J × J matrix M over R such that (1) M is SSE over R to N, (2) tr (Mk) = 0 for 1 ≤ k ≤ K, and (3) M∞ < ǫ . Proof. Because N is nilpotent, tr(N k) = 0 for all positive integers k. Set B0 = tN. We define matrices B1, . . . , BK recursively, letting I − Bk+1 be the matrix I − B provided by Lemma 3.3 from input I − A = I − Bk. The conditions of the lemma are satisfied recursively, because the (zero) trace of the kth power of the nilpotent matrix (Bk)♯ ii . The matrix BK is n × n with must be (in the terminology of the lemma) Pi a(k) entries of degree at most d := 1 + 3(1) + 3(2) + · · · + 3(K) = 1 + 3K(K + 1)/2 . Let (BK)i be the matrices, 1 ≤ i ≤ d, such that BK = Pd i=1(BK)iti . Define M to be the matrix (BK)♯, an nd × nd matrix over R which is SSE over R to N. Set J = nd. It is now clear from condition (2) of Lemma 3.3 and induction that given ǫ > 0, there is a δ > 0 such that N < δ implies (BK) < ǫ. (We are not trying to optimize estimates.) With K > 1 (without loss of generality), we have BK = (BK)♯∞. This finishes the proof. (cid:3) Example 3.5. There are subrings of R with nontrivial NK1. For example, let R = Q[t2, t3, z, z−1]. By the Bass-Heller-Swan Theorem (see [19], 3.2.22) for any ring S, there is a splitting K1(S[z, z−1]) ∼= K1(S) ⊕ K0(S) ⊕ NK1(S) ⊕ NK1(S), which implies NK1(S[z, z−1]) always contains a copy of NK0(S). An elementary argument (see for example exercise 3.2.24 in [19]) shows that NK0(Q[t2, t3]) 6= 0, so NK1(Q[t2, t3, z, z−1]) is non-zero. Since Q[t2, t3, z, z−1] can be realized as a subring of R (by an embedding sending t, z to algebraically independent transcendentals in R) this provides an exam- ple of a subring R of R for which NK1(R) is not zero, and therefore shift equivalence over R does not imply strong shift equivalence over R. It is possible to produce explicit examples by tracking through the exact sequences behind the argument of the last paragraph. This is done in [20], and for R = Q[t2, t3, z, z−1] yields the following matrix over R[s], I − M = (cid:18) 1 − (1 − z−1)s4t4 (1 − z−1)(s2t2)(1 + st + s2t2 + s3t3) (z − 1)(s2t2 − s3t3) 1 + (z − 1)(s4t4) (cid:19) , SSE AND GSC 11 which is nontrivial as an element of NK1(R). Writing M as M = (cid:18) (1 − z−1)s4t4 (z−1 − 1)(s2t2)(1 + st + s2t2 + s3t3) (1 − z)(s2t2 − s3t3) (1 − z)(s4t4) (cid:19) = 5 Xi=1 siMi with the Mi over R, we obtain (see [5]) a nilpotent matrix N over R, N = M1 M2 M3 M4 M5 0 I 0 0 0 0 0 0 0 0 I 0 0 I 0 0 0 0 0 I   0 0 0 0 0 (z−1 − 1)t2 1 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0   (1 − z)t2 0 0 0 0 1 0 0 0 0   = 0 (1 − z)(−t3) (z−1 − 1)t3 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 (1 − z−1)t4 (z−1 − 1)t4 0 (1 − z)t4 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 (z−1 − 1)t5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0   which is nontrivial as an element of Nil0(R), as is the matrix N ′ obtained by removing the last row and the last column from N. The matrix N ′ is 9 × 9. We don't have a smaller example, and we don't have a decent example of two positive matrices which are shift equivalent but not strong shift equivalent over a subring of R. Remark 3.6. Suppose R is a subring of R and N is a nonnegative nilpotent matrix over R. Then there is a permutation matrix P such that P −1NP is triangular with zero diagonal. Using elementary SSEs of the block form (cid:18)X Y 0(cid:19) = (cid:18)I 0 0(cid:19)(cid:0)X Y(cid:1) and (cid:0)X(cid:1) = (cid:0)X Y(cid:1)(cid:18)I 0(cid:19) we see that P −1NP (and hence N) is SSE over R to [0]. By Theorem 2.1, with A = 0, it follows that a nilpotent matrix N is SSE over R to a nonnegative matrix if and only if [I − tN∞] is trivial in NK1(R). Therefore, if (and only if) NK1(R) is nontrivial, there will be nilpotent matrices over R which cannot be SSE over R to a nonnegative matrix. The matrix N in Example 3.5 is one such example. 12 MIKE BOYLE AND SCOTT SCHMIEDING 4. Reflections on the Generalized Spectral Conjecture Is the Generalized Spectral Conjecture true? For R = Z, the Spectral Conjecture is true [11]. The GSC is true for R = Z for a given ∆ if every entry of ∆ is a rational integer [3]. There is not much more direct evidence for the GSC for R = Z, but we know of no results which cast doubt. From here, suppose R is a dense subring of R. As noted earlier, the Spectral Con- jecture is almost surely true. Theorem 1.1 removes the possibility that the very subtle algebraic invariants following from Theorem 2.1 could be an obstruction to the GSC. The GSC was proved in [3] in the following cases: (1) when the nonzero spectrum is contained in R, and R is a Dedekind domain with a nontrivial unit; (2) when the nonzero spectrum has positive trace and either (i) the spectrum is real or (ii) the minimal and characteristic polynomials of the given matrix are equal up to a power of the indeterminate. The following Proposition (almost explicit in [2, Appendix 4]) is more evidence for the GSC in the positive trace case. Proposition 4.1. Suppose the Generalized Spectral Conjecture holds for matrices of positive trace for the ring R. Then it holds for matrices of positive trace for every dense subring R of R. Proof. Let A be a square matrix over R of positive trace which over R is SSE to a primitive real matrix B. We need to show that A is SSE over R to a primitive matrix. By [12] (or the alternate exposition [4, Appendix B]), because B is primitive with positive trace, there is a positive matrix B1 SSE over R (in fact over R+) to B. And then, by arguments in [12], for some m there are m × m matrices A2, B2 (obtained through row splittings of A and B1 ), with B2 positive, such that A is SSE over R to A2; B1 is SSE over R (in fact over R+) to a positive matrix B2; and there is a matrix U in SL(m, R) such that U −1A2U = B2. Because SL(m, R) is dense in SL(m, R), and B2 is positive, there is a V in SL(m, R) such that V −1A2V is positive. This matrix (V −1A2)(V ) is SSE over R to the matrix (V )(V −1A2) = A. (cid:3) After more than 20 years, the GSC remains open even in the case R = R. Still, the GSC seems correct. What we lack is a proof. References [1] Mike Boyle. Symbolic dynamics and matrices. In Combinatorial and graph-theoretical problems in linear algebra (Minneapolis, MN, 1991), volume 50 of IMA Vol. Math. Appl., pages 1 -- 38. Springer, New York, 1993. SSE AND GSC 13 [2] Mike Boyle and David Handelman. The spectra of nonnegative matrices via symbolic dynamics (including Appendix 4 joint with Kim and Roush). Ann. of Math. (2), 133(2):249 -- 316, 1991. [3] Mike Boyle and David Handelman. Algebraic shift equivalence and primitive matrices. Trans. Amer. Math. Soc., 336(1):121 -- 149, 1993. [4] Mike Boyle, K. H. Kim, and F. W. Roush. Path methods for strong shift equivalence of positive matrices. Acta Appl. Math., 126:65 -- 115, 2013. [5] Mike Boyle and Scott Schmieding. Strong shift equivalence and algebraic K-theory. arXiv:1501.04695, 2015. [6] Edward G. Effros. Dimensions and C ∗-algebras, volume 46 of CBMS Regional Conference Series in Mathematics. Conference Board of the Mathematical Sciences, Washington, D.C., 1981. [7] Richard Ellard and Helena Smigoc. Constructing new realisable lists from old in the niep. arXiv:1306.2998, 2013. [8] Charles R. Johnson, Thomas J. Laffey, and Raphael Loewy. The real and the symmetric non- negative inverse eigenvalue problems are different. Proc. Amer. Math. Soc., 124(12):3647 -- 3651, 1996. [9] K. H. Kim and F. W. Roush. Free Zp actions on subshifts. Pure Math. Appl., 8(2-4):293 -- 322, 1997. [10] K. H. Kim and F. W. Roush. The Williams conjecture is false for irreducible subshifts. Ann. of Math. (2), 149(2):545 -- 558, 1999. [11] Ki Hang Kim, Nicholas S. Ormes, and Fred W. Roush. The spectra of nonnegative integer matrices via formal power series. J. Amer. Math. Soc., 13(4):773 -- 806 (electronic), 2000. [12] Ki Hang Kim and Fred W. Roush. On strong shift equivalence over a Boolean semiring. Ergodic Theory Dynam. Systems, 6(1):81 -- 97, 1986. [13] Thomas Laffey, Raphael Loewy, and Helena Smigoc. Power series with positive coefficients arising from the characteristic polynomials of positive matrices. arXiv:1205.1933, 2013. [14] Thomas J. Laffey. A constructive version of the Boyle-Handelman theorem on the spectra of nonnegative matrices. Linear Algebra Appl., 436(6):1701 -- 1709, 2012. [15] Thomas J. Laffey, Raphael Loewy, and Helena Smigoc. Nonnegative matrices that are similar to positive matrices. SIAM J. Matrix Anal. Appl., 31(2):629 -- 649, 2009. [16] Douglas Lind and Brian Marcus. An Introduction to Symbolic Dynamics and Coding. Cambridge University Press, Cambridge, 1995. [17] M. Maller and M. Shub. The integral homology of Smale diffeomorphisms. Topology, 24(2):153 -- 164, 1985. [18] Andrew Ranicki. High-dimensional knot theory. Springer Monographs in Mathematics. Springer- Verlag, New York, 1998. Algebraic surgery in codimension 2, With an appendix by Elmar Winkelnkemper. [19] Jonathan Rosenberg. Algebraic K-theory and its applications, volume 147 of Graduate Texts in Mathematics. Springer-Verlag, New York, 1994. [20] Scott Schmieding. The Nil group of a ring: examples. arXiv, to be posted, 2015. [21] J. B. Wagoner. Strong shift equivalence and K2 of the dual numbers. J. Reine Angew. Math., 521:119 -- 160, 2000. With an appendix by K. H. Kim and F. W. Roush. [22] Charles A. Weibel. The K-book, volume 145 of Graduate Studies in Mathematics. American Mathematical Society, Providence, RI, 2013. An introduction to algebraic K-theory. [23] R. F. Williams. Classification of subshifts of finite type. Ann. of Math. (2), 98:120 -- 153, 1973; erratum, ibid. 99:380 -- 381, 1974. 14 MIKE BOYLE AND SCOTT SCHMIEDING [24] R. F. Williams. Strong shift equivalence of matrices in GL(2, Z). In Symbolic dynamics and its applications (New Haven, CT, 1991), volume 135 of Contemp. Math., pages 445 -- 451. Amer. Math. Soc., Providence, RI, 1992. Appendix A. Correction The preceding version of "Strong shift equivalence and the generalized spectral con- jecture for nonnegative matrices" is essentially the same as the arxiv version 1. The purpose of this post is to communicate a correction, which we state separately since the paper has been published (DOI 10.1016/j.laa.2015.06.004), in Linear Algebra and its Applications. Theorem 2.1 states a result claimed in the version 1 post of [5]. This quoted result was corrected in the version 2 post of [5] (to appear in Crelle's Journal): the bijec- tion of Theorem 2.1(2) in general is not to NK1(R), but to a certain quotient group NK1(R)/E(A, R). The "elementary stabilizer" E(A, R) need not be trivial, but is trivial in many cases (for example, if A is invertible over R). With the corrected reference, the proof of equivalence of the strong and weak forms of the Spectral Conjecture (i.e., Theorem 1.1) goes through without change. Also, because E(A, R) is in many cases trivial, it still holds (as discussed after the statements of the weak and strong forms of the conjecture, bottom of page 3) that SE-R in general doesn't imply SSE-R, and therefore the strong form of the conjecture was not vacuously equivalent to the weak form.
1711.09435	1	1711	2017-11-26T18:08:54	Almost nilpotency of an associative algebra with an almost nilpotent fixed-point subalgebra	[ "math.RA" ]	Let A be an associative algebra of arbitrary dimension over a field F and G a finite soluble group of automorphisms of A oforder n, prime to the characteristic of F. We prove that if the fixed-point subalgebra of A under the action of G contains a two-sided nilpotent ideal of nilpotency index d and of finite codimension m in I, then A contains a nilpotent two-sided ideal of nilpotency index bounded by a function of n and d and of finite codimension bounded by a function of m, n and d.	math.RA	math	Almost nilpotency of an associative algebra with an almost nilpotent fixed-point subalgebra N. Yu. Makarenko∗ Sobolev Institute of Mathematics, Novosibirsk, 630 090, Russia natalia [email protected] Abstract Let A be an associative algebra of arbitrary dimension over a field F and G a finite group of automorphisms of A of order n, prime to the characteristic of F . Denote by AG = {a ∈ A ag = a for all g ∈ G} the fixed-point subalgebra. By the classical Isaacs -- Bergman theorem, if AG is nilpotent of index d, i.e. (AG)d = 0, then A is also nilpotent and its nilpotency index is bounded by a function depending only on n and d. We prove, under additional assumption of solubility of G, that if AG contains a two-sided nilpotent ideal I ⊳ AG of nilpotency index d and of finite codimension m, then A contains a nilpotent two-sided ideal H ⊳ A of nilpotency index bounded by a function of n and d and of finite codimension bounded by a function of m, n and d. An even stronger result is provided for graded associative algebras: if G is a finite (not necessarily soluble) group of order n and A = Lg∈G Ag is a G-graded associative algebra over a field F , i.e. AgAh ⊂ Agh, such that the identity component Ae has a two-sided nilpotent ideal Ie ⊳ AG of index d and of finite codimension m in Ae, then A has a homogeneous nilpotent two-sided ideal H ⊳A of index bounded by a function on n and d and of finite codimension bounded by a function on n, d and m. Keywords. associative algebra, actions of finite groups of automorphisms, finite grading, graded associative algebra, fixed-point subalgebra, almost nilpotency. 1 Introduction By the classical Isaacs -- Bergman theorem [1], if an associative algebra A over a field F admits a finite group of automorphisms G of order n, prime to the characteristic of F , and the fixed-point subalgebra AG = {a ∈ A ag = a for all g ∈ G} is nilpotent of index d, i.e. (AG)d = 0, then A is nilpotent of index bounded by a function of n and d. Starting from this work, a great number of paper was devoted to study of properties of algebras ∗The research is supported by RSF (project N 14-21-00065) Mathematics Subject Classification (2010): Primary 16W20, 16W22, 16W50 1 (or rings) subject to corresponding properties of fixed-point algebra under finite group actions. In this paper we prove, under the additional assumption of the solubility of the automorphism group, that the "almost nilpotency" of the fixed-point subalgebra implies the "almost nilpotency" of the algebra itself. Namely, the following theorem holds. Theorem 1.1. Let A be an associative algebra of arbitrary (possibly infinite) dimension over a field F acted on by a finite soluble group G of order n. Suppose that the character- istic of F does not divide n. If the fixed-point subalgebra AG has a nilpotent two-sided ideal I ⊳ AG of nilpotency index d and of finite codimension m in AG, then A has a nilpotent two-sided ideal H ⊳ A of nilpotency index bounded by a function of n and d and of finite codimension bounded by a function of m, n and d. The restrictions on the order of the automorphism group are unavoidable. There are examples showing that no results of this kind are possible either for infinite automorphism groups or for algebras with n-torsion. Theorem 1.1 follows by induction on the order of G from the Bergman-Isaacs theorem and the following statement on graded associative algebras, in which we do not suppose either G to be soluble or the order of G to be prime to the characteristic of the field. Theorem 1.2. Let G be a finite group of order n and let A = Lg∈G Ag be a G-graded associative algebra over a field F , i.e. AgAh ⊂ Agh. If the identity component Ae has a nilpotent two-sided ideal Ie ⊳ Ae of nilpotency index d and of finite codimension m in Ae, then A has a homogeneous nilpotent two-sided ideal H ⊲ A of nilpotency index bounded by a function on n and d and of finite codimension bounded by a function on n, d and m. The proof of Theorem 1.2 is based on the method of generalized centralizers, originally created by Khukhro in [3] for nilpotent groups and Lie algebras with an almost regular automorphism of prime order. In [4, 5] the approach was significantly revised and new techniques were introduced to study a more complicated case of an almost regular auto- morphism of arbitrary (not necessarily prime) finite order. In particular, it was proved that if a Lie algebra L admits an automorphism ϕ of finite order n with finite-dimensional fixed-point subalgebra of dimension dim CL(ϕ) = m, then L has a soluble ideal of derived length bounded by a function of n whose codimension is bounded by a function of m and n. The fundamental combinatorial nature of the construction in [5] makes possible to apply it to a wide range of situations. For example, the approach was used to study Lie type algebras (a large class of algebras which includes associative, Lie algebras, color Lie superalgebras) with an almost regular automorphism of finite order in [6]. In the proof of Theorem 1.2 we use virtually the same construction as in [5]. But the strong condition of associativity simplifies the reasoning and allows to provide much stronger results than in the case of Lie algebras. In particular, we do not need to suppose that the automorphism group is cyclic. We give some definitions and auxiliary lemmas in § 2. In § 3 we prove Theorem 1.2. For this, we set N = d2 + 1 and for each g ∈ G \ e we construct by induction generalized centralizers Ag(i) of levels i = 1, 2, . . . N, which are some subspaces of the homogeneous components Ag. Then we demonstrate that the ideal generated by all the Ag(N), g ∈ G\e 2 is the required one. In § 4 by induction on the order of G we derive the Theorem 1.1 from Theorem 1.2 and the Bergman-Isaacs Theorem. Throughout the paper we will say that a number is "(a, b, . . .)- bounded" if it is "bounded above by some function depending only on a, b, . . .". 2 Preliminaries If G is a group of automorphisms of A, then AG = {a ∈ A ag = a for all g ∈ G} will denote the fixed-point subalgebra. The two-sided ideal H of A is denoted by H ⊳ A. If I and J are subspaces of A, IJ will denote the subspace spanned by all products ab with a ∈ I and b ∈ J, and I d will denote the d-fold product I . . . I . We say that an algebra {z }d is nilpotent of (nilpotency) index d if the product of any d elements of the ring A equals zero, i.e. Ad = 0. The subalgebra generated by subspaces B1, B2, . . . , Bs is denoted by hB1, B2, . . . , Bsi, and the two-sided ideal generated by B1, B2, . . . , Bs is denoted by If H is an algebra, then H # will denote the algebra obtained by idhB1, B2, . . . , Bsi. adjoining 1 to H. The (two-sided) ideal of H generated by a subspace I is sometimes written as H #IH #. We now state some facts needed in the proof. Lemma 2.1. (Bergman -- Isaacs Theorem [1]). Let G be a finite group of automorphisms of an associative ring (algebra ) R of order n. If R has no n-torsion and RG is nilpotent of index d then R is nilpotent of index at most hd, where h = 1 + Qn i=0(C i n + 1). The following two lemmas are known. We give their proofs for the convenience of readers. Lemma 2.2. (Bergman -- Isaacs [1, lemma 1.1]). Let G be a finite group of order n and let A = Lg∈G Ag be a G-graded associative algebra over a field F , i.e. AgAh ⊂ Agh. If the identity component Ae is nilpotent of index d, then A is nilpotent of index at most nd. Proof. It suffices to prove that a product ag1ag2 . . . agnd in homogeneous elements agi ∈ Agi, i = 1, . . . , nd of length nd is trivial. We consider nd + 1 products h0 = e, h1 = g1, hi = g1 . . . gi, i = 1, . . . , nd. Since the order of G is n, some d + 1 elements must be equal. If hi = hj with i < j, then gi+1 . . . gj = e. We obtain that ag1ag2 . . . agnd can be represented as P1Q1Q2 . . . QdP2, where P1 and P2 are (possibly empty) products in homogeneous agi, and each Qi is a non-empty product of the form Qi = agi+1 . . . agj with gi+1 . . . gj = e. It follows that Qi ∈ Ae for all i = 1, . . . , d. Since (Ae)d = 0, we have that Q1Q2 . . . Qd = 0, and therefore ag1ag2 . . . agnd = 0. Lemma 2.3 ([2, Lemma 1.3.7 ]). Let A be an associative algebra over a field F acted on by a finite group G of order n. Suppose that the characteristic of F does not divide n. If the subalgebra of invariants AG contains a nilpotent ideal I ⊳ AG of nilpotency index d, then A contains a G-invariant ideal J > I of (n, d)-bounded nilpotency index. 3 Proof. Consider the right-sided ideal B = IA# generated by I (recall that A# is the algebra obtained from A by joining the unit). Let b ∈ BG be an element of B fixed by G, i. e. bg = b for all g ∈ G. There exist elements s ∈ B, im ∈ I, am ∈ A# such that b = ns = nPm imam. Since Pg∈G ag m ∈ AG and I is an ideal in AG, we have b = ns = X g∈G sg = X m imX g∈G ag m ∈ I, i. e. BG 6 I and BG is nilpotent of index 6 d. Applying Bergman -- Isaacs Theorem to the algebra B we obtain that B is nilpotent of index at most hd, where h = 1 +Qn n + 1). Finally, two-sided G-invariant ideal J > I generated by B is also nilpotent of index at most hd: (A#IA#)hd = A#(IA#)hd = 0. i=0(C i 3 Proof of Theorem 1.2 Let G be an arbitrary finite group of order n and A = Lg∈G Ag be a G-graded associative algebra over a field F , i.e. AgAh ⊂ Agh. Suppose that the identity component Ae has a nilpotent ideal Ie of nilpotency index d and dim Ae/Ie = m. Index Convention. In what follows, unless otherwise stated, a small letter with an index g will denote an element of the homogeneous component Ag. The index only indicates which component this element belongs to: xg ∈ Ag. To lighten the notation, we shall not be using numbering indices for elements of the Ag, so that different elements can be denoted by the same symbol. For example, xg and xg can be different elements of Ag. Construction of generalized centralizers and representatives. We fix N = d2 + 3. In each homogeneous component Ag, g ∈ G \ {e} we construct by induction a descending chain of subspaces: Ag = Ag(0) > Ag(1) > · · · > Ag(N). The subspaces Ag(s) are called generalized centralizers of level s. Simultaneously we fixe some homogeneous elements in Ag(s), s = 0, . . . , N which are referred to as representatives of level s. The total number of representatives will be (n, d, m)-bounded. Definition. For a monomial ag1ag2 . . . agk, where agi ∈ Agi, the record (∗g1∗g2· · · ∗gk) is called the pattern of the monomial. The length of a pattern is the degree of the monomial. The monomial is said to be the value of its pattern on the given elements. For example, agagav and bgcgbv are values of the same pattern (∗g ∗g ∗v). (Under the Index Convention the elements ag in the first product can be different.) Definition. Let g ∈ G \ {e}. For every ordered tuple of elements ~x = (xg1, . . . , xgk), xgs ∈ Ags, such that g1g2 . . . gl−1 g gl . . . gk = e for some l ∈ {1, . . . k + 1} we define the mappings: ϑ~x,l : Ag → Ae/Ie; 4 ϑ~x,l : yg → xg1xg2 . . . xgl−1ygxgl . . . xgkIe, where Ie is the nilpotent ideal of Ae of nilpotency index d and of codimension m in Ae. We use index l to distinguish eventual cases of g1g2 . . . gk−1 g gk . . . gk = e and g1g2 . . . gl−1 g gl . . . gk = e with k 6= l which lead to different mappings. By linearity, the mapping ϑ~x,l is a homomorphism of the subspace Ag into factor-space Ae/Ie. Since dim Ae/Ie 6 m, we have dim Ag/Ker ϑ~x,l 6 m. Definition of level 0. We set Ag(0) = Ag for all g ∈ G \ {e}. To construct representatives of level 0 we fix some elements xe ∈ Ae whose images form a basis of Ae/Ie. These elements are called representatives of level 0 and are denoted by xe(0) (under the Index Convention). In addition we consider a pattern P = (∗g ∗g−1) of length 2 with g ∈ G \ {e}. The dimension of the subspace of the factor-space Ae/Ie spanned by all images of values of P on homogeneous elements of Ag, Ag−1 is at most m by hypothesis. Hence we can choose at most m products c = xgxg−1 ∈ Ae whose images form a basis of this subspace. The elements xg, xg−1 involved in these representations of the elements c are also called representatives of level 0 and are denoted by xg(0), xg−1(0) (under the Index Convention). The same is done for every pattern P of the form (∗g ∗g−1), g ∈ G \ {e}. Since dim Ae/Ie 6 m and the total number of patterns P is n − 1, the number of representatives of level 0 is at most 2(n − 1)m + m. Definition of level 111. Let W1 = 2d3(n − 1) + 2. For each g ∈ G \ {e} we set Ag(1) = \ ~z \ l Ker ϑ~z,l, where ~z = (zg1(0), . . . , zgk(0)) runs over all possible ordered tuples of all lengths k 6 W1 consisting of representatives of level 0 such that g1 . . . g . . . gk = e; if for a fixed tuple ~z = (zg1(0), . . . , zgk(0)) of length k there are several different integers l 6 k + 1 such that g1 . . . gl−1ggl . . . gk = e, we take the intersection over all such integers l. The subspaces Ag(1) are referred to as the generalized centralizers of level 1, elements of the Ag(1) are called centralizers of level 1 and are denoted by yg(1) (under the Index Convention). The subspace Ag(1) has (n, d, m)-bounded codimension in Ag since the intersection here is taken over an (n, d, m)-bounded number of subspaces of m-bounded codimension in Ag. The representatives of level 1 are constructed in two different ways. First, for each g ∈ G \ {e} we fix some elements of Ag whose images form a basis of of the factor-space Ag/Ag(1). These elements are called b-representatives of level 1 are denoted by bg(1) ∈ Ag (under the Index Convention). Since the dimensions Ag/Ag(1) are (n, d, m)-bounded for all g ∈ G \ {e}, the total number of b-representatives of level 1 is (n, d, m)-bounded. Second, for each pattern P = (∗g ∗g−1) of length 2 with indices g, g−1 ∈ G \ {e} we consider the subspace of the factor-space Ae/Ie spanned by all images of the values of P on homogeneous elements of Ag(1), Ag−1(1). Since dim Ae/Ie 6 m, we can choose at most m products c = yg(1)yg−1(1) ∈ Ae whose images form a basis of this subspace in Ae/Ie and fix the elements yg(1), yg−1(1) involved in these representations. These elements are called x-representatives of level 1 and are denoted by xg(1) (under the Index Condition). 5 Since the number of patterns under consideration is equal to n − 1, the total number of x-representatives of level 1 is at most 2(n − 1)m. By construction, if g1 . . . gt−1 g gt . . . gk = e, for some t 6 k+1 and k 6 W1, a centralizer yg(1) has the following property with respect to representatives xgj (0) of level 0: xg1(0) . . . xgt−1(0) yg(1) xgt(0) . . . xgk (0) ∈ Ie. (1) s > 0 Definition of level s > 0 s > 0. Suppose that we have already fixed representatives of level < s, which are either x-representatives or b-representatives and its number is (m, n, d)- bounded. We now define the generalized centralizers of level s. Let Ws = Ws−1 + 1 = 2d3(n − 1) + 1 + s. For each g ∈ G \ {e} we set \ Ag(s) = \ Ker ϑ~z,l, ~z l where ~z = (zg1(ε1), . . . , zgk (εk)) runs over all possible ordered tuples of all lengths k 6 Ws consisting of representatives of (possibly different) levels < s (i. e., zgu(εu) denote elements of the form xgu(εu) or bgu(εu), εu < s, in any combination) such that g1 . . . g . . . gk = e; if for a fixed tuple ~z = (zg1(ε1), . . . , zgk(εk)) of length k there are several different integers l 6 k + 1 such that g1 . . . gl−1ggl . . . gk = e, we take the intersection over all such integers l. Elements of the Ag(s) are also called centralizers of level s and are denoted by yg(s) (under the Index Convention). The intersection here is taken over an (n, d, m)-bounded number of subspaces of m- bounded codimension in Ag, since the number of representatives of all levels < s is (n, d, m)-bounded and dim Ag/Ker ϑ~z,l 6 m for all ~z. Hence Ag(s) also has (n, d, m)- bounded codimension in the subspace Ag. We now fix representatives of level s. First, for each g ∈ G \ {e} we fix some elements of Ag whose images form a basis of of the factor-space Ag/Ag(s). These elements are denoted by bg(s) ∈ Ag (under the Index Convention) and are called b-representatives of level s. The total number of b-representatives of level s is (n, d, m)-bounded, since the dimensions Ag/Ag(s) are (n, d, m)-bounded for all g ∈ G \ {e}. Second, for each pattern P = (∗g ∗g−1) of length 2 with indices g ∈ G \ {e} we consider the subspace of the factor-space Ae/Ie spanned by all images of values of P on homogeneous elements of Ag(s), Ag−1(s). Since dim Ae/Ie 6 m, we can choose at most m products c = yg(s)yg−1(s) ∈ Ae whose images form a basis of this subspace in Ae/Ie and fix the elements yg(s), yg−1(s) involved in these representations. These fixed elements are called x-representatives of level s and are denoted by xg(s) (under the Index Condition). The total number of x-representatives of level s is at most 2(n − 1)m. Note that x-representatives of level s, elements xg(s), are also centralizers of level s. It is clear from the construction that Ag(k + 1) 6 Ag(k) (2) for all g ∈ G \ {e} and any k. 6 By definition, if g1 . . . gt−1 g gt . . . gk = e, for some t 6 k + 1 and k 6 Ws, then a centralizer yg(s) has the following property with respect to representatives of lower levels: zg1(ε1) . . . zgt−1(εt−1) yg(s) zgt(εt) . . . zgk(εk) ∈ Ie, (3) where the elements zgj (εj) are representatives (that is, either bgj (εj) or xgj (εj), in any combination) of any (possible different) levels εl < s. The following lemmas are direct consequences of the inclusions (2), (3) and the defi- nitions of representatives. Lemma 3.1. Let g ∈ G \ {e}. Then 1) every homogeneous element ae ∈ Ae can be represented modulo Ie as a linear com- bination of representatives xe(0) of level 0. 2) every product ag bg−1 in homogeneous elements can be represented modulo Ie as a linear combination of products of the same pattern in representatives of level 0. 3) every product yg(k1)yg−1(k2) in centralizers of levels k1, k2 can be represented modulo Ie as a linear combination of products xg(s)xg−1(s) of the same pattern in x-representatives of any level s satisfying 0 6 s 6 min{k1, k2}. Lemma 3.2. Let yg(l + 1) be a centralizer of level l + 1, bh(l) be b-representative of level l with g, h, gh ∈ G \ e. Then elements of the form ugh = yg(l + 1)bh(l) or vhg = bh(l)yg(l + 1) are centralizers of level l. Proof. The proof follows directly from (3) and the definitions of Wi. Lemma 3.3. Any product of the form ag−1 yg(k + 1) or yg(k + 1) ag−1, where yg(k + 1) is a centralizer of level k > 0, is equal modulo Ie to a product of the form yg−1(k) yg(k) or accordingly yg(k) yg−1(k), where yg−1(k), yg(k) are centralizers of level k − 1. Proof. We represent ag−1 as a sum of a centralizer yg−1(k) of level k and a linear com- bination of b-representatives bg−1(k) of level k and substitute this sum into the product ag−1 yg(k + 1). We obtain a sum of the element yg−1(k) yg(k + 1) and a linear combi- nation of elements of the form bg−1(k) yg(k + 1). By (3) the product bg−1(k) yg(k + 1) belongs to Ie. Hence ag−1 yg(k + 1) = yg−1(k) yg(k + 1) (mod Ie). Similarly, yg−1(k + 1) ag = yg−1(k + 1) yg(k) (mod Ie). Since Ag(k) > Ag(k + 1), both products have the required form. Construction of nilpotent ideal. Recall that N = d2 + 3 is the fixed notation for the highest level. We have constructed the generalized centralizers Ag(N) for g ∈ G \ {e}. Let G \ {e} = {g1, . . . , gn−1} We set Z =id (cid:10)Ag1(N), Ag2(N), . . . , Agn−1(N), Ie(cid:11) . This ideal has (n, d, m)-bounded codimension in A, since each subspace Ah(N), h ∈ G \ {e}, has (n, d, m)-bounded codimension in Ah, while the dimension of Ae/Ie is at most m by hypothesis. To prove the Theorem 1.2 we show that the ideal Z is nilpotent of (n, d)-bounded class. Definition. For every g ∈ G we set Zg = Z ∩ Ag. 7 Lemma 3.4. The subspace Ze is contained modulo Ie in the subspace spanned by products of the form yh−1(N −2) yh(N −2) and by products of the form ag−1 ie ag, where yh−1(N −2), yh(N − 2) are centralizers of level N − 2, ag−1 ∈ Ag−1, ag ∈ Ag, ie ∈ Ie, h, g ∈ G \ {e}. Proof. An element of Ze is modulo Ie a linear combination of products of the forms: ag−1 ie ag, where ag−1 ∈ Ag, ie ∈ Ie, ag ∈ Ag, g 6= e{e} ag−1 yg(N), where ag−1 ∈ Ag−1, g 6= e, yg(N) ∈ Ag(N), yg(N) ag−1, where g 6= e, yg(N) ∈ Ag(N), ag−1 ∈ Ag−1, ag1 yg(N) ag2, where ag1 ∈ Ag1, ag2 ∈ Ag2, yg(N) ∈ Ag(N), g1gg2 = e, (4) (5) (6) (7) The product (4) is already of the required form. By Lemma 3.3 the products yg(N) ag−1 and ag−1 yg(N) can be represented modulo Ie as linear combinations of products of the form yg(N − 1) yg−1(N − 1) and therefore have also the required representation since Ag(N − 1) 6 Ag(N − 2). Consider the product (7). Since g1gg2 = e and g 6= e, at least one gi, i = 1, 2 is not equal to e. Let, for example, g1 6= e. We represent ag1 as a sum of a centralizer yg−1(N −1) of level N − 1 and a linear combination of b-representatives bg−1(N − 1) of level N − 1 and insert this expression into (7). We obtain a linear combination of products of the following two forms and yg1(N − 1) yg(N) ag2 bg1(N − 1) yg(N) ag2. (8) (9) In (8) we set ag−1 1)ag−1 2) yg−1 := yg(N) ag2. Applying Lemma 3.3 and the inclusions (2) to yg1(N − we obtain that (8) is equal modulo Ie to a product of the required form yg1(N − (N − 2). 1 1 1 Let us now consider the product (9). If g2 = e, then g1g = e and bg1(N − 1) yg(N) ∈ Ie by (3). Since Ie is an ideal of Ae and g2 = e, bg1(N − 1) yg(N) ae ∈ Ie. If g2 6= e, then g1g 6= e and bg1(N −1) yg(N) is a a centralizer of level N −1 by Lemma 3.2: bg1(N − 1) yg(N) ag2 = yg1g(N − 1) ag2. Again by Lemma 3.3 the product yg1g(N − 1) ag2 is equal modulo Ie to the product of the require form yg1g(N − 2) yg2(N − 2). The case where g1 = e, g2 6= e in (7) can be treated in the same manner. Proof of Theorem 1.2. We set H = d2 + 1, T = d(H − 1) + 1 = d3 + 1, S = (T − 1)(n − 1) + 1 = d3(n − 1) + 1, U = d(n − 1), and Q = (U + 1)(S − 1) + 1 = (d(n − 1) + 1)d3(n − 1) = d4(n − 1)2 + d3(n − 1). By Lemma 2.2 it suffices to show that (Ze)Q = 0. 8 We consider an arbitrary product of length Q in elements ci from Ze: c1 c2 . . . cQ, (10) (here the indices are numbering). By Lemma 3.4 we can represent modulo Ie every ck as a linear combination of products of some special form. Substituting these expressions into (10) we obtain a linear combination of elements z1 z2 . . . zQ, (11) where the zk (here the indices are also numbering) are either elements ie ∈ Ie or products ce = ag−1 we ag ∈ Ae, we ∈ Ie or products ve = yg−1 (N − 2) ygk(N − 2) ∈ Ae, in centralizers yg−1 (N − 2), ygk(N − 2) of level N − 2. If in (11) among zk there are at least d occurrences of elements ie ∈ Ie the summand k k is trivial, since Ie is an ideal of Ae and (Ie)d = 0. Suppose now that in (11) there are at least (d − 1)n + 1 entries of products ce = ie agk ∈ Ae with the same pair ag−1 ie ag. Among them we can choose d products ce = ag−1 of indices g−1 k k , gk: z1 . . . zl1 ag−1 ie agk {z } k Since the products agk zls+1 . . . zls+1 ag−1 ideal in Ae and (Ie)d = 0, then the product (11) is equal to 0. k zl1+1 . . . zl2 ag−1 ie agk {z } k zl2 . . . zlk ag−1 ie agk {z } k zlk+1 . . . zQ. between the elements ie belong to Ae, Ie is an Consider the case where the number of ie-occurrences in (11) is at most d and the number of ce-occurrences is at most U = d(n − 1). Since Q = (U + d + 1)(S − 1) + 1, the product (11) has at least one subproduct consisting of S elements ve (going one after another): 1 2 (N − 2) yg1(N − 2)(cid:1) (cid:0)yg−1 (N − 2) yg2(N − 2)(cid:1) . . .(cid:0)yg−1 (cid:0)yg−1 where ygi(N − 2) ∈ Agi(N − 2), yg−1 (N − 2) are (possibly different) centralizers of level N − 2. Since S = (T − 1)(n − 1) + 1 in (12) there are at least T entries of products yg−1 k , gk. We choose any T such products and represent modulo Ie all the other pairs as linear combinations of products of representatives of level 0 by Lemma 3.1: (N − 2)ygi(N − 2) with the same pair of indices, say, g−1 (N − 2) ygS(N − 2)(cid:1), (N − 2) ∈ Ag−1 (12) S i i i we . . . we(cid:0)yg−1 k (N − 2) ygk(N − 2)(cid:1) we . . . we(cid:0)yg−1 k (N − 2) ygk(N − 2)(cid:1) . . . , (13) where there are T occurrences of (possibly different) products yg−1 (N − 2), ygk(N − 2) with the same pair of indices g−1 k , gk, the we are possibly different elements of Ae: either ie ∈ Ie or representatives xe(0). If in (13) among we there are at least d occurrences of elements of Ie the summand is trivial, since Ie is an ideal of Ae and (Ie)d = 0. In the opposite case, as T = d(H − 1) + 1, there is a subproduct of the form k (cid:0)yg−1 k (N − 2) ygk(N − 2)(cid:1) xe(0) . . . xe(0)(cid:0)yg−1 k (N − 2) ygk(N − 2)(cid:1) . . . , 9 where there are H = d2 + 1 occurrences of products yg−1 (N − 2) ygk(N − 2) and between them there are only x-representative of level 0 and no elements from Ie. By lemma 3.1 we represent modulo Ie the first entry yg−1 (N − 2) ygk(N − 2) as a linear combination of the products of the same pattern in representatives in level 1, the second -- in level 2, and so on, the last one -- in level H. We obtain a linear combination k k (cid:0)xg−1 k (1) xgk(1) + ie(cid:1) xe(0) . . . xe(0)(cid:0)xg−1 k (2) xgk(2) + ie(cid:1) . . .(cid:0)xg−1 k (H) xgk(H) + ie(cid:1). Expanding this expression we get a linear combination of products of the form c1 xe(0) . . . xe(0) c2 xe(0) . . . xe(0) . . . cH, k (nk) xgk(nk) of different levels with one and the same pair of indices g−1 (here the indices are numbering) where the ck are either elements ie ∈ Ie or products xg−1 k , gk ∈ G. If in a summand there are at least d entries of ie ∈ Ie it is trivial by assumptions. In the summands with less than d entries of the ie, we can find an interval long enough without ie-entries. More precisely, since H = d2 + 1 there is a subproduct of the form (cid:0)xg−1 k (s) xgk(s)(cid:1) xe(0) . . . xe(0)(cid:0)xg−1 k (s + 1) xgk(s + 1)(cid:1) . . .(cid:0)xg−1 k (s + d) xgk(s + d)(cid:17), where there are d + 1 products yg−1 t = 0, . . . , d − 1, the product k (l) ygk(l) of different levels l = s, . . . , s + d. For each xgk(s + t) xe(0) . . . xe(0) xg−1 k (s + t + 1) includes exactly one centralizer of level s+t+1, all the other elements are representatives of lower levels, and the weight of the product is at most 2S = 2d3(n−1) + 2 = W1 6 Ws+t+1. By (3) xgk (s + t) xe(0) . . . xe(0) xg−1 k (s + t + 1) ∈ Ie for all t = 0, . . . , d − 1. It follows that (??) is equal to product xg−1 k (s) (ie ie . . . ie) xgk(s + d) = 0, {z d } which is trivial, since (Ie)d = 0. (cid:3) 4 Proof of the main result In this section we prove Theorem 1.1. Recall that we are given an associative algebra A over a field F that admits a finite soluble automorphism group G of order n prime to the characteristic of F such that the fixed-point subalgebra AG has a two-sided nilpotent ideal I ⊳ AG of nilpotency index d and of finite codimension m in AG. The aim is to find a nilpotent ideal in A of (n, d)-bounded nilpotency index and of finite (n, d, m)-bounded codimension. Proof of Theorem 1.1. First, we consider the case where G is a cyclic group of prime order p. Let g be a generator of G. Then g induces an automorphism of the algebra 10 A ⊗Z Z[ω], where ω is a primitive p-th root of unity. The fixed-point subalgebra of this automorphism denoted by the same letter has the same dimension m over the field extended by ω. It suffices to prove Theorem 1.1 for the algebra A ⊗Z Z[ω]. Hence in what follows we can assume that the ground field F contains ω. We define the homogeneous components Ak for k = 0, . . . , p − 1 as the subspaces Since the characteristic of F does not divide p, we have Ak = (cid:8)a ∈ A ag = ωka(cid:9) . A = A0 ⊕ A1 ⊕ · · · ⊕ Ap−1. This decomposition determines a grading on A by a cyclic group of prime order p, with A0 = AG in view of the obvious inclusions AsAt ⊆ As+t, where s + t is computed modulo p. Hence the case G = p in Theorem 1.1 follows from Theorem 1.2. Let now G be any finite soluble group of automorphisms of A, and suppose that its order n is not divisible by the characteristic of F . We use induction on G. We may assume that n is not a prime number. This means that in G there is a non-trivial normal subgroup H. We consider the subalgebra C = AH of its fixed points. Since H ⊳ G, we have C g 6 C for any g ∈ G. The subalgebra C admits a finite solvable group of automorphisms of order 6 G/H which is strictly less than G and not divisible by the characteristic of F . By induction C has a nilpotent ideal J ⊳ C of (G/H, d, m)- bounded codimension t = t(G/H, d, m) and of (G/H, d)-bounded nilpotency index h = h(G/H, d). By Lemma 2.3 there exists a nilpotent G-invariant ideal K > J in A of nilpotency index h1 = h1(H, h) bounded by H and by the nilpotency index of J. The subgroup H acts on the factor-algebra ¯A = A/K and subalgebra of fixed points ¯AH has dimension at most t. We apply induction hypothesis to the algebra ¯A and the automorphism group H of Aut ¯A whose order is strictly less than G. The algebra ¯A has a nilpotent ideal Z of (H, t)-bounded codimension and of H-bounded nilpotent index h2 = h2(H). The image of Z in A is a required ideal since its nilpotency index is at most h1h2, which is a (n, d)-bounded number, and the codimension is (n, d, m)-bounded. (cid:3) References [1] G. M. Bergman, I. M. Isaacs, Rings with fixed-point-free group actions, Proc. London. Math. Soc. 27, N 3 (1973), 69 -- 87. [2] V. K. Kharchenko, Automorphisms and Derivations of Associative Rings, KLUWER ACADEMIC PUBLISHERS, 1991. [3] E. I. Khukhro, Groups and Lie rings admitting an almost regular automorphism of prime order, Math. USSR Sb. 71 (1992) 51 -- 63. 11 [4] E. I. Khukhro, N. Yu. Makarenko, Lie rings with almost regular automor- phisms, J. Algebra, 264, N 2 (2003), 641 -- 664. [5] N. Yu. Makarenko, E. I. Khukhro, Almost solubility of Lie algebras with almost regular automorphisms, J. Algebra, 277, N 1 (2004), 370 -- 407. [6] N. Yu. Makarenko, Lie type algebras with an automorphism of finite order, J. Algebra, 439 (2015), 33-66. 12
1702.02598	3	1702	2017-02-16T12:34:31	On $\mathbb{Z}_{2}$-graded polynomial identities of $sl_{2}(F)$ over a finite field	[ "math.RA" ]	Let $F$ be a finite field of $char F > 3$ and $sl_{2}(F)$ be the Lie algebra of traceless $2\times 2$ matrices over $F$. In this paper, we find a basis for the $\mathbb{Z}_{2}$-graded identities of $sl_{2}(F)$.	math.RA	math	ON Z2-GRADED POLYNOMIAL IDENTITIES OF sl2(F ) OVER A FINITE FIELD LU´IS FELIPE GONC¸ ALVES FONSECA 7 1 0 2 b e F 6 1 ] . A R h t a m [ 3 v 8 9 5 2 0 . 2 0 7 1 : v i X r a Abstract. Let F be a finite field of charF > 3 and sl2(F ) be the Lie algebra of traceless 2 × 2 matrices over F . In this paper, we find a basis for the Z2-graded identities of sl2(F ). Keywords: 16R10,17B01,15A72,17B70. Mathematics Subject Classification 2010: Graded identities; Lie algebras; finite basis identities. 1. Introduction The well-known Ado-Iwasawa' theorem posits that any finite-dimensional Lie al- gebra over an arbitrary field has a faithful finite-dimensional representation. Briefly, any finite dimensional Lie algebra can be viewed as a subalgebra of a Lie algebra of square matrices under the commutator brackets. Thus, the study of Lie algebras of matrices is of considerable interest. A task in PI-theory is to describe the identities of sl2(F ), the Lie algebra of traceless 2 × 2 matrices over a field F of charF 6= 2. The first breakthrough in this area was made by Razmyslov [12], who described a basis for the identities of sl2(F ) when charF = 0. Vasilovsky [16] found a single identity for the identities of sl2(F ) when F is an infinite field of charF 6= 2, and Semenov [13] described a basis (with two identities) for the identities of sl2(F ) when F is a finite field of charF > 3. The Lie algebra sl2(F ) can be naturally graded by Z2 as follows: sl2(F ) = (sl2(F ))0 ⊕ (sl2(F ))1 where (sl2(F ))0, (sl2(F ))1 contain diagonal and off-diagonal matrices respectively. A recent development in PI-theory is the description of the graded identities of sl2(F ). Using invariant theory techniques, Koshlukov [8] de- scribed the Z2-graded identities for sl2(F ) when F is an infinite field of charF 6= 2. Several further papers on graded identities of sl2(F ) have appeared in recent years (cf. e.g., [4] and [5]). In these studies, the ground field is of characteristic zero. To date, no basis has been found for the Z2-graded identities of sl2(F ) when F is a finite field. In this paper we give a basis for the graded identities sl2(F ) when F is a finite field of charF > 3. 2. Preliminaries Let F be a fixed finite field of charF > 3 and size F = q, let N0 = {1, 2, . . . , , n, . . .}, let G = (Z2, +), and let L be a Lie algebra over F . In this study (un- less otherwise stated), all vector spaces and Lie algebras are considered over F . The +·, ⊕, spanF {a1, . . . , an}, ha1, . . . , ani, (a1, . . . , an ∈ L) signs denote the direct sum of Lie algebras, the direct sum of vector spaces, the vector space generated 1 2 LU´IS FELIPE GONC¸ ALVES FONSECA by a1, . . . , an, and the ideal generated by a1, . . . , an respectively, while an asso- ciative product is represented by a dot: ".′′. The commutator ([, ]) denotes the multiplication operation of a Lie algebra. We assume that all commutators are left-normed, i.e., [x1, x2, . . . , xn] := [[x1, x2, . . . , xn−1], xn] n ≥ 3. We use the convention [x1, xk 2] = [x1, x2, . . . , x2], where x2 appears k times in the expanded commutator. We denote by gl2(F ) the Lie algebra of 2 × 2 matrices over F . Let sl2(F ) denote the Lie algebra of traceless 2 × 2 matrices over F . Here, eij ⊂ gl2(F ) denotes the unitary matrix unit whose elements are 1 in the positions (ij) and 0 otherwise. The basic concepts of Lie algebra adopted in this study can be found in Chapters 1 and 2 of [6]. We denote the center of L by Z(L). If x ∈ L, we denote by adx the linear map with the function rule: y 7→ [x, y]. L is said to be metabelian if it is solvable of class at most 2. As is known, if L (L over a finite field of charF > 3) is a three-dimensional simple Lie algebra, then L ∼= sl2(F ). L is regarded as a Lie A-algebra if all of its nilpotent subalgebras are abelian. A Lie algebra L is said to be G-graded (a graded Lie algebra or graded by G) when there exist subspaces {Lg}g∈G ⊂ L such that L = Lg∈G Lg, and [Lg, Lh] ⊂ Lg+h for any g, h ∈ G. G-graded associative algebras are defined in the same way. In that context, {Lg}g∈G is said to be a grading for L. An element a is called homogeneous when a ∈ Sg∈G Lg. We say that a 6= 0 is a homogeneous element of G-degree g when a ∈ Lg. A G-graded homomorphism of two G-graded Lie algebras L1 and L2 is a homomorphism φ : L1 → L2 such that φ(L1g) ⊂ L2g for all g ∈ G. g}g∈G on L are called isomorphic when there Two gradings on L {Lg}g∈G and {L′ g for all g ∈ G. An exists a G-graded isomorphism φ : L → L such that φ(Lg) = L′ ideal I ⊂ L is graded when I = Lg∈G(I ∩ Lg) (we define graded Lie subalgebras similarly). Likewise, if I is a graded ideal of L, CL(I) = {a ∈ L[a, I] = {0}} is also a graded ideal of L. Furthermore, Z(L), Ln (the n-th term of descending central series), and L(n) (the n-th term of derived series) are graded ideals of L. We use the convention that L(1) = [L, L] and L1 = L. Let L be a finite-dimensional Lie algebra. Denote by N il(L) the greatest nilpo- tent ideal of L and by Rad(L) the greatest solvable ideal of L. Clearly, N il(L) is the unique maximal abelian ideal of L when L is a Lie A-algebra. Furthermore, every subalgebra and every factor algebra of L is a Lie A-algebra when L is also a Lie A-algebra (see Lemma 2.1 in [15] and Lemma 1 in [10]). The next theorem is a structural result on solvable Lie A-algebras. Theorem 2.1 (Towers, Theorem 3.5, [15]). Let L be a (finite-dimensional) solvable Lie A-algebra (over an arbitrary field F ) of derived length n + 1 with nilradical N il(L). Moreover, let K be an ideal of L and B a minimal ideal of L. Then we have the following: : K = (K ∩ An) ⊕ (K ∩ An−1) ⊕ . . . ⊕ (K ∩ A0); : N il(L) = An +· (An−1 ∩ N il(L)) +· . . . +· (A0 ∩ N il(L)); : Z(L(i)) = N il(L) ∩ Ai for each 0 ≤ i ≤ n; : B j N il(L) ∩ Ai for some 0 ≤ i ≤ n. An = L(n), An−1, . . . , A0 are abelian subalgebras of L defined in the proof of Corol- lary 3.2 in [15]. Remark 2.2. Assuming Theorem 2.1, we can prove that, if L = Lg∈G Lg is a (finite-dimensional) solvable graded Lie A-algebra (over an arbitrary field F ) of sl2 OVER A FINITE FIELD 3 derived length n+1 with nilradical N il(L) , then N il(L) is a graded ideal. Moreover, if L is finite-dimensional metabelian Lie A-algebra (over an arbitrary field), then N il(L) = [L, L] +· Z(L). A finite-dimensional Lie algebra L is called semisimple if Rad(L) = {0}. Recall that L (finite-dimensional and non solvable) has a Levi decomposition when there exist a semisimple subalgebra S 6= {0} (termed a Levi subalgebra) such that L is a semidirect product of S and Rad(L). We now present a result. Proposition 2.3 (Premet and Semenov, Proposition 2, adapted, [10]). Let L be a finite-dimensional Lie A-algebra over a finite field F of charF > 3. Then, : [L, L] ∩ Z(L) = {0}. : L has a Levi decomposition. Moreover, each Levi subalgebra S is represented as a direct sum of F -simple ideals in S, each one of which splits over some finite extension of the ground field into a direct sum of the ideals isomorphic to sl2(F ). A Lie algebra L is said to be G-simple if [L, L] 6= {0}, and L does not have any proper non-trivial graded ideals. By mimicking the arguments of Zaicev et al. in [11] (Lemma 2.1; Section 3; Proposition 3.1, items i and ii), we have the following. Proposition 2.4. Let L be a finite dimensional graded Lie algebra. The ideal Rad(L) is a graded ideal. If L is G-simple, then L is a direct sum of simple Lie algebras. If L is direct sum of simple Lie algebras, then L is a direct sum of G- simple Lie algebras. 3. Graded identities and varieties of graded Lie algebras 1, . . . , xg Let X = {Xg = {xg n, . . .}g ∈ G} be a class of pairwise-disjoint enu- merable sets, where Xg denotes the variables of G-degree g. Let F hXi be the free associative unital algebra and let L(X) be the Lie subalgebra of F hXi gen- erated by X. L(X) is known to be isomorphic to the free Lie algebra with a set of free generators X. The algebras L(X) and F hXi have natural G-grading. A graded ideal I ⊂ L(X) invariant under all graded endomorphisms is called a graded verbal ideal. Let S ⊂ L(X) be a non empty set. The graded ver- bal ideal generated by S, hSiT , is defined as the intersection of all verbal ideals containing S. A polynomial f ∈ L(X) is called a consequence of g ∈ L(X) when f ∈ hgiT , and it is called a graded polynomial identity for a graded Lie algebra L if f vanishes on L whenever the variables from Xg are substituted by elements of Lg for all g ∈ G. We denote by IdG(L) the set of all graded identities of L. The variety determined by S ⊂ L(X) is denoted by V(S) = {A is a G-graded Lie algebraIdG(A) ⊃ hSiT }. The variety generated by a graded Lie algebra L is denoted by varG(L) = {A is a G-graded Lie algebraIdG(L) ⊂ IdG(A)}. We say that a class of graded Lie algebras {Li}i∈Γ, where Γ is an index set, generates V(S) when hSiT = Ti∈Γ IdG(Li). We denote by Id(L) the set of all ordinary identities of a Lie algebra L, and by var(L) the variety generated by L. The variety of metabelian Lie algebras over F is denoted by A2. A set S ⊂ L(X) of ordinary polynomials (respectively graded polynomials) is called a basis for the ordinary identities (respectively graded identities) of a Lie algebra (respectively a graded Lie algebra) A when Id(A) = hSiT (respectively IdG(A) = hSiT ). 4 LU´IS FELIPE GONC¸ ALVES FONSECA Example 1. In 1990's, Semenov (Proposition 1, [13]) proved that Sem1(x1, x2) = (x1)f (ad(x2)), f (t) = tq2+2 − t3, Sem2(x1, x2) = [x1, x2]−[x1, x2, (x1)q2 (x1)), [x1, x2]q−2, (x2)q2 are polynomial identities of sl2(F ). −1]−[x1, (x2)q]+[x1, x2, (x1)q2 − (x2)] − [x2, ([(x1)q2 −1, (x2)q−1]+[x1, x2, ((x1)q2 −2 − (x2)q−2)]. − (x1), x2])q, ((x2)q2 − A finite-dimensional ordinary (respectively graded) Lie algebra L is critical if var(L) (respectively varG(L)) is not generated by all proper subquotients of L. It is monolithic if it contains a single ordinary (respectively graded) minimal ideal. This single ideal is termed a monolith. It is known that if L is an ordinary (respectively graded) critical Lie algebra, then L is monolithic. Notice that if L = ⊕g∈GLg is a critical ordinary Lie algebra, then L is critical as a G-graded Lie algebra. Example 2. If L is a critical abelian (respectively graded) Lie algebra, then dimL = 1. If L is a two-dimensional (non abelian) metabelian Lie algebra, then L is critical. Furthermore, sl2(F ) is a critical Lie algebra. Proposition 3.1. Let L = Lg∈G Lg be a finite-dimensional (non abelian) metabelian graded Lie A-algebra over an arbitrary field F . If L is monolithic, then N il(L) = [L, L]. Proof. According to Theorem 2.1, N il(L) = [L, L] +· Z(L). By hypothesis, L is monolithic. Thus, Z(L) = {0} and N il(L) = [L, L]. (cid:3) The next theorem describes the relationship between critical metabelian Lie A- algebras and monolithic Lie A-algebras. Theorem 3.2 (Sheina, Theorem 1,[14]). A finite-dimensional monolithic Lie A- algebra L over an arbitrary finite field is critical if, and only if, its derived algebra cannot be represented as a sum of two ideals strictly contained within it. A locally finite Lie algebra is a Lie algebra for which every finitely generated subalgebra is finite. A variety of Lie algebras (respectively graded Lie algebras) is said to be locally finite when every finitely generated Lie algebra (respectively graded Lie algebra) has finite cardinality. It is known that a variety generated by a finite Lie algebra (respectively a graded finite Lie algebra) is locally finite. As in the ordinary case, if a variety of graded Lie algebras is locally finite, then it is generated by its critical algebras. For more details about varieties of Lie algebras, see Chapters 4 and 7 of [1]. The next result will prove useful for our purposes. Theorem 3.3 (Semenov, Proposition 2, [13]). Let B be a variety of ordinary Lie algebras over a finite field F . If there exists a polynomial f (t) = a1t + . . . + antn ∈ F [t] such that yf (adx) := a1[y, x] + . . . + an[y, xn] ∈ Id(B), then B is a locally finite variety. Let L1 and L2 be two graded Lie algebras (finite -dimensional), and I1 ⊂ L1 and I2 ⊂ L2 be graded ideals. We say that I1 (in L1) is similar to I2 (in L2) (I1 E A1 ∼ I2 E A2) if there exist isomorphisms α1 : I1 → I2 and α2 : such that for all a ∈ I1 and b + CL1 (I1) ∈ CL1 (I1) → L2 CL2 (I2) L1 L1 CL1 (I1) : α1([a, c]) = [α1(a), d], sl2 OVER A FINITE FIELD 5 where c + CL1 (I1) = b + CL1(I1) and d + CL2 (I2) = α2(b + CL1 (I1)). By proceeding as in [9] (cf. pages 162 to 166), we have the following. Proposition 3.4. If two critical graded Lie algebras L1 and L2 generate the same variety, then their monoliths are similar. 4. Z2-graded identities of sl2(F ) From now on, we denote by Y = {y1, . . . , yn, . . .} the even variables, by Z = {z1, . . . , zn, . . .} the odd variables. Lemma 4.1. Let sl2(F ) be the Lie algebra of traceless 2 × 2 matrices over F endowed with the natural grading. The following polynomials are graded identities of sl2(F ) [y1, y2], [z1, yq 1] = [z1, y1]. Proof. It is clear that [y1, y2] ∈ IdG(sl2(F )), because the diagonal is commutative. Choose ai = λ11,ie11 − λ11,ie22 and bj = λ12,j e12 + λ21,je21, so: [bj, aq i ] = λq 11,i[bj, hq] = λq λ11,i(−2λ12,je12 + 2λ21,je21) = [bj, ai]. 11,i((−2)qλ12,je12 + 2qλ21,j e21) = Thus, [z1, yq 1] = [z1, y1] ∈ IdG(sl2(F )). The proof is complete. (cid:3) We now cite two papers. First we present a corollary of Bahturin et al. in [2]. Proposition 4.2 (Bahturin et al., Corollary 1, [2]). Let R = Mn(F ), charF = p > 0, p 6= 2. Let G be an elementary abelian p-group. Suppose that R = Lg∈G Rg is a grading on R(−). Then R = Lg∈G Rg is a grading on R if and only if 1 ∈ Re. Remark 4.3. Here, 1 denotes the identity matrix of Mn(F ), and e denotes the identity element of G. In this study, F is a finite field of charF = p > 3 and size q. So, there exists b ∈ F − {0} that is not a perfect square. Notice that if sl2(F ) = (sl2(F ))0 ⊕ (sl2(F ))1 is a Z2-grading on sl2(F ), then ((sl2(F ))0 ⊕ F (e11 + e22)) ⊕ (sl2(F ))1 is a Z2-grading on sl2(F ). By Proposition 4.2, ((sl2(F ))0 ⊕ F (e11 + e22)) ⊕ (sl2(F ))1 is a Z2-grading on M2(F ). The next proposition describes the Z2-grading on M2(F ). Proposition 4.4 (Khazal et al., Theorem 1.1, adapted, [7]). Let F be a field of charF 6= 2. Then any Z2-grading of M2(F ) is isomorphic to one of the following. : (M2(F )0, M2(F )1) = (M2(F ), 0); : (M2(F )0, M2(F )1) = (F e11 ⊕ F e22, F e12 ⊕ F e21); : (M2(F )0, M2(F )1) = (F (e11 + e22) ⊕ F (e12 + be21), F (e11 − e22) ⊕ F (e12 − be21)), where b ∈ F − F 2. Remark 4.5. It is well known that (F − {0}, .) is a cyclic group of order q − 1. By elementary theory of groups, for every divisor d of q − 1, there exists a unique subgroup H ′ of (F − {0}, .) of order d. Let H be the subgroup of order q−1 2 . It is easy to see that there exists b′ ∈ (F − F 2) ∩ (F − H). Finally, note that [(e11 − e22), (e12 + b′e21)] 6= [(e11 − e22), (e12 + b′e21), . . . , (e12 + b′e21)], where (e12 + b′e21) appears q times in the expanded commutator. 6 LU´IS FELIPE GONC¸ ALVES FONSECA Proposition 4.6. Let sl2(F ) = (sl2(F ))0 ⊕ (sl2(F ))1 be a Z2-grading on sl2(F ) having the following characteristics: : dim (sl2(F ))0 = 1, : [a, cq] = [a, c] for all a ∈ (sl2(F ))1 and c ∈ F (e11 + e22) ⊕ (sl2(F ))0. Then the Z2-gradings ((sl2(F ))0, (sl2(F ))1) and (F (e11 − e22), F e12 ⊕ F e21) are isomorphic. Proof. First, note that (((sl2(F ))0 ⊕ F (e11 + e22)) ⊕ (sl2(F ))1) is a Z2-grading on gl2(F ). According to Proposition 4.2, (((sl2(F ))0 ⊕ F (e11 + e22)) ⊕ (sl2(F ))1) is a Z2-grading on M2(F ). It is clear that this grading on M2(F ) is not isomor- phic to the first grading presented in Proposition 4.4. Notice also that (F (e11 + e22) ⊕ (sl2(F ))0, (sl2(F ))1) cannot be isomorphic to the third grading presented in Proposition 4.4, because [z1, y1] = [z1, yq 1] is not a polynomial identity for M2(F ) endowed with third grading (Remark 4.5). According to Proposition 4.4, there exists a G-graded isomorphism φ : M2(F ) → M2(F ) such that φ((sl2(F ))0 ⊕ F (e11 + e22)) = F e11 ⊕ F e22 and φ((sl2(F ))1) = F e12 ⊕ F e21. Note that φ : gl2(F ) → gl2(F ) is an isomorphism of Lie algebras and φ(sl2(F )) = sl2(F ). Thus, ((sl2(F ))0, (sl2(F ))1) and (F (e11 − e22), F e12 ⊕ F e21) are isomorphic. (cid:3) The proof is complete. Henceforth we consider only sl2(F ) and F e11 ⊕ F e12 endowed with the natural grading by (Z2, +). Recall that Sem1(x1, x2), Sem2(x1, x2) ∈ Id(sl2(F )). We denote by S the set with following polynomials. Sem1(y1 + z1, y2 + z2), Sem2(y1 + z1, y2 + z2), [y1, y2], and [z1, yq 1] = [z1, y1]. Corollary 4.7. The variety V(S) is locally finite. Proof. Let L = L0 ⊕ L1 ∈ V(S) be a finitely generated algebra. By definition of S, Sem1(y1 + z1, y2 + z2) ∈ IdG(L). Hence, Sem1(x1, x2) ∈ Id(L). So, by Theorem (cid:3) 3.3, it follows that L is a finite Lie algebra. Corollary 4.8. Let L ∈ V(S) be a finite-dimensional Lie algebra. Then every nilpotent subalgebra of L is abelian. Proof. From the definition of S, it follows that Sem2(x1, x2) ∈ Id(L). Let M 6= {0} be a nilpotent (unnecessarily graded) subalgebra of L. If M t = {0} for a positive integer t ≤ q + 1, it is clear that M is abelian. If the index of nilpotency is equal to q + 2, then M Z(M) is abelian. Consequently, M is abelian. Induction on the index (cid:3) of nilpotency will give the desired result. It is well known that a verbal ideal (and, respectively, a graded verbal ideal) over an infinite field is multi homogeneous. This fact can be weakened, as stated in the next lemma. Lemma 4.9. Let I be a graded verbal ideal over a field of size q. If f (x1, . . . , xn) ∈ I and 0 ≤ degx1 f, . . . , degxnf < q, then each multi homogeneous component of f belongs to I as well. Lemma 4.10. If L = spanF {e11, e12} ⊂ gl2(F ), then the Z2-graded identities of L follow from [y1, y2], [z1, z2] and [z1, yq 1] = [z1, y1]. sl2 OVER A FINITE FIELD 7 Proof. It is clear that L satisfies the identities [y1, y2], [z1, z2] and [z1, yq 1] = [z1, y1]. We will prove that the reverse inclusion holds true. Let f be a polynomial identity of L. We may write f = g + h, where h ∈ h[y1, y2], [z1, z2], [z1, yq 1] = [z1, y1]i and g(x1, . . . , xn) ∈ IdG(L), with 0 ≤ degx1g, . . . , degxng < q. In this way, we may suppose that g is a multi homogeneous polynomial. If g(y1) = α1.y1 or g(z1) = α2z1 we can easily see that α1 = α2 = 0. In the other case, we may assume that g(z1, y1, . . . , yl) = α3.[z1, ya1 1 , . . . , yal l ], 1 ≤ a1, . . . , al < q. However, g(e12, e11, . . . , e11) is a non zero multiple scalar of e12, and consequently, (cid:3) α3 = 0. Hence, f = h and the proof is complete. Lemma 4.11. Let L = L0 ⊕ L1 ∈ A2 ∩ V(S) be a critical Lie A-algebra. Then L ∈ varG(spanF {e11, e12}). Proof. According to Lemma 4.10, it is sufficient to prove that L satisfies the identity [z1, z2]. By assumption, L is critical and therefore L is monolithic. If L is abelian, then dimL = 1. In this case L ∼= spanF {e11} or L ∼= spanF {e12}. In the sequel, we suppose that L is non abelian . From Proposition 3.1, we have [L, L] = N il(L) = [L1, L1] ⊕ [L0, L1]. From the identity [z1, y1] = [z1, yq 1], {0} = [L1, [L1, L1]] = −[L1, L1, L1]. So, by the identity Sem2(y1 + z1, y2 + z2), we have [z1, z2] ∈ IdG(spanF {e11, e12}), as required. The proof is complete. (cid:3) Corollary 4.12. A2∩varG(sl2(F )), A2∩V(S)and varG(spanF {e11, e12}) coincide. Proof. First, notice that A2∩varG(sl2(F )) ⊂ A2∩V(S) which is a locally finite vari- ety. By Lemma 4.11, all critical algebras of A2∩V(S) belong to varG(spanF {e11, e12}) ⊂ A2 ∩ varG(sl2(F )). Therefore, A2 ∩ V(S) ⊂ varG(spanF {e11, e12}). Thus, we have A2 ∩ varG(sl2(F )) = A2 ∩ V(S) = varG(spanF {e11, e12}). (cid:3) Lemma 4.13. Let L be a critical solvable Lie A-algebra belonging to V(S). Then L is metabelian. Proof. Let L be a critical (non abelian) solvable Lie algebra that belongs to V(S) with monolith W . By Proposition 2.3, we have [L, L] ∩ Z(L) = {0}. Consequently, Z(L) = {0}. Notice that Z(CL(N il(L))) = N il(L). If (N il(L))1 = L1, then L is metabelian. Now, we suppose that (N il(L))1 L1. We assert that (N il(L))0 = {0}. Suppose, on the contrary, that there exists a 6= 0 ∈ (N il(L))0. Hence, there exists b ∈ L1 − (N il(L))1 such that [b, a] 6= 0, because Z(L) = {0}. However, [b, a] = [b, aq] = 0. This is a contradiction. Thus, [L1, N il(L)] = {0}. Consequently CL(N il(L)) ⊃ L1 ∪ [L1, L1]. By Proposition 2.3 Z(CL(N il(L))) ∩ [CL(N il(L)), CL(N il(L))] = {0}. Hence, [CL(N il(L)), CL(N il(L))] = {0}. So, L(2) = {0} and the proof is complete. (cid:3) 8 LU´IS FELIPE GONC¸ ALVES FONSECA Lemma 4.14. Let L be a critical non-solvable Lie A-algebra belonging to V(S). Then L is G-simple. Proof. Let W be the monolith of L. We claim that L is semisimple. Suppose on the contrary that Rad(L) 6= {0}. Thus, W ⊂ Rad(L) ∩ [L, L]. The non trivial subspace W is an abelian ideal and it is contained in L(n), where n is the least nonnegative integer such that L(n) = L(n+1). According to Proposition 2.3, [L, L] ∩ Z(L) = {0}. So Z(L) = {0} and [L, W ] = W . The identities [y1, y2] and [z1, y1] = [z1, yq 1] mean that the subspace W0 = {0}. Notice that [W, [L, L]] = {0} = [W, L(n)] and Z(L(n)) ⊃ W . By Proposition 2.3, Z(L(n)) ∩ L(n) = {0}. This is a contradiction, so L is semisimple. By Propositions 2.3 and 2.4, L is a direct sum of G-simple Lie (cid:3) algebras. Given that L is monolithic, we conclude that L is G-simple. The next theorem was proved by Drensky in ([3] Lemma, page 991). Lemma 4.15. Let V be a finite dimensional vector space over F and let A be an abelian Lie algebra of the linear transformations φ : V → V , where each has the equality Then, every φ ∈ A is diagonalizable. φq = φ. Definition 4.16. Let L be a finite dimensional Lie algebra with a diagonalizable operator T : L → L. We denote by V (T ) a basis of L formed by the eigenvectors of T . Moreover, we denote V (T )λ = {v ∈ V (T )T (v) = λ.v}. We denote EV (w) the eigenvalue associated with the eigenvector w ∈ V (T ). Let L ∈ V(S) be a finite dimensional Lie algebra. It is not difficult to see that ad(L0) = {ada : L → La ∈ L0} is an abelian subalgebra of linear transformations of L. Moreover, (ada0)p = ada0 for all a0 ∈ L0. By Lemma 4.15, we have the following. Corollary 4.17. Let L ∈ V(S) be a finite dimensional Lie algebra. Let a0 ∈ L0. Then there exists V (ada0) ⊂ L0 ∪ L1. Proposition 4.18. Let L ∈ V(S) be a finite dimensional G-simple algebra. Let a0 ∈ L0. Then there exists V (ada0) ⊂ L0 ∪ L1. Moreover, V (ada0)0 ∩ L1 = ∅ for any basis V (ada0) ⊂ L0 ∪ L1. Proof. According to Corollary 4.17, there exists V (ada0) ⊂ L0 ∪ L1. Let b1, b2 ∈ V (ada0) ∩ L1. Notice that if [b1, b2] 6= 0, then EV (b1) = −EV (b2). It is clear that ha0i is a graded ideal, and that it is equal to L. Notice also that L = spanF {[a0, b1, . . . , bn]b1, . . . , bn ∈ V (ada0), n ≥ 1}. If there was a non zero element b ∈ V (ada0)0 ∩ L1, we could easily check that [a0, b1, b] = 0 for any b1 ∈ V (ad(a0)). More generally, by an inductive argument and routine calculations, we would have [a0, b1, . . . , bn, b] = 0 for any n ≥ 1 and b1, . . . , bn ∈ V (ada0). However, an element such as b cannot be, because Z(L) = {0}. So, V (ada0)0 ∩ L1 = ∅. (cid:3) Lemma 4.19. Let L ∈ V(S) be a critical non solvable algebra, then L ∼= sl2(F ). Proof. First of all, notice that dimL0 ≥ 1 and dimL1 ≥ 2. According to Lemma 4.14 L is G-simple. Let a0 ∈ L0. By Proposition 4.18, there exists V (ada0) = {b1, . . . , bn} ⊂ L0 ∪ L1. Moreover, V (ada0)0 ∩ L1 = ∅. sl2 OVER A FINITE FIELD 9 Let −λ1 ≤ . . . ≤ −λm < 0 < λm ≤ . . . ≤ λ1 be the eigenvalues associated with the eigenvectors of V (ada0). Notice that L0 = m P i=1 spanF {[V (ada0)λi , V (ada0)−λi ]}. Without loss of generality, suppose that spanF {[V (ada0)λ1 , V (ada0)−λ1 ]} 6= {0}. We assert that spanF {[V (ada0)λ1 , V (ada0)−λ1 ]} ⊕ spanF {V (ada0)λ1 } is a subalge- bra of L. In fact, let a ∈ V (ada0)λ1 and b ∈ spanF {[V (ada0)λ1 , V (ada0)−λ1 ]}. Consider [a, b] = αibi. So, n P i=1 On the other hand, Hence [a, b, a0] = − n P i=1 αi.EV (bi)bi. [a, b, a0] = −λ1[a, b] = −λ1( n P i=1 αibi). (−EV (bj).αj + λ1.αj)bj = 0. Consequently, if αj 6= 0, then λ1 = EV (bj). Similarly, the subspace spanF {[V (ada0)λ1 , V (ada0)−λ1 ]} ⊕ spanF {V (ada0)−λ1 } is a subalgebra. Notice that spanF {[V (ada0)λ1 , V (ada0)−λ1 ]} ⊕ spanF {V (ada0)λ1 } ⊕ spanF {V (ada0)−λ1 } is a graded ideal of L. Therefore, L0 = spanF {V (ada0)0} = spanF {[V (ada0)λ1 , V (ada0)−λ1 ]} and the subspace L1 is equal to spanF {V (ada0)λ1 } ⊕ spanF {V (ada0)−λ1 }. Notice that spanF {V (ada0)λ1 } is an irreducible L0-module, because L is G- simple. Moreover, it is not difficult to see that L0 ⊕ spanF {V (ada0)λ1 } is a mono- lithic metabelian Lie algebra with monolith spanF {V (ada0)λ1 } when viewed as ordinary Lie algebra. Notice that [L0 ⊕ spanF {V (ada0)λ1 }, L0 ⊕ spanF {V (ada0)λ1 }] = spanF {V (ada0)λ1 } cannot be represented by the sum of two ideals strictly contained within it. By Theorem 3.2, L0 ⊕ spanF {V (ada0)λ1 } is critical when viewed as an ordinary Lie algebra. Thus, it is critical when viewed as a graded algebra as well. Following the arguments of Lemma 4.10, we can prove that IdG(L0 ⊕ spanF {V (ada0)λ1 }) = h[y1, y2], [z1, y1] = [z1, yq 1], [z1, z2]iT . Consequently, it follows from Proposition 3.4 that spanF {V (ada0)λ1 } is a one- dimensional vector space. Analogously, we have dim(spanF {V (ada0)−λ1 }) = 1. Therefore, L0 ⊕ spanF {V (ada0)λ1 } ⊕ spanF {V (ada0)−λ1 } is a three-dimensional G-simple Lie algebra. So, L is simple and isomorphic to sl2(F ) (as ordinary Lie algebras). Hence, by Proposition 4.6, L ∼= sl2(F ) (as graded Lie algebras), where (cid:3) sl2(F ) is naturally graded by Z2. The proof is complete. 10 LU´IS FELIPE GONC¸ ALVES FONSECA We now prove the main theorem of this paper. 5. Main theorem Theorem 5.1. Let F be a field of char(F ) > 3 and size F = q. The Z2-graded identities of sl2(F ) follow from [y1, y2], Sem1(y1 + z1, y2 + z2), Sem2(y1 + z1, y2 + z2), and [z1, y1] = [z1, yq 1]. Proof. It is clear that varG(sl2(F )) ⊂ V(S). To prove that the reverse inclusion holds, it is sufficient to prove that all critical algebras of V(S) are also critical alge- bras of varG(sl2(F )). According to Corollary 4.12, A2 ∩ V(S) = A2 ∩ varG(sl2(F )). By Lemma 4.13, any critical solvable Lie algebra of V(S) is metabelian. By Lemma 4.19, any critical non solvable Lie algebra of V(S) is isomorphic to sl2(F ). There- fore, V(S) ⊂ varG(sl2(F )), and the theorem is proved. (cid:3) 6. Acknowledgments The author thanks CNPq for his Ph.D. scholarship. Moreover, the author thanks the reviewer for his/her comments. References [1] Yu. A. Bahturin. Identical Relations in Lie Algebras. VNU Science Press BV (1987). [2] Yu. A. Bahturin, M. Kochetov and S. Montgomery. Group gradings on simple Lie algebras in positive characteristic. Proceedings of The American Mathematical Society 137, 1245-1254 (2009). [3] V.S. Drensky. Identities in Matrix Lie algebras. Trudy seminara imeni I.G. Petrovskogo 6, 987-994 (1981). [4] A. Giambruno and M.S. Souza. Graded polynomial identities and Specht property of the Lie algebra sl2. Journal of Algebra 389, 6-22 (2013). [5] A. Giambruno and M.S. Souza. Minimal varieties of graded Lie algebras of exponential growth and the Lie algebra sl2. Journal of Pure and Applied Algebra 218 (8), 1517-1527 (2014). [6] J.E. Humphreys. Introduction to Lie algebras and Representation Theory. Springer-Verlag. Third Edition (1972). [7] P. Khazal, C. Boboc and S. Dascalescu. Group gradings of M2(K). Bulletin of Australian Mathematical Society 68 , 285-293 (2003). [8] P.E. Koshlukov. Graded polynomial identities for the Lie algebra sl2(K). International Journal of Algebra and Computation 18 (5), 825-836 (2008). [9] H. Neumann. Varieties of Groups. Springer-Verlag Berlin (1967). [10] A.A. Premet and K.N. Semenov. Varieties of residually finite Lie algebras. Mathematics of the USSR-Sbornik 137 (1), 103-113 (1988). [11] D. Pagon, D. Repovs and M. Zaicev. Group gradings on finite dimensional Lie algebras. Algebra Colloquim 20 (4), 573-578 (2013). [12] Yu. P. Razmyslov. Finite basing of the identites of a matrix algebra of second order over a field of characteristic zero. Algebra and Logic 12, 43-63 (1973). [13] K.N. Semenov. Basis of identities of the algebra sl2(K) over a finite field. Matematicheskie Zametski 52, 114-119 (1992). [14] G.V. Sheina. Metabelian Varieties of Lie A-algebras. Russian Math Surveys 33, 249-250 (1978). [15] D.A. Towers. Solvable Lie A-algebras. Journal of Algebra 340 (1), 1-12 (2011). [16] S. Yu. Vasilovsky. Basis of identities of a three-dimensional simple Lie algebra over an infinite field. Algebra and Logic 28 (5), 355-368 (1990). Departamento de Matem´atica, Universidade Federal de Vic¸osa - Campus Florestal, Rodovia LMG 818, km 06, Florestal, MG, Brazil E-mail address: [email protected]
1707.06748	3	1707	2017-11-24T03:19:48	Burnside graphs, algebras generated by sets of matrices, and the Kippenhahn Conjecture	[ "math.RA" ]	Given a set of matrices, it is often of interest to determine the algebra they generate. Here we exploit the concept of the Burnside graph of a set of matrices, and show how it may be used to deduce properties of the algebra they generate. We prove two conditions regarding a set of matrices generating the full algebra; the first necessary, the second sufficient. An application of these results is given in the form of a new family of counterexamples to the Kippenhahn conjecture, of order $8 \times 8$ and greater.	math.RA	math	BURNSIDE GRAPHS, ALGEBRAS GENERATED BY SETS OF MATRICES, AND THE KIPPENHAHN CONJECTURE BEN LAWRENCE Abstract. Given a set of matrices, it is often of interest to determine the algebra they generate. Here we exploit the concept of the Burnside graph of a set of matrices, and show how it may be used to deduce properties of the algebra they generate. We prove two conditions regarding a set of matrices generating the full algebra; the first necessary, the second sufficient. An application of these results is given in the form of a new family of counterexamples to the Kippenhahn conjecture, of order 8 × 8 and greater. 1. Introduction One of our main goals is to determine whether or not a set of n×n matrices over a field F generates the full matrix algebra Mn(F ). Various authors have looked at this problem. Here is a small sample of the vast literature on this question. Kostov [7] placed minimum bounds on the number of complex matrices required to generate a subalgebra of Mn(C). For the case of two matrices, one of which has distinct eigenvalues, George and Ikramov gave a criterion for when they cannot generate the full algebra [4]. Again using the assumption that one of the generating matrices has distinct eigenvalues, Laffey gave two separate criteria for generation of the full algebra [9, 10]. Aslaksen and Sletsjøe in their 2009 paper [1] published some criteria for n = 2 or 3. We will go beyond the distinct eigenvalue requirement, and give criteria for the case of repeated eigenvalues. In accordance with Burnside's theorem for matrix algebras, given for instance as Corollary 5.23 of Bresar [2], a set of n×n complex matrices generates the full algebra Mn(C) if and only if they have no invariant subspaces in common. In the appopriate basis, these invariant subspaces are immediately apparent. We will now define the Burnside graph of a set of matrices to help the invariant subspaces emerge. This definition is adapted from [9]. Definition 1 (Burnside graph). Let A = {A1, ..., Ak} be a set of n × n matrices over a field F . The Burnside graph B(A) of A is a directed graph of n nodes {1, ..., n}, with a directed edge existing from node i to node j if and only if there is some matrix Am with a non-zero entry at the (i, j) position. Self-loops are not considered. A Burnside graph B(A) is strongly connected if every pair of nodes in B(A) is path-connected in both directions. Date: June 27, 2021. 2010 Mathematics Subject Classification. Primary 05C50, 15A22, 16S50; Secondary 15A15, 15B57. Key words and phrases. Burnside graph, matrix subalgebra, linear pencil, double eigenvalues, Kippenhahn conjecture. Supported by the Marsden Fund Council of the Royal Society of New Zealand. 1 2 BEN LAWRENCE The Burnside graph B(A) as defined above is formed by treating each Ai as an ad- jacency matrix without regard to weighting, constructing all of the associated graphs, and merging them all. See section 2 for an example of a Burnside graph. When the graph has certain non- connectivity properties, it is guaranteed that the set of matrices does not generate the full algebra. However, the Burnside graph will change if the basis is changed. It is necessary in a sense to be in the correct basis in order to see the invariant subspaces. This brings us to our first theorem, which connects the algebra generated by a tuple of matrices A with the strong connectedness of the Burnside graph B(A). Since a tuple of n × n matrices over a field F generates Mn(F ) if and only if they generate over the algebraic closure F of F the full matrix algebra Mn(F ), we will focus mainly on algebraically closed fields of characteristic 0, and C in particular. We will occasionally also work over R. Theorem 1 (Obstacle to full algebra). Let A = {A1, ..., Ak} be a set of n × n matrices over C, and let A be the algebra generated by A. If B(A) is not strongly connected, then A (cid:54)= Mn(C). The proof of this theorem will be given in section 2. This leads to our main theorem. Here we present a simplified version, to give an impression of the full version which can be found along with its proof in Section 3. Theorem 2 (Even-order constructibility - special case). Let H and K be 2n × 2n hermitian matrices over C. If (1) K is diagonal with eigenvalues each of multiplicity 2, (2) B(H, K) is strongly connected, (3) the top row of 2 × 2 blocks of H are all invertible, (4) there exist distinct top-row 2 × 2 blocks H1j and H1k of H such that H1jH T 1j and H1kH T 1k do not commute, then the algebra A generated by H and K is the full algebra M2n(C). Since the full algebra can always be 2-generated [7], it is of particular interest to tell whether or not a given pair of matrices generates the full algebra. In the case of a pair of hermitian matrices, we can diagonalise one of them, and find a basis within which to assess the Burnside graph. The Burnside graph illustrates the decomposition, according to Burnside's theorem, of a set of matrices into blocks corresponding to invariant subspaces. Our work was motivated by the 1951 conjecture of Rudolf Kippenhahn [5]: Conjecture 1 (Kippenhahn [5]). Let and H, K be hermitian 2n× 2n matrices, and let f = det(xH + yK + I) ∈ R[x, y]. Let A be the algebra generated by H and K. If there exists g ∈ C[x, y] such that f = gk then there is some unitary matrix U such that U∗(xH + yK)U is block diagonal, and thus A (cid:54)= Mn(C). Our goal has been to further understanding of the counterexamples to this conjec- ture. Kippenhahn orginally gave a more general form of this conjecture where f is permitted to be a product of more than one irreducible polynomial. Kippenhahn's BURNSIDE GRAPHS AND THE KIPPENHAHN CONJECTURE 3 conjecture linked the multiplicity of eigenvalues of a certain type of matrix polyno- mial to the algebra generated by the coefficients of that polynomial. The claim was that, given hermitian H and K, if the polynomial Hx + Ky for scalar x and y has, for all x and y, eigenspaces each of even dimension, then H and K cannot generate the full algebra. More intuitively, if Hx + Ky had square characteristic polynomial f 2, then the conjecture was that one could transform H and K simultaneously into block diagonal form, and each block would have characteristic polynomial f . In his paper [5], Kippenhahn proved that his conjecture holds for n ≤ 2. Shapiro extended the validity range of the conjecture in a series of 1982 papers. In her first paper [14] she demonstrated that if f has a linear factor of multiplicity greater than n/3, then the conjecture holds. This proves the conjecture for n = 3. Her second paper [15] shows that if f = gn/2 where g is quadratic, then the conjecture holds. This, combined with [14], proves the conjecture for n = 4 and 5. Her final paper [16] showed that the conjecture holds if f is a power of a cubic factor. This is sufficient to prove the conjecture for order 6 in the form we are interested in, where f is a power of an irreducible polynomial, but not Kippenhahn's orginal form. Buckley recently gave a proof of the same result as a corollary to a more general result about Weirstrass cubics [3]. In his 1987 paper [8], Laffey disproved the simple form of the conjecture for n = 8 with a single counterexample: , 10 −16 12 22 −8 36 0 −120 (1)  H = −122 0 12 −122 −6 0 −6 −218 12 −12 18 −30 −16 −28 18 20 26 −16 10 0 44 8 24 12 −30 26 18 18 −12 −16 −28 20 24 8 0 44 −218 −2 −34 −10 −2 −216 −12 0 −216 −8 −34 0 −10 −12 −8 −120 −8 36 22 −4 0 0 0 0 0 0 −4 0 0 0 0 0 4 0 0 0 0 0 0 0 0 4 0 0 0 0 −8 0 0 0 0 −8 0 0 0 0 0 8 0 0 0 0 0 0 0 0 0 0 8 . 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0  K = In 1998 Li, Spitkovsky, and Shukla disproved Kippenhahn's more general form of the conjecture for n = 6 by constructing a family of counterexamples of the form f = det(I + xH + yK) = g2h, where both g and h are quadratics [11]. We have used our results to construct an one-parameter family of counterexam- ples of Kippenhahn's conjecture, for square matrices of order n ≥ 8, in Section 4. This places Laffey's single counterexample into a wider context, and provides a novel way for constructing polynomial matrices which non-trivially have square determi- nant. We refer to [6] for a positive solution to a quantized version of Kippenhahn's conjecture. 4 BEN LAWRENCE Linear matrix polynomials, also called linear pencils, are a key tool in matrix theory and numerical analysis (e.g. the generalized eigenvalue problem), and they frequently appear in (real) algebraic geometry (cf. [17, 12]). Furthermore, linear pencils whose coefficients are hermitian matrices give rise to linear matrix inequal- ities (LMIs). LMIs produce feasible regions of semidefinite programs (SDPs) [18], which are currently a hot topic in mathematical optimization. 2. Obstacles to generating the full algebra In this section we prove Theorem 1, and another theorem clearly linking the Burnside graph of a set of matrices to Burnside's theorem for matrix algebras. To begin with, we add an extra definition pertaining to Burnside graphs, and present an example. Definition 2 (Strongly connected component). A strongly connected compo- nent of a Burnside graph B(A) is a maximal subset of nodes in which every pair of nodes are path connected in both directions. Example 1. Suppose that  A1 = 1 1 0 0 1 0 0 1 0 1 0 0 0 0 1 1 1 1 1 1 0 1 1 0 1 0 0 1 0 0 1 0 1 0 1 0  and A2 =   . 1 0 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 2 0 0 0 0 0 0 2 0 0 0 0 0 0 2 The Burnside graph of {A1, A2} is as follows: The graph B(A) is not strongly connected. Nodes 6 and 3 form a strongly connected component, but there is no way to get to these nodes from the rest of the graph. The strongly connected components are {1, 2, 4, 5} and {6, 3}. Now we will prove Theorem 1: Proof of Theorem 1. Since B(A) is not strongly connected, there are two possibili- ties: (1) B(A) consists of at least 2 disconnected components, (2) B(A) is connected but not strongly connected. BURNSIDE GRAPHS AND THE KIPPENHAHN CONJECTURE 5 Strictly speaking, both of these cases can be dealt with at once, but we have sepa- rated them for clarity. Case 1 : The graph can be split into two subsets which do not connect to each other. Renumbering the nodes corresponds to symmetrically permuting the columns and rows of the set of matrices A and so does not affect the dimension of A. Suppose then that nodes 1, ..., j form one subset, and nodes j + 1, ..., k form the other, and there are no edges joining one subset to the other. This means that, for every Ai: • for each row up to and including row j, every entry beyond the jth column • for each column up to and including column j, every entry beyond the jth is zero, row is zero. This means that every matrix Ai is in block diagonal form, and so A (cid:54)= Mn(C). By way of illustration, if A1 and A2 are as follows: A1 = 1 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 1 0 0 0 0 1  , A2 =  0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 then the corresponding Burnside graph (and its permutation) is leading to the following block diagonal matrices obtained for the Ai: A1 = 1 0 1 0 0 0 1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 1 0  , A2 =  0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 1 0 0 Case 2 : Organise B(A) into strongly connected components, which may consist of a single node. Reorder columns and rows so that these strongly connected com- ponents consist of consecutive nodes. Reordering columns and rows will not affect the dimension of the algebra A. If every strongly connected component had at least one inbound edge and at least one outbound edge, then a cycle would exist and B(A) would be strongly connected.    ,  . 6 BEN LAWRENCE Therefore, without loss of generality we can assume that there is at least one node with no inbound edge. By symmetric reordering of rows and columns we can suppose that this is the first strongly connected component, associated with nodes {1, ..., j}, where j may be equal to 1. Recall how edges are defined: a directed edge i → j exists if and only if there is at least one matrix in A with a non-zero entry at the (i, j) position. Therefore, since the first strongly connected component is not the endpoint for any edge, every matrix in A must have all zeros in the first j columns, beyond the jth position:  j × j j × (n − j) 0 (n − j) × (n − j)  . In other words, it must be possible to rearrange the matrix rows and columns into this block form, with a zero block in the lower left. Therefore, the subspace spanned by the first j canonical basis vectors {e1, e2, ..., ej} of F k is invariant under every matrix in A and therefore also invariant under the algebra A, and so A (cid:54)= Mn(C). For example:  0 0 0 0 1 0 1 0 1 0 0 1 0 0 0 0  , A2 =  0 0 0 1 0 1 0 0 0 0 0 0 0 0 0 0  , A1 = To handle the situation where the strongly connected component has no outbound edge, take the transpose of every Ai, which will not affect the dimension of A, and (cid:3) repeat the argument. Now we will see the origin of the term Burnside graph. An algebra of linear transformations on a vector space is irreducible if the only invariant subspaces with respect to the algebra are {0} and the entire vector space. Recall Burnside's theorem for matrix algebras: Theorem 3 (Burnside [2]). Let V be a finite dimension vector space over C, with dim(V ) > 1. The only irreducible algebra of linear transformations on V is the full algebra Mdim(V )(C). We have then the following corollary. Corollary 1. Every matrix algebra A over C can be put into a block upper triangular form, where the diagonal blocks are full sub-algebras Mni(C) for some ni ∈ N. Proof. Let V be the vector space upon which A acts. Take the smallest A-invariant subspace of V , denoted U1. Then we have V = U1 ⊕ U⊥. Note that if the smallest BURNSIDE GRAPHS AND THE KIPPENHAHN CONJECTURE 7 invariant subspace is all of V , then straight away we have that A is the full algebra (by Theorem 3), which would trivially satisfy the corollary. So assume that U1 is a proper subspace of V . Let u1 be an orthonormal basis for U1, and v1 and orthonormal basis for U⊥. Then (u1, v1) is an orthonormal basis for V . With respect to this basis, then, A must have the form  ←− dim(U1) −→ ↑ ↓ 0 . . . . . . to ensure that U1 is invariant. Examine the dim(U1) block in the top left. Since we have assumed that U1 is the smallest invariant subspace of V , there can be no A invariant subspaces of U1. Treating this upper left block as a sub-algebra acting on u1 embedded within V , Burnside's theorem then tells us that this sub-algebra can only be the full algebra Mdim(U1)(C):   .  A = Mdim(U1)(C) 0 . . . . . . 1 , we can work our way through all of V and put A in Repeating this process for U⊥ the required block upper triangular form, with copies of the full algebra of various sizes down the diagonal:  A = Mdim(U1)(C) . . . Mdim(U2)(C) . . . . . . Mdim(U3)(C) 0 . . . . . . . . . . . . . . . . . . . . . Mdim(Un)(C)  . (cid:3) The connection with the Burnside graph is now clear. Putting a set of matrices A into what we may call Burnside form, as in Corollary 1, gives a Burnside graph where the strongly connected components correspond to the diagonal blocks, and the non-zero upper blocks correspond to the connections between strongly connected components, making allowance for left and right invariance corresponding to inbound and outbound edges. Corollary 2. If every Ai in A is hermitian, and B(A) is not strongly connected, then every Ai in A can be put simultaneously in block diagonal form with an orthogonal transformation. 8 BEN LAWRENCE Proof. If every matrix in A is hermitian, then every element Ai,jk of each matrix Ai contributes the same edge to B(A) as does Ai,kj, but with the direction reversed. Therefore in this case B(A) will partition into strongly connected components, be- cause there will be no one-way edges. This means that, for hermitian A, if B(A) as a whole is not strongly connected, it must be disconnected, and as Case 1 in the proof of Theorem 1, a basis exists which simultaneously block-diagonalises ev- ery Ai. We can make the disconnected nature of B(A) visible with an orthogonal (cid:3) transformation. 3. Burnside graphs and their associated algebras Laffey [9] shows for a pair of matrices, one of which is diagonal with distinct eigen- values, a strongly connected Burnside graph is all that is needed to guarantee the generation of the full matrix algebra. Without this distinct eigenvalue assumption, additional conditions are necessary. In this section, we provide a set of conditions on the submatrix blocks which guarantee generation of the full algebra. Of particular interest is the case where a pair of matrices have eigenvalues all of multiplicity 2. This is related to Kippenhahn's conjecture [5], to which we will construct a family of counterexamples in Section 4. First we will need some definitions, which will allow us to prove an expanded version of Theorem 2 given in the introduction, which ensures that the algebra A generated by a set of matrices A is the full matrix algebra. Definition 3 (p-block). Let p ∈ N, and let H be a real symmetric matrix of size pn × pn. A p-block of H is a p × p submatrix occupying columns p(i − 1) + 1 to pi and rows p(j − 1) + 1 to pj for some i, j = 1, ..., n. We denote such a p-block by Hij and say that it is in the ij-position in H. Definition 4 (p-word). Given a set {H (1), ..., H (m)} of pn× pn matrices over a field F , a p-word based at i1 and ending at iq is a matrix product H (k1) i1j1 H (k2) j1i1 ....H (kq) jq−1iq , where the second index of each entry matches the first index of the subsequent entry. For example H (1) 12 H (1) 24 H (2) 43 H (3) 33 is a p-word based at 1 and ending at 3 over the matrices {H (1), H (2), H (3)}. Definition 5 (Condition Multp). Let p ∈ N. A hermitian matrix K satisfies Con- dition Multp if its eigenvalues have maximum multiplicity p. For example, if K has eigenvalues {1,−1, 2, 3, 4, 4}, then K satisfies Condition Mult2. Definition 6 (Condition L− p). Let p ∈ N, and let H be a hermitian matrix of size pn × pn. Take a partition of n as (l1 = 1, l2 = 1, l3, l4...., lm) so that(cid:80)m Suppose also that each lj = (cid:80)k i=1 li for some k, with k < j. For example, i=1 li = n. (1, 1, 2, 4, 8) or (1, 1, 1, 1) or (1, 1, 2, 2). Identify square blocks along the top row of H, of non-decreasing size pl1 = p, pl2 = p, pl3, pl4..., plm, starting at the top left and proceeding along to the right. Note that the importance of p is that it sets the minimum size of the smallest pair of blocks with which the block sequence starts. BURNSIDE GRAPHS AND THE KIPPENHAHN CONJECTURE 9 If such a partition exists so that each of these blocks is invertible, we say that H satisfies condition L − p. An example to illustrate these partitions: Example 2. Consider the matrix  0 1 1 1 1 0 1 0 1 1 1 0 1 0 0 1 0 1 1 1 0 1 0 1 0 1 1 0 0 0 1 1 1 0 0 1 1 1 0 1 H = ... ... ... ...  . Only some entries of H are shown for clarity. Let p = 2, and take the partition (1, 1, 2, 2) of 6. Then H satisfies condition L − 2. We could also have taken the partition (1, 1, 1, 1, 1, 1) and H would still have satisfied Condition L − 2. In this example, notice how we could have used two different partitions of 6. The point is that the condition requires only the existence of a suitable partition. It does not refer to a specific partition. We will now we can present the full version of Theorem 2. Theorem 4 (Even-order constructibility). Let H and K be 2n × 2n hermitian matrices over C. Then if, in the basis in which K is diagonal with weakly ascending diagonal entries, (1) K satisfies Condition Mult2, (2) H satisfies Condition L − 2, (3) there exist distinct 2-words w1 and w2 both based at 1 so that w1wT 1 and w2wT 2 do not commute, then the algebra A generated by H and K is the full algebra M2n(C). Proof. Since K satisfies Condition Mult2, we can find an orthogonal transformation to put K in the form  k1 K =  , k2 . . . k2n where each of the ki appear with multiplicity at most 2, and there is at least one such pair. With a symmetric permuation of rows and columns, we can arrange for all such pairs eigenvalues to be consecutive. Since the diagonal entries {k1, ..., k2n} of K now all lie along the diagonal and all have multiplicity at most 2, with there being at least one such pair which must be consecutive (due to the ascending order assumption), we can define n distinct 10 BEN LAWRENCE polynomials {q1, ..., qn} of order 2n − 1 such that  qi(K) =  ∈ A, 02 ... ... 02 02 . . . . . . . . . . . . 02 ... ... . . . . . . I2 that is, for every i = 1, ..., n, qi(K) has I2 at the (i, i) 2-block position and zeroes elsewhere and such matrices are elements of the algebra A. Now consider qi(K) H qj(K). Conjugation in this way produces a matrix of the form  02 ... ... 02 02 . . . . . . Hij . . . . . . . . . 02 ... ... . . .  ∈ A, a matrix which consists of only the (i, j) 2-block of H at the (i, j) location, with zeros elsewhere. Since we are conjugating by elements of A, these isolated 2-blocks are themselves elements of A. The 2-words w1 and w2 can therefore be obtained by isolating each 2-block in each word and multiplying the word out. Since H is 1 and w2wT symmetric, wT 2 are symmetric, and moreover lie in the (1, 1) position. We have assumed that w1wT 1 and w2wT 2 do not commute, and so by considering available dimensions we can see that 1 and w2wT w1wT 2 are also both valid 2-words. Both w1wT 2 generate M2(C). Therefore, 1 and wT  M2(F ) ... ... . . . . . .  ⊆ A. Now we can use Condition L − 2 of H. Condition L− 2 requires that H12 is invertible and therefore so is H21 (since H is hermitian), and so we can move the copy of M2(F ) around by multiplying with  as follows: M2(C) ... ...  ,  02 02 H21 02 ... . . . 02 H12 02 02 ... . . . . . . . . . . . .   02 H12 02 02 ... . . .  = . . .  ∈ A  02 M2(C) . . . ... ...  , . . . BURNSIDE GRAPHS AND THE KIPPENHAHN CONJECTURE   02 02 H21 02 ... . . . . . . 02 02 H21 02 ... . . . ... ...   M2(C)   M2(C)  02 ... ... = . . . . . . . . . . . . . . . M2(C) ...  =  02    . . . . 02 H12 02 02 ... 02 02 M2(C) ... 11  ,  . . . . . . . . . . . . Since H12 and its transpose H21 is invertible, each of these products is equal to M2(C). Thus there are copies of M2(C) in the 11, 12, 21, and 22 positions. Therefore every possible 4×4 matrix over F exists in the top-left corner, since every such matrix can be partitioned into 4 blocks which are elements of M2(C). Therefore  M4(C) ... ... . . . . . .  ⊆ A. Condition L − 2 ensures that we can simply keep repeating the process. If the next invertible block is of size 1, we repeat the process on the copy M2(C) which lies in the top left. If on the other hand it is of size 4, we use our new copy of M4(C). The entire matrix is filled out in this way with copies of each full algebra of lesser order. (cid:3) Therefore A = M2n(C).  0 1 0 1 1 0 2 0 0 2 0 0 1 0 0 0  and C2 =  1 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0  . Example 3. Let C1 = Let C be the algebra generated by C1 and C2. The Burnside graph B({C1, C2}) is as follows: We can check off the requirements of Theorem 4 one by one: (1) C2 satsifies condition Mult2, (cid:19) (cid:18) 0 1 1 0 L-2, (cid:19) (cid:18) 0 1 2 0 (2) and are both invertible, and so C1 satisfies condition 12 (cid:18) 0 1 (cid:19) (cid:18) 0 1 (cid:19)(cid:18) 0 2 (cid:19) BEN LAWRENCE (cid:18) 1 0 0 4 = (cid:19) do not commute. (3) and 1 0 1 0 Therefore by Theorem 4, C = M4(C). 2 0 The key element of Theorem 4 is the non-commutativity requirement. This is what sets the process going by giving us a sub-algebra to start with. We can obtain a few more corollaries using the following theorem of Laffey: Theorem 5 (Laffey's Generation Theorem [9]). Let A = {A1, ..., Ak} be a set of n × n matrices over a field F . Let B be an n × n diagonal matrix over a field F with distinct diagonal entries. Let A be the algebra generated by A∪{B}. Then the following statements are equivalent for n > 1: (1) A is simple, (2) A = Mn(F ), (3) B(A) is strongly connected. Laffey's theorem helps us establish the following theorem: Theorem 6 (q-block constructibility). Let H and K be qn× qn hermitian matrices over C. Then if, in the basis in which K is diagonal with weakly ascending diagonal entries, (1) There is a set {v1, ..., vk} of q-words starting and ending at 1 such that the Burnside graph B({v1, ..., vk}) is strongly connected, (2) K satisfies Condition Multq, (3) H satisfies Condition L − q, (4) there is some q-word w based at 1 so that wwT is diagonal and has q unique eigenvalues, then the algebra A generated by H and K is the full algebra M2n(C). Proof. Since H is hermitian, wwT will also be hermitian and can therefore be di- agonalised. Diagonalising wwT can be done by conjugating H and K with the Kronecker product P ⊗ In, where P is the unitary diagonalisation matrix of wwT . Because K itself is a Kronecker product Iq ⊗ D, where D is some diagonal matrix, it will not be affected. We have assumed that wwT has unique eigenvalues - denote them by w1, ..., wq. Use the same block-isolation process as in the proof of Theorem for each j = 1, ..., k. Since we have assumed that B({v1, ..., vk}) is strongly con- nected, Laffey's Generation Theorem gives us a copy of Mq(C) in the top left corner: 4, to obtain that w1 . . . wq  ∈ A,  ∈ A, . . . . . .  vj  ⊆ A.  Mq(C) ... ... BURNSIDE GRAPHS AND THE KIPPENHAHN CONJECTURE 13 The rest of the proof follows as in Theorem 4, using Condition L − q to distribute copies of Mq(C) all around the remaining positions. (cid:3) Theorem 1 actually gives a condition that is in a generic sense necessary and sufficient. Recall the notion of 'generic' from algebraic geometry: a property holds generically if it holds except on a proper Zariski-closed subset. Corollary 3. Let A = {A1, . . . , Ak}, where k > 1, be a set of n × n matrices over C with strongly connected Burnside graph B(A). Let A denote the algebra generated by A. If A is a generic tuple, then A = Mn(C). Proof. If A (cid:40) Mn(C), then the σ × n2 matrix, where σ = kn2−1−k , obtained by k−1 forming all products of the Aj matrices of length less than or equal to n2 and flattening them has rank strictly less than n2. This can be expressed with the vanishing of all n2 × n2 minors, so is a Zariski closed condition. Thus it suffices to find a tuple A with the given graph B(A) that generates the full matrix algebra. By Laffey's Theorem 5, we can simply take an n × n diagonal matrix A1 with n distinct eigenvalues, and a single matrix A2 which is the adjacency matrix of the strongly connected Burnside graph B(A), and A({A1, A2}) = Mn(C). (cid:3) Likewise, condition (4) in Theorem 6 holds generically. We thus have: Corollary 4. Let H and K be real qn× qn symmetric matrices over C generic with respect to the following properties: (1) There is a set {v1, ..., vk} of q-words starting and ending at 1 such that the Burnside graph B({v1, ..., vk}) is strongly connected, (2) K satisfies Condition Multq, (3) H satisfies Condition L − q. Then the algebra generated by H and K is the full algebra Mqn(C). Proof. The matrix H1iHi1 is q × q and the condition of having q unique eigenvalues is Zariski open. It thus suffices to find an example where (1)-(3) hold and H1iHi1 has q unique eigenvalues. We will construct a suitable H and K. Take the Burnside graph B({v1, ..., vk}), and construct H11 as the adjacency matrix of this graph. Place this q × q matrix in top left position of H. Then construct H12 = H21 = diag(h1, ..., hq), where the hi are distinct and positive. Place a copy of H12 in every remaining position, to fill out H. Choose n different eigenvalues ki, and define K = diag(k1Iq, ..., knIq). Together H and K satisfy all the conditions of Theorem (cid:3) 6, and so the theorem holds generically. 4. Kippenhahn's Conjecture Here we put Theorem 4 to use, and construct a one-parameter family of coun- terexamples to Kippenhahn's conjecture. Remark 1. The interest in Kippenhahn's conjecture can also be illustrated geomet- rically. Given a linear pencil L = I + xH + yK as in Conjecture 1, its determinant f = det L gives rise to the affine scheme Spec C[x, y]/(f ). If condition (1) in Con- jecture 1 holds, then then this scheme is obviously nonreduced - see for example Chapter 5, Section 3.4 of [13]. 14 BEN LAWRENCE 4.1. Existing counterexample. Recall Laffey's counterexample (1) from the in- troduction. It satisfies the requirements of Theorem 4. In particular, H satisfies Condition L − 2, via the partition (1, 1, 2) of 4. In the rest of this paper, we will use Theorem 4 to present the construction of an entire family of counterexamples to the strong form of Kippenhahn's conjecture, of order 8 and above. 4.2. Family of counterexamples for 8 × 8 and above. Let us define: (cid:18) 1 (cid:19) α = 0 0 −1 (cid:19) (cid:18) 0 b b 0 , β = , U = (cid:18) 0 −1 (cid:19) 1 0 , where b ∈ R and b (cid:54)= 0. The key properties of these matrices are listed in the following lemma: Lemma 1. These properties follow directly from the definitions of α, β, and U : (cid:18) 0 (cid:19) 2b 0 (cid:54)= 0, −2b (1) [α, β] = αβ − βα = (2) α2 = −U 2 = I2, (3) (α + β)2 = (1 + b2)I2, and , (4) αU + U α = βU + U β = 0. Define 2n × 2n matrices α 2U α −α 3U −α − β −α 0 A = U −α −α −α ... −α α α + β α . . . α α 0 4U 5U . . . nU  ,  U B =  . U . . . U   Both A and B are skew-symmetric. Recall our notation; we denote 2 × 2 blocks of a matrix by an ij subscript, so for example A14 = α + β. Now, define symmetric matrices H = A2, K = AB + BA. Use { , } to denote the anti-commutator, and consider the block form of K: {α + β, U} {α, U} . . . {α, U} 2U 2 {α, U} {α, U} 6U 2 −{α + β, U} −{α, U} −{α, U} {α, U} 4U 2 −{α, U} −{α, U} −{α, U} −{α, U} {α, U} {α, U} 8U 2 ... −{α, U} 10U 2 . . . 2nU 2  BURNSIDE GRAPHS AND THE KIPPENHAHN CONJECTURE 15 By Lemma 1, U α = −αU , U β = −βU , and U 2 = −I2. Therefore every off-diagonal block of K vanishes, and  K = −2I2 −4I2  , . . . −2nI2 and likewise H is of the form 0 −2 1 −2 b − 1 −b2 − n −b − 1 −2 −2 1 −b − 1 −7 0 −1 −1 −2 −7 −1 −1 0 −2 −1 −1 −11 0 −1 −1 0 −11 1 −3 b −b2 − 18 b − 2 −2 −1 1 − 3b −b − 2 −1 −b −2 3b + 1 −3 −1 b − 2 −2 b 0 . . . −3 1 − 3b −b − 2 −1 −b −2 0 0 1 − n −1 0 −1 0 −1 −b2 − 18 . . . −b  . 1 − n 0 0 −1 0 −1 b −1 ... 0  −b2 − n 0 −2 b − 1 1 −2 3b + 1 −3 ... 0 1 − n Note the regular structure of the 2-blocks of H. Where j ≥ 5, the H1j and Hj1 2-blocks are of the form , Hjj blocks are of the form 1 − n 0 −1 0 0 −1 −1 0 0 −1 −1 b (cid:18) 0 1 − j (cid:19) (cid:19) 1 − j 0 (cid:18) −1 − j2 0 0 −1 − j2 other Hij 2-blocks where i, j ≥ 5 are of the form that H and K violate Kippenhahn's conjecture. ... −b −1 . . . −1 − n2 . . . 0 −1 − n2 (cid:18) −1 (cid:19) b b −1 . All . We will now show (cid:18) −1 0 0 −1 (cid:19) , and H4j and Hj4 blocks are of the form Lemma 2. All of the eigenvalues of xH + yK have even multiplicity. Proof. Consider first Ax + By. Take some x0 and y0 in R, where x0 (cid:54)= 0. The eigenvalues of of the skew-symmetric matrix Ax0 + By0 will be purely imaginary and will exist in conjugate pairs. Note that a given pair may appear more than once. Denote such pairs by ±iλk, where λk is real and k ranges from 1 to n. Then the eigenvalues of (Ax0 + By0)2 will be −λ2 k, obviously coming in pairs. Since the same pair of eigenvalues of Ax0 + By0 may occur several times, we cannot say for sure that each −λ2 k has multiplicity 2, but we can be sure that it has even multiplicity. Let vk be an eigenvector of Ax0 + By0 with eigenvalue iλk, and wk be an eigenvector of Ax0 + By0 with eigenvalue −iλk. Since vk and wk belong to different eigenspaces of Ax0 + By0, the subspace span{v, w} which they generate is two-dimensional. AB + BA = K, and B2 = −I2n, so we have (Ax0 + By0)2 = Hx2 kvk and (Ax0 + By0)2wk = −λ2 Then (Ax0 + By0)2vk = −λ2 0 + Kx0y0 − I2ny2 0, kwk. But A2 = H, so (Hx2 0 + Kx0y0)vk = (y0 − λ2 k)vk. 16 BEN LAWRENCE Dividing through by x0 (cid:54)= 0 we have that (Hx0 + Ky0)vk = (y0 − λ2 k)vk. 1 x0 (y0 − λ2 Therefore, vk is an eigenvector of Hx0 + Ky0 with eigenvector 1 k). Repeat- x0 ing this process for wk, we see that wk is also an eigenvector of Hx0 + Ky0 with eigenvector 1 k). Therefore vk and wk span a two-dimensional eigenspace of x0 Hx0 + Ky0. (y0 − λ2 For the case where x0 = 0, we simply have Ky0, which clearly has paired eigen- values because of the diagonal structure of K. Therefore, for every x0 and y0 in R (cid:3) Hx0 + Ky0 has eigenvalues all of even multiplicity. Denote by A the algebra generated by H and K. The Kippenhahn Conjecture claims that A cannot be the full algebra M2n(C). We will show that it is in fact the full algebra. 4.3. The algebra generated by H and K. We will show that H and K satisfy the requirements of Theorem 4. H is clearly symmetric, and p = 2. We will first evaluate the characterstics of the 2-blocks of H, and draw the associated Burnside graph. Let us evaluate H12: H12 = U α + 2αU − 2α2 − βα. Applying the definitions of α, β and U , we see that H12 = αU − βα − 2I2. Evaluating this gives us H12 = (cid:18) −2 −1 − b (cid:19) , −1 + b −2 which has determinant 3 + b2 (cid:54)= 0. In the Burnside graph B(H, K) there must be a connection between nodes 1 and 3, nodes 2 and 4, and either nodes 1 and 4 or nodes 2 and 3, or both. Recall that because H is symmetric, all connections are bi-directional: Here the dashed edges indicate that at least one of them has to exist, potentially both. Now evaluate H13: which becomes H13 = U α + α2 + 3αU, (cid:18) 1 −2 (cid:19) −2 1 H13 = 2αU + I2 = = H31, determinant equal to -3. Add the extra connections to the Burnside graph: BURNSIDE GRAPHS AND THE KIPPENHAHN CONJECTURE 17 Likewise H14 = 3(α + β)U + I2 = with determinant −8 − 9b2 (cid:54)= 0 and for j = 5, ..., n, (cid:19) = H41, (cid:18) 1 + 3b −3 (cid:19) (cid:18) 0 −3 1 − 3b 1 − j 0 1 − j H1j = (j − 1)αU = = Hj1, all of which have non-zero determinant. Add these new nodes and edges, and we see that the Burnside graph is strongly connected: Now we will check off the requirements of Theorem 4 one by one. (1) K clearly satisfies condition Mult2. (2) We have already established that every H1j is invertible. H therefore satisfies Condition L − 2 via the partition (1, 1, ..., 1) of n. (3) Consider the single element 2-words H13 and H14. Then, and likewise H14H41 = (cid:19) H13H31 = 5 −4 (cid:18) 5 −4 (cid:18) 9 − (1 + 3b)2 (cid:18) 0 −6 , −6 (cid:19) 9 + (1 + 3b)2 (cid:19) [H13H31, H14H41] = −48b 48b 0 (cid:54)= 0. Now directly evaluate their commutator. After simplifying, we have: Therefore, H13 and H14 satisfy the third requirement of Theorem 4. The pair of matrices H and K therefore satsify the requirements of Theorem 4, and so A = M2n(C). Thus H and K as defined are a one-parameter family of counterexamples to Kippenhahn's Conjecture, for order 8 × 8 and greater. 18 BEN LAWRENCE References [1] Helmer Aslaksen and Arne B. Sletsjøe, Generators of matrix algebras in dimensions 2 and 3, Linear Algebra and its Applications 430: 1-6 (2009) [2] Matej Bresar, Introduction to Noncommutative Algebra, Springer (2014). [3] Anita Buckley, Indecomposable matrices defining plane cubics, Operators and Matrices Vol. 10, No. 4: 1059-1072 (2016) [4] Alan George and Khakim D. Ikramov, Common invariant subspaces of two matrices, Linear Algebra and its Applications 287: 171-179 (1999) [5] Rudolf Kippenhahn, Uber der Wertevorrat einer Matrix, Mathematische Nachrichten 6: 193- 228 (1951-52) [6] Igor Klep, Jurij Volcic, Free loci of matrix pencils and domains of noncommutative rational functions, Comment. Math. Helv. 92: 105-130 (2017). [7] Vladimir P. Kostov, The minimal number of generators of a matrix algebra, Journal of Dynam- ical and Control Systems Vol. 2, No. 4: 549-555 (1996) [8] Thomas J. Laffey, A counterexample to Kippenhahn's conjecture on hermitian pencils, Linear Algebra and its Applications 51: 179-183 (1983) [9] Thomas J. Laffey, A structure theorem for some matrix algebras, Linear Algebra and its Ap- plications 162-164: 205-215(1992) [10] Thomas J. Laffey, A criterion for generating full matrix algebras, Linear Algebra and its Applications 37: 191-198(1981) [11] Chi-Kwong Li, Ilya Spitkovsky, and Sudheer Shukla, Equality of higher numerical ranges of matrices and a conjecture of kippenhahn on hermitian pencils, Linear Algebra and its Applica- tions 270: 323-349 (1998) [12] Tim Netzer, Daniel Plaumann, and Andreas Thom, Determinantal representations and the Hermite matrix, Michigan Math. J. 62: 407-420 (2013). [13] Igor R. Shafarevich, Basic algebraic geometry. 2. Schemes and complex manifolds (Translated from the 2007 Russian edition by Miles Reid), 3rd edition, Springer (2013). [14] Helene Shapiro, On a conjecture of Kippenhahn about the characteristic polynomial of a pencil generated by two hermitian matrices (Part 1), Linear Algebra and its Applications 43: 201-221 (1982) [15] Helene Shapiro, On a conjecture of Kippenhahn about the characteristic polynomial of a pencil generated by two hermitian matrices (Part 2), Linear Algebra and its Applications 45: 97-108 (1982) [16] Helene Shapiro, Hermitian pencils with a cubic minimal polynomial, Linear Algebra and its Applications 48: 81-103 (1982) [17] Victor Vinnikov, Selfadjoint determinantal representations of real plane curves, Math. Ann. 296: 453-479 (1993). [18] Henry Wolkowicz, Romesh Saigal and Lieven Vandenberghe (eds), Handbook of semidefinite programming: theory, algorithms, and applications, Kluwer Academic Publishers (2000). Department of Mathematics, University of Auckland E-mail address: [email protected]
1511.00877	2	1511	2016-06-06T10:45:31	X-simple image eigencones of tropical matrices	[ "math.RA" ]	We investigate max-algebraic (tropical) one-sided systems $A\otimes x=b$ where $b$ is an eigenvector and $x$ lies in an interval $X$. A matrix $A$ is said to have $X$-simple image eigencone associated with an eigenvalue $\lambda$, if any eigenvector $x$ associated with $\lambda$ and belonging to the interval $X$ is the unique solution of the system $A\otimes y=\lambda x$ in $X$. We characterize matrices with $X$-simple image eigencone geometrically and combinatorially, and for some special cases, derive criteria that can be efficiently checked in practice.	math.RA	math	X-simple image eigencones of tropical matrices J´an Plavkaa,1, Sergeı Sergeevb,∗ aDepartment of Mathematics and Theoretical Informatics, Technical University, bUniversity of Birmingham, School of Mathematics, Edgbaston B15 2TT, UK B. Nemcovej 32, 04200 Kosice, Slovakia Abstract We investigate max-algebraic (tropical) one-sided systems A ⊗ x = b where b is an eigenvector and x lies in an interval X. A matrix A is said to have X-simple image eigencone associated with an eigenvalue λ, if any eigenvector x associated with λ and belonging to the interval X is the unique solution of the system A ⊗ y = λx in X. We characterize matrices with X-simple image eigencone geometrically and combinatorially, and for some special cases, derive criteria that can be efficiently checked in practice. Keywords: Max algebra, one-sided system, weakly robust, interval analysis AMS classification: 15A18, 15A80, 93C55 1. Introduction 1.1. Problem statement and main results In this paper, by max algebra we mean the set of nonnegative numbers R+ equipped with usual multiplication a⊗ b := a· b and idempotent addition a ⊕ b := max(a, b). Algebraically, R+ equipped with these operations forms a semifield. The operations of max algebra are then extended to matrices and vectors in the usual way, giving rise to an analogue of nonnegative linear algebra. Max-algebraic one-sided systems A ⊗ x = b and max-algebraic eigen- problem A ⊗ x = λx are two fundamental problems of max algebra whose ∗Corresponding author. Email: [email protected] Email addresses: [email protected] (J´an Plavka ), [email protected] (Sergeı Sergeev) 1Supported by APVV grant 04-04-12. Preprint submitted to Elsevier October 21, 2018 solution goes back to the works of Cuninghame-Green [9, 11], Vorobyev [21] and Zimmermann [22] and these two topics are thoroughlly discussed in any textbook of the max-plus (tropical) linear algebra [2, 3, 13]. Our intention is to consider the situation when the right-hand side of A ⊗ x = b is an eigenvector, and also when the solution has to lie in some interval of Rn +. + of the form X = ×n X i, i=1 where each X i is an arbitrary interval belonging to R+, with its upper end possibly equal to +∞ (and lower end possibly equal to 0). In particular, Rn + is an interval of itself. For each i we denote xi := inf X i and xi = sup X i. Then we also have x := (xi)n + we mean a subset of Rn By an interval of Rn i=1 = inf X and x := (xi)n i=1 = sup X. The notion of X-simple image eigencone, which we introduce next, is related to the concept of simple image set [4]. By definition, simple image set of A is the set of vectors b such that the system A ⊗ x = b has a unique solution. If the only solution of the system A ⊗ x = b is x = b, then b is called a simple image eigenvector. A matrix A is said to have X-simple image eigencone associated with a (fixed) eigenvalue λ, if any eigenvector x associated with the eigenvalue λ and belonging to the interval X is the unique solution in X for the system A⊗y = λx. The characterization of a matrix with X-simple image eigencone is described as the main result of the paper in Section 4. Let us now give more details on the organization of the paper and on the results obtained there. Section 2 is devoted to basic notions of max algebra and its connections to the theory of digraphs and max-algebraic (tropical) convexity. In particular, we revisit here the spectral theory, focusing on the eigencone associated with an arbitrary eigenvalue, the generating matrix and the critical graph. Some aspects of the diagonal similarity scaling are also briefly discussed. Section 3 starts by discussing the problem of covering the node set of a digraph by ingoing edges. We proceed with the theory of one-sided systems A ⊗ x = b where we describe the solution set to such systems and start analysing the case when b is an eigenvector of A. The main result of that section is Theorem 3.10, which characterizes matrices that have at least one simple image eigenvector corresponding to an eigenvalue λ. In the beginning of Section 4 we first discuss the relation between X- simple image eigencone and X-weak robustness of a matrix. We then develop an interval version of theory of one-sided systems A ⊗ x = b, i.e., when x has to belong to an interval X. The second part of the section contains the main results of the paper: Theorem 4.16 and Theorem 4.17. More precisely, 2 Theorem 4.16 characterizes when A has X-simple image eigencone in general, and Theorem 4.17 focuses on the case when X is of a certain special type. In the end of the paper we formulate some conclusions and discuss some directions for further research. 1.2. Motivations In the literature, max algebra often appears as max-plus semiring devel- oped over the set R ∪ {−∞} equipped with operations a ⊗ b := a + b and a ⊕ b := max(a, b). However, this semiring is isomorphic to the semiring defined above, via a logarithmic transform. Max-plus algebra plays the cru- cial role in the study of discrete-event dynamic systems connected with the optimization problems such as scheduling or project management in which the objective function depends on the operations maximum and plus. The main principle of discrete-event dynamic systems consisting of n entities (ma- chines [9, 11], processors [7], computers, etc.) is that the entities work inter- actively, i.e., a given entity must wait before proceeding to its next event until certain others have completed their current events. Cuninghame-Green [9] and Butkovic [3] discussed a hypothetical industrial discrete-event dynamic system and a multiprocessor interactive system, respectively, which can be described by the interferences using recurrence relations xi(r + 1) = max(x1(r) + a1i, x2(r) + a2i, . . . , xn(r) + ani), i ∈ {1, 2, . . . , n}. The formula expresses the fact that entity i must wait with its r + 1st cycle until entities j = 1, . . . , n have finished their rth cycle. The symbol xi(r) denotes the starting time of the rth cycle of entity i, and aij is the corre- sponding activity duration at which entity ej prepares the outputs (products, components, data, atc.) for entity ei. The steady states of such systems cor- respond to eigenvectors of max-plus matrices, therefore the investigation of properties of eigenvectors is important for the above mentioned applications. In max-plus algebra the matrices for which the steady states of the sys- tems are reached with any nontrivial starting vector are called robust. Such matrices have been studied in [3], [19]. The matrices for which the steady states of the systems are reached only if a nontrivial starting vector is an eigenvector of the matrix are called weakly robust. Efficient characteriza- tions of such matrices are described in [6]. In practice, the values of starting vector are not exact numbers and usually they are rather contained in some intervals. Considering matri- ces and vectors with interval entries is therefore of practical importance. 3 See [12, 14, 15, 16, 18] for some of the recent developments. In particular, the weak robustness of an interval matrix is studied in [17]. The aim of this paper is to characterize the weak X-robustness, i.e., the weak robustness of matrices with initial times confined in an interval vector X, using X-simplicity of image eigencone. 2. Preliminaries 2.1. Matrices and graphs Many problems of max algebra can be described and resolved in terms of digraphs (i.e., directed graphs). Let us give some of the relevant definitions here. Definition 2.1 (Associated digraphs). The weighted digraph associated with A ∈ Rn×n is the digraph G(A) = (N, E) with the node set N := {1, . . . , n} and the edge set E such that (i, j) ∈ E (edge from i to j) if and only if aij > 0. The number aij is called the weight of (i, j). + Definition 2.2 (Paths). A path in the digraph G(A) = (N, E) is a sequence of nodes p = (i1, . . . , ik+1) such that (ij, ij+1) ∈ E for j = 1, . . . , k. A path p is closed if i1 = ik+1, elementary if all nodes are distinct, and a cycle if it is closed and elementary. The number k is the length of the path p and is denoted by l(p). Definition 2.3 (Strongly connected components). By a strongly con- nected component (for brevity s.c.c.) of G(A) = (N, E) we mean a subdigraph G′ = (N ′, E′) with N ′ ⊆ N and E′ ⊆ E, such that any two distinct nodes i, j ∈ N ′ are contained in a common cycle and N ′ is a maximal subset of N with that property. Particularly, G(A) is strongly connected if G′ = (N ′, E′) is strongly connected component of G(A) with N ′ = N and E′ = E. Powers of max algebraic matrices are closely related to optimization on is digraphs. Observe that the i, jth entry of the power A⊗k := A ⊗ . . . ⊗ A } the biggest weight among all paths of length k connecting i to j. If we define the formal series A+ =L∞ k=1 A⊗k then the i, jth entry of A+ (possibly diverging to +∞) equals to the greatest weight among all paths connecting i to j. Such weight is guaranteed to be finite if the weight of any cycle in G(A) does not exceed 1. k {z 4 Definition 2.4 (Irreducibility). A matrix A ∈ Rn×n if G(A) is strongly connected, and reducible otherwise. Definition 2.5 (Graph restrictions). For arbitrary K ⊆ N, we denote by G(A)K the subgraph of G(A) consisting of all nodes of K and all edges of G(A) between the nodes of K. 2.2. Geometry is called irreducible + Max algebra also gives rise to the max-algebraic (tropical) analogue of convexity. Definition 2.6 (Max cone). A subset K ⊆ Rn have 1) λx ∈ K for any λ ≥ 0 and x ∈ K, 2) x ⊕ y ∈ K for any x, y ∈ K. + is called a max cone if we The name "max cone" was suggested in [6]. In the literature this object also appears as tropical cone or max-plus linear space. Definition 2.7 (Column span). For A ∈ Rm×n column span as + define its max-algebraic span⊕(A) :=( n Mj=1 A·jxj : xj ≥ 0 ∀j) , (1) where A·j, for j = 1, . . . , n denotes the j-th column of A. The set of all positive vectors in span⊕(A) will be denoted by span+ ⊕(A). It is easily shown that span⊕(A) is a max cone. Furthermore, span⊕(A) is always closed in the Euclidean topology [6]. Consider now the following operator. Definition 2.8 (Projector). Let W be a closed max cone. Define PW (x) := max{y ∈ W : y ≤ x}. (2) brevity. In the case W = span⊕(A) we will write PA instead of Pspan⊕(A), for PW is a nonlinear projector on the max cone W. It is homogeneous (PW(λx) = λPWx) and isotone (x ≤ y ⇒ PW x ≤ PW y.) These operators are crucial for tropical convexity: see, e.g., [8]. 5 2.3. Eigenvalues and eigenvectors Definition 2.9 (Eigencone). The set V (A, λ) = {x : A ⊗ x = λx}, (3) + where A ∈ Rn×n and λ ≥ 0, is called the (max-algebraic) eigencone of A associated with λ. The nonzero vectors of V (A, λ) are (max-algebraic) eigenvectors of A associated with λ. The set of all positive vectors in V (A, λ) will be denoted by V +(A, λ). Note that V (A, λ) consists of the eigenvectors associated with λ and vec- tor 0. Obviously, V (A, λ) is a max cone. Definition 2.10 (Maximum cycle geometric mean). Let A = (aij) ∈ Rn×n + . For any i1, . . . , ik ∈ N, the geometric mean of the cycle (i1, i2 . . . , ik) is defined as k√ai1i2 · . . . aiki1. The maximum cycle geometric mean of A ∈ Rn×n equals to + max k=1 max 1≤i1,...,ik≤n k√ai1i2 · . . . aiki1. λ(A) := n (4) λ(A) is the greatest max-algebraic eigenvalue of A, for any A ∈ Rn×n + . If A is irreducible then λ(A) is the only max-algebraic eigenvalue of A(e.g. [3], Theorem 4.4.8). Reducible A ∈ Rn×n + may have up to n max-algebraic eigenvalues, in general. We next give some elements of the spectral theory of reducible matrices. Although that theory is usually developed in terms of the Frobenius normal form and spectral classes [3] following a similar development of the spectral theory of nonnegative matrices, we choose not to do so, since in this paper we only need 1) the relation between the critical graph and the saturation graph, 2) the generating matrix of V (A, λ). Denote by Λ(A) the set of (max-algebraic) eigenvalues of A. General λ ∈ Λ(A) can be characterized as maximum cycle geometric mean of a certain subgraph of G(A). Definition 2.11 (Support). For x ∈ Rn called the support of x. + the set supp(x) = {i : xi > 0} is The proof of the following statement is standard, but we give it for the reader's convenience. 6 Proposition 2.12. Let λ ∈ Λ(A) and x ∈ V (A, λ) be nonzero. Then λ is the maximum cycle geometric mean of G(A)supp(x). Proof: Take any cycle (i1, . . . , ik) with all indices belonging to supp(x). Multiplying the inequalities ailil+1xil+1 ≤ λxil for l = 1, . . . , k − 1 and aiki1xi1 ≤ λxik , and cancelling all the coordinates of x we obtain that the cycle mean of (i1, . . . , ik) does not exceed λ. Now, start with any j0 ∈ supp(x) and find j1 such that aj0j1xj1 = λxj0. We again have j1 ∈ supp(x). Proceeding this way we obtain a cy- cle (jt, . . . jt+l) (for some t and l) with the cycle mean equal to λ. (cid:3) Let us also recall a useful link between the support of an eigenvector of A and the zero-nonzero pattern of A. and J, L ⊆ N. AJ,L denotes the submatrix of A with row + , x ∈ V (A, λ) and N ′ := Let A ∈ Rn×n + index set J and column index set L. Proposition 2.13 (e.g. [3], p.96 ). Let A ∈ Rn×n supp(x). Then AN \N ′,N ′ = 0. Definition 2.14 (Critical graph Gc(A, x, λ)). For x ∈ V (A, λ), define the critical graph Gc(A, x, λ) as the subgraph of G(A) consisting of all nodes and edges belonging to the cycles of G(A)supp(x) whose geometric mean is equal to λ. Definition 2.15 (Saturation graph). For x ∈ V (A, λ), the saturation graph Sat(A, x, λ) is the subgraph of G(A) with set of nodes N and set of edges ESat = {(i, j) : aijxj = λxi 6= 0}. (5) Proposition 2.16 ([2], Theorems 3.96 and 3.98). For any x ∈ V (A, λ), (i) Every node i ∈ N such that xi 6= 0 has an outgoing edge in Sat(A, x, λ); (ii) Any cycle in Sat(A, x, λ) belongs to Gc(A, x, λ); (iii) Gc(A, x, λ) is a subgraph of Sat(A, x, λ); Definition 2.17 (Critical graph Gc(A, λ)). The critical graph Gc(A, λ) as- sociated with λ and the set of nodes N λ associated with λ: Gc(A, λ) := Gc(A, x′, λ), N λ := supp(x′), where for x′ ∈ V (A, λ) such that supp(x) ⊆ supp(x′)∀x ∈ V (A, λ). (6) 7 lated from each other. Each Gc(A, x, λ) consists of several strongly connected components iso- In the following proposition let us collect some facts about the relation between Gc(A, x, λ) and Gc(A, λ). Proposition 2.18. Let A ∈ Rn×n (i) Gc(A, x, λ) ⊆ Gc(A, λ); (ii) Gc(A, x, λ) = Gc(A, λ)supp(x); (iii) Gc(A, x, λ) consists of entire strongly connected components of Gc(A, λ). + , λ ∈ Λ(A) and x ∈ V (A, λ). Then Proof: We only prove (ii) and (iii) since (i) follows from any of them. (ii): Observe that both Gc(A, x, λ) and Gc(A, λ)supp(x) consist of the nodes and edges of the cycles on supp(x) that have the cycle geometric mean λ, hence Gc(A, x, λ) = Gc(A, λ)supp(x). (iii): Since Gc(A, x, λ) is defined as a subgraph consisting of all nodes and edges of some critical cycles, it consists of several isolated strongly connected components, and each of these components is a subgraph of a component of Gc(A, λ). It remains to prove that none of these subgraphs is proper. By the contrary, suppose that one of these components is a proper sub- graph of a component of Gc(A, λ). Since Gc(A, x, λ) = Gc(A, λ)supp(x), the component of Gc(A, λ) should contain a node in supp(x) and a node not in supp(x), otherwise it coincides with the component of Gc(A, xλ). However, this contradicts with Proposition 2.13. Hence the claim. (cid:3) We further define a generating matrix of V (A, λ). For that, first define the matrix Aλ with the columns (Aλ)·i =(A·i/λ if i ∈ N λ, otherwise. 0, (7) Here N λ is as in (6). A+ not exceed 1. λ is finite, since the weight of any cycle in Aλ does Definition 2.19 (Generating Matrix). Let N λ,1 be the node sets of the strongly connected components of Gc(A, λ), and let j1, . . . , jk be the first , . . . , N λ,k c c 8 indices in those components. Define the generating matrix of V (A, λ) as the matrix resulting from stacking the columns (A+ λ )·j1, . . . , (A+ λ )·jk together: GA,λ = [(A+ λ )·j1, . . . , (A+ λ )·jk] Proposition 2.20 ([3] Coro. 4.6.2). For any nonzero λ ∈ Λ(A) V (A, λ) = V (Aλ, 1) = span⊕(GA,λ). 2.4. Invertible matrices and diagonal similarity scaling (8) (9) The class of invertible matrices in max algebra is quite thin. In fact it coincides with that in nonnegative algebra, consisting of all products of positive diagonal and permutation matrices. The positive diagonal matrices will be especially interesting to us since they give rise to a particularly useful visualization scaling, also known as a Fiedler-Pt´ak scaling. For a positive vector x ∈ Rn + whose ith diagonal entry is xi and all off-diagonal entries are 0. +, denote by diag(x) matrix X ∈ Rn×n Proposition 2.21 ([20], Theorem 3.7). Let A ∈ Rn×n x ∈ Rn + . For a positive +, let X = diag(x) and A = X −1AX with entries aij for i, j = 1, . . . , n. (i) If x satisfies A ⊗ x ≤ x then aij ≤ 1 and, in particular, aij = 1 for (i, j) ∈ Ec(A, 1) (visualization scaling) . (ii) There exists a positive x satisfying A ⊗ x ≤ x such that aij ≤ 1, and aij = 1 if and only if (i, j) ∈ Ec(A, 1) (strict visualization scaling). Definition 2.22 (Visualization). A matrix A = (aij) ∈ Rn×n is called visualized if aij ≤ λ(A) for all i, j ∈ N and strictly visualized if it is visualized and aij = λ(A) holds only for (i, j) ∈ Gc(A, λ). + We will also use the following observation about the diagonal similarity scaling, where by Sis V (A, λ) we denote the set of vectors in V (A, λ) that belong to the simple image set of A. Lemma 2.23. Let A ∈ Rn×n and A := X −1AX. Then (i) Λ(A) = Λ( A), and Gc(A, λ) = Gc( A, λ) for every λ ∈ Λ(A); (ii) y ∈ V (A, λ) ⇔ X −1y ∈ V ( A, λ); 9 + , let X ∈ Rn×n be a positive diagonal matrix. + (iii) y ∈ Sis V (A, λ) ⇔ X −1y ∈ Sis V ( A, λ). Proof: The facts described in part (i) are well-known. Parts (ii) and (iii) follow from the observation that A ⊗ y = b ⇔ X −1A ⊗ y = X −1b ⇔ X −1AX ⊗ (X −1y) = X −1b for all y, b ∈ Rn×n + . (cid:3) Let us recall some properties of A+ λ and GA,λ when A is visualized. By E we denote a matrix of the same dimensions as A whose every element is 1. Proposition 2.24 ([20], Proposition 4.1). Let A ∈ Rn×n be visualized and let λ = λ(A). Let N 1 c be the node sets of the components of Gc(A, λ), and let K = {j1, . . . , jk} be the index set of the columns of A+ forming GA,λ. (i) For each r, s ∈ K there exists αrs ≤ 1 such that (A+ c = αrsEN r c , . . . , N k c N s c λ )N r c N s + λ and (GA,λ)N r c ,s = αrsEN r c ,js. (ii) If r = s then αrs = 1. (iii) There exists s such that αrs < 1 for all r 6= s. Proof: Parts (i) and (ii) follow from [20] Proposition 4.1, part 2. For part (iii), note that if it does not hold, then there exist indices i1, . . . , il belonging to K such that αi1i2 = 1, . . . , αili1 = 1. This implies existence of a cycle in Gc(A, λ) going through different strongly connected components, contradicting the fact that they are isolated. (cid:3) 3. One-sided systems and simple image eigenvectors 3.1. Solving max-algebraic one-sided systems In this section we shall suppose that A ∈ Rm×n is a given matrix and recall the crucial results concerning a system of linear equations A ⊗ x = b. Our notation is similar to that introduced in [3], [22]. However, unlike for example in [3], Section 3.1, we do not assume that b has full support (i.e., is positive), or even that every row and column of A contains a nonzero element. Denote + S(A, b) = {x ∈ Rn 10 + : A ⊗ x = b}. (10) For any j ∈ N denote γ∗ j (A, b) = max{α ∈ R+ ∪ {+∞} : αA·j ≤ b} = min i : aij 6=0 bia−1 ij , assuming that 0 · (+∞) = (+∞) · 0 = 0, M = {1, . . . , m} and Mj(A, b) = {i ∈ M : aijγ∗ j (A, b) = bi 6= 0}. (11) (12) j=1 is closely related to the projection: PA(b) = j (A, b))n Vector γ∗(A, b) = (γ∗ A ⊗ γ∗(A, b). the corresponding column is zero: A·j = 0. γ∗(A, b) can have a +∞ component in general, but only in the case when The following lemma is crucial for the theory of A ⊗ x = b. Lemma 3.1 (e.g. [3] Theorem 3.1.1). x ∈ S(A, b) if and only if x ≤ γ∗(A, b) and ∪j : xj =γ∗ j (A,b)Mj(A, b) = supp(b). We now give a description of the solution set of A ⊗ x = b in terms of minimal coverings. Theorem 3.2. Let Ω be a collection of minimal subsets N ′ ⊆ N such that Sj∈N ′ Mj(A, b) = supp(b). Then S(A, b) = [N ′∈Ω SN ′(A, b), (13) where SN ′(A, b) = {x ∈ Rn + : xj = γ∗ j (A, b) for j ∈ N ′, xj ≤ γ∗ j (A, b) for j ∈ N\N ′}. (14) Proof: We first show that S(A, b) ⊆SN ′⊆Ω SN ′(A, b). If x ∈ S(A, b) then by Lemma 3.1 we have x ≤ γ∗(A, b) and Sj : xj=γ∗ j (A,b) Mj(A, b) = supp(b). Hence x ∈ SN ′(A, b) for some minimal N ′ ⊆ {j : xj = γ∗ j (A, b)} such that ∪j∈N ′Mj(A, b) = supp(b). Let x ∈ SN ′(A, b). Then Lj∈N ′ A·jxj = b and Lj∈N \N ′ A·jxj ≤ b, hence A ⊗ x = b. (cid:3) 11 Corollary 3.3. Let us have ∪j∈N ′Mj(A, b) = supp(b) for some proper N ′ ⊆ N, and let i /∈ N\N ′. Then for each α ≤ γ∗ i (A, b) there exists x ∈ SN ′(A, b) ⊆ S(A, b) with xi = α. Proof: We can assume without loss of generality that N ′ is a minimal subset such that ∪j∈N ′Mj(A, b) = supp(b). The claim then follows from (14) and (13). (cid:3) Let us now formulate, without proof, conditions for existence and unique- ness of a finite solution to A⊗x = b. Here S(A, b) := {x ∈ (R∪{+∞})n : A⊗ x = b}. Proposition 3.4 (e.g., [3], Coro. 3.1.2). Let A ∈ Rm×n Then the following conditions are equivalent: and b ∈ Rm + . + Mj(A, b) = supp(b). (i) S(A, b) 6= ∅, (ii) γ∗(A, b) ∈ S(A, b), (iii) Sj∈N Proposition 3.5 (e.g., [3], Coro. 3.1.3). Let A ∈ Rm×n and b ∈ Rm let the solution to A ⊗ x = b exist. Then the following are equivalent: + + , and (i) A ⊗ x = b has unique solution; (ii) S(A, b) = {γ∗(A, b)}; (iii) ∪j6=iMj(A, b) 6= supp(b) ∀i : γ∗ 3.2. Digraph coverings and systems with eigenvector on the right-hand side i (A, b) 6= 0. Let G = (N, E) be a strongly connected digraph. Define Mj(G) = {i ∈ N : (i, j) ∈ E} (15) Theorem 3.6. Let G = (N, E) be a strongly connected digraph. Then, i ∈ N with the property ∪j∈N \{i}Mj(G) = N exists if and only if G is not a cycle. Proof: Observe first that if G is a cycle then there are no nodes with such property. 12 To prove the converse, observe that if G is not a cycle, then it contains two intersecting cycles, and one of the nodes in the intersection will have at least two ingoing edges. Take a node with at least two incoming edges, and number this node as 1. Put N0 = ∅ and N1 := {1}. For each l ≥ 1 let Mj(G). Observe that since Sj∈Nl−1 Nl+1 = Nl ∪ [j∈Nl Mj(G) ⊆ Nl, we can replace (16) with Nl+1 = Nl ∪ [j∈Nl\Nl−1 Mj(G). (16) (17) (18) Hence for each l ≥ 1 we have Nl+1\Nl ⊆ [j∈Nl\Nl−1 Mj(G). Since G is finite, there is t > 1 that satisfies Nt+1 = Nt and Nt−1 6= Nt. Observe that Nt = N, for if Nt is a proper subset of N, then this contradicts the assumption that G is strongly connected. Consider the case when Nt\Nt−1 > 1. Observe that a covering of N can be built by taking all nodes in Nt−1 and a node in Nt\Nt−1 that has an ingoing edge from 1, or just the nodes in Nt−1 if one of the nodes of Nt−1 has an ingoing edge from 1. Therefore in both of these cases there exists i ∈ Nt\Nt−1 with ∪j∈N \{i}Mj(G) = N. Consider the remaining case when Nt\Nt−1 = 1 and when the node forming Nt\Nt−1 is the only node that has an ingoing edge from 1. Since N2\N1 > 1 there exists an s < t with Ns+1\Ns < Ns\Ns−1. Since Ns\Ns−1 does not contain a node to which 1 is connected with an edge, for some node i ∈ Ns\Ns−1 we have Ns+1\Ns = ∪j∈Ns\Ns−1,j6=iMj(G). (19) Combining this with (18) for all other l 6= s we obtain that This completes the proof. (cid:3) N = ∪j6=iMj(G). 13 Corollary 3.7. For each x ∈ V (A, λ), an i with the property ∪j6=iMj(Gc(A, λ, x)) = Nc(A, x, λ) exists if and only if Gc(A, x, λ) is not a union of disjoint cycles. We now briefly examine the link to one-sided systems with eigenvector on the right-hand side. Proposition 3.8. Let x ∈ V (A, λ), and let j ∈ N be such that there exists l with (l, j) ∈ Sat(A, x, λ). Then (i) x ≤ γ∗(A, λx); (ii) γ∗ j (A, x) = xj and Mj(A, λx) = Mj(Sat(A, x, λ)); (iii) Mj(Gc(A, λ)) ⊆ Mj(A, λx). Proof: follows by Lemma 3.1. (ii): Since x ∈ V (A, λ), we have akjxj ≤ λxk for all k, so γ∗ j (A, λx) = mini∈N{λxi(aij)−1} = λxl(alj)−1 = xj. Mj(A, λx) = {i : aijxj = λxi} follows from the definition of these sets, sub- stituting γ∗ (i): j (A, λx) = xj. (iii): By (ii) we have Mj(A, λx) = Mj(Sat(A, x, λ), and also Mj(Gc(A, x, λ)) ⊆ Mj(Sat(A, x, λ)) = Mj(A, λx) since Gc(A, x, λ) ⊆ Sat(A, x, λ) by Proposition 2.16 part (iii). Graph Gc(A, x, λ) consists of entire components of Gc(A, λ) (Proposition 2.18 part (iii)), hence Mj(Gc(A, λ)) = Mj(Gc(A, x, λ)) ⊆ Mj(Sat(A, x, λ)) = Mj(A, λx). This concludes the proof. (cid:3) 3.3. Simple image eigenvectors Denote by A(i) the matrix which remains after the ith column of A is removed. and + . Then, b belongs to the simple image set of A if and only if there is Lemma 3.9 (e.g. [3] Theorems 3.1.5 and 3.1.6). Let A ∈ Rm×n b ∈ Rm no i for which γ∗ i (A, b) 6= 0 and b ∈ span⊕(A(i)). + 14 Proof: By Proposition 3.4 b ∈ span⊕(A(i)) if and only if ∪j∈N \{i}Mj(A, b) = supp(b). By Proposition 3.5 the non-uniqueness of solution to A ⊗ x = b is equivalent to the existence of i with ∪j∈N \{i}Mj(A, b) = supp(b) and γ∗ i (A, b) 6= 0. (cid:3) Thus, x is not a simple image eigenvector if and only if there exists an i (and λ) such that x ∈ span⊕(A(i)) ∩ V (A, λ) and γ∗ Theorem 3.10. V (A, λ) contains simple image vectors if and only if there exists a subset N ′ ⊆ N with the following properties: i (A, λx) > 0. (i) There exists x ∈ V (A, λ) with supp(x) = N ′; (ii) γ∗ i (A, x) = 0 for all i /∈ N ′; (iii) N ′ = Nc(A, x, λ); (iv) All components of Gc(A, x, λ) = Gc(A, λ)N ′ are cycles. Proof: By Proposition 2.20 we can assume without loss of generality that λ = 1. "Only if": Let us first argue that (i) and (ii) are necessary. Let x be a simple image vector in V (A, λ), and take N ′ = supp(x). Condition (i) is immediate, and for (ii) we observe that that γ∗ i (A, x) > 0 for some i /∈ N ′ would imply that x is not the only solution of the system A ⊗ y = x, since γ∗(A, x) 6= x. As condition (ii) depends only on the support of x and on the zero-nonzero pattern of A, it also holds for arbitrary x ∈ V (A, 1) with supp(x) = N ′. That condition also implies supp(y) ⊆ N ′ for all y such that A ⊗ y = x. (20) Moreover, we can further consider the system AN ′N ′ ⊗ yN ′ = xN ′ with xN ′ ∈ V (AN ′N ′), since we have the following correspondence: xN ′ ∈ V +(AN ′N ′, 1) ↔ (x ∈ V (A, 1) & supp(x) = N ′), yN ′ ∈ S(A, xN ′) ↔ y ∈ S(A, x). (21) In that correspondence, x and y arise from xN ′ and yN ′ by setting xN \N ′ = yN \N ′ = 0, and xN ′ and yN ′ are formed as the usual subvectors of x and y 15 with indices in N ′. The first part of that correspondence follows from Propo- sition 2.13, and the second part follows from (20). Therefore, to show that (iii) and (iv) hold we can further assume that x is positive, that is, N ′ = N. Assume for the contrary that (iii) does not hold. If Nc(A, 1) 6= N, consider the indices in N\Nc(A, 1). Every node with index in N\Nc(A, 1) has an outgoing edge in Sat(A, x, 1). It cannot be that all of these edges also end in N\Nc(A, 1), because then there would be critical cycles and hence critical components in N\Nc(A, 1), a contradiction. Therefore there exists (i, j) ∈ ESat(A, x, 1) with i /∈ Nc(A, 1) and j ∈ Nc(A, 1) implying that Mj(Sat(A, x, 1)). (22) Nc(A, 1) ∪ {i} ⊆ [j∈Nc(A,1) For any i′ /∈ Nc(A, 1) there exists j′ such that i′ ∈ Mj(Sat(A, x, 1), and therefore Nc(A, 1) ∪ {i} ∪ {i′} ⊆ [j∈Nc(A,1)∪{j ′} Mj(Sat(A, x, 1)). Thus adding indices to the left hand side of (22) one by one, we obtain that there exists k /∈ Nc(A, 1) such that N = ∪j6=kMj(Sat(A, x, 1)), i.e., N = ∪j6=kMj(A, x), a contradiction. Now assume that (iii) holds but (iv) does not hold, i.e., one of the com- ponents of Gc(A) is not a cycle. In this case by Corollary 3.7 there exists i such that ∪j6=iMj(Gc(A, x, 1)) = N. However, as Gc(A, x, 1) ⊆ Sat(A, x, 1) by Proposition 2.16, we also have Mj(Gc(A, x, 1)) ⊆ Mj(A, x) for all j ∈ N, implying ∪j6=iMj(A, x) = N, a contradiction. "If": Let us now prove that (i)-(iv) are sufficient. Due to bijection (21) we can assume that N ′ = N. Then, by the main result of [20], there ex- ists a diagonal matrix X such that A := X −1AX is strictly visualised, that is aij ≤ 1, with the equality aij = 1 if and only if (i, j) is a criti- cal edge, that is, if and only if (i, j) belongs to one of the disjoint critical cycles. Let u be the vector whose every component is 1. For this vec- tor we obtain Sat( A, u, 1) = Gc( A, u, 1) = Gc( A, 1) and hence Mj( A, u) = Mj(Gc( A, 1)) for all j ∈ N. Since Gc(A) consists of disjoint cycles only, by Corollary 3.7 we have ∪j6=iMj(Gc( A, 1)) 6= N for any i ∈ N, and since Mj( A, u) = Mj(Gc( A, 1)) for all j ∈ N we obtain that u is a simple image eigenvector of A. By Lemma 2.23, Xu is a simple image eigenvector of A. (cid:3) 16 Remark 3.11. Condition (i) of Theorem 3.10 can be expressed in terms of the Frobenius normal form of A, see for instance [10] Ch. IV Theorem 2.2.4. Condition (ii) of Theorem 3.10 is equivalent to the following: ∀i /∈ N ′∃l /∈ N ′ : ali 6= 0. (23) The next result is mostly needed for Theorem 4.17, but it is also of in- dependent interest. We formulate it for positive vectors only, for the sake of simplicity. Theorem 3.12. Let Nc(A, λ) = N and let all the s.c.c of Gc(A, λ) be cycles and K = {j1, . . . , jk} be the index set of the columns of A+ λ forming GA,λ. Then ⊕(G(s) A,λ). span+ (24) ([i∈N span⊕(A(i)) ∩ V +(A, λ) = ([s∈K Proof: Assume λ = 1. For any x ∈ V (A, 1) we have x ∈ span⊕(A) and x ∈ span⊕(GA,1), and if x is positive then we also have coverings ∪i∈N Mi(A, x) = N and ∪r∈KMr(GA,1, x) = N. The claim of the theorem, reducing to ∃i : x ∈ span⊕(A(i)) ⇔ ∃s : x ∈ span⊕(G(s) A,1) for positive x ∈ V (A, 1), then amounts to equivalence between the following statements: (a) ∃i ∈ N : Mi(A, x) ⊆Sj6=i Mj(A, x) and (b) ∃s ∈ K : Ms(GA,1, x) ⊆Sr6=s Mr(GA,1, x). Let X = diag(x) and consider the matrix A := X −1AX. Matrix G A,1 then generates the cone V ( A, 1) which is the same as {y : Xy ∈ V (A, 1)}, by Lemma 2.23 part (ii). Proposition 3.8 part (ii) implies that Mi(A, x) = Mi(Sat(A, x, 1) for all i, since all nodes are critical and Gc(A, 1) ⊆ Sat(A, x, 1). i,j=1 has the property aij ≤ 1 with the equality if and only if (i, j) ∈ Sat(A, x), that is, if and only if i ∈ Mj(A, x). In A+, all columns with indices belonging to the same component of Gc( A) are equal to each other (Proposition 2.24). By Proposition 2.21 A =: (aij)n Denote the entries of GA,1 and G A,1 by gis and gis respectively. Assume (a). It means that there exists i ∈ N s c for some s ∈ K such that for each i′ ∈ Mi(A, x) = Mi(Sat(A, x)) there is an index l(i′) with l(i′) ∈ N s′ with s′ 6= s, such that i′ ∈ Ml(i′)(A, x) = Ml(i′)(Sat(A, x)). In terms of matrix A it means that c ai′i = 1 ⇒ ai′l(i′) = 1 with l(i′) ∈ N s′ c , i ∈ N s c and s 6= s′. (25) 17 el(i′) = 1 and hence ges′ = 1 with s′ 6= s. We will show that for s ∈ K as above and for every e ∈ N such that ges = 1 we can find s′ with ges′ = 1. Firstly if ges = 1 then a+ ejs = 1 and a+ ei = 1 since i and js are in the same cycle of Gc(A). This implies that for some i′ there is a path P of weight 1 connecting e to i′, and edge (i′, i) of weight 1, in digraph G( A). By (25) there exists index l(i′) such that ai′l(i′) = 1 with l(i′) ∈ N s′ c and s′ 6= s. Concatenating P with edge (i′, l(i′)) we obtain a path of weight 1 such that a+ Assume (b). By Proposition 2.24 part (i) we have ges = 1 for any s ∈ K c , and statement (b) implies that there also exist e and r ∈ K c with s 6= r. Moreover, by Proposition 2.24 c , we have a+ c and j ∈ N r c . such that ai1i′ = 1, where and e ∈ N s such that ger = 1 and e ∈ N s part (ii) we can also assume that gis < 1 for all i /∈ N s c . As ger = 1 for all e ∈ N s This implies that there exist i1 ∈ N s s 6= s′. But as i1 ∈ N s ei2 < 1 implying also that aei2 < 1 for all e /∈ N s c . Since each component of Gc(A, 1) is a cycle, we conclude that Mi2(Sat(A, x)) = {i1}, and it is covered by Mi′(Sat(A, x)), implying (a). (cid:3) c , i′ ∈ N s′ c , there also exists i2 ∈ N s c , we have a+ ej = 1 for all e ∈ N s c c such that ai1i2 = 1. As ges < 1 for all e /∈ N s 4. Interval problems 4.1. Weak X-robustness and X-simple image eigencone In this section we consider an interval extension of weak robustness and its connection to X-simplicity, the main notion studied in this paper. Definition 4.1 (Weak X-robustness). Let A ∈ Rn×n interval. + and X ⊆ Rn + be an (i) attr(A, λ) := {x : A⊗t ⊗ x ∈ V (A, λ) for some t}. (ii) A is called weakly (X, λ)-robust if attr(A, λ) ∩ X ⊆ V (A, λ). If X = Rn, then the notion of weak robustness can be described in terms of simple image eigenvectors. The proof is omitted. + . The following are equivalent: Proposition 4.2. A ∈ Rn×n (i) A is weakly λ-robust; (ii) S(A, x) = 1 for all x ∈ V (A, λ); 18 (iii) Each x ∈ V (A, λ) is a simple image eigenvector. Definition 4.3 (Invariance). Let A ∈ Rn×n We say that X is invariant under A if x ∈ X implies A ⊗ x ∈ X. Theorem 4.4. Let A ∈ Rn×n (i) If A is weakly (X, λ)-robust then A has X-simple image eigencone and X ⊆ Rn + be an interval. + and X be an interval. + associated with λ. (ii) If A has X-simple image eigencone associated with λ and if X is in- variant under A then A is weakly (X, λ)-robust. Proof: (i) Suppose that A is weakly (X, λ)-robust and x ∈ V (A, λ)∩ X. If the system A⊗y = λx has a solution y 6= x in X, then y is not an eigenvector but belongs to attr(A, λ) ∩ X, which contradicts the weak X-robustness. (ii) Assume that A has X-simple image eigencone and x is an arbitrary element of attr(A, λ)∩X. As X is invariant under A, we have that A⊗k⊗x ∈ X for all k. Moreover from the definition of X-simple image eigencone we get A ⊗ x ∈ V (A, λ) ⇒ x ∈ V (A, λ). Then A⊗k ⊗ x ∈ V (A, λ) for some k implies A⊗(k−1) ⊗ x ∈ V (A, λ),..., x ∈ V (A, λ). (cid:3) Thus the X-simplicity is a necessary condition for weak X-robustness. It is also sufficient if X is invariant under A. Observe that A is order-preserving (x ≤ y ⇒ A ⊗ x ≤ A ⊗ y), which is due to the following arithmetic properties: x ≤ y ⇒ αx ≤ αy ∀α, x, y ∈ R+ α1 ≤ α2, β1 ≤ β2 ⇒ α1 ⊕ β1 ≤ α2 ⊕ β2 ∀α1, α2, β1, β2 ∈ R+. Since A is order-preserving the invariance of X under A admits the fol- lowing simple characterization: Proposition 4.5. Let X be closed. Them X is invariant under A if and only if x ≤ A ⊗ x ≤ A ⊗ x ≤ x. 4.2. X-simple image of a matrix Let us first introduce the following bits of notation: Definition 4.6. c(x) = {i ∈ N : xi ∈ X i}, o(x) = N\c(x) (26) 19 Definition 4.7. X ↑ i , where i=1 X ↑ = ×n X ↑ i = X i ∪ [sup(X i), +∞). (27) We illustrate the definitions by the following example. Example 4.8. Suppose that X = [1, 2] × [3, 5) × (7, 9] be given. Then we have x = (1, 3, 7), c(x) = (1, 2) and X ↑ = [1,∞) × [3,∞) × (7,∞). Lemma 4.9. Let A ∈ Rm×n exists if and only if + . A solution to A ⊗ x = b in X and b ∈ Rm + γ∗(A, b) ∈ X ↑ & [j : γ∗ j (A,b)∈X j Mj(A, b) = supp(b). (28) j (A,b)∈X j j (A, b) for all j such that γ∗ Proof: "Only if". Assume that X contains a solution to A⊗x = b, but (28) If γ∗(A, b) /∈ X ↑, then since every solution to A ⊗ x = b does not hold. satisfies x ≤ γ∗(A, b), we have x /∈ X ↑ and hence x /∈ X for every solution. Let γ∗(A, b) ∈ X ↑, Sj : γ∗ Mj(A, b) 6= supp(b). Since x ∈ X we have xj < γ∗ j (A, b) /∈ X j. This implies that A ⊗ x 6= b, a contradiction. "If". Assume that (28) holds. Then by Corollary 3.3 any x with xj = j\X j is j (A, b) ∈ X ↑ γ∗ j (A, b) for j : γ∗ a solution to A ⊗ x = b. Hence A ⊗ x = b has a solution in X. (cid:3) Theorem 4.10. Let A⊗x = b have a solution in X. That solution is unique in X if and only if the following condition is satisfied for each i: j (A, b) ∈ X j, and xj ≤ γ∗ j (A, b) for j : γ∗ Mj(A, b) = supp(b) ⇒ [i ∈ c(x) & xi = min(xi, γ∗ i (A, b))] (29) [j6=i Proof: As A ⊗ x = b has a solution in X, condition (28) holds, and by Corollary 3.3 we have In particular if γ∗ ∀α ∈ X i s.t. α ≤ γ∗ i (A, b) ∈ Xi ∀α ∈ X i ∃x ∈ S(A, b) ∩ X : xi = α. i (A, b) ∃x ∈ S(A, b) ∩ X : xi = α. ↑\X i then (30) (31) 20 i (A, b) ∈ Xi If we assume that (29) does not hold for some i with γ∗ "If": Assume that (29) holds and, by contradiction, that there is more "Only if": Suppose that there is a unique solution belonging to X. By (31) if X i does not reduce to one point, then S(A, b) also contains more ↑\X i, which than one vector. Thus, xi = xi for all i such that γ∗ satisfies (29). i (A, b) ∈ Xi then Sj6=i Mj(A, b) = supp b and (31) implies that the solution is non-unique since the interval X i ∩ {α : α ≤ γ∗ than one solution to A ⊗ x = b belonging to X. Since the solution is non-unique, it follows that there exists a proper subset N ′ of N such that ∪j∈N ′Mj(A, b) = supp(b). Assume that N ′ is a minimal such subset, with respect to inclusion. We have ∪j6=iMj(A, b) = supp(b) for all i ∈ N\N ′. By (29) N\N ′ ⊆ c(x), and xi = min(xi, γ∗ i (A, b)) for all i ∈ N\N ′. This condition implies that there exists only one N ′-solution to A ⊗ x = b belonging to X, and it has coordinates i (A, b)} contains more than one point. xi =(xi, γ∗ i (A, b), if xi = xi, otherwise. As this solution is the same for any minimal subset N ′, system A⊗ x = b has a unique solution, contradicting the non-uniqueness of it. (cid:3) Corollary 4.11. Let x ∈ V (A) ∩ X. Then x is an X-simple image eigen- vector if and only if Mj(A, b) = supp(x) ⇒ [i ∈ c(x) & xi = min(xi, γ∗ i (A, b))] (32) [j6=i If Nc = N then this condition can be replaced with the following one: Mj(A, b) = supp(x) ⇒ [i ∈ c(x) & xi = min(xi, xi)] (33) [j6=i Proof: As x ∈ V (A) ∩ X, system A ⊗ y = x has a solution in X, which is y = x. So we can apply Theorem 4.10 yielding (32) for the uniqueness of this solution. Further if all nodes are critical, then γ∗ i (A, x) = xi for all x, so (32) gets replaced with (33). (cid:3) 21 4.3. Characterizing matrices with X-simple image eigencone We begin with the following definition and key lemma of geometric kind. Definition 4.12. An interval X ⊆ Rn called x-closed if it x ∈ X. Lemma 4.13. Let X be an x-closed interval and let A ∈ Rm×n. Then X ∩ span⊕(A) 6= ∅ if and only if PAx ∈ X. + is called x-open if o(x) = N. It is Proof: If PAx ∈ X then X ∩ span⊕(A) 6= ∅ (since PAx ∈ span⊕(A)). If X ∩ span⊕(A) 6= ∅, take y ∈ X ∩ span⊕(A) 6= ∅. Vector z = y ⊕ PAx belongs to span⊕(A) and satisfies y ≤ z ≤ x. It follows then that z ∈ X. However, PAx ≤ z ≤ x while PAx is the greatest vector of span⊕(A) bounded from above by x. This implies that z = PAx and PAx ∈ X. (cid:3) Definition 4.14. For any l, i ∈ N, let i =(xl, if i = l, otherwise, (34) xhli xi, and let xhli = (xhli i )i∈N . Definition 4.15. For any l ∈ N, let λxj > ajlxl ∀j}, (35) X l A,λ = {x ∈ X : xl = xl, X (l = {x ∈ X : xl > xl}. Observe that if X is x-closed then X l for every l. Also note that all < λ is a necessary condition for X l non-empty. A is xhli-closed (if it is nonempty) A,λ to be Theorem 4.16. Let λ > 0 and V (A, λ) ∩ X 6= ∅. (i) A has X-simple image eigencone corresponding to λ if and only if span⊕(A(i)) ∩ V (A, λ) ∩ X (i = ∅ ∀i, V (A, λ) ∩ X l A,λ = ∅ ∀l ∈ c(x)\Nc(A, λ) s.t. xl < xl. (36) 22 (ii) If X is x-closed then (36) is equivalent to PW ix /∈ X (i ∀i ∈ N where W (i) = span⊕(A(i)) ∩ V (A, λ), PV (A,λ)xhli /∈ X l A,λ, ∀l ∈ c(x)\Nc(A, λ) s.t. xl < xl. (37) Proof: Assume without loss of generality that λ = 1. (i): Let x ∈ V (A, 1) ∩ X. i (A, x) we may have γ∗ In general, x ≤ γ∗(A, x). More precisely, for γ∗ i (A, x) = xi (for all i ∈ Nc(A, 1) and some other nodes), or γ∗ i (A, x) > xi (for at least one node in N\Nc(A, 1) if N 6= Nc(A, 1)). "Only if": Let us show that the conditions (36) are necessary. For this assume by contradiction that one of these conditions is violated but A has X-simple image eigencone. i (A, x′), xi). l (equivalent with the condition x′′ (a) If the first condition does not hold then take x′ ∈ X (i ∩ V (A, 1) ∩ i > xi and also ∪j6=iMj(A, x′) = supp(x′), span⊕(A(i)). Then x′ satisfies x′ hence x′ is not an X-simple image eigenvector by Corollary 4.11: either i ∈ o(x) and ∪j6=iMj(A, x′) = supp(x′), or i ∈ c(x), ∪j6=iMj(A, x′) = supp(x′) and xi < min(γ∗ (b) If the second condition does not hold then take x′′ ∈ V (A, 1) ∩ X l A l = xl, xl < xl and l ∀j from (35)). The l implies that ∪j6=lMj(A, x) = supp(x′′), and we also l (A, x′′), xl) and l ∈ c(x). By Corollary 4.11, this shows that "If": By contradiction, suppose that the conditions hold but A does not have simple image eigencone. Let x be an X-simple image eigenvector. Then either xi > xi and ∪j6=iMj(A, x) = supp(x) for some i, which implies x ∈ X (i ∩ V (A) ∩ span⊕(A(i)), or xl = xl, ∪j6=lMj(A, x) = supp(x) and xl < min(γ∗ l (A, x), xl) for some l ∈ c(x). In this case necessarily xl < xl and xl < γ∗ for some l ∈ c(x)\Nc(A, 1) such that xl < xl. We have x′′ j > ajlx′′ γ∗ l (A, x′′) > x′′ inequality γ∗ have xl < min(γ∗ x is not an X-simple image eigenvector. l (A, x), which is only possible for l ∈ N\Nc. l (A, x′′) > x′′ This shows the sufficiency of (36). (ii) By Lemma 4.13, W i ∩ X (i V (A, 1) ∩ X l A 6= ∅ if and only if PV (A,1)xhli ∈ X l A. (cid:3) 6= ∅ if and only if PW ix ∈ X (i, and In general, the basis of span⊕(A(i)) ∩ V (A, λ) can be computed algorith- mically, using the method of Butkovic, Hegedus [5] or the more recent and efficient methods of Allamigeon et al. [1] Let us now examine the case when X is x-open. 23 Theorem 4.17. Let X be an x-open interval and V (A, λ)∩X be non-empty. (i) A has X-simple image eigencone corresponding to λ if and only if span⊕(A(i)) ∩ V (A, λ) ∩ X = ∅ ∀i = 1, . . . , n. (38) (ii) In that case Nc(A, λ) = N and Gc(A, λ) consists of disjoint cycles c1, . . . , ck for some k. (iii) Condition (38) is equivalent to span⊕(GA(s),λ) ∩ X = ∅ ∀s = 1, . . . , k. (iv) If X is also x-closed then (38) is also equivalent to PG A(s),λ x 6> x ∀s = 1, . . . , k. (39) (40) Proof: Assume λ = 1. (i): Follows from Theorem 4.16 part (i), where we take into account that c(x) is empty and X (i = X for each i. (ii): Note that each vector in V (A, 1) ∩ X is positive. By Theorem 3.10, if Gc(A) does not consist of disjoint cycles or Nc 6= N then each vector x ∈ V (A, 1) belongs to span⊕(A(i)) for some i ∈ N. Hence span⊕(A(i)) ∩ V (A, 1) ∩ X 6= ∅ for some i, a contradiction. (iii): By Theorem 3.12 we have span⊕(A(i)) ∩ V (A, 1) = ∪k s=1 span⊕(GA(s),1). ([i∈N Hence (38) and (39) are equivalent. (iv): By applying Lemma 4.13 to (39), we obtain that (39) is equivalent to PG A(s) ,1 x /∈ X ∀s = 1, . . . , k, (41) and that is the same as (40). (cid:3) 24 References [1] X. Allamigeon, ´E. Goubault and S. Gaubert. Computing the Vertices of Tropical Polyhedra using Directed Hypergraphs. Discrete and Compu- tational Geometry, 49:2 (2013), 247-279. [2] F.L. Baccelli, G. Cohen, G.J. Olsder and J.P. Quadrat. Synchro- nization and Linearity. Wiley and Sons, 1992. Available online: https://www.rocq.inria.fr/metalau/cohen/SED/book-online.html [3] P. Butkovic. Max-linear Systems: Theory and Algorithms, Springer Monographs in Mathematics, Springer-Verlag 2010. [4] P. Butkovic. Simple image set of (max, +) linear mappings. Discrete Appl. Math. vol. 105 (2000), 73 -- 86. [5] P. Butkovic and G. Hegedus. An elimination method for finding all solutions of the system of linear equations over an extremal algebra Ekonomicko-Matematick´y Obzor, 20 (1984), pp. 203214 [6] P. Butkovic, H. Schneider and S. Sergeev. Generators, extremals and bases of max cones. Linear Algebra Appl. 421, 2007, 394-406. [7] P. Butkovic, H. Schneider and S. Sergeev. Recognizing weakly stable matrices. SIAM J. Control Optim. 50 (5), 2012, 3029-3051. [8] G. Cohen, S. Gaubert, J.P. Quadrat and I. Singer. Max-plus convex sets and functions. In: G.L. Litvinov and V.P. Maslov (eds.), Idempotent Mathematics and Mathematical Physics, Cont. Math. 377 , AMS, 2005, pp. 105 -- 129, [9] R. A. Cuninghame-Green. Minimax Algebra. Lecture Notes in Eco- nomics and Mathematical Systems 166, Springer, Berlin, 1979. [10] S. Gaubert. Th´eorie des les dioıdes. PhD Thesis, L'´Ecole des Mines de Paris, 1992. Available online: http://www.cmap.polytechnique.fr/~gaubert/PAPERS/ALL.pdf syst`emes lin´eaires dans [11] R. A. Cuninghame-Green. Minimax algebra and applications. Advances in Imaging and Electron Physics 90, (1995) 1 -- 121. 25 [12] M. Gavalec, K. Zimmermann. Classification of solutions to systems of two-sided equations with interval coefficients. Inter. J. of Pure and Ap- plied Math. 45 (2008), 533 -- 542. [13] B. Heidergott, G.-J. Olsder, and J. van der Woude. Max-plus at Work. Princeton Univ. Press, 2005. [14] H. Myskov´a. Interval systems of max-separable linear equations. Linear Algebra and Its Applications 403 (2005) 263 -- 272. [15] H. Myskov´a. Control solvability of interval systems of max-separable linear equations. Linear Algebra and Its Applications 416 (2006) 215 -- 223. [16] J. Plavka. The weak robustness of interval matrices in max-plus algebra Discrete Appl. Math. 173 (2014) 92-101. [17] J. Plavka. The weak robustness of interval matrices in maxplus algebra. Discrete Applied Mathematics 173 (2014) 92 -- 101. [18] J. Rohn. Systems of Linear Interval Equations. Linear Algebra and Its Applications 126 (1989) 39 -- 78. [19] S. Sergeev. Max-algebraic cones of nonnegative irreducible matrices, Linear Algebra and its Applications 435 (2011), 1736-1757. [20] S. Sergeev, H. Schneider and P. Butkovic. On visualization scaling, subeigenvectors and Kleene stars in max algebra. Linear Algebra and Its Applications 431 (2009) 2395 -- 2406. [21] N.N. Vorobyev. Extremal algebra of positive matrices (in Russian). Elek- tronische Informationsverarbeitung und Kybernetik, 3 (1967) 39 -- 71. [22] K. Zimmermann. Extrem´aln´ı algebra (in Czech), Ekon. ´ustav CSAV Praha, 1976. 26
1910.10778	1	1910	2019-10-23T19:43:10	Tied links in the solid torus	[ "math.RA", "math.GN", "math.GT" ]	We introduce the concept of tied links in the solid torus, which generalize naturally the concept of tied links in $S^3$ previously introduced by Aicardi and Juyumaya. We also define an invariant of these tied links by using skein relations, and subsequently we recover this invariant by using Jones' method over the bt-algebra of type $\mathtt{B}$ and the Markov trace defined on this.	math.RA	math	TIED LINKS IN THE SOLID TORUS MARCELO FLORES1 ABSTRACT. We introduce the concept of tied links in the solid torus, which generalize naturally the concept of tied links in S3 previously introduced by Aicardi and Juyumaya. We also define an invariant of these tied links by using skein relations, and subsequently we recover this invariant by using Jones' method over the bt-algebra of type B and the Markov trace defined on this. 1. INTRODUCTION In [11, 12], Jones constructs the famouus Jones polynopmial by using the Markov trace function on the tower of classical Temperley-Lieb algebras. These algebras can be regarded as quotients of the associated Hecke algebras of type A. Subsequently, in [13], he applies this procedure to Hecke algebras of type A, obtaining as result the Homplypt polynomial, which had been defined previously in [8] by using skein relations. These procedure led to the idea of knot algebras, that are towers of algebras which support a Markov trace which may be rescaled and allow thus the construction new of invariants for knotted objects. The Hecke algebra, the Temperley-Lieb algebra and the BMW algebra are the most well-known examples of knot algebras. The Yokonuma -- Hecke algebra, which was originally introduced by T. Yokonuma [21] in the context of Chavalley gruops, is other significant example of knot algebra. Indeed, in [14], Juyumaya proves that the tower of Yokonuma -- Hecke algebras support a unique Markov trace. Subsequently, by using Jones' method invariants for: framed links [17], classical links [15] and singular links [16] were constructed. Moreover, recently it was proved that the invariants for classical links constructed in [15] are not topologically equiv- alent either to the Homflypt polynomial or to the Kauffman polynomial, see [5]. In [1], Aicardi and Juyumaya introduces the algebra of braids and ties (or bt -- algebra), denoted by En, that is also a knot algebra. The term 'braids and ties' refers to the generators of this algebra, which have a diagrammatical interpretation in terms of braids and ties (see [2, Section 6]). This algebra is defined by abstractly considering it as a certain subalgebra of the Yokonuma -- Hecke algebra Yd,n := Yd,n(u). Subsequently, in [3], a Markov trace for En is constructed by implementing the method of relative traces (see [2], cf. [4, 7, 10]). Then using Jones' method [13] over this trace, the invariant for classical knots ∆(u, A, B) and the invariant for singular knots Γ(u, A, B) are define. It is worth noting that, for links, the invariant ∆ is more powerful than the Homflypt polynomial (see [2, Addendum]). In [3], the authors introduce the concept of tied links in S3, which generalize classical links. These objects are the closure of tied braids that are come from the diagrammatical inter- pretation of the defining generators of the bt -- algebra. Subsequently, they construct the 2010 Mathematics Subject Classification. 05A18, 57M27, 57M25, 20C08, 20F36, 20F55. Key words and phrases. Braids of type B, bt-algebra of type B, framization, Markov trace, link invariants, knots and links in solid torus. 1 This project was partially supported by FONDECYT 11170305. 1 2 MARCELO FLORES1 invariant F for these new objects via skein relations. Finally, they prove an analogue of the Alexander and Markov theorem for tied links and recover the invariant F by applying Jones' method to the bt -- algebra together with the Markov trace defined in [2, Section 4]. All the results that are mentioned above are related to Coxeter groups of type A. How- ever, there has been a growing interest also in knot algebras related to Coxeter systems of type B. Indeed, the affine and cyclotomic Yokonuma-Hecke algebra is introduced in [4], and recently the first author together Juyumaya and Lambopoulou [7] introduced the alge- bra YB d,n(u, v), that can be regarded as an analogue of the Yokonuma -- Hecke algebra in the context of Coxeter groups of type B. This algebra supports a Markov trace (see [7, Theorem 3]), consequently invariants for framed knots and links in the solid torus are obtained by applying Jones' method. In order to generalize the classical bt -- algebra, in [6], a braids and ties algebra of type B is introduced, denoted by E B n(u, v), for n ≥ 1. This algebra n := E B is defined in analogy to the construction of the bt-algebra of type A, that is, it is obtained d,n(u, v). We further prove that E B by considering it abstractly as a certain subalgebra of YB n supports a Markov trace, and using this trace as the main ingredient in Jones' method, we then define an invariant of classical links in the solid torus ST . Then, it is natural try to define a analogue to the concept of tied links, though in the context of Coxeter groups of type B. Thus, the purpose of this article is to introduce the concept of tied links in the solid torus, which generalize naturally the concept of tied links in S3 previously introduced by Aicardi and Juyumaya. We do so by considering the dia- grammatic interpretation of the bt -- algebra of type B [6, Section 3.1]. We then define the invariant FB for tied links in ST via skein relations. Finally, we prove analogues of the Alexander and Markov theorems in order to recover the invariant FB by applying Jones' method to the algebra E B n. The article is organized as follows. In Section 2, we provide the notation and necessary results. In Section 3, using an analogy to the classical case, we introduce the concept of tied links in ST . In Section 4 we define the invariant FB, which coincide with F considering tied links in S3 as affine tied links in ST (see Remark 4). We prove that this invariant is unique by defining it via skein relations (see Theorem 3). In Section 5, we introduce the n, which contains the monoid T Bn originally Tied braid monoid of type B, denoted T BB n plays the role that Bn does in the context defined in [3, Section 3.1]. The monoid T BB of classical links in S3. That is, we use this monoid to prove in Section 6 analogues of Alexander and Markov theorem for tied links. In Section 7, using the natural the represen- tation of T BB n (see Proposition 3), we define an invariant by using Jones' method on the Markov trace defined in [6, Section 5]. Finally, we prove that the invariant obtained by this procedure is equivalent to the invariant FB from Section 4. n in E B 2. PRELIMINARIES In this section, we recall some basic results to be used. We begin recalling some basic notions about braids of type B and links in the solid torus. 2.1. Gruops of type Bn. Set n ≥ 1. We denote the Coxeter group of type Bn by Wn. This group is the finite Coxeter group associated to the Dynkin diagram c r1 c s1 q q q c c sn−2 sn−1 TIED LINKS IN THE SOLID TORUS 3 The group Wn can be realized as a subgroup of the permutation group of the set Xn := {−n, . . . ,−1, 1, . . . , n}. More specifically, the elements of Wn are the permutations w such that w(−m) = −w(m), for all m ∈ Xn. Note that there is a natural projection τ : Wn → Sn, defined by r1 (cid:55)→ 1 and si (cid:55)→ si, The corresponding braid group of type Bn associated to Wn is defined as the group(cid:102)Wn where si are the Coxeter generators of the symmetric group Sn. generated by ρ1, σ1, . . . , σn−1 subject to the following relations: σiσj = σjσi, σiσjσi = σjσiσj, ρ1σi = σiρ1, ρ1σ1ρ1σ1 = σ1ρ1σ1ρ1. for for for i − j > 1, i − j = 1, i > 1, (1) Geometrically, braids of type Bn can be regarded as classical braids of type An with n + 1 strands, such that the first strand is identically fixed. This strand is called 'the fixed strand'. The 2nd, . . . , (n + 1)st strands are renamed from 1 to n and are called 'the moving strands'. The 'loop' generator ρ1 stands for the looping of the first moving strand around the fixed strand in the right-handed sense, see [18, 19]. Figure 1 illustrates a braid of type B6. FIGURE braid of type B6. 1. A FIGURE 2. Generators of(cid:102)Wn. 2.2. Knots and links in the solid torus. It is well known that the solid torus ST may be regarded as the complement in S3 of another solid torus I, i.e. ST = S3/ I. So links in ST can be regarded as mixed links in S3 containing the complementary solid torus. Therefore, any link L in ST is represented by a mixed link I ∪ L(cid:48) in S3, consisting of a standard link L(cid:48) in S3, which is linked in some way with the fixed complementary torus part I (see Figure 3). Consequently, a mixed link diagram is the projection of I ∪ L(cid:48) on the plane of the projection of I. These facts also stand for oriented links in ST . Thus, from now on, any oriented link L in ST with n components will be seen as an oriented mixed link in S3 with n + 1 components. The component that represents the complementary solid torus, which is fixed and unknotted, will be called the fixed component of L. The others n components will be called the standard components of L. Thus, this mixed link diagram has crossings between standard components, called stan- dard crossings or simply crossings, and eventually it also has some loopings between the standard components and the fixed component, which are called loops (see Figure 4). Remark 1. A link L in ST is called affine if it lies in 3 -- ball in ST . Or in other words, the link L does not have loops around the fixed component. Thus, classical links in S3 can be regarded as an affine links in the solid torus. 1i,(i 1)-th strandi-th strand. . .. . .. . .4 MARCELO FLORES1 FIGURE 3. A link in the solid torus and its corresponding mixed link diagram. FIGURE 4. Standard crossing and loops. Two links in ST are isotopic if and only if any two corresponding mixed links diagrams in S3 differ by a planar isotopy and a finite sequence of mixed moves (see Figure 5) to- gether with the three Reidermeister moves for the standard part of the link (see [18] for details). FIGURE 5. Reidemiester moves between mixed components. Observe that Reidemeister moves in Figure 5 imply the following move and the analogue one for the negative loop. The closure of a braid α in the group(cid:102)Wn is defined by joining with simple (unknotted and unlinked) arcs its corresponding endpoints, and it is denoted by(cid:98)α. The result of closure, (cid:98)α, is a link in the solid torus. Thus, we have the following analogues of Alexander and Markov theorems for links in ST (see [18] for details). Theorem 1. Any oriented link in ST is isotopic to a closure of a braid of type B. positive loopnegative looppositive crossingnegative crossing,TIED LINKS IN THE SOLID TORUS 5 classes of(cid:83) Theorem 2. Isotopy classes of oriented links in ST are in bijection with equivalence n(cid:102)Wn, the inductive limit of braid groups of type B, respect to the equivalence relation ∼B: (i) αβ ∼B βα, (ii) α ∼B ασn and α ∼B ασ−1 n , for all α, β ∈ Wn. 3. TIED LINKS IN THE SOLID TORUS In this section, we introduce the concepts of tied links in the solid torus and their dia- grams. Indeed, a tied link in ST is simply a standard link in ST whose set of components are related in some way. We use ties as a formalism to indicate that two components are related. The ties will be drawn as a wavy line between two such components. These new knotted objects naturally generalize links in ST and classical tied links in S3 (see [3]). Definition 1. A tied (oriented) link in ST with n components is a pair L(I) := (L, I), where L is a link in ST and I is a collection of unordered pairs of points (pi, pj) of L (points in the fixed component are allowed). We called I the set of ties. Thus, a pair (pi, pj) ∈ I is represented as an wavy arc called tie that connects the points pi and pj, which may belong to different components or to the same one. Ties they are not embedded arcs, they are just a notational device. Consequently, the arcs of L can cross through the ties. We will denote TST the set of oriented tied links in ST. Remark 2. If I is empty, then L(I) is nothing else that a classical link in ST . In the same fashion, if L is an affine link in ST , and I only contains pairs of points that belong to the standard components, then according to Remark 1, L(I) can be regarded as a tied link in S3. Thus, we have that the set of classical tied links T from [3] is embedded in TST . Note that the set I induces a partition on the set of the components of L, where two components of L belong to the same class if they are connected by a tie. Definition 2. Let L(I) a tied link in ST . A diagram of L(I) is a corresponding mixed link diagram of L in S3 provided with ties connecting pairs of points in the set of ties I. Definition 3. Let L(I), L(cid:48)(I(cid:48)) be two oriented tied links. We say that L(I) and L(cid:48)(I(cid:48)) are tie isotopic if: (i) L and L(cid:48) are isotopic in ST (Section 2.2). (ii) I and I(cid:48) define the same partition in the set of components of L and L(cid:48), respec- tively. It is not difficult to check that tie isotopy is an equivalence relation, which is denoted symply by ∼t. ST . Additionally, we just write L instead L(I). From now on, without risk of confusion, when we say tied link we will refer tied link in Note that tie isotopy says that we can move any tie between two components letting its extremes move along the whole component. Additionally, we can add or remove ties as long as these do not modify the induced partition on the set of components. For instance, we can add or remove: • ties connecting two points of the same component, • ties between components that are already in the same class. 6 MARCELO FLORES1 Let L be a tied link, and let ci, cj, ck be three different components of L. Set points ps, p(cid:48)s ∈ cs for s ∈ {i, j, k}. The tie isotopy also stand that if we have two ties (pi, pj), (p(cid:48)j, pk) ∈ I, we then can change indistinctly these ties for (pi, pj)(p(cid:48)i, pk) or (pk, pi)(p(cid:48)k, pj). For in- stance, Figure 6 shows two tie isotopic links. It is clear that the components are ambient isotopic. On the other hand we also have that the corresponding set of ties induces the same partition into their respective components. FIGURE 6. Two links tied -- isotopic in ST. Definition 4. We say that a tie is essential if this cannot be removed, i.e. removing this tie we obtain a different partition in the set of components. For instance, in Figure 6, the tied link on the left has two essentials ties and one that is not (the tie connecting points in the green component). 4. AN INVARIANT FOR TIED LINKS IN ST In this section, we construct an invariant for tied links in ST . In order to do that, we need to set notation. From now on, let u, v, x, y, w, z be indeterminates, and set K := C(u, v, x, y, w, z). An invariant of ties links is nothing else that a function FB : TST → K that is constant in the classes of tie -- isotopic links. We define this invariant via skein rela- tions. The following theorem is obtained by readjusting the arguments in [3]. Theorem 3. There exists an invariant of oriented tied links FB : TST → K that is uniquely defined by the following conditions: Let (cid:13), (cid:13),(cid:13)+ (cid:13)+ be the tied unknots in the Figure 9. (i) Initial conditions: FB((cid:13)) = 1; FB( (cid:13)) = x; FB((cid:13)+) = y; FB( (cid:13)+) = w. (ii) Let L be a tied link. Then we have FB(L (cid:116) (cid:13)) = 1 zλFB(L), FB(L (cid:116) (cid:13)) = x zλFB(L), FB(L (cid:116) (cid:13)+) = y zλFB(L) FB(L (cid:116) (cid:13)+) = FB(L(cid:116)(cid:13)+) = w zλFB(L). where (cid:116) means that we add the corresponding unknot tied together to some stan- dard component of L. Additionally, we have: FB(L(cid:116) (cid:13)+) = w zλFB( L), FB(L(cid:116) (cid:13)) = x zλFB( L) where L is the tied link obtained from L by adding a tie from the component that is connected with the unknot added to the fixed component. TIED LINKS IN THE SOLID TORUS 7 (iii) Skein rule I: Let L+, L−, L∼, L+,∼, L−,∼ be the diagrams of tied links, that are identical outside the small disk, whereas inside the disk the diagram looks as shown in Figure 7. Then the following identity holds: 1 λFB(L+) − λFB(L−) = (u − u−1)FB(L∼) (cid:113) z−(u−u−1)x . z + , LM − , LM where λ := (iv) Skein rule II: Let LM be the diagrams of tied links, that are identical outside the small disk, whereas inside the disk the diagram looks as shown in Figure 8. Then the following identity holds: +,∼, LM −,∼ ∼ , LM FB(LM + ) − FB(LM − ) = (v − v−1)FB(LM ∼ ) FIGURE 7. The disks where L+, L−, L+,∼, L−,∼, L∼ differ. FIGURE 8. The disks where LM + , LM − , LM +,∼, LM −,∼, LM ∼ differ. FIGURE 9. The six tied -- unknots in ST. L+,∼L−,∼L∼L+L−LM+,∼LM−,∼LM∼LM+LM−(cid:31)(cid:31)−(cid:31)+(cid:31)(cid:31)−(cid:31)+8 MARCELO FLORES1 Remark 3. Skein rules (ii) and (iii) imply the following skein rules: (v) 1 (vi) FB(LM λFB(L+,∼) − λFB(L−,∼) = (u − u−1)FB(L∼), +,∼) − FB(LM −,∼) = (v − v−1)FB(LM ∼ ), which are obtained by adding a tie between the two strands inside the disc in each case. Proof. We proceed by following the proof of [20]. The proof has some slight changes when ties and loops around the fixed component are involved. Let T n ST be the set of diagram of n crossings (recall Season 2.2), and let L be in T n ST . It is well known that we can associate to L an ascending diagram L(cid:48). To obtain this diagram, we first have to order the components and fix a base point on each of them. Then L(cid:48) is ob- tain by starting at the base point of the first component and changing all the overpasses to underpasses along the component. We then do the same process for the subsequent compo- nents. Thus, we obtain a diagram that every crossing is first encountered as an underpass. This process separates and unknots the components. Eventually the components of L(cid:48) have loops around the fixed component. Without loss of generality, we can assume that all are positive or negative, since two consecutive loops with opposite sign are isotopic to a seg- ment that does not have loops around the fixed component (see Figure 5). Then, we define the positive ascending diagram L(cid:48)+ as the diagram that is obtain from L(cid:48) by changing the loops of the components of L(cid:48). We proceed as follows: • If a component of L(cid:48) has only positive loops. We leave the first loop (according to the orientation of the component) unaltered, the second one is change by a negative loop. We then do the same with the fourth and so on. • If the component has only negative loops we proceed analogously. Thus, if a component has 2n loops in the diagram L(cid:48), this will have n couples of con- secutive loops with opposite sign in the diagram L(cid:48)+. Analogously, if the number of loops is 2n + 1, the corresponding component L(cid:48)+ will have n couples of consecutive loops with opposite sign and a positive or a negative loop at the end. We thus have that L(cid:48)+ is a disjoint union of tied unknots (cid:13),(cid:13)−,(cid:13)+, which are tied together according to the initial ties in the link L. It is clear that L and L(cid:48)+ just differ in a finite number of crossings and loops, called "deciding crossings" (deciding loops, respectively), where the signs in those cross- ing and loops are opposites. This procedure allows to get an ordered sequence of deciding points, whose order depends from the ordering of the components, and the choice of base points. We now proceed by induction in the number of standard crossings. We thus assume that the ST → K satisfies the relations (i)-(iv), is independent of the ordering of the function FB : T n points, and of the choices of base points as well. Also, FB is invariant under Reidemeister moves. Moreover, for any disjoint union of tied unknots on Figure 9, the value of FB may be computed by using rules (i) and (ii). We start with zero crossings. Thus, the tied link is a disjoint union of tied unknots. And Let L be in T n+1 we know the value of FB in this case. ST . If L is a disjoint union of tied -- unknots the result follows. Otherwise, consider the first deciding crossing p. If in a neighborhood of p the tied link looks like L+ (or L−), we can use the skein rule (iii) for writing the value of FB in terms of L− (or L+) and L∼. Then we apply the same procedure on the second deciding crossing and so on. Finishing this process, we proceed to do the same with the deciding loops though using TIED LINKS IN THE SOLID TORUS 9 ), we can use the . Analogously, ), we can use skein rules (vi) for deducing the value of skein rule (iv) (or (vi)). Remember that if the a loop looks like LM skein rule (iv) for writing the value of FB in terms of LM (or LM − if the loop looks like LM FB in terms of LM −,∼ Thus, at the end of the process, we have express FB(L) in terms of FB(L(cid:48)+) and two other tied links that are a disjoint union of unknots tied together in some way. For these unions the value of FB is known and only depends of the number of components and the number of essential ties. Thus, it remains to prove that: + (or LM − + ) and LM ∼ −,∼ ) and LM ∼ (resp. LM (or LM . +,∼ +,∼ (i) the procedure is independent of the order of the deciding crossings and deciding loops. of base points. (ii) the procedure is independent of the order of the components, and from the choice (iii) the function is invariant under Reidemeister moves. The skein rule (iii) is similar to the skein rule used in [20] (Homflypt type). Indeed, just the link of right part of the equality changes, including a tie between the strands. We then omit the proofs of (i) -- (iii), since these follow almost directly by slightly modifying (cid:3) the corresponding proofs given in [20]. Remark 4. Let F be the invariant defined for tied links in S3 in [3, Section 2]. By Re- mark 2, we have that FB restricted to affine tied links in TST is equivalent to invariant F. Recall from [3] that F holds the following properties: (i) F is multiplicative with respect to the connected sum of tied links. (ii) The value of F does not change if the orientations of all curves of the link are (iii) Let L be a link diagram whose components are all tied together, and L+ be the link diagram obtained from L by changing the signs of all crossings. Thus, F(L+) is obtained from F(L) by the following changes: λ → 1/λ and u → 1/u. reversed. It is not difficult to check that the invariant FB just satisfies an analogue of property (iii) above. More precisely, let L be a tied link whose standard components are all tied together, and L∗ be the link diagram obtained from L by changing the signs of all crossings. Thus FB(L∗) is obtained from FB(L) by doing the change: λ → 1/λ, u → 1/u., v → 1/v. On the other hand, unlike the classical case, there is no well-defined operation of con- nected sum for knots in ST (see [9]). Thus, we do not have an analogous for (ii). Addi- tionally, observe that the value of FB is not invariant if we reverse the orientation of all the components of the links. Indeed, we have that FB((cid:13)2) (cid:54)= FB((cid:13)4). However, if we con- sider a tied link L without loops, then FB(L) does not change if we reverse the orientation (cf. [3, Section 2.2]). Example 1. Let H +, H +, H +, H + (cid:96) , H + We next compute the value of FB( H + (cid:96) be the tied links in the Figure 10. (cid:96) , H + (cid:96) ). Using (ii), we obtain FB( H + (cid:96) ) = λ2FB(H1) + λ(u − u−1)FB(H2), where H1 and H2 are the tied links in Figure 11. 10 MARCELO FLORES1 FIGURE 10. Different Hopf links in ST FIGURE 11 Therefore, we have: FB( H + (cid:96) ) = λ2 wx zλ x z = λw( + λ(u − u−1)w + u − u−1) = λw(xu + u2z − z) uz We can compute the polynomial of the others tied links in Figure 10 by an analogous way. More precisely, we have: FB(H +) = FB( H +) = FB( H + (cid:96) ) = zu λ(u + u2z − z) λx(xu + u2z − z) λw(u + u2z − z) zu uz , FB( H +) = , FB(H + (cid:96) ) = λx(u + u2z − z) λy(u + u2z − z) zu zu H+H+ H+(cid:31)H+(cid:31)H+H+(cid:31)H1H2TIED LINKS IN THE SOLID TORUS 11 5. THE TIED BRAID MONOID OF TYPE B In this section, we introduce the tied braid monoid of type B in order to obtain analogues for Alexander and Markov theorems for tied links in ST . This, with the aim of recovering FB via Jones' method using the algebra of braids and ties of type B and the respective Markov trace defined in [6] . We begin introducing the tied braid monoid of type B and giving the corresponding diagrammatical interpretation. φ1, η1, . . . , ηn−1, called ties, satisfying the relations (1) of(cid:102)Wn together with the following Definition 5. We define the tied braid monoid of type B, denoted by T BB n, as the monoid generated by ρ1, σ1, . . . , σn−1, the usual braid generators of B -- type, and the generators relations: ηiηj = ηjηi ηiσi = σiηi ηiσj = σjηi ηiσjσi = σjσiηj = σjσ−1 ηiσjσ−1 i i ηj for all i, j, for all 1 ≤ i ≤ n − 1, for all i − j > 1, for all i − j = 1, for all i − j = 1, ηiηjσi = σiηiηj = ηjσiηj for all i − j = 1, for all i, η2 i = ηi φ2 1 = φ1 for all i, ρ1ηi = ηiρ1 ρ1φ1 = φ1ρ1 φ1ηi = ηiφ1 φ1σi = σiφ1 i−1 = σ−1 i−1 . . . σ−1 φ1η1 = φ1σ1φ1σ−1 for all i, for all 2 ≤ i ≤ n − 1, 1 φ1σ1 . . . σi 1 = σ1φ1σ−1 1 η1 (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) σi−1 . . . σ1φ1σ−1 1 . . . σ−1 for all 2 ≤ i ≤ n. Remark 5. Note that relations (2) -- (8) are exactly the defining relations of T Bn, the briad tied monoid defined in [3]. Therefore, we have T Bn ≤ T BB n . Remark 6. For n ≥ 1, we have that T BB inductive limit (cid:116)n≥1T BB n. n+1. Then, we can define T BB ∞ n ⊆ T BB as the From now on, the relations of T Bn will be called type -- A relations, and the rest of them type -- B relations. In terms of diagrams, the generators ρ1, σ1, . . . , σn−1 represent the usual braid genera- tors of type B (see Figure 2). On the other hand, the defining generator φ1 corresponds to the braid of type Bn that has a tie connecting the fixed strand and the first moving strand, whereas ηi is represented by the B -- type braid that has a tie connecting the i -- th and (i+1) -- st moving strands. See Figure 12 for this identification. The defining relations may also be expressed in terms of diagram. For instance, recall from [3] that relation (5) corresponds to move the tie from top to bottom behind or in front of the strand (see Figure13). For more details about type-A relations see [3, Section 3.1]). 12 MARCELO FLORES1 FIGURE 12. Diagrams corresponding to the generators of T BB n FIGURE 13. Relations (5) and (6) in terms of diagrams (n = 3) FIGURE 14. Equivalent diagrams of η1,3 according Eq. (17) 5.1. Generalized ties. Let ηi,j, φj be the elements T BB n defined as follows: ηi,j = σi ··· σj−2ηj−1σ−1 φj = σj−1 ··· σ1φ1σ−1 j−2 ··· σ−1 i ··· σ−1 j−1 1 for for i − j > 1, 2 ≤ j ≤ 1. (16) where, by convention ηi,i = 1 and ηi,i+1 = ηi We begin recalling some known facts about the elements ηi,j's from [3]. By definition, we have that n1,3 corresponds to the diagram in the top left of Figure 14. Then, observed that, if the tie is provided with elasticity, we may transform such diagram into the diagram in top right of Figure 14 by using a Reidermeister move of second type. Thus, we consider ties as elastic objects, and therefore, they are represented as a spring. More generally, using the defining relations of T BB n, we know that there are 2k−i−1 equivalent expression for ni,k. Specifically, given a pair i, k, such that k − 1 > 1, we have that: (17) ni,k = sisi+1 ··· sk−2ηk−1s−1 k−2 ··· s−1 for all possible choices of sl = σl or sl = σ−1 (see [3, Section 3.2] for details). Thus, we have that the elements ηi,j diagrammatically corresponds to an elastic tie joining the i-th moving strand with the j-th moving strand. i+1s−1 i l ...,,,. . .. . .i+1i. . .. . .ii+1. . .φ1ηiρ1σi∼,∼∼∼∼∼TIED LINKS IN THE SOLID TORUS 13 FIGURE 15. Relation (14) in terms of diagrams FIGURE 16. Relations (21) and (22) in terms of diagrams Additionally, we have that the following relations hold: σiηi,j = ηi+1,jσi σjηi,j = ηi,j+1σi σi−1ηi,j = ηi−1,jσi−1 σj−1ηi,j = ηi,j−1σj−1 (18) and ni,knk,m = ni,kni,m = nk,mni,m for all 1 ≤ i, k, m ≤ n (19) On the other hand, we can obtain similar results for the ties that are connected to the fixed strand. Indeed, for i = 2 the relation (14) corresponds to the diagram in Figure 15. Then, by using a Reidermeister move of second type, we also may consider that the tie is elastic (as in the type A case). Thus, using induction, we have that the element φj diagrammatically corresponds to a tie joining the fixed strand and the j-th moving strand. Additionally, the elements φ(cid:48)js satisfy the following relations: φjηi = ηiφj φjσi = σiφsi(j) where si is the transposition (i i + 1) ηi,jφi = φiφj = φjηi,j for all 1 ≤ i ≤ n − 1 and 1 ≤ j ≤ n (20) (21) (22) for all 1 ≤ i (cid:54)= j ≤ n Indeed, (20) and (21) follow directly by using defining relations (12) -- (14). And, we obtain (22) by conjugating the defining relation (15) by the element (σi−1 . . . σ1)(σj−1 . . . σ1), whenever i < j, which we can suppose without loss of generality. Then, these relations correspond to the diagrams in Figure 16. ∼∼. . .. . .. . .. . .∼,. . .. . .. . .ij. . .. . .. . .ij. . .. . .. . .ij∼∼14 MARCELO FLORES1 FIGURE 17. Rearrange of ties in Alexander Theorem Let T B∼n be the submonoid of T BB n generated by the elements ηi,j, φj with 1 ≤ i, j ≤ n. Thus, using the preceding results, we have the following proposition Proposition 1. Let α be a tied briad in T BB n. Then, α can be written by α = γβ (or α = βγ(cid:48)), where β is a braid of type B, and γ ( or γ') is in T B∼n . (cf. [3, Proposition 3.2]) (cid:3) Proof. The result follows easily by using relations (18) and (21). Proposition 2. Let γ be an element of T B∼n . Then, γ defines a equivalence relation in the set of n+1 strands (including the fixed strand). (cf. [3, Proposition 3.3]) The closure of a tied braid α in T BB in order to prove a Markov theorem for tied links. 6. THE ALEXANDER AND MARKOV THEOREMS FOR TIED LINKS IN ST Proof. The properties of reflexivity and symmetry are direct, whereas the transitivity prop- (cid:3) erty is implied by relations (19) and (22) Remark 7. Let α = γβ be an element of T BB n, where β is a braid and γ ∈ T B∼n . By Proposition 2, γ induces a partition in the set of strands of β, or equivalently, a set- partition of {0, 1, . . . , n}, where 0 represents the fixed strand, and i represents the i-th moving strand, for 1 ≤ i ≤ n. (cf. [6, Proposition 3]) n, denoted by(cid:98)α, is defined analogously as closure in (cid:102)Wn (see Section 2.2). Clearly, the result of closure(cid:98)α, is a tied link in ST . Thus, we have a map(cid:98) : T BB ∞ → TST . In this section, we prove that this map is surjective. We then define n such that L =(cid:98)α. (Alexander theorem for tied links in ST) Let L be a link in TST . Then, there a set of Markov moves in T BB ∞ Theorem 4. is α ∈ T BB Proof. Let L be a tied link in TST . Recall from Section 2.2 that L can be regarded as a mixed link with ties. Then, we apply the algorithm proposed by S. Lambropoulou (see [18, Section 2.1]) ignoring the ties. To do that, roughly speaking, we fix O as the center of the fixed component. Then, we apply the Alexander procedure, thought maintaining the fixed component unaltered. Eventually, the resulting link could have ties connecting points in opposite sides from O. However, using that the ties ends can move freely along the strands and the transparency property, we can arrange them such that they lie in an annulus centered in O (see Figure 17). Finally, we obtain a tied braid by cutting along a half line (cid:3) with origin O. This tied braid is by construction tie isotopic to L. In the following τα denotes the image of α through the natural homomorphism from Wn into the symmetric group Sn (see Section 2.1). OOTIED LINKS IN THE SOLID TORUS 15 FIGURE 18. Two different tied braids that have the same closure Definition 6. Two tied braids in T BB ∞ other by applying a finite sequence of the following moves: are ∼M -equivalent if one can be obtained from the (i) αβ can be exchanged by βα (ii) ασ can be exchanged by ασn or ασ−1 n (iii) α can be exchanged by ηi,jα, if τα(i) = j. (iv) α can be exchanged by φjα, whenever τα(i) = j and α constains φi. If α and β in T BB ∞ are ∼M -equivalent, we write α ∼M β. Let α1 and α2 be tied braids in T BB (Markov theorem for tied links in ST) Let α1, α2 be tied braids in T BB n. Theorem 5. Then, the links L1 = α1 and L2 = α2 are tied isotopic if and only if α1 ∼M α2. Proof. Firstly, note that considering the ties properties (elasticity, transparency), we can proceed for tied links as in the proof of Markov theorem for classical links in ST (see [18, Theorem 3]). n. Thus, we have that α1 = γ1β1 and α2 = γ2β1 according to Proposition 1. Set Li = (cid:98)αi, for i = 1, 2, and suppose that L1 and L2 are isotopic tied links. We have to prove that α1 ∼M α2. By Definition 3, we have that L1 and L2 are isotopic as links in ST . Thus, we have that α1 and α2 are related by moves of type (i) and (ii), which coincide with the classical Markov moves in ST. Thus, we have that β1 and β2 are ∼M -- equivalent. More precisely, we can transform β1 into β2 by using (i) and (ii) moves. Thus, after applying this moves, we have that β1 and β2 consist in the same braid, denoted by β. Therefore, by now, we have that αi ∼M γiβ, for i = 1, 2. Since L1 ∼t L2, we also of components of (cid:98)β. However, this fact does not imply that γ1 = γ2 (for instance, see know that the set of ties corresponding to α1 and α2 define the same partition in the set Figure 18 ). Therefore, it is enough to prove that we can transform γ1 into γ2 by applying moves of type (iii) and (iv). If γ1 and γ2 just contain ties joining the moving strands. By [3, Theorem 3.7], we have that γγ1 = γγ2, where γ = ηi,j. (23) (cid:89) τβ (i)=j That is, γ1 ∼M γ2 by using move (iii). We now suppose that γ1 and γ2 have some tie interacting with the fixed strand. Let us say that γ1 contains φi. Let ci be the cycle of τβ define the same partition in the set of components of(cid:98)β. For a cycle ck of τβ, we define that contains i. Then, γ2 must contain φi or φj, for some j in the cycle ci, since, γ1 and γ2 (cid:89) j∈ck δck := φj. 16 Now, set MARCELO FLORES1 (cid:89) δci, δ = where ci is the cycle of τβ containing i. Then, we have that δγγ1 = δγγ2, where γ is the (cid:3) element from (23). Thus, γ1 ∼M γ2 by using moves (iii) and (iv). φi∈γ1 7. THE INVARIANT FB VIA JONES' METHOD The goal of this section is to recover the invariant FB by using Jones method. Firstly, observe that, by applying Theorems 4 and 5, we have a correspondence between isotopy (according to ∼M ). Secondly, classes of TST and the set of equivalence classes in T BB ∞ we define a natural representation from T BB n into the algebra an algebra of braids and ties of type B [6]. This algebra supports a Markov trace, hence we may apply Jones' method to obtain the invariant ∆B. We then probe that this invariant is equivalent to FB from Section 4. 7.1. An algebra of braids and ties of type B. We begin recalling the definition of the algebra introduced in [6], which is an analogous of the classical bt -- algebra in the context of Coxeter groups of type B. Definition 7. Let n ≥ 2. We define a bt -- algebra of type B, denoted by E B n(u, v), as the algebra generated by B1, T1 . . . , Tn−1 and F1, . . . Fn, E1 . . . , En−1, subject to the following relations n = E B TiTj = TjTi TiTi+1Ti = Ti+1TiTi+1 i = 1 + (u − u−1)EiTi T 2 E2 i = Ei for all i, EiEj = EjEi EiTi = TiEi EiTj = TjEi for all i − j > 1, for all i − j = 1, for all i − j = 1, for all 1 ≤ i ≤ n − 2, for all 1 ≤ i ≤ n − 1, for all i, j for all 1 ≤ i ≤ n − 1, for all i − j > 1, (24) (25) (26) (27) (28) (29) (30) (31) (32) (33) (34) (35) (36) (37) (38) (39) (40) (41) 1 as the algebra generated by 1, B1 and F1 subject to for all 1 ≤ i ≤ n − 1. EiEjTi = TiEiEj = EjTiEj EiTjTi = TjTiEj B1T1B1T1 = T1B1T1B1, B1Ti = TiB1 for all i > 1, 1 = 1 + (v − v−1)F1B1, B2 B1Ei = EiB1 for all i, F 2 i = Fi for all i, for all j, for all i, j, B1Fj = FjB1 FiEj = EjFi FjTi = TiFsi(j) where si = (i, i + 1), EiFi = FiFi+1 = EiFi+1 For n = 1, we define the algebra E B the relations (35), (37) and (38). Proposition 3. The mapping σi (cid:55)→ Ti, ρ1 (cid:55)→ B1, ηi (cid:55)→ Ei and φ1 (cid:55)→ F1 defines a representation from T BB n, denoted by θ. n into E B TIED LINKS IN THE SOLID TORUS 17 Proof. It is enough to prove that Ti, B1, Ei and F1, for 1 ≤ i ≤ n − 1 satisfy the defining n. By [3, Proposition 4.2], we have that the relations of type A are satisifed relations of T BB by the elements Ti's and Ei's. On the other hand, using the defining relations (33) -- (41) of n, we obtain that the generators of E B E B (cid:3) n also satisfy type B relations. We now recall the definition of the Markov trace supported by the algebra E B n, which is n}n≥1. That is, for all n ≥ 1, the linear map the main ingredient of the Jones method. Theorem 6. tr is a Markov trace on {E B trn : E B n → K satisfies the following properties: (i) trn(1) = 1 (ii) trn+1(XTn) = trn+1(XEnTn) = ztrn(X) (iii) trn+1(XEn) = trn+1(XFn+1) = xtrn(X) (iv) trn+1(XBn) = ytrn(X) (v) trn+1(XBn+1En) = trn+1(XBn+1Fn+1) = wtrn(X) (vi) trn(XY ) = trn(Y X), n. where X, Y ∈ E B In [6, Section 6], we define the invariant ∆B for classical links in the solid torus, by using Jones method. This, invariant is essentially the composition of π, the natural repre- n, and the Markov trace from Theorem 6 (up to normalization and sentation of Wn into E B re -- escalation). Analogously, we now construct an extension of such invariant, which is also denoted by ∆B, to simplify notation. Set L := z − (u − u−1)x z n in E B Let θL be the representation of T BB ηi (cid:55)→ Ei and φ1 (cid:55)→ F1. Then, for α ∈ T BB 1 z√L and D := . (42) n, defined by the mapping σi (cid:55)→ √LTi, ρ1 (cid:55)→ B1, n, we define It is well know that the previous expression can be rewritten as follows ∆B(α) := (D)n−1(trn ◦ θL)(α). ∆B(α) = (D)n−1(√L)e(α)(trn ◦ θ)(α), (43) (44) where e(α) is the exponent sum of the σi's appearing in the braid α, and θ is the represen- tation from Proposition 3. Similarly to [6, Theorem 4], we obtain that ∆B is an invariant for tied links in ST . Moreover, we have the following result. Theorem 7. Let L be a tied link in TST obtained by closing a braid α ∈ T BB have ∆B(α) = FB(L). Theorem 3). Firstly, note that the unknots (cid:13), (cid:13),(cid:13)+ (cid:13)+ correspond to(cid:98)1, (cid:99)φ1, (cid:98)ρ1 and Proof. It is enough to prove that the invariant ∆B satisfies the skein relations of FB (see (cid:91)ρ1φ1, respectively. Thus, by trace conditions, we have that ∆B(α) = FB((cid:98)α) for all α ∈ n, and set L = (cid:98)α. Let α+, α− and α∼ be the tied braids 1, that is, ∆B satisfies the initial conditions (i) from Theorem 3. Let α be a tied braid in T BB that are identical outside the small disk, whereas inside the disk look according to Figure 7. Then, we have that L+ = α+, L− = α− and L∼ = α∼. n. Then, we T BB 18 MARCELO FLORES1 Therefore, using the quadratic relation of E B n, we have that trn(θ(α+)) = trn(θ(α−)) + (u − u−1)trn(θ(αsim)). (45) Note now that e(α∼) = e(α+) − 1 = e(α−) + 1. Thus, we have ∆B(α+) =Dn−1λe(α∼)+1trn(θ(α+)), ∆B(α−) = Dn−1λe(α∼)−1trn(θ(α−)) and ∆B(α∼) =Dn−1λe(α∼)trn(θ(α∼)) Therefore, from Eq. (45) we obtain 1 λ ∆B(α+) − λ∆B(α−) = (u − u−1)∆Bα∼), as we wanted. Finally, we can prove analogously that ∆B(L+) satisfies the second skein (cid:3) relation by using the quadratic relation (35) of E B n. ACKNOWLEDGEMENTS. The author wishes to thank the Math section of ICTP, where the paper was written, for the invitation and hospitality. In particular, the author wishes to express his gratitude to Francesca Aicardi for their helpful comments, which were essential to develop this work. [1] AICARDI, JUYUMAYA, https://arxiv.org/abs/1709.03740 (2000). F., AND REFERENCES J. An algebra involving braids and ties. See [2] AICARDI, F., AND JUYUMAYA, J. Markov trace on the algebra of braids and ties. Moscow Mathematical Journal 16, 3 (2016), 397 -- 431. [3] AICARDI, F., AND JUYUMAYA, J. Tied links. Journal of Knot Theory and its Ramifications 25, 9 (2016), 28 pages. [4] CHLOUVERAKI, M., AND D'ANDECY, L. P. Markov trace on affine and cyclotomic Yokonuma -- Hecke algebras. Int. Math. Res. Notices 2016 (2016), 4167 -- 4228. [5] CHLOUVERAKI, M., JUYUMAYA, J., KARVOUNIS, K., AND LAMBROPOULOU, S. Identifying the invari- ants for classical knots and links from the Yokonuma -- Hecke algebras. Submitted for publication. See also arXiv:1505.06666 (2015). [6] FLORES, M. A braids and ties algebra of type b. Journal of Pure and Applied Algebra 224, 1 (2020), 1 -- 32. [7] FLORES, M., JUYUMAYA, J., AND LAMBROPOULOU, S. A framization of the Hecke algebra of type B. J. Pure and Appl. Algebra 222 (2018), 778 -- 806. [8] FREYD, P., YETTER, D., HOSTE, J., LICKORISH, W., MILLETT, K., AND OCNEANU, A. A new polyno- mial invariant of knots and links. Bull. AMS 12 (1985), 239 -- 246. [9] GRABOVSEK, B. Knots in the solid torus up to 6 crossings. Journal of Knot Theory and its Ramifications 21 (2012), 43 pages. [10] ISAEV, A. P., AND OGIEVETSKY, O. On baxterized solutions of reflection equation and integrable chain models. Nuclear Phys. B 760, no. 3 (2007), 167 -- 183. [11] JONES, V. Index for subfactors. Inventiones Mathematicae 72 (1983), 1 -- 25. [12] JONES, V. A polynomial invartiant for knots via von neumann algebras. Bulletin of the AMS 12 (1985), 103 -- 111. [13] JONES, V. Hecke algebra representations of braid groups and link polynomials. Annals of Mathematics 126 (1987), 335 -- 388. [14] JUYUMAYA, J. Markov trace on the Yokonuma-Hecke algebra. J. Knot Theory and its Ramifications 13 (2004), 25 -- 39. [15] JUYUMAYA, J., AND LAMBROPOULOU, S. An adelic extension of the jones polynomial. In Mathematics of knots, M. Banagl and D. Vogel, Eds., Contributions in the Mathematical and Computational Sciences, Vol. 1. Springer, 2009, pp. 825 -- 840. [16] JUYUMAYA, J., AND LAMBROPOULOU, S. An invariant for singular knots. J. Knot Theory and its Ramifi- cations 18, 6 (2009), 825 -- 840. [17] JUYUMAYA, J., AND LAMBROPOULOU, S. p-adic framed braids II. Advances in Mathematics 234 (2013), 149 -- 191. TIED LINKS IN THE SOLID TORUS 19 [18] LAMBROPOULOU, S. Solid torus links and Hecke algebras of B -- type. Proceeding of the conference on quantum Topology, D.N. Yetter ed., World Scientific Press (1994), pp. 225 -- 245. [19] LAMBROPOULOU, S. Knot theory related to generalized and cyclotomic hecke algebras of type b. J. Knot Theory Ramifications 8 (1999), No. 5, 621 -- 658. [20] LICKORISH, W., AND MILLET, K. C. A polynomyal invariant for oriented links. Topology 26, 1 (1987), 107 -- 141. [21] YOKONUMA, T. Sur la structure des anneux de Hecke d'un group de Chevalley fin. C.R. Acad. Sc. Paris 264 (1967), 344 -- 347. 1 INSTITUTO DE MATEMÁTICAS, UNIVERSIDAD DE VALPARAÍSO, GRAN BRETAÑA 1091, VALPARAÍSO, CHILE E-mail address: [email protected]
1703.05072	1	1703	2017-03-15T10:42:03	Irreducible weight modules with a finite-dimensional weight space over the twisted N=1 Schr\"{o}dinger-Neveu-Schwarz algebra	[ "math.RA", "math.RT" ]	It is shown that there are no simple mixed modules over the twisted N=1 Schr\"{o}dinger-Neveu-Schwarz algebra, which implies that every irreducible weight module over it with a nontrivial finite-dimensional weight space, is a Harish-Chandra module.	math.RA	math	Irreducible weight modules with a finite-dimensional weight space over the twisted N = 1 Schrodinger-Neveu-Schwarz algebra Huanxia Fa1), Jianzhi Han2), Junbo Li1) 1)School of Mathematics and Statistics, Changshu Institute of Technology, Changshu 215500, China 2)Department of Mathematics, Tongji University, Shanghai, 200092, China E-mail: sd [email protected], [email protected], sd [email protected] Abstract. It is shown that there are no simple mixed modules over the twisted N = 1 Schrodinger- Neveu-Schwarz algebra, which implies that every irreducible weight module over it with a nontrivial finite-dimensional weight space, is a Harish-Chandra module. Key Words: the twisted N = 1 Schrodinger-Neveu-Schwarz algebra, weight modules, irreducible modules Mathematics Subject Classification (2010): 17B10, 17B65, 17B68. 1 Introduction The Schrodinger-Virasoro type Lie algebras and their supersymmetric counterparts are closely related to mathematics and physics. The Schrodinger-Virasoro algebra was originally introduced by M. Henkel in 1994 during the process of trying to apply the concepts and methods of conformal field theory to models of statistical physics (see [2] for reference). The deformative counterparts were introduced in [8] and their supersymmetric extensions were investigated in the context of supersymmetric quantum mechanics. Schrodinger-Neveu- Schwarz Lie superalgebras sns(N ) with N supercharges were introduced in [3] generally, which appear as a semi-direct product of Lie algebras of super-contact vector fields with infinite-dimensional nilpotent Lie superalgebras. The twisted N = 1 Schrodinger-Neveu- Schwarz algebra gtsns (introduced firstly in [3]) is an infinite-dimensional Lie superalgebra over C with the basis {Ln, Gr, Yp, Mp, c n ∈ Z, r ∈ 1 non-vanishing super brackets: Z} and the following 2 + Z, p ∈ 1 2 [Ln, Lm] = (m − n)Lm+n + n3 − n 12 δm,−nc, [Ln, Gr ] = (r − n 2 )Gr+n, [Gr, Gs] = 2Lr+s + 1 − 4r2 12 δr,−sc, 7 1 0 2 r a M 5 1 ] . A R h t a m [ 1 v 2 7 0 5 0 . 3 0 7 1 : v i X r a 2 )Yp+n, pYp+n, [Ln, Yp] =( (p − n [Gr, Yp] =( 1 2(p − r)Yp+r, 2Yp+r, if p ∈ Z, if p ∈ 1 2 + Z, if p ∈ Z, if p ∈ 1 2 + Z, [Ln, Mp] =( [Gr, Mp] =( p pMp+n, if p ∈ Z, (p + n 2 )Mp+n, if p ∈ 1 2 + Z, 2 Mp+r, 2Mp+r, if p ∈ Z, if p ∈ 1 2 + Z, [Yp, Yq] = 1 2(q − p)Mq+p, q 2 Mq+p, 2Mq+p, if p, q ∈ Z, if p ∈ Z, q ∈ 1 2 + Z, if p, q ∈ 1 2 + Z. 1 It is easy to see that gtsns is Z2-graded with gtsns = gtsns¯0 ⊕gtsns¯1, where gtsns¯0 = SpanC{Ln, Yn, Mn, c n ∈ Z}, gtsns¯1 = SpanC{Gr, Yr, Mr r ∈ 1 2 + Z}. The Cartan subalgebra (exactly the maximal toral subalgebra) of gtsns iseh = CL0 ⊕ CM0 ⊕ Cc. It should be noted that gtsns¯0 is precisely the well-known twisted Schrodinger-Virasoro Lie algebra and the subalgebra ens spanned by {Ln, Gr, c n ∈ Z, r ∈ 1 Neveu-Schwarz algebra. DenoteeI = I ⊕ c with I = SpanC{Yp, Mp p ∈ 1 see that botheI and I are ideals of gtsns. Aneh-diagonalizable module over gtsns is usually called weight module. If all weight spaces of a weight gtsns-moudule are finite-dimensional, the module is called a Harish-Chandra 2 + Z} is the N = 1 Z}. It is easy to module. We introduce the following notations: 2 1 2 1 2 1 2 1 2 Z, r ∈ Z, r ∈ + Z, p, r < 0}, + Z, p, r > 0}, gtsns+ = SpanC{Ln, Yp, Mp, Gr n ∈ Z+, p ∈ gtsns− = SpanC{Ln, Yp, Mp, Gr n ∈ Z−, p ∈ gtsns0 = SpanC{L0, Y0, M0, c}. Then gtsns admits the following triangular decomposition: gtsns = gtsns+ ⊕ gtsns0 ⊕ gtsns−. For any irreducible weight gtsns-module V , L0, M0 and c must act as some complex weight gtsns-module, then there exists some λ ∈ C such that supp(V ) ⊆ λ + 1 numbers on it. Furthermore, V has the weight space decomposition V = ⊕λ∈CVλ, where Vλ = {v ∈ V L0v = λv} is called a weight space with weight λ. Denote the set of weights λ of V by supp(V ) := {λ ∈ C Vλ 6= 0}, which is called the support of V . If V is an irreducible An irreducible weight module V is called a pointed module if there exists a weight λ ∈ C such that dim Vλ = 1. The following natural problem was firstly referred in [10]: Z. 2 Problem 1.1 Is any irreducible pointed module over the Virasoro algebra a Harish-chandra module? An irreducible weight module V is called a mixed module if there exist λ ∈ C and k ∈ Z∗ such that dim Vλ = ∞ and dim Vλ+k < ∞. The following conjecture was given in [6]: Conjecture 1.2 There are no irreducible mixed modules over the Virasoro algebra. The positive answers to the above question and conjecture were given in [7]. Such question and conjecture have already been solved on the truncated Virasoro algebras, the W - algebra W (2, 2), the twisted Schrodinger-Virasoro algebra, the twisted Heisenberg-Virasoro 2 algebra and the Neveu-Schwarz algebra in [1, 4, 5, 9, 11]. In this note, we also give the positive answers to the above question and conjecture for the twisted N = 1 Schrodinger- Neveu-Schwarz algebra. Our main result is the following: that dim Vλ = ∞, then supp(V ) = λ + 1 2 Theorem 1.3 For any irreducible weight gtsns-module V , if there exists some λ ∈ C such words, for any irreducible weight gtsns-module V , the condition that there exists some λ ∈ C irreducible mixed gtsns-modules. Throughout this paper, we respectively denote Z∗, Z+, Z−, Z+ and Z− the sets of the such that 0 < dim Vλ < ∞ implies that V is a Harish-Chandra module. Then there are no Z and dim Vλ+k = ∞ for every k ∈ 1 2 Z. In other nonzero, positive, negative, nonnegative and non-positive integers. 2 Proof of Theorem 1.3 We first recall a main result about the irreducible weight Neveu-Schwarz-modules given in [11], which we cited here as the following lemma: Z, we have dim Vλ+k = ∞. Z and for every k ∈ 1 2 Since I = SpanC{Yp, Mp p ∈ 1 2 that dim Vλ = ∞. Then supp(V ) = λ + 1 2 Lemma 2.1 Let V be an irreducible weight ens-module. Assume that there exists λ ∈ C such Z} is a nontrivial ideal of gtsns, IV is a submodule of V for any gtsns-module V , which gives the following lemma (such lemma was not used in [11]). Lemma 2.2 For any irreducible weight gtsns-module V , we have IV = 0 or IV = V . For any irreducible weight gtsns-module V , if we can prove IV = 0, then V will degenerate to be an irreducible weight ens-module. Then Theorem 1.3 follows from Lemma 2.1 in this case. Denote the universal enveloping algebra of gtsns by U(gtsns). For any irreducible weight gtsns-module V , if there exists some 0 6= v ∈ V such that Iv = 0, then U(gtsns)v = V . And Lemma 2.3 Assume that V is an irreducible weight module over gtsns. If there exists some weight module over ens. erate gtsns+ and gtsns−. Then the following lemma follows from the fact that both highest and lowest weight gtsns-modules are Harish-Chandra modules (similar to the Virasoro algebra 0 6= v ∈ V such that Iv = 0. Then I acts trivially on V and V is simply an irreducible the following lemma follows: It is easily to see that {G 3 case investigated in [7]). } respectively gen- } and {G− , M− , G 1 , Y 1 , M 1 , G− , Y− 2 2 2 2 1 2 3 2 1 2 1 2 Lemma 2.4 Assume that there exists µ ∈ C and a non-zero element v ∈ Vµ, such that G 3 2 v = G 1 2 v = Y 1 2 v = M 1 2 v = 0 or G− v = G− v = Y− 1 2 1 2 3 2 v = M− 1 2 v = 0. Then V is a Harish-Chandra module. 3 Proof of Theorem 1.3 We shall prove this theorem step by step by several lemmas. Assume now that V is an irreducible weight gtsns-module such that there exists λ ∈ C satisfying dim Vλ = ∞. Denote the set Sλ = {p dim Vλ+p < ∞, p ∈ 1 2 Z∗}. Lemma 2.5 For any λ ∈ C satisfying dim Vλ = ∞, there are at most two adjacent elements in Sλ. For convenience, we can suppose Sλ ⊆ { 1 2 , 1}. Proof Suppose there are two different elements p and q in Sλ. Without loss of generality, we can assume p = 1 Case 1 p = 1 2, q > 1 2, q > 1 2. 2 and q ∈ 1 2 + Z. Let W be the intersection of the kernels of the linear maps X 1 2 : Vλ → Vλ+ 1 2 and Xq : Vλ → Vλ+q for X = G, Y, M. Then X 1 assumptions dim Vλ = ∞, dim Vλ+ 1 G 1 W = XqW = 0 for all X ∈ {G, Y, M}. The < ∞ and dim Vλ+q < ∞ force dim W = ∞. Using W = 0, GqW = YqW = MqW = 0 and the given brackets, we obtain W = M 1 W = Y 1 2 2 2 2 2 LnW = GrW = YpW = MpW = 0 for all n = 1, 2q, 2q+1, 2q+2, · · · , r = 1 p ∈ 1 2 2 , q, q+1, q+2, · · · , w 6= 0 for all 0 6= w ∈ W . Otherwise, W would be a Z>0. We claim that q > 3 2 and G 3 2 Harish-Chandra module by Lemma 2.4. Thus dim G 3 2 W = ∞. Since dim Vλ+ 1 2 < ∞, there exists some w ∈ W such that v = G 3 2 w and L−1v = Y−1v = M−1v = 0. It is easily to verify that gtsns+v = 0, which forces V to be a Harish-Chandra module according to Lemma 2.4. This contracts with our assumptions. Hence this case is impossible. Case 2 p = 1 2, q > 1 and q ∈ Z. Let W be the intersection of the kernels of the linear maps X 1 2 : Vλ → Vλ+ 1 2 and Zq : Vλ → Vλ+q for Z = L, Y, M. Then X 1 Z ∈ {L, Y, M}. The assumptions dim Vλ = ∞, dim Vλ+ 1 dim W = ∞. Using G 1 < ∞ and dim Vλ+q < ∞ force W = 0, LqW = YqW = MqW = 0 and the given brackets, we obtain LnW = GrW = YpW = MpW = 0 for all n = 1, q, q + 1, q + 2, · · · , r = 1 w 6= 0 for all 0 6= w ∈ W . Otherwise, W = ZqW = 0 for all X ∈ {G, Y, M} and Z+. We claim that G 3 W = M 1 W = Y 1 2 , q + 3 2 , · · · , p ∈ 1 2 , q + 1 2 2 2 2 2 2 2 W would be a Harish-Chandra module by Lemma 2.4. Thus dim G 3 2 W = ∞. Similar to the proof of the first case, one can prove that this case is also impossible. Then this lemma follows. (cid:3) The following lemma can be easily verified, partially given in Lemma 2 of [11]. Lemma 2.6 (i) Let 0 6= v ∈ V be such that G 1 2 v = Y 1 2 v = M 1 2 v = 0. Then L1G 1 2 − G 3 v = 0 = Ypv = Mpv for all p ∈ Z+. 1 2 (cid:0) 1 2 (ii) Let 0 6= v ∈ V be such that G− v = Y− v = M− 1 2 1 2 v = 0. Then L−1G− 1 2 + G− (cid:0) 1 2 v = 0 = Y−pv = M−pv for all p ∈ 3 2 1 2 Z−. 4 1 2 2(cid:1)G 3 2(cid:1)G− 2 3 According to Lemma 2.5, we can assume V is an irreducible weight gtsns-module such that there exists some µ ∈ C satisfying dim Vµ < ∞, dim Vµ+ 1 all p ∈ Z∗ ∪ { 1 < ∞ and dim Vµ+p = ∞ for 2 2 + Z∗}. Lemma 2.7 µ ∈ {−1, 1 2}. Proof Let U be the intersection of the kernels of the linear maps X 1 → Vµ for U = 0. The assumptions dim Vµ < ∞ and u 6= 0 for all 0 6= u ∈ U. Otherwise, M : Vµ− 2 1 2 X = G, Y, M. Then G 1 2 U = Y 1 2 U = M 1 2 1 = ∞ force dim U = ∞. We claim that G 3 dim Vµ− would be a Harish-Chandra module by Lemma 2.4. Thus dim G 3 2 2 ∞, there exists some u ∈ U such that v = G 3 Lemma 2.6, we obtain ( 1 2 L1G 1 2 − G 3 2 u and G− v = M− )v = 0 and Ypu = Mpu = 0 for all p ∈ 1 2 v = Y− 2 1 2 2 2 Z+. Then U is 2 U = ∞. Since dim Mµ < v = 0. Using 1 1 simply a ens-module. Then we can use the same discussions as those given in Lemma 3 of [11], and have the following observations: 0 = L−1G− ( 1 2 L1G 1 2 − G 3 2 )w = (2L2 1 2 0 − 3L0)w =(cid:0)2(µ + 1)2 − 3(µ + 1)(cid:1). Then the lemma follows. (cid:3) According to Lemma 2.7, any irreducible weight gtsns-module V possessing only two different finite-dimensional weight spaces can be assumed to be of the following type: 2 + Z∗}. Without loss dim V 1 of generality, we will just prove that this case can not happen, which can be formulated as the following lemma: < ∞, dim V1 < ∞ and dim Vp+ 1 = ∞ for all p ∈ Z∗ ∪ { 1 2 2 Lemma 2.8 The irreducible weight gtsns-module V satisfying dim V 1 2 +p = ∞ for all p ∈ Z∗ ∪ { 1 2 + Z∗} does not exist. and dim V 1 2 < ∞, dim V1 < ∞ Proof Let W be the intersection of the kernels of the linear maps X 1 2 : V0 → V 1 2 for all X ∈ {G, Y, M}. Then G 1 2 W = Y 1 2 W = M 1 2 W = L0W = 0 and dim W = ∞. According to the following facts: [G 1 2 , G 1 2 ] = 2L1, [L1, Yr] = rY1+r, [L1, Mr] = (r + we can prove the following identities: L1W = YrW = MrW = 0, ∀ r ∈ )Mr+1, ∀ r ∈ 1 2 + Z+, + Z+. (2.1) 1 2 1 2 Recalling the following identities: [G 1 2 , Yp] = 2Yp+ 1 2 , [Yp, Yq] = 2Mp+q, ∀ p, q ∈ 1 2 + Z, 5 we can obtain the following identities: Then combining (2.1) and (2.2), we arrive at the following identities: YnW = MnW = 0, ∀ n ∈ Z+. YpW = MpW = 0, ∀ p ∈ ( 1 2 Z+)∗. Using the above obtained results, we can prove the following identities: YpG 3 2 W = MpG 3 2 W = 0, ∀ p ∈ ( 1 2 Z+)∗. (2.2) (2.3) (2.4) We claim that G 3 2 w 6= 0 for all 0 6= w ∈ W . Otherwise, V would become a Harish-Chandra module by Lemma 2.4, which forces dim G 3 we can deduce that there exists 0 6= v ∈ G 3 2 2 1 2 2 W = ∞. Recalling our suppose dim V1 < ∞, v = 0. Then W such that G− v = M− v = Y− 1 1 2 L−1v = G− G− 1 2 1 2 v = 0. According to the following identities: [L−1, Yp] = pYp−1, [L−1, Mp] = (p − 1 2 )Mp−1, [G− 1 2 , Yp] = 2Yp− , 1 2 [Yp, Yq] = 2Mp+q, ∀ p, q ∈ 1 2 + Z−, we can deduce the following results: Ypv = Mpv = 0, ∀ p, q ∈ ( The identities given in (2.4), imply Ypv = Mpv = 0, ∀ p, q ∈ ( 1 2 1 2 Z−)∗. Z+)∗, Combining Y0 = 1 2[G− , Y 1 2 1 2 ], M0 = 1 2 [G− , M 1 2 1 2 ], (2.5) and (2.6), we can deduce which together with (2.5) and (2.6), gives Y0v = M0v = 0, Ypv = Mpv = 0, ∀ p ∈ 1 2 Z. (2.5) (2.6) Then recalling Lemma 2.3, we can finally deduce that Yp, Mp act trivially on the whole V for all p ∈ 1 2 follows from Lemma 2.1. Z and V is simply an irreducible weight module over ens. Then this lemma By Lemmas 2.5, 2.7 and 2.8, we get the following lemma immediately. (cid:3) Lemma 2.9 There is at most one element in Sλ. 6 According to Lemma 2.9, we can suppose there exists some p ∈ 1 2 ∞. For convenience, we denote λ + p by µ in the following lemma. Z∗, such that dim Vλ+p < Lemma 2.10 Sλ = ∅. Proof According to our suppose, we know dim Vµ < ∞ with µ ∈ Sλ. Let W be the → Vµ for X ∈ {G, Y, M}. Then intersection of the kernels of the linear maps X 1 W = L0W = 0 and dim W = ∞. Similar to the corresponding proof of W = M 1 W = Y 1 : Vµ− 1 2 2 G 1 2 2 2 Lemma 2.8, we can prove the following identities: dim G 3 2 W = ∞, YpG 3 2 W = MpG 3 2 W = 0, ∀ p ∈ ( 1 2 Z+)∗. (2.7) (2.8) Recalling our suppose that dim Vµ < ∞, we claim that there exists some nonzero element v ∈ G 3 W such that 2 According to (2.8), we also know that L−1v = Y−1v = M−1v = 0. Ypv = Mpv = 0, ∀ p, q ∈ ( 1 2 Z+)∗. The following identities hold: [L−1, Y1−p] =( ( 3 2 − p)Y−p, (1 − p)Y−p, if p ∈ Z, if p ∈ 1 2 + Z, from which we can deduce Combining (2.10) and (2.12), we arrive at the following result: Ypv = 0, ∀ p ∈ 1 2 Z−. The identity Mpv = 0 for all p ∈ 1 2 Ypv = 0, ∀ p ∈ 1 2 Z. Z follows from 2.12 and the follows identities: [Yp, Yq] =( q 2 Mq+p, 2Mq+p, if p ∈ Z, q ∈ 1 2 + Z, if p, q ∈ 1 2 + Z. (2.9) (2.10) (2.11) (2.12) Then again recalling Lemma 2.3, we can finally deduce that Yp, Mp act trivially on the whole V for all p ∈ 1 2 follows from Lemma 2.1. Z and V is simply an irreducible weight module over ens. Then this lemma Then Theorem 1.3 follows immediately from Lemma 2.10. (cid:3) (cid:4) Acknowledgements Supported by a NSF grant BK20160403 of Jiangsu Province and NSF grants 11501417, 11671056, 11271056, 11101056 of China, Innovation Program of Shanghai Municipal Education Commission, Program for Young Excellent Talents in Tongji Univer- sity and the Fundamental Research Funds for the Central Universities. 7 References [1] X. Guo, X. Liu, Weight modules with a finite-dimensional weight space over the truncated Virasoro algebras, J. Math. Phys, 51 (2010), 123522. [2] M. Henkel, Schrodinger invariance and strongly anisotropic critical systems, J. Stat. Phys., 75(1994), 1023 -- 1029. [3] M. Henkel, J. Unterberger, Supersymmetric extensions of Schrodinger-invariance, Nucl. Phys. B, 746 (2006), 155 -- 201. [4] D. Liu, S. Gao, L. Zhu, Classification of irreducible weight modules over W -algebra W (2, 2), J. Math. Phys, 49 (2008), 113503. [5] J. Li, Y. Su, Irreducible weight modules over the twisted Schrodinger-Virasoro algebra, Acta Mathematica Sinica, English Series, 25(4) (2009), 531 -- 536. [6] V. Mazorchuk, On simple mixed modules over the Virasoro algebra, Mat. Stud., 22 (2004), 121 -- 128. [7] V. Mazorchuk, K. Zhao, Classification of simple weight Virasoro modules with a finite- dimensional weight space, J. Algebra, 307 (2007), 209 -- 214. [8] J. Unterberger, C. Roger, The Schrodinger-Virasoro Lie algebra, Springer Texts and Mono- graphs in Physics, Springer (Heidelberg 2012). [9] R. Shen, Y. Su, Classification of irreducible weight modules with a finite-dimensional weight space over twisted Heisenberg-Virasoro algebra, Acta Math. Sinica, English Series, 23 (2007), 189 -- 192. [10] X. Xu, Pointed representations of Virasoro algebra, A Chinese summary appears in Acta Math. Sinica, Chinese Series, 40(3), 479; Acta Math. Sinica, New Series, 13 (1997), 161 -- 168. [11] X. Zhang, Z. Xia, Classification of simple weight modules for the Neveu-Schwarz algebra with a finite-dimensional weight space, Comm. Alg. 40 (2012), 2161 -- 2170. 8
1706.04412	1	1706	2017-06-14T11:21:56	Towards a Chevalley-style non-commutative algebraic geometry	[ "math.RA" ]	We aim to construct a non-commutative algebraic geometry by using generalised valuations. To this end, we introduce groupoid valuation rings and associate suitable value functions to them. We show that these objects behave rather like their commutative counterparts. Many examples are given and a tentative connections with Dubrovin valuation rings is established.	math.RA	math	TOWARDS A CHEVALLEY-STYLE NON-COMMUTATIVE ALGEBRAIC GEOMETRY NIKOLAAS D. VERHULST Abstract. We aim to construct a non-commutative algebraic geometry by using generalised valuations. To this end, we introduce groupoid valuation rings and associate suitable value functions to them. We show that these objects behave rather like their commutative counterparts. Many examples are given and a tentative connections with Dubrovin valuation rings is established. Introduction Although valuation theory has reached a venerable age by now, it is still fertile ground for new research, in connection with e.g. resolution of singularities or trop- ical geometry. In the commutative case, the theory of valuations acts as a kind of translation mechanism, turning a field extension of transcendence degree one1 into its Zariski-Riemann surface and associating to a curve a collection of valuation rings in a field. We refer to Chevalley's classic work [2] for an in-depth study of this correspondence. It seems reasonable to hope that, with a good non-commutative analogue of a valuation ring, one could mimic the commutative construction to ar- rive at non-commutative abstract Riemann surfaces and general non-commutative algebraic geometry. Over the years, many suggestions have been made for what this good analogue could be. Besides Schilling's original definition of a non-commutative valuation ring (see below), the most important candidates are perhaps Dubrovin valuation rings (cfr. e.g. [7] or [8]) and gauges (cfr. [11]). The classical definition of a valuation ring as a subring R of a field k such that (R is total ) (1) ∀x ∈ k∗ : x /∈ R ⇒ x−1 ∈ R can easily be adapted for skewfields by adding the condition (2) ∀x ∈ k∗ : xRx−1 = R Schilling has shown that, if these two conditions hold true in a given ring R, then one can associate to R an equivalence class of valuation functions in much the same way as in the commutative case. Yet this definition is not completely satisfactory. For example, even in a very well-behaved non-commutative extension, valuations on the centre might not extend to the whole skewfield. A p-valuation on Q, for instance, does not extend to a valuation on the skewfield of Hamilton quaternions H (unless p = 2, see [13]). (R is stable). This problem can be solved by considering H as a (Z/2Z)2-graded skewfield ex- tension of R. Just like a field is defined as a ring wherein every non-zero element is invertible, so is a graded skewfield defined as a graded ring wherein every homo- geneous element is invertible. Similarly, a graded valuation ring is a homogeneous subring R of a graded skewfield such that (1) and (2) hold for any homogeneous x ∈ k∗. With this definition, it is easy to check that, if we write Rp for the ring of positives of some extension of the p-adic valuation to R, the ring Rp⊕Rpi⊕Rpj⊕Rpk 1If the transcendence degree is higher things become more complicated, but valuation theory is still useful e.g. to verify properness of maps. 1 2 NIKOLAAS D. VERHULST is a graded valuation ring on H considered as a (Z/2Z)2-graded skewfield. Graded valuation rings were introduced by Johnson in the late seventies (cfr. [5]) and are still an active field of research (cfr. e.g. [1]). However, in a certain sense, the correct non-commutative counterpart for the notion of a field is not that of a skewfield but rather that of a simple artinian ring and, unfortunately, such rings need not be graded skewfields. We do know, on the other hand, that they can be equipped with a natural groupoid grading since they are isomorphic to matrix rings over skewfields -- and matrix rings are groupoid graded (see section 1). In this paper, we will introduce the notions of a G-skewfield (of which simple artinian rings will be examples) and a G-valuation ring by adapting conditions (1) and (2) to a groupoid-graded context. We will also introduce G-valuations -- a natural generalisation of valuations -- and we will show that, as in the commutative case, G-valuation rings are in one-one correspondence with equivalence classes of G-valuations. Since our versions of conditions (1) and (2) are far less rigid, we will be able to associate G-valuations to many relatively ill-behaved subrings as well. Finally, we will establish a tentative connection with Dubrovin valuation rings. The next step in the non-commutative algebraic geometry programme hinted at in the first paragraph, would be proving approximation theorems and developing G-divisor theory. Once so far, a G-graded version of the Riemann-Roch theorem should be within reach; an ambitious goal, perhaps, but a worthy one. 1. Terminology and basic properties Remember that a groupoid is a (small) category wherein all morphisms are in- vertible. Alternatively, a groupoid can be thought of as a group for which the multi- plication is only partially defined. As a concrete example, let eij be the n×n-matrix with a one on place i, j and zeroes everywhere else. Then ∆n = {eij 1 ≤ i, j ≤ n} equipped with the partial multiplication (. · .) : ∆n × ∆n → ∆n : (eij , ekl) 7→(eil if j = k undefined if j 6= k is a groupoid. In fact, we can do this more generally: if G is any group, then the set G [∆n] = G × ∆n with the partial multiplication (. · .) : G [∆n] × G [∆n] → G [∆n] : ((g, eij), (g′, ekl)) 7→((gg′, eil) if j = k undefined if j 6= k is a groupoid as well. For the remainder of this paper, G will be a groupoid and R will be a ring. We will use the notation s(g) = gg−1 and t(g) = g−1g for the source and the target of g ∈ G respectively. Note that the multiplication gg′ of two elements g, g′ ∈ G is defined if and only if t(g) = s(g′). Two elements g and g′ of G are called connected if there is a morphism from t(g) to s(g′). This is a reflexive and transitive property which, since G is a groupoid, is also symmetric. The connected components are the equivalence classes with respect to connectedness, i.e. the maximal subsets of G in which any two elements are connected. Definition 1.1. R is said to be G-graded if there are abelian subgroups (Rg)g∈G such that R =Lg∈G Rg and RgRg′ ⊆ Rgg′ if gg′ exists while RgRg′ = 0 otherwise. If RgRg′ = Rgg′ whenever gg′ exists, then the grading is called strong. Example 1.2. The groupoid ring R[G] is constructed by endowing the set R[G] = {f : G → R #{g ∈ G f (g) 6= 0} < ∞} , TOWARDS A CHEVALLEY-STYLE NON-COMMUTATIVE ALGEBRAIC GEOMETRY 3 with a sum and a multiplication as follows: (f + f ′)(g) = f (g) + f ′(g), (f f ′)(g) = Xg′g′′=g f (g′)f ′(g′′). These operations are well-defined since f and f ′ have finite support. In a similar fashion as for group rings it can be checked that they define a ring structure on R[G]. This ring is canonically G-graded by putting R[G]g = {f : G → R ∀g′ 6= g : f (g′) = 0} for all g ∈ G. The most important example of groupoid graded rings are matrix rings: Mn(R) is isomorphic to R[∆n]. An element h is in a groupoid-graded ring R =Lg∈G Rg is called homogeneous if it is in S Rg. An ideal or a subring is called homogeneous if it is generated by homogeneous elements. We will call a homogeneous ideal G-maximal if it is maximal among proper homogeneous ideals. Similarly, we will call a G-graded ring G-simple if it contains no proper homogeneous ideals. The support of an element of the supports of its elements. r =Pg∈G rg is the set of g ∈ G for which rg 6= 0. The support of a set is the union We use G0 for the principal component, i.e. the set of idempotent elements of G. It is harmless to assume that G0 consists of but finitely many elements and that, if 1 = Pe∈G0 1e is the homogeneous decomposition of 1, we have 1e 6= 0 for all e ∈ G0 (cfr. [6]). Proposition 1.3. If R is G-graded, then the following elementary properties hold: (1) Re is a ring for any idempotent e of G. (2) If I is a G-ideal of R, then Ie is an ideal of Re for every idempotent e. (3) Rg is a left Rs(g), right Rt(g)-module. (4) G is a group if and only if there is some invertible homogeneous element. Proof. Re is by definition closed under addition and, since e is an idempotent, it is also closed under multiplication. Since the product of two distinct idempotents e and e′ of G is never defined, we have r = r1 = Pe∈G0 r1e = r1e for all r ∈ Re. Hence 1e is the unit of Re. For (3) it suffices to note that the map (. · .) : Rs(g) × Rg → Rg : (x, y) 7→ xy defines a left Rs(g)-multiplication on Rg, the right Rt(g)-multiplication being defined analogously. (2) is a special case of (3) in disguise where Ie ⊆ Re. To prove (4), note that, since ee′ is undefined for idempotents e 6= e′, any homogeneous element h ∈ Rg must be a zero divisor if there is some unit e 6= t(h) or e 6= s(h). If G is a group with unit e, then 1 ∈ Re is homogeneous and invertible. Proposition 1.4. Let R be a strongly G-graded ring. The homogeneous ideals of R are in 1-1 correspondence with ideals (cid:3) Ie1 ⊆ Re1 , ..., Ien ⊆ Ren where the ei are representatives of the connected components of G. Proof. Suppose g is in the connected component of e, i.e. g = g′eg′′. Since R is strongly graded, we must have that Ig = Rg′ Rg′−1 IgRg′′−1 Rg′′ ⊆ Rg′ IeRg′′ ⊆ Ig′eg′′ = Ig so any two homogeneous ideals of R restricting to the same ideals on Re1 , ..., Ren must be equal. Suppose, on the other hand, if Ie1 ⊆ Re1 , ..., Ien ⊆ Ren are ideals in their respective rings. If g is in the same connected component as e, then we can define Ig = Rg′ IeRg′′ where g′ and g′′ are connecting elements for g and e. (cid:3) I =Pg∈G Ig is then a homogeneous ideal of R. 4 NIKOLAAS D. VERHULST As an immediate consequence of the preceding proposition, the G-maximal ideals of a strongly G-graded R are those corresponding to a maximal ideal in one of the connected components and to Rg for any g not in that component. Therefore, the intersection of the G-maximal ideals -- which we call the G-Jacobson radical -- is the homogeneous ideal corresponding to the Jacobson radical in every connected component. We write, for any a in a G-graded R, 2. G-skewfields t(a) = Xe∈G0 a1e6=0 1e and s(a) = Xe∈G0 1ea6=0 1e. A G-inverse of a is an element b satisfying s(a) = ab = t(b) and s(b) = ba = t(a). If a has a G-inverse, we say that it is G-invertible. We will use the notation a−1 for the G-inverse of a, but one should keep in mind that the G-inverse of a may exist even if a is not invertible in R. In a desperate attempt to avoid confusion, we will denote the set of G-invertible elements of a G-graded ring R by R∗, while the set of invertible elements will be denoted by U (R). Proposition 2.1. If R is G-graded, then: (1) The G-inverse of a ∈ R, if it exists, is unique. (2) (ab)−1 = b−1a−1 if all terms involved exist. (3) The G-inverse of a ∈ Rh, if it exists, is in Rh−1. (4) If a is invertible in R, say ba = ab = 1, then b is the G-inverse of a. (5) The grading on R is strong if and only if Rg ∩ R∗ 6= ∅ for all g ∈ G. Proof. If b and b′ are G-inverses of a, then b = bt(b) = bs(a) = bab′ = s(a)b′ = b′ab′ = b′t(b′) = b′ which proves (1). (2) is obvious. Suppose a ∈ Rh is homogeneous and let a−1 = Pg∈G bg. For all g 6= h−1 we have that aa−1 = s(a) implies abg = 0 and a−1a = t(a) implies bga = 0. Therefore, bh−1 is a G-inverse and by (1) it must be unique. If a is invertible with inverse b, then 1ea 6= 0 for all e ∈ G0, which establishes that s(a) = 1. Similarly, we find t(a) = 1 and by symmetry the same holds for b. Consequently, a and b are each others G-inverses. To show (6), suppose that Rg ∩ R∗ 6= ∅ for all g ∈ G and assume gg′ exists. Then we have Rgg′ = Rs(g)Rgg′ = RgRg−1 Rgg′ ⊆ RgRg′ ⊆ Rgg′ so the grading is strong. If we assume the grading to be strong, then RgRg−1 must contain 1s(g) which implies that some element of Rg is G-invertible. (cid:3) A (group) graded ring is called a (group) grade skewfield if the homogeneous elements form a group (cfr. [9]). Similarly, we will call a G-graded ring a G- skewfield if the homogeneous elements form a groupoid, in other words, if every homogeneous element is G-invertible. In view of the preceding proposition, G- skewfields are necessarily strongly graded. Example 2.2. For a (skew)field k, the groupoid ring k[G] is a G-(skew)field. This means in particular that the matrix ring Mn(k) is a ∆n-skewfield. Proposition 2.3. If Q is a G-skewfield, then (1) If Q is a G-skewfield, then Qe is a skewfield for any idempotent e. TOWARDS A CHEVALLEY-STYLE NON-COMMUTATIVE ALGEBRAIC GEOMETRY 5 (2) If Q is a G-skewfield then, for non-zero h ∈ Qg, h′ ∈ Qg′ , hh′ = 0 if and only if t(g) 6= s(g′). Proof. Since for any a ∈ Qg we have a−1 ∈ Qg−1 (1) follows. Assume that (2) does not hold, then we can take non-zero h, h′ with hh′. Then 0 = h−1hh′ = 1t(h)h′ = h′ which is a contradiction. (cid:3) Proposition 2.4. A G-skewfield is G-simple in the sense that it has no non-trivial homogeneous ideals if and only if G is connected. Proof. Clearly, G being connected is a necessary condition for a G-skewfield to have no homogeneous ideals sinceLg∈C Qg for a connected component C ⊆ G is always a homogeneous ideal. On the other hand, if every element is G-invertible, then an ideal I with a homogeneous a ∈ I ∩ Rh necessarily contains all 1e for e ∈ G0 in the connected component of h, so a connected G-skewfield is G-simple. (cid:3) Example 2.5. Let k be a field and let G be ∆2 [Z/2Z]. Then Q = k[G] is an example of a G-skewfield for which supp(Q) is connected (so it is G-simple) but which is not simple. Indeed, S(e11,0) ≃ S(e22,0) ≃ k[Z/2Z] and this ring contains non-trivial ideals. The first building block is firmly in place, now: G-graded skewfields will play the same role that fields play in classical valuation theory. The most important exam- ples are of course the matrix rings, but there are more as the following construction shows. Suppose k is a field and assume that a partial function α : G × G → k∗ and a map σ : G → Aut(k) have been given such that for all a ∈ k and f, g, h ∈ G (1) σ(f )(σ(g)(a)) = α(f, g)σ(f g)(a)α(f, g)−1, (2) α(f, g)α(f g, h) = σ(f )(α(g, h))α(f, gh), (3) α(f, t(f )) = 1 = α(s(f ), f ), (4) α(f, g) exists if f g exists. Let k[G, α, σ] denote the free k-module with basis G and define a multiplication by demanding (ag)(bh) =(aσ(g)(b)α(f, g)f g 0 if f g is defined otherwise and distributivity. Proposition 2.6. This is indeed a G-skewfield and, if G is connected, then every G-skewfield Q is of this form. Proof. To show the latter statement, notice first that, due to the connectedness of G, we have Qe ≃ Qe′ for any e, e′ idempotent. Choose such an isomorphism and call it ιe,e′ . Take for any g ∈ G a G-invertible ug ∈ Qg. We assume ue to be the identity of Qe for any e ∈ G0. Define a map σ : G → Aut(Qe) by putting σ(g) : k → k : a 7→ ιs(g),e(cid:0)ugιe,t(g)(a)u−1 g (cid:1) and a partial function α : G × G → k∗ by α(f, g) =(uf ugu−1 undefined otherwise f g if f g is defined To check that these functions satisfy the necessary conditions, that Q is isomorphic to Qe[G, α, σ], and that any k[G, α, σ] is indeed a G-skewfield, it suffices to sprinkle the phrase "if f g is defined" liberally throughout the group-graded proof from [9]. Since this is relatively straightforward but rather tedious we omit it here. (cid:3) 6 NIKOLAAS D. VERHULST Example 2.7. Take a proper field extension k ֒→ k(√a), then the G-skewfield Q =(cid:18) k √ak √ak k (cid:19) is by Artin-Wedderburn isomorphic to M2(k) but, if both rings are endowed with their respective canonical G-gradings, not as a G-graded ring. This is an example of a (non-trivially) twisted groupoid ring. Proposition 2.8. A G-skewfield Q is artinian if and only if all Se are artinian. Proof. If some Se is not artinian, then there exists an infinite descending chain I0,e ) I1,e ) ··· of Se-ideals. This induces a chain I0 ) I1 ) ··· of Q-ideals by putting In = Lt(g)=e=s(g′) hgIe,nhg′ ⊕Lg /∈Ce Qg where Ce is the connected component of e and hg and hg′ are arbitrary non-zero elements in Qg and Qg′ respectively.2 Suppose on the other hand that all Se are artinian and that I0 ) I1 ) ··· is an infinite chain of descending ideals in Q. Then, for any e ∈ G0, 1eI01e ⊇ 1eI11e ⊇ ··· gives a descending chain of ideals in Se. Such a chain must stop, so there is some n with 1eIn1e = 1eIn+11e = ··· for all e ∈ G0. Take x ∈ In\In+1, then 1ex1e′ /∈ In+1 for some e, e′ ∈ G0. In fact, these e and e′ are in the same connected component, so Se ≃ Se′ and 1ex1e′ h1e /∈ Ie where h is an arbitrary homogeneous element connecting e′ and e. (cid:3) 3. G-valuation rings and G-valuations For the remainder of this section, we let R be a G-graded subring of a G-skewfield Q. If for every homogeneous h ∈ Q we have either h ∈ R or h−1 ∈ R, then we say that R is G-total . This is the canonical generalisation of totality (as referred to in the introduction) and gives rise to somewhat similar results. Note that if R is a G-total subring of the G-skewfield Q, then Re is a total subring of the skewfield Qe for any idempotent e ∈ G. This implies that any G-total subring of a G-skewfield contains 1e for all idempotents e. Proposition 3.1. Suppose R is G-total. If I and J are homogeneous left (resp. right) ideals, then Jg * Ig implies Ig′ ⊆ Jg′ for any g′ with the same right (resp. left) unit as g. In particular, we have Ig ⊆ Jg or Jg ⊆ Ig. Proof. Suppose I and J are homogeneous and Jg * Ig, so there exists some non- zero h ∈ Jg \ Ig. Suppose t(g′) = t(g), and assume h′ 6= 0 is in Ig′ (if no such h′ exists the claim is certainly true). This means that hh′−1 and h′h−1 are defined and at least one of these is in R. If hh′−1 is in R, then hh′−1h′ is in I ∩ Rg = Ig which is a contradiction, so h′h−1 must be in R and consequently h′h−1h is in J ∩ Rg′ = Jg′ . The other case is similar. (cid:3) Corollary 3.2. If R is a G-total subring, then any left (resp. right) ideal generated by homogeneous elements h1, ..., hn with the same target (resp. source) is cyclic. If G happens to be group, then the previous statements reduce to ideals are totally ordered and finitely generated ideals are cyclic respectively, both well-known results from (non-commutative) valuation theory which are generally not true in the G-graded case as the following example demonstrates. 2This is a slight abuse of notation, since the sum Lt(g)=e=s(g ′) hgIe,nhg ′ is only direct up to repetitions. TOWARDS A CHEVALLEY-STYLE NON-COMMUTATIVE ALGEBRAIC GEOMETRY 7 Example 3.3. Let k be a field and let Rv be a valuation ring in k with unique maximal ideal mv. Consider the subring k R =(cid:18)Rv 0 Rv(cid:19) of the ∆2-skewfield Q = M2(k). R is a G-total subring of Q, but the homogeneous ideals are not totally ordered since k I =(cid:18)mv 0 Rv(cid:19) and J =(cid:18)Rv k 0 mv(cid:19) are incomparable. Note that the fact that I1,1 ( J1,1 and J2,2 ( I2,2 implies I1,0 = J1,0 as well as J1,0 = J0,1. Note also that the ideal generated by e11 and e22 is not cyclic. Proposition 3.4. Let R be G-total, and put M the (homogeneous) ideal generated by the set of homogeneous elements which are not G-invertible in R. Then R/M is a G-skewfield and M is the maximal homogeneous ideal with the property that it contains no 1e for e ∈ G0. Proof. If x 6= 0 is some homogeneous element of R/M , then x = h + p where h is a non-zero homogeneous element of R \ M and p ∈ M . Let p = h1 + ··· + hn be the homogeneous decomposition of p. Then h−1 is also in R \ M and xh−1 = hh−1 + ph−1 = 1s(h) = 1s(x) since ph−1 must be in M -- otherwise some hih−1 is not in M and we would have hh−1 i ∈ R\, which is a contradiction. Analogously, we find h−1x = 1t(x) which implies that R/M is a G-skewfield. If M ′ is an ideal which contains M strictly, then there is some homogeneous h ∈ M ′ \ M so h is G-invertible in R and consequently hh−1 is in M , which implies that 1e ∈ M for some e ∈ G0. i ∈ R \ M and consequently h−1hh−1 i = h−1 (cid:3) R will be called G-stable if hRt(h)h−1 = Rs(h) for any homogeneous h. This implies that Re is stable for all e ∈ G0. In particular, if R is a G-total G-stable subring of the G-skewfield Q, then Re is a graded valuation ring in Qe for every e ∈ G0. Proposition 3.5. Any homogeneous right (resp. right) ideal if R is G-stable. Proof. Let I be a right G-ideal of R, let h ∈ I ∩ Qg be homogeneous and pick r ∈ R ∩ Qg′ arbitrary. If g′g does not exist rh = 0 ∈ I follows, so suppose g′g does exist. Because of G-stability of R, h−1rh is in R -- whether it is zero or not. If h−1r exists, we have hh−1rh = rh ∈ I since I is a right G-ideal. If h−1r does not exist, hh−1rh = 0 so it is again in I. This proves the claim for right G-ideals; the reasoning for left G-ideals is similar. (cid:3) left) ideal of R is a left (resp. Definition 3.6. If R is G-total and G-stable, we call it a G-valuation ring. With that, the second important concept is in place. Next on the menu are the G-valuation functions but, as a small intermezzo, we will consider some examples first: if Rv is a valuation ring in a skewfield D, then Mn(Rv) is a ∆n-valuation ring in the ∆n-skewfield Mn(D). This already yields a vast class of examples of G- valuation rings but there are many more, like example 3.3 or the following example. Example 3.7. Consider the rational Hamilton quaternions H(Q). There is a natural (Z/2Z)2-grading on H(Q) and consequently Mn(H(Q)) is a (Z/2Z)2[∆n]- skewfield. Any p-valuation on Q extends to a graded valuation ring R = Zp ⊕ Zpi⊕ Zpj ⊕ Zpk in H(Q). Mn(R) is then a (Z/2Z)2[∆n]-valuation ring on Mn(H(Q)). 8 NIKOLAAS D. VERHULST We say that a groupoid G is partially ordered by some partial order relation ≤ if g ≤ g′ implies hg ≤ hg′ and gh ≤ g′h when the multiplications are defined. We will say that G is ordered if every g ∈ G is comparable to s(g) and t(g). If G is a group, ordered in this sense is the same as totally ordered. Definition 3.8. Let G be a groupoid and let Q be a G-skewfield. A G-valuation on Q is a surjective map v : Q → Γ ∪ {∞} for some ordered groupoid Γ (with, as usual, ∞ > γ and ∞γ = γ∞ = ∞ for all γ ∈ Γ) satisfying: (1) v(x) = ∞ ⇔ x = 0, (2) v(x + y) ≥ v(z) if v(y) ≥ v(z) ≤ v(x), (3) v(hh′) = v(h)v(h′) for h ∈ Qg, h′ ∈ Qg′ if gg′ is defined. We have all the components now, but we still have to make sure everything fits smoothly together. If v : Q → Γ ∪ {∞} is a G-valuation, then we let Tv be the ring generated by homogeneous elements h with v(h) ≥ v(t(h)). Note that, since G0 is a finite set, 1 ∈ Tv follows. Since Tv is generated by homogeneous elements, it inherits the G-grading from Q. Proposition 3.9. For any G-valuation v : Q → Γ ∪ {∞} the ring Tv is G-stable and G-total. Proof. Suppose h is a homogeneous element of Q and suppose v(h) < v(t(h)). Then v(h−1) = v(t(h))v(h−1) > v(h)v(h−1) = v(s(h)) = v(t(h−1)) showing that h−1 ∈ Tv. To show G-stability, pick some homogeneous element h ∈ Qg and suppose that r ∈ Tt(g), then v(hrh−1) ≥ v(h)v(1t(h))v(h−1) = v(1s(h)). Since s(h) is the target of hrh−1, this shows that hrh−1 ∈ T ∩ Qs(h) = Ts(h). Similarly, if r′ is in Ts(g), it follows that h−1r′h is in Tt(h), so r′ ∈ hTt(h)h−1 which establishes the G-stability of Tv. (cid:3) Obviously, one can define another ring, Sv say, as the ring generated by homo- geneous elements with v(h) ≥ v(s(h)). Mutatis mutandis, proposition 3.9 can be proven for Sv instead of Tv. Example 3.10. The groupoid ∆2 can be ordered by letting e11 be the maximum and e22 the minimum of ∆2, the elements e12 and e21 remaining incomparable. Using this, we can define an ordering on the groupoid G of non-zero elements of Z[∆2] by putting Xδ∈∆2 aδδ ≥ Xδ∈∆2 bδδ if, for all δ ∈ ∆2, aδ ≥ bδ or ∃δ′ > δ : aδ′ ≥ bδ′. If v is a discrete valuation on a field k, then v : M2(k) → G ∪ {∞} : Xδ∈∆2 mδ 7→ X∄δ′<δ mδ′ 6=0 v(mδ)eδ is a groupoid valuation. In this case, we have Tv =(cid:18)Rv 0 Rv(cid:19) while Sv =(cid:18)Rv k Rv(cid:19) . k 0 We will now show that, although Tv and Sv might be different, there exists for any G-valuation ring R some G-valuation v with Tv = R = Sv. First, we briefly remind the reader how quotients of groupoids can be defined. Suppose F is subgroupoid of G containing all idempotents and such that gFt(g)g−1 = Fs(g) for all g, then one can construct a factor groupoid G/F = G/ ∼ where g ∼ h ⇔ ∃fs, ft ∈ F : g = fshft. TOWARDS A CHEVALLEY-STYLE NON-COMMUTATIVE ALGEBRAIC GEOMETRY 9 It can easily be verified that this is an equivalence relation which is compatible with the multiplication on G, so there is a canonical induced multiplication on G/F . Proposition 3.11. For any G-stable, G-total subring R there is some ordered groupoid Γ and some partial G-valuation v : Q → Γ ∪ {∞} with Tv = R = Sv. Proof. H(Q)∗ is a groupoid for the multiplication and H(R)∗ is a subgroupoid containing all 1e for e ∈ G0, which are exactly the idempotents of H(Q)∗. Moreover, because of the G-stability of R, we have hH(R)∗ s(h) for all h ∈ H(Q)∗. Denote the quotient groupoid H(Q)∗/H(R)∗ by Ω. We can define an ordering on Ω by t(h)h−1 = H(R)∗ x ≥ y ⇔ ∃rs, rt ∈ H(R) : x = rsyrt. It is a standard verification that this is a well-defined partial order relation relation on Ω. Pick ω = q in Ω for some q ∈ H(Q)∗. If q ∈ H(R), then ω = q1t(q) and If q /∈ H(R), then q−1 ∈ H(R) 1t(q) = t(ω) so ω is comparable to its target. because of the G-totality of R. Therefore, t(ω) = q−1q so ω is again comparable to its target. Of course, a similar argument holds for sources instead of targets. Suppose now χ, ψ, ω ∈ Ω such that ψ ≤ ω and both χψ and χω are defined. We have x, y, z ∈ H(Q)∗ and rs, rt ∈ H(R) \ {0} with x = χ, y = ψ, z = ω and rsyrt = z. Clearly, ψ and ω have the same source, so we can assume y and z to have equal source as well. Therefore t(rs) = s(rs) holds and, by the G-stability of R, we have yRt(y)y−1 = Rs(y) so there is some r′ trt hence t ∈ Rt(y) with rsy = yr′ t. Thus xz = xyr′ The other compatibilities can be checked in a similar fashion. This means Ω is ordered. Note that if h and h′ are in the same Qg they must be comparable and χψ = xy ≤ xz = χω. h + h′ ≥ min(cid:8)h, h′(cid:9). Let g ∼ g′ if for any hg′ ∈ Q∗ g. This is an equivalence relation compatible with multiplication, so G = G/ ∼ is a groupoid which can be ordered by putting for all g, g′ ∈ G, g < g′ if and only if h < h′ for all non-zero h ∈ Rg, h′ ∈ Rg′ . Set Γ the groupoid of non-zero elements of Ω[G] ordered by g with hg ≤ hg′ ≤ h′ g′ there are hg, h′ g ∈ Q∗ if, for all g ∈ G, ag ≥ bg or ∃g′ > g : ag′ ≥ bg′ . Xg∈G Define bgg agg ≥Xg∈g v : Q → Γ ∪ {∞} : Xg∈G mg 7→ X∄g′<g m g′ 6=0 min{mg g ∈ g} g. This will be our G-valuation. It is clear that v(cid:16)Pg∈G mg(cid:17) = ∞ can only happen if all mg are zero, and it is just as clear that v(hh′) = v(h)v(h′) if h and h′ are homogeneous. Suppose x, y and z are such that v(x) ≥ v(z) ≤ v(y), then v(x + y) ≥ v(z) since this property holds in Ω. We certainly have R ⊆ Tv and we know that 1e ∈ R for any idempotent e ∈ G0. Suppose now that v(h) ≥ v(t(h)) for some homogeneous h. If h is not in R, then h−1 is in R, whence v(h−1) ≥ v(t(h−1)) = v(s(h)), leading to v(t(h)) = v(h−1)v(h) ≥ v(s(h))v(h) = v(h), i.e. v(h) = v(t(h)). This means that h = 1t(h), so there are rs, rt in H(R) with h = rs1t(h)rt hence h ∈ R. (cid:3) In view of this theorem, it is harmless to restrict attention to canonical G- valuation, i.e. G-valuations which satisfy (4) v(x) ≥ v(t(x)) ⇔ v(x) ≥ v(s(x)). 10 NIKOLAAS D. VERHULST in addition to the previously mentioned conditions. From now on, we will assume for the sake of simplicity that all G-valuations are canonical. Corollary 3.12. In the same context as 3.11, we have (cid:8)h ∈ H(R) h−1 /∈ R(cid:9) = {h ∈ H(R) v(h) > v(t(h))} . Proof. Take a homogeneous h with v(h) > v(t(h)), then v(h−1) = v(t(h))v(h−1) < v(h)v(h−1) = v(s(h)) = v(t(h−1)) so h−1 /∈ R. On the other hand, if h ∈ H(R) and h−1 /∈ R, then v(h−1) < v(t(h−1)) = v(s(h)) so v(t(h)) = v(h−1)v(h) < v(s(h))v(h) = v(h). (cid:3) This set will be denoted by the P of positive. As in the classical case, a G-valuation ring is completely determined by its set of positives. Indeed, R is the ring generated by {h ∈ H(Q) hP ⊆ P}. If R is a G-valuation ring in a G-skewfield Q, then R/P inherits a canonical G-grading and, in view of the preceding corollary, it will again be a G-skewfield, Q say. The map π : Q ∪ {∞} → Q{∞} : x 7→(x if x ∈ R ∞ otherwise can then be reasonably be called a G-place of Q in Q. In the classical case, there is a one-one correspondence between valuation ring and places (cfr. e.g. [3]). No doubt, this could be generalised to the G-graded context as well. For the sake of conciseness we will not go into this here. isomorphism if Let Γ and ∆ be ordered groupoids. A bijection f : Γ → ∆ is an order-preserving (1) ∀γ, γ′, γ′′ ∈ Γ : γγ′ = γ′′ ⇔ f (γ)f (γ′) = f (γ′′), (2) ∀γ, γ′ ∈ Γ∀δ, δ′ ∈ ∆ : γ ≤ γ′ ⇔ f (γ) ≤ f (γ′). Two G-valuations v : Q → Γ and w : Q → ∆ are called equivalent if there ex- ists an order-preserving isomorphism f : Γ → ∆ with v(h) = f (w(h)) for every homogeneous h. Proposition 3.13. Two G-valuations v : Q → Γ and w : Q → ∆ are equivalent if and only if Tv = Tw. Proof. Suppose f is an order-preserving isomorphism with v(h) = f (w(h)). Then h ∈ Tv if v(h) ≥ v(t(h)) i.e. f (w(h)) ≥ f (w(t(h))) which happens precisely if w(h) ≥ w(t(h)), or in other words, when h ∈ Tw. If, on the other hand, Tv = Tw, then we can define f : Γ → ∆ : γ 7→ w(v−1(γ)). We must first check that this is indeed a function, so suppose γ = v(h) = v(h′), then there are some rs, rt ∈ H(Tv)∗ with h = rsh′rt. These rs, rt must necessarily be in H(Tw)∗ whence w(h) = w(rs)w(h′)w(rt) = w(h′), so f is well-defined. Essentially the same argument proves injectivity and surjectivity we get for free because w is surjective. Moreover, we have f (γγ′) = w(v−1(γγ′)) = w(v−1(γ)v−1(γ′)) since v is a G-valuation. The right hand side of the last equality is in turn equal to w(v−1(γ))w(v−1(γ′)) which is f (γ)f (γ′) as had to be proven. Finally, if v(x) = γ ≤ γ′ = v(x′), then there are rs, rt ∈ H(Tv) with rsxrt = x′. Since Tv = Tw, we have w(x′) = w(rsxrt) ≥ w(x). Consequently, f is an order-preserving isomorphism hence v and w are equivalent. (cid:3) Proposition 3.14. If supp(Q) is connected and equals supp(R) then the G- valuation from 3.11 takes values in a group . TOWARDS A CHEVALLEY-STYLE NON-COMMUTATIVE ALGEBRAIC GEOMETRY 11 Proof. Suppose supp(Q) is connected and equal to supp(R). Take e, e′ in G0, then there are r, r′ ∈ R with 1e = r1e′ r′, so 1e = 1e′ . Consequently, Ω has but one it is a group. An ordered groupoid which is a group is a totally idempotent, i.e. ordered group, so G is the trivial group whence Γ is a group. (cid:3) Example 3.15. Let us first consider an example of the simplest kind: the ∆2- valuation ring Rv Rv(cid:19) contained in the ∆2-skewfield (cid:18)k k (cid:18)Rv Rv k k(cid:19) for a valuation ring Rv with maximal ideal P in a field k. In this case, G is the trivial group and Ω ≃ Γ ≃ Rv/P . The associated value function is v : Q → Γ ∪ {∞} :(cid:18)a b d(cid:19) 7→ min{v(a), v(b), v(c), v(d)} . c This and similar value functions have been studied in [11]. Whether a deeper connection exists between groupoid valuations and the value functions considered there would be an interesting topic for future research. Matters get a bit more complicated if we consider the situation from example 3.3. Here, ∆2 = ∆2 and we find that Ω ≃ E [∆2] where E is the value group of the valuation v. For example, 11,1 and 12,2 are incomparable in Ω, while 11,2 is larger and 12,1 is smaller than both. Let a and b be in k with v(a) > v(b). Then we have, if we denote by some abuse of notation the valuation on the G-skewfield with v as well, v(cid:18)(cid:18)b a b(cid:19)(cid:19) ≥ v(cid:18)(cid:18)a b a a(cid:19)(cid:19) while e.g. v(cid:18)(cid:18)a b b b b(cid:19)(cid:19) and v(cid:18)(cid:18)b b a(cid:19)(cid:19) b are incomparable. Consider a simple artinian Q which is finite dimensional over its centre Z(Q) and a complete discrete valuation ring R on Z(Q). By the Artin-Wedderburn theorem, we have Q ≃ Mn(D) for some skewfield D which is finite dimensional over Z(Q). It is known (cfr. [10]) that R is contained in a unique maximal order O of D and that any maximal order in Mn(D) is of the form qMn(O)q−1 for some invertible q. Lemma 3.16. If R is a G-valuation ring on a G-skewfield Q and if q ∈ U (Q), then qRq−1 is a G-valuation ring as well. Proof. Suppose h ∈ H(Q) \ qRq−1. Then q−1hq /∈ R so its G-inverse, which is q−1h−1q, must be in R. Consequently, h−1 ∈ qRq−1. On the other hand, hqRt(h)q−1h−1 = hqh−1Rs(h)hq−1h−1. This is furthermore equal to Rs(h) since hqh−1 has s(hqh−1) = t(hqh−1) = s(h). (cid:3) Remark 3.17. It might be worth pointing out that this qRq−1 is a G-valuation ring for a different G-grading. Indeed, the homogeneous elements will be of the form qhq−1 now, where h is homogeneous with respect to the original G-grading.. We find that, in this case at least, any maximal order is a G-valuation ring for a suitable G-grading. Note that by 3.14 the associated value function takes values in a group. It seems doubtful that completeness and discreteness are really necessary, which inspires the following question: Question 3.18. Is every maximal order a G-valuation ring? 12 NIKOLAAS D. VERHULST 4. G-valuations and Dubrovin valuation rings One of the most important concepts in non-commutative valuation theory is the Dubrovin valuation ring. These rings were introduced by Dubrovin in the eighties and have been studied quite extensively, in no small part due to their excellent extension properties. For the general theory of Dubrovin valuation rings we refer the interested reader to, for example, [7] or [8]. In this section we establish a tentative connection between Dubrovin valuation rings and G-valuation rings, but there is still work to be done here. Definition 4.1. Recall that a subring R of a simple artinian ring Q is called a Dubrovin valuation ring if (1) R/J(R) is simple artinian (2) for every q /∈ R there are r, r′ in R such that both rq and qr′ are in R\J(R). where J(R) denotes the Jacobson radical of R. It is clear from e.g. example 3.3 that not every G-valuation ring is a Dubrovin valuation ring. The reason is that the ring under consideration there does not have full support, as the following proposition shows. Proposition 4.2. If Q ≃ Mn(D) for a skewfield D and R is a ∆n-valuation ring containing all 1kδij , then R is a Dubrovin valuation ring. Proof. By 3.4, the ideal M generated by homogeneous elements of R which are not in R∗ is the unique maximal ideal which does not contain 1ii for any i. Con- sider some ideal I containing some 1ii. Because R contains all 1ij, it follows that 1jj ∈ I for any j. Therefore I must be R, so M is maximal. We have R/M ≃Lδ∈∆n Let v be the G-valuation as constructed in 3.11. Note that, by 3.14, v takes values in a group. If a = P aδδ is not in R, then there is some δ with v(aδδ) < 0 Rδ/Mδ, so R/M is simple Artinian. minimal. We find v(a(aδδ−1)) = min γ∈∆n v(aγγ)v((aδδ)−1) = v(aδδ)v((aδδ)−1) = v(1l(δ)) which implies that a(aδδ)−1 is in R \ M . In a similar fashion we find an r with ra ∈ R \ M , so R is a Dubrovin valuation ring. (cid:3) This suggests the following question: is every Dubrovin valuation ring a G-valuation ring? It is known that the property of being a Dubrovin valuation ring is invariant under Morita equivalence, so it would suffice to answer this question for Dubrovin valuation rings in skewfields, i.e. Question 4.3. Is every Dubrovin valuation ring in a skewfield a G-valuation ring (for some suitable grading)? In [4], it was shown that this is certainly true in sufficiently nice cases, e.g. if the skewfield is a crossed product and the Dubrovin valuation ring lies above an unramified valuation. Moreover, it is known that the set of divisorial ideals of a Dubrovin valuation ring R forms a groupoid (cfr. [8]). This groupoid probably induces a more or less canonical grading on the Ore localisation Q of R with respect to which Q will be a G-skewfield. It is to be expected that the Dubrovin valuation ring will then be a G-valuation ring in this G-skewfield. TOWARDS A CHEVALLEY-STYLE NON-COMMUTATIVE ALGEBRAIC GEOMETRY 13 Acknowledgement Some of this research was carried out during the author's doctoral studies at the University of Antwerp under supervision of Freddy Van Oystaeyen. The author wants to thank Arno Fehm for his helpful suggestions concerning the presentation of the material. References [1] D. D. Anderson, D. F. Anderson, C. W. Chang, Graded-valuation domains, Comm. Algebra Vol.45 No.9, 2017, 4018-4029 [2] C. Chevalley, Introduction to the theory of algebraic functions of one variable, Math. Surveys Monogr. Vol.6, 1951 [3] O. Endler, Valuation theory, Springerm 1972 [4] D. Haile & P. Morandi, On Dubrovin valuation rings in crossed product algebras, Trans. Amer. Math. Soc., Vol.338 No.2, 1993, 723-751 [5] J. L. Johnson, The graded ring R [X1, ..., Xn], Rocky Mountain J. Math. Vol.9 No.3, 1979, 415-424 [6] P. Lundstrom, Separable groupoid rings, Comm. Algebra, Vol.34 No.8, 2006, 3029-3041 [7] H. Marubayashi, H. Miyamoto & A. Ueda, Noncommutative valuation rings and semi- hereditary orders, K-monographs in math. Vol.3, Kluwer Acad. Publ., 1997 [8] H. Marubayashi & F. Van Oystaeyen, General theory of primes, Springer LNM Vol.2059, 2012 [9] C. Nastasescu & F. Van Oystaeyen, Methods of graded rings, Springer LNM Vol.1836, 2004 [10] I. Reiner, Maximal orders, London Math. Soc. Monogr. (N.S.) Vol. 28, Oxford University Press, 2003 [11] J.-P. Tignol & A. Wadsworth, Value Functions on Simple Algebras, and Associated Graded Rings, Springer Monogr. Math., 2015 [12] F. Van Oystaeyen & N. Verhulst, Arithmetical pseudo-valuations associated to Dubrovin valuation rings and prime divisors of bounded Krull orders, Bull. Belg. Math. Soc. Simon Stevin, Vol.23 No.1, 2016, 115-131 [13] A. Wadsworth, Extending valuations to finite-dimensional division algebras, Proc. Amer. Math. Soc., Vol.98 No.1, 1986, 20-22 TU Dresden, Fachrichtung Mathematik, Insitut fur algebra, 01062 Dresden, Ger- many E-mail address: nikolaas [email protected]
1904.02728	1	1904	2019-04-04T18:07:52	A Universal Algebraic Survey of $\mathcal{C}^{\infty}-$Rings	[ "math.RA" ]	In this paper we present some basic results of the Universal Algebra of $\mathcal{C}^\infty$-rings which were nowhere to be found in the current literature. The outstanding book of I. Moerdijk and G. Reyes,[24], presents the basic (and advanced) facts about $\mathcal{C}^\infty$-rings, however such a presentation has no universal algebraic "flavour". We have been inspired to describe $\mathcal{C}^\infty$-rings through this viewpoint by D. Joyce in [15]. Our main goal here is to provide a comprehensive material with detailed proofs of many known "taken for granted" results and constructions used in the literature about $\mathcal{C}^\infty$-rings and their applications - such proofs either could not be found or were merely sketched. We present, in detail, the main constructions one can perform within this category, such as limits, products, homomorphic images, quotients, directed colimits, free objects and others, providing a "propaedeutic exposition" for the reader's benefit.	math.RA	math	A Universal Algebraic Survey of C∞−Rings Jean Cerqueira Berni1 and Hugo Luiz Mariano2 1 Institute of Mathematics and Statistics, University of Sao Paulo, Rua do Matao, 2 Institute of Mathematics and Statistics, University of Sao Paulo, Rua do Matao, 1010, Sao Paulo - SP, Brazil. [email protected] 1010, Sao Paulo - SP, Brazil, [email protected] Abstract. In this paper we present some basic results of the Universal Algebra of C∞−rings which were nowhere to be found in the current lit- erature. The outstanding book of I. Moerdijk and G. Reyes, [24], presents the basic (and advanced) facts about C∞−rings, however such a presen- tation has no universal algebraic "flavour". We have been inspired to describe C∞−rings through this viewpoint by D. Joyce in [15]. Our main goal here is to provide a comprehensive material with detailed proofs of many known "taken for granted" results and constructions used in the literature about C∞−rings and their applications - such proofs either could not be found or were merely sketched. We present, in detail, the main constructions one can perform within this category, such as limits, products, homomorphic images, quotients, directed colimits, free objects and others, providing a "propaedeutic exposition" for the reader's ben- efit. Introduction As observed by E. Dubuc in [13], a C∞−ring is a model of the algebraic theory which has as n−ary operations all the smooth functions from Rn into R, and whose axioms are all the equations that hold between these functions. Since every polynomial is smooth, all C∞−rings are, in particular, R−algebras. This point of view allows us to regard this theory as an extension of the concept of R−algebra. The theory of C∞−rings has been originally studied in view of its applica- tions to Singularity Theory and in order to construct topos-models for Synthetic Differential Geometry (the Dubuc Topos, for instance), which "grew out of ideas of Lawvere in the 1960s" (cf. [15]). Recently, however, this theory has been ex- plored by some eminent mathematicians like David I. Spivak and Dominic Joyce in order to extend Jacob Lurie's program of Derived Algebraic Geometry to De- rived Differential Geometry (cf. [15] and [16] ). Just as any Lawvere theory, C∞−rings can be interpreted within any topos. In this specific case, a C∞−ring in a topos E is a finite product preserving func- tor from the category whose objects are the Cartesian products of R and whose 2 J. C. Berni and H. L. Mariano morphisms are the smooth functions between them, into E (see, for example, [21]). In this work, however, we focus on set-theoretic C∞−rings (i.e., C∞−rings in the topos of sets, Set). This is the first paper of a series that presents a detailed account of some algebraic aspects of C∞−rings. Some categorial and logical aspects were given in [5]. In the next papers we are going to present some topics on Smooth Commuta- tive Algebra and von Neumann regular C∞−rings. Here we present and analyze a C∞−ring as a universal algebra whose functional symbols are the symbols for all smooth functions from Cartesian powers of R to R. Such an approach em- phasizes the power of a C∞−ring in interpreting a broader language than the algebraic one, which is expressed in terms of the R−algebra structure. It also has the advantage of giving us explicitly many constructions, such as products, coproducts, directed colimits, among others, as well as simpler proofs of the main results. We make a detailed exposition of the description of free C∞−rings in terms of a colimit, and we use it to account the often used description of an arbitrary C∞−rings in terms of generators and relations. Our idea of describing C∞−rings from a universal-algebraic point of view was mainly inpired by the clear and elegant presentation made by Dominic Joyce in [15] - which we found very enlightening. Overview of the paper: We begin by presenting the equational theory of C∞−rings in terms of a first order language with a denumerable set of variables. We define the class of C∞−structures and the (equationally defined) subclass of C∞−rings. In the Section 2 we present a detailed description of the main constructions involving C∞−rings: C∞−subrings (Definition 8), intersections (Proposition 1), the C∞−subring generated by a set (Definition 9), the directed union of C∞−rings (Proposition 2), products (Definition 10), C∞−congruences (Definition 11) and quotients (Definition 12), homomorphic images (Propo- sition 7), directed colimits (Theorem 4) and small projective limits (Theorem 5). We present and prove the "Fundamental Theorem of C∞−Homomorphism" (Theorem 3). We prove also that the category of C∞−rings is a reflective sub- category of the category of all C∞−structures (Theorem 6). We dedicate Section 3 to describe the free C∞−rings, first with a finite set of generators and then with an arbitrary set of generators. We use this construc- tion in order to describe an adjunction between the category of all C∞−rings and C∞−homomorphisms, C∞Rng and the category of sets, Set (Proposition 11). A Universal Algebraic Survey of C∞−Rings 3 In Section 4 we describe other constructions, and in Subsection 4.1 we use Hadamard's lemma (Theorem 7) to prove that the ring-theoretic ideals of any finitely generated C∞−ring classify their congruences (Proposition 13). We extend this result by presenting a proof for the general case (Proposition 17). We prove that any C∞−ring can be expressed as a directed colimit of finitely generated C∞−rings (Theorem 8) and in Subsection 4.2 we present an ex- plicit description for the C∞−coproduct of C∞−rings (Definition 13). We end up this work presenting an ubiquitous construction in Algebra in the Subsec- tion 4.3, namely the C∞−ring of C∞−polynomials, constructed in terms of the C∞−product. Such a construction will play an important role in the future pa- pers on Smooth Commutative Algebra and von Neumann Regular C∞−rings. 1 Preliminaries: The equational theory of C∞-rings The theory of C∞−rings can be described within a first order language L with a denumerable set of variables (Var(L) = {x1, x2, · · · , xn, · · · }) whose nonlogical symbols are the symbols of C∞−functions from Rm to Rn, with m, n ∈ N, i.e., the non-logical symbols consist only of function symbols, described as follows: For each n ∈ N, the n−ary function symbols of the set C∞(Rn, R), i.e., F(n) = {f (n)f ∈ C∞(Rn, R)}. So the set of function symbols of our language is given by: C∞(Rn) F = [n∈N F(n) = [n∈N Note that our set of constants is R, since it can be identified with the set of all 0−ary function symbols, i.e., Const(L) = F(0) = C∞(R0) ∼= C∞({∗}) ∼= R. The terms of this language are defined, in the usual way, as the smallest set which comprises the individual variables, constant symbols and n−ary function symbols followed by n terms (n ∈ N). Definition 1. A C∞−structure on a set A is a pair A = (A, Φ), where: Φ :Sn∈N C∞(Rn, R) → Sn∈N Func (An; A) 7→ Φ(f ) := (f A : An → A) (f : Rn C∞ → R) , that is, Φ interprets the symbols3 of all smooth real functions of n variables as n−ary function symbols on A. We call a C∞−struture A = (A, Φ) a C∞−ring if it preserves projections and all equations between smooth functions. We have the following: 3 here considered simply as syntactic symbols rather than functions. 4 J. C. Berni and H. L. Mariano Definition 2. Let A = (A, Φ) be a C∞−structure. We say that A (or, when there is no danger of confusion, A) is a C∞−ring if the following is true: • Given any n, k ∈ N and any projection pk : Rn → R, we have: A = (∀x1) · · · (∀xn)(pk(x1, · · · , xn) = xk) • For every f, g1, · · · gn ∈ C∞(Rm, R) with m, n ∈ N, and every h ∈ C∞(Rn, R) such that f = h ◦ (g1, · · · , gn), one has: A = (∀x1) · · · (∀xm)(f (x1, · · · , xm) = h(g(x1, · · · , xm), · · · , gn(x1, · · · , xm))) Definition 3. Let (A, Φ) and (B, Ψ ) be two C∞−rings. A function ϕ : A → B is called a morphism of C∞−rings or C∞-homomorphism if for any n ∈ N and any f : Rn C∞ → R the following diagram commutes: An Φ(f ) A ϕ(n) ϕ / Bn Ψ (f ) / B i.e., Ψ (f ) ◦ ϕ(n) = ϕ ◦ Φ(f ). Since L does not contain any relational symbol, the set of the atomic for- mulas, AF, is given simply by the equality between terms, that is AF = {t1 = t2t1, t2 ∈ T} Finally, the well formed formulas, WFF are constructed as one usually does in any first order theory. One possible set of axioms for the theory of the C∞-rings can be given by the following two sets of equations: (E1) For each n ∈ N and for every k ≤ n, denoting the k-th projection by pk : Rn → R, the equations: Eqn,k (1) = {(∀x1) · · · (∀xn)(pk(x1, · · · , xn) = xk)} (E2) for every k, n ∈ N and for every (n+2)−tuple of function symbols, (f, g1, · · · , gn, h) such that f ∈ F(n), g1, · · · , gn, h ∈ F(k) and h = f ◦ (g1, · · · , gn), the equa- tions: Eqn,k (2) = = {(∀x1) · · · (∀xk)(h(x1, · · · , xk) = f (g1(x1, · · · , xk), · · · , gn(x1, · · · , xk)))} / / A Universal Algebraic Survey of C∞−Rings 5 The class of C∞−ring is, thus, equational, and as such it a Now we present an alternative description of a C∞−rings making use of uni- versal algebraic concepts,inspired by [15]. We first define the notion of a C∞−structure: Definition 4. A C∞−structure on a set A is a pair (A, Φ), where: Φ :Sn∈N C∞(Rn, R) → Sn∈N Func (An; A) 7→ Φ(f ) := (f A : An → A) (f : Rn C∞ → R) , that is, Φ is a function that interprets the symbols4 of all smooth real functions of n variables as n−ary function symbols on A. Definition 5. Let (A, Φ) and (B, Ψ ) be two C∞−structures. A function ϕ : A → B is called a morphism of C∞−structures (or, simply, a C∞−morphism) if for any n ∈ N and any f ∈ C∞ (Rn, R) the following diagram commutes: An Φ(f ) A ϕ(n) ϕ / Bn Ψ (f ) / B i.e., Ψ (f ) ◦ ϕ(n) = ϕ ◦ Φ(f ). Theorem 1. Let (A, Φ), (B, Ψ ), (C, Ω) be any C∞−structures, and let ϕ : (A, Φ) → (B, Ψ ) and ψ : (B, Ψ ) → (C, Ω) be two morphisms of C∞−structures. We have: (1) idA : (A, Φ) → (A, Φ) is a morphism of C∞−structures; (2) ψ ◦ ϕ : (A, Φ) → (C, Ω) is a morphism of C∞−structures. Proof. Ad (1): Given any n ∈ N and any f ∈ C∞ (Rn, R), since idA we have: (n) = idAn , id(n) A idA An Φ(f ) A / An Φ(f ) / A so Φ(f )◦idA (n) = Φ(f )◦idAn = Φ(f ) = idA ◦Φ(f ), thus idA : (A, Φ) → (A, Φ) is a morphism of C∞−structures. Ad (2): Suppose that ϕ : (A, Φ) → (B, Ψ ) and ψ : (B, Ψ ) → (C, Ω) are two morphisms of C∞−structures, so given any m ∈ N and any f ∈ C∞(Rm, R) the following diagrams commute: 4 here considered simply as syntactic symbols rather than functions. / / / / 6 J. C. Berni and H. L. Mariano Am Φ(f ) A ϕ(m) ϕ / Bm Bm ψ(m) / Cm Ψ (f ) Ψ (f ) Ω(f ) / B B ψ / C Since (ψ ◦ ϕ)(m) = ψ(m) ◦ ϕ(m), the following diagram commutes: Am Φ(f ) A (ψ◦ϕ)(m) (ψ◦ϕ) / Cm Ω(f ) / C so ψ ◦ ϕ : (A, Φ) → (C, Ω) is a morphism of C∞−structures. Theorem 2. Let (A, Φ), (B, Ψ ), (C, Ω), (D, Γ ) be any C∞−structures, and let ϕ : (A, Φ) → (B, Ψ ), ψ : (B, Ψ ) → (C, Ω) and ν : (C, Ω) → (D, Γ ) be morphisms of C∞−structures. We have the following equations between pairs of morphisms of C∞−structures: ν ◦ (ψ ◦ ϕ) = (ν ◦ ψ) ◦ ϕ; ϕ ◦ idA = idB ◦ ϕ. Proof. Since, as functions, we have: and ν ◦ (ψ ◦ ϕ) = (ν ◦ ψ) ◦ ϕ ϕ ◦ idA = idB ◦ ϕ and the composition of morphisms of C∞−structures is again a morphism of C∞−structures, it follows that the equality between morphisms of C∞−structures holds. Definition 6. We are going to denote by C∞Str the category whose objects are the C∞−structures and whose morphisms are the morphisms of C∞−structures. As a full subcategory of C∞Str we have the category of C∞−rings, defined as follows: Definition 7. Let (A, Φ) be a C∞−structure. We say that (A, Φ) is a C∞−ring if the following two conditions are fulfilled: (1) For each n ∈ N and for every k ≤ n, denoting the k-th projection by pk : Rn → R, we have: Φ(pk) = πk : An → A, that is, each projection is interpreted as the canonical projection on the k−th coordinate, πk : An → A / / / / / / A Universal Algebraic Survey of C∞−Rings 7 (2) for every k, n ∈ N and for every (n + 2)−tuple of smooth functions: (f, g1, · · · , gn, h) such that f ∈ C∞(Rn, R), g1, · · · , gn, h ∈ C∞(Rk, R), we have: h = f ◦ (g1, · · · , gn) ⇒ Φ(h) = Φ(f ) ◦ (Φ(g1), · · · , Φ(gn)) Given two C∞−rings (A, Φ) and (B, Ψ ), a morphism of C∞−rings from (A, Φ) to (B, Ψ ), or a C∞−homomorphism from (A, Φ) to (B, Ψ ), is simply a morphism of C∞−struc- tures from (A, Φ) to (B, Ψ ). We will denote the category of C∞−rings and C∞−homomorphisms by C∞Rng. Thus, in the context of first order theories, a model of a C∞−ring is a C∞−structure, A = (A, Φ) such that: • For every n ∈ N, k ≤ n, denoting the projection on the k−th coordinate by pk : Rn → R: A = (∀x1) · · · (∀xn)(pk(x1, · · · , xn) = xk) that is, Φ(pk) = πk : An → A; • For every n, k ∈ N, f ∈ C∞(Rn, R), h, g1, · · · , gn ∈ C∞(Rk, R) such that h = f (g1, · · · , gn): A = (∀x1) · · · (∀xk)(h(x1, · · · , xk) = f (g1(x1, · · · , xk), · · · , gn(x1, · · · , xk)) that is, Φ(h) = Φ(f )(Φ(g1), · · · , Φ(gn)). As we are going to see later on, the category of C∞−rings and its morphisms has many constructions, such as arbitrary products, coproducts, directed colim- its, quotients and many others. It also "extends" the category of commutative unital rings, CRing, in the following sense: Remark 1. Since the sum + : R2 → R, the opposite, − : R → R, · : R2 → R and the constant functions 0 : R → R and 1 : R → R are particular cases of C∞−functions, any C∞−ring (A, Φ) may be regarded as a commutative ring (A, Φ(+), Φ(·), Φ(−), Φ(0), Φ(1)), where: Φ(+) : A × A → A (a1, a2) 7→ Φ(+)(a1, a2) = a1 + a2 Φ(−) : A → A a 7→ Φ(−)(a) = −a 8 J. C. Berni and H. L. Mariano Φ(0) : A0 → A ∗ 7→ Φ(0) Φ(1) : A0 → A ∗ 7→ Φ(1) where A0 = {∗}, and: Φ(·) : A × A → A (a1, a2) 7→ Φ(·)(a1, a2) = a1 · a2 Thus, we have a forgetful functor: eU : C∞Rng → (A Φ) CRing 7→ (A, Φ(+), Φ(·), Φ(−), Φ(0), Φ(1)) ϕ ϕ↾ (B, Ψ ) (B, Ψ (+), Ψ (·), Ψ (−), Ψ (0), Ψ (1)) eU : C∞Rng → CRing Analogously, we can define a forgetful functor from the category of C∞−rings and C∞−homomorphisms into the category of commutative R−algebras with unity, U : C∞Rng → R − Alg These functors will be analyzed with detail in the next paper, [3]. 2 The main constructions in the category of C∞−rings Since the theory of C∞−rings is equational, the class C∞Rng is closed in C∞Str under many algebraic constructions, such as substructures, products, quotients, directed colimits and others. In this section we give explicit descrip- tions for some of these constructions. 2.1 C∞−Subrings We begin defining what we mean by a C∞−subring. Definition 8. Let (A, Φ) be a C∞−ring and let B ⊆ A. Under these circum- stances, we say that (B, Φ′) is a C∞−subring of (A, Φ) if, and only if, for any n ∈ N, f ∈ C∞(Rn, R) and any (b1, · · · , bn) ∈ Bn we have: Φ(f )(b1, · · · , bn) ∈ B A Universal Algebraic Survey of C∞−Rings 9 That is to say that B is closed under any C∞−function n-ary symbol. Note that the C∞−structure of B is virtually the same as the C∞−structure of (A, Φ), since they interpret every smooth function in the same way. However Φ′ has a different codomain, as: Φ′ :Sn∈N C∞ (Rn, R) → Sn∈N Func (Bn, B) 7→ (Φ(f ) ↾Bn: Bn → B) (Rn f → R) We observe that Φ′ is the unique C∞−structure such that the inclusion map: is a C∞−homomorphism. ιA B : B ֒→ A We are going to denote the class of all C∞−subrings of a given C∞−ring (A, Φ) by Sub (A, Φ). Next we prove that the intersection of any family of C∞−subrings of a given C∞−ring is, again, a C∞−subring. Proposition 1. Let {(Aα, Φα)α ∈ Λ} be a family of C∞−subrings of (A, Φ), so (∀α ∈ Λ)(∀n ∈ N)(∀f ∈ C∞(Rn, R))(Φα(f ) = Φ(f ) ↾Aα n : Aα n → Aα) We have that: where: \α∈Λ Aα, Φ′! Φ′ :Sn∈N C∞(Rn, R) → Sn∈N Func(cid:0)(cid:0)Tα∈Λ Aα(cid:1)n 7→ Φ(f ) ↾(Tα∈Λ Aα)n :(cid:0)Tα∈Λ Aα(cid:1)n ,Tα∈Λ Aα(cid:1) →Tα∈Λ Aα (Rn f → R) is a C∞−subring of (A, Φ). Proof. Let n ∈ N, f ∈ C∞(Rn, R) and (a1, · · · , an) ∈(cid:0)Tα∈Λ Aα(cid:1)n Since for every i ∈ {1, · · · , n} we have ai ∈ Tα∈Λ Aα, we have, for every α ∈ Λ, ai ∈ Aα for every i ∈ {1, · · · , n}. . Since every (Aα, Φα) is a C∞−subring of (A, Φ), it follows that for every α ∈ Λ we have: thus Φα(f )(a1, · · · , an) := Φ(f ) ↾An α (a1, · · · , an) ∈ Aα 10 J. C. Berni and H. L. Mariano Φ′(f )(a1, · · · , an) = Φα(f ) ↾(Tα∈Λ Aα)n (a1, · · · , an) ∈ \α∈Λ so the result follows. Aα As an application of the previous result, we can define the C∞−subring gen- erated by a subset of the carrier of a C∞−ring: Definition 9. Let (A, Φ) be a C∞−ring and X ⊆ A. The C∞−subring of (A, Φ) generated by X is given by: hXi = (Ai, Φi), \X⊆Ai (Ai,Φi)(cid:22)(A,Φ) where (Ai, Φi) (cid:22) (A, Φ) means that (Ai, Φi) is a C∞−subring of (A, Φ) together with the C∞−structure given in Proposition 1. We note that, given any C∞−ring (A, Φ), the map of partially ordered sets given by: σ : (℘(A), ⊆) → (Sub (A), ⊆) X 7→ hXi satisfies the axioms of a closure operation. 2.2 Directed union of C∞−rings In general, given an arbitrary family of C∞−subrings, (Aα, Φα)α∈Λ of a given C∞−ring (A, Φ), its union,Sα∈Λ Aα, together with Φ ↾∪α∈ΛAα, needs not to be a C∞−subring of (A, Φ). However, there is an important case in which the union of a family of C∞−subrings of a C∞−ring (A, Φ) is again a C∞−ring. This case is discussed in the following: Proposition 2. Let (A, Φ) be a C∞−ring and let {(Aα, Φα)α ∈ Λ}, Λ 6= ∅, be a directed family of C∞−subrings of (A, Φ), that is, a family such that for every pair (α, β) ∈ Λ × Λ there is some γ ∈ Λ such that: and We have that: Aα ⊆ Aγ Aβ ⊆ Aγ [α∈Λ Aα, Φ′! A Universal Algebraic Survey of C∞−Rings 11 where: Φ′ :Sn∈N C∞(Rn, R) → Sn∈N Func(cid:0)(cid:0)Sα∈Λ Aα(cid:1)n 7→ Φ(f ) ↾(Sα∈Λ Aα)n :(cid:0)Sα∈Λ Aα(cid:1)n ,Sα∈Λ Aα(cid:1) →Sα∈Λ Aα (Rn f → R) is a C∞−subring of (A, Φ). Proof. First we note that since Λ is directed, we have: (Aα)n = [α∈Λ [α∈Λ Aα!n Let n ∈ N, f ∈ C∞(Rn, R) and (a1, · · · , an) ∈(cid:0)Sα∈Λ Aα(cid:1)n the unionSα∈Λ Aα, for every i ∈ {1, · · · , n} there is some αi ∈ Λ such that: . By definition of ai ∈ Aαi . Since {(Aα, Φα)α ∈ Λ} is directed, given (Aα1 , Φα1 ), · · · , (Aαn , Φαn ), there is some γ ∈ Λ such that for every i ∈ {1, · · · , n} we have: Aαi ⊆ Aγ Thus, (a1, · · · , an) ∈ An γ , and since (Aγ, Φγ) is a C∞−subring of (A, Φ), it follows that: Φγ(f )(a1, · · · , an) := Φ(f ) ↾Aγ n (a1, · · · , an) ∈ Aγ ⊆ [α∈Λ Aα Since Λ is directed and (Aγ)n ⊆(cid:0)Sα∈λ Aα(cid:1)n , we have: Φ(f ) ↾(Sα∈Λ Aα)n (a1, · · · , an) = Φ(f ) ↾(Aγ )n (a1, · · · , an) so: Φ′(f )(a1, · · · , an) := Φ(f ) ↾(Sα∈Λ Aα)n (a1, · · · , an) = = Φ(f ) ↾(Aγ )n (a1, · · · , an) ∈ Aγ ⊆ [α∈Λ Aα Thus, for every (a1, · · · , an) ∈ (cid:0)Sα∈Λ Aα(cid:1)n Sα∈Λ Aα. This proves that(cid:0)Sα∈Λ Aα, Φ′(cid:1) is a C∞−subring of (A, Φ). , we have Φ′(f )(a1, · · · , an) ∈ 12 J. C. Berni and H. L. Mariano 2.3 Products, C∞−Congruences and Quotients Next we describe the products in the category C∞Rng, that is, products of ar- bitrary families of C∞−rings. Definition 10. Let {(Aα, Φα)α ∈ Λ} be a family of C∞−rings. The product of this family is the pair: Yα∈Λ Aα, Φ(Λ)! where Φ is given by: Φ(Λ) :Sn∈N C∞ (Rn, R) → (f : Rn C∞ → R) 7→ Φ(Λ)(f ) : ((x1 Sn∈N Func(cid:0)(cid:0)Qα∈Λ Aα(cid:1)n (cid:0)Qα∈Λ Aα(cid:1)n α)α∈Λ, · · · , (xn → ,Qα∈Λ Aα(cid:1) Qα∈Λ Aα α, · · · , xn α)α∈Λ) 7→ (Φα(f )(x1 α))α∈Λ Remark 2. In particular, given a C∞−ring (A, Φ), we have the product C∞−ring: (A × A, Φ(2)) where: and: Φ(2) :Sn∈N C∞(Rn, R) → Sn∈N Func ((A × A)n, A × A) 7→ (Φ × Φ)(f ) : (A × A)n → A × A (f : Rn C∞ → R) Φ(2)(f ) : (A × A)n → A × A ((x1, y1), · · · , (xn, yn)) 7→ (Φ(f )(x1, · · · , xn), Φ(f )(y1, · · · , yn)) We turn now to the definition of congruence relations in C∞−rings. As we shall see later on, the congruences of a C∞−rings will be classified by their ring- theoretic ideals. Definition 11. Let (A, Φ) be a C∞−ring. A C∞−congruence is an equivalence relation R ⊆ A × A such that for every n ∈ N and f ∈ C∞(Rn, R) we have: (x1, y1), · · · , (xn, yn) ∈ R ⇒ Φ(2)(f )((x1, y1), · · · , (xn, yn)) ∈ R In other words,a C∞−congruence is an equivalence relations that preserves C∞−function symbols. A characterization of a C∞−congruence can be given using the product C∞−structure, as we see in the following: A Universal Algebraic Survey of C∞−Rings 13 Proposition 3. Let (A, Φ) be a C∞−ring and let R ⊆ A × A be an equivalence relation. Under these circumstances, R is a C∞−congruence on (A, Φ) if, and only if, (R × R, Φ(2)′ ), where: Φ(2)′ :Sn∈N C∞ (Rn, R) → Sn∈N Func (Rn, R) 7→ Φ(2)(f ) ↾Rn : Rn → R (Rn f → R) is a C∞−subring of (A×A, Φ(2)), with the structure described in the Remark 2. Proof. Given any n ∈ N, f ∈ C∞(Rn, R), ((x1, y1), · · · , (xn, yn)) ∈ Rn we have, by definition: Φ(2)(f )((x1, y1) · · · , (xn, yn)) := (Φ(f )(x1, · · · , xn), Φ(f )(y1, · · · , yn)) and since R is a C∞−congruence, we have: ((x1, y1), · · · , (xn, yn)) ∈ Rn ⇒ Φ(2)(f )((x1, y1), · · · , (xn, yn)) ∈ R Since: Φ(2)(f )((x1, y1), · · · , (xn, yn)) = (Φ(f )(x1, · · · , xn), Φ(f )(y1, · · · , yn)) it follows that ((x1, y1), · · · , (xn, yn)) ∈ Rn ⇒ (Φ(f )(x1, · · · , xn), Φ(f )(y1, · · · , yn)) ∈ R Also, since Rn ⊆ (A × A)n, Φ(2)(f ) ↾Rn ((x1, y1), · · · , (xn, yn)) = Φ(2)(f )((x1, y1), · · · , (xn, yn)) so (f )((x1, y1), · · · , (xn, yn)) = Φ(2)(f ) ↾Rn ((x1, y1), · · · , (xn, yn)) = Φ(2)′ = Φ(2)(f )((x1, y1), · · · , (xn, yn)) = (Φ(f )(x1, · · · , xn), Φ(f )(y1, · · · , yn)) ∈ R Thus (R, Φ(2)′ ) is a C∞−subring of (A × A, Φ(2)). Conversely, let R be an equivalence relation on A such that (R, Φ(2)′ ) is a C∞−subring of (A × A, Φ(2)). Given any n ∈ N, f ∈ C∞(Rn, R) and (x1, y1), · · · , (xn, yn) ∈ R, since (f )((x1, y1), · · · , (xn, yn)) ∈ (R, Φ(2)′ ) is a C∞−subring of (A×A, Φ(2)), we have Φ(2)′ J. C. Berni and H. L. Mariano 14 R. By definition, Φ(2)′ (f )((x1, y1), · · · , (xn, yn)) = Φ(2)(f ) ↾Rn ((x1, y1), · · · , (xn, yn)) and since Rn ⊆ (A × A)n, we have: Φ(2)(f ) ↾Rn ((x1, y1), · · · , (xn, yn)) = Φ(2)(f )((x1, y1), · · · , (xn, yn)) and Φ(2)(f )((x1, y1), · · · , (xn, yn)) = (Φ(f )(x1, · · · , xn), Φ(f )(y1, · · · , yn)) Thus, (x1, y1), · · · , (xn, yn) ∈ R ⇒ (Φ(f )(x1, · · · , xn), Φ(f )(y1, · · · , yn)) ∈ R and R is a C∞−congruence. Remark 3. Given a C∞−ring (A, Φ) and a C∞−congruence R ⊆ A × A, let: A R = {aa ∈ A} be the quotient set. Given any n ∈ N, f ∈ C∞(Rn, R) and (a1, · · · , an) ∈ (cid:18) A R(cid:19)n we define: Φ :Sn∈N C∞ (Rn, R) →Sn∈N Func(cid:18)(cid:18) A R(cid:19)n R(cid:19)n 7→ (cid:18)Φ(f ) :(cid:18) A (f : Rn C∞ → R) → , A R(cid:19) R(cid:19) A where: R(cid:19)n Φ(f ) : (cid:18) A → A R (a1, · · · , an) 7→ Φ(f )(a1, · · · , an) Note that the interpretation above is indeed a function, that is, its value does not depend on any particular choice of the representing element. This means that given (a1, · · · , an), (a′ 1, · · · , a′ such that (a1, a′ 1), · · · , (an, a′ n) ∈ R, we n) ∈(cid:18) A R(cid:19)n have: Φ(f )(a1, · · · , an) = Φ(f )(a1, · · · , an) A Universal Algebraic Survey of C∞−Rings 15 and since R is a C∞−congruence, (a1, a′ 1), · · · , (an, a′ n) ∈ R ⇒ (Φ(f )(a1, · · · , an), Φ(f )(a′ 1, · · · , a′ n)) ∈ R so: Φ(a1, · · · , an) = Φ(f )(a1, · · · , an) = Φ(f )(a′ 1, · · · , a′ n) = Φ(f )(a′ 1, · · · , a′ n) The above construction leads directly to the following: Definition 12. Let (A, Φ) be a C∞−ring and let R ⊆ A×A be a C∞−congruence. The quotient C∞−ring of A by R is the ordered pair: (cid:18) A R , Φ(cid:19) A R = {aa ∈ A} where: and Φ :Sn∈N C∞ (Rn, R) →Sn∈N Func(cid:18)(cid:18) A R(cid:19)n R(cid:19)n 7→ (cid:18)Φ(f ) :(cid:18) A (f : Rn C∞ → R) → , A R(cid:19) R(cid:19) A where Φ(f ) is described in Remark 3. The following result shows that the canonical quotient map is, again, a C∞−homomorphism. Proposition 4. Let (A, Φ) be a C∞−ring and let R ⊆ A×A be a C∞−congruence. The function: q : (A, Φ) →(cid:18) A R 7→ a a , Φ(cid:19) is a C∞−homomorphism. Proof. Let n ∈ N and f ∈ C∞(Rn, R). We are going to show that the following diagram commutes: An Φ(f ) A qn q (cid:18) A R(cid:19)n R(cid:19) /(cid:18) A Φ(f ) / / / 16 J. C. Berni and H. L. Mariano Given (a1, · · · , an) ∈ An, we have on the one hand: Φ(f )◦qn(a1, · · · , an) = Φ(f )(q(a1), · · · , q(an)) = Φ(f )(a1, · · · , an) := Φ(f )(a1, · · · , an) On the other hand, q ◦ Φ(f )(a1, · · · , an) = Φ(f )(a1, · · · , an) so Φ ◦ qn = q ◦ Φ(f ), and q is a C∞−homomorphism. We remark that the structure given above is the unique C∞−structure such that the quotient map is a C∞−homomorphism. Proposition 5. Let (A, Φ) and (B, Ψ ) be two C∞−rings and let ϕ : (A, Φ) → (B, Ψ ) be a C∞−homomorphism. The set: ker(ϕ) = {(a, a′) ∈ A × Aϕ(a) = ϕ(a′)} is a C∞−congruence on (A, Φ). Proof. It is easy to check that ker(ϕ) is an equivalence relation on A. Let n ∈ N, f ∈ C∞(Rn, R) and (a1, a′ n) ∈ ker(ϕ), that is: 1), · · · , (an, a′ (∀i ∈ {1, · · · , n})(ϕ(ai) = ϕ(a′ i)) We are going to show that (Φ(f )(a1, · · · , an), Φ(f )(a′ 1, · · · , a′ n)) ∈ ker(ϕ). Since ϕ is a C∞−homomorphism, we have the following commutative dia- gram: An Φ(f ) A ϕn ϕ Bn Ψ (f ) / B thus, we have: ϕ(Φ(f )(a1, · · · , an)) = Ψ (f )(ϕ(a1), · · · , ϕ(an)) = 1), · · · , ϕ(a′ = Ψ (f )(ϕ(a′ n)) = ϕ(Φ(f )(a′ 1, · · · , a′ n)) and (Φ(f )(a1, · · · , an), Φ(f )(a′ 1, · · · , a′ n)) ∈ ker(ϕ). Corollary 1. Let (A, Φ) and (B, Ψ ) be two C∞−rings and let ϕ : (A, Φ) → (B, Ψ ) be a C∞−homomorphism. Then (ker(ϕ), Φ(2)′ ) is a C∞−subring of (A × A, Φ(2)). / / / A Universal Algebraic Survey of C∞−Rings 17 Proposition 6. For every C∞−congruence R ⊆ A×A in (A, Φ), there are some C∞−ring (B, Ψ ) and some C∞−homomorphism ϕ : (A, Φ) → (B, Ψ ) such that R = ker(ϕ). Proof. It suffices to take (B, Ψ ) =(cid:18) A R , Φ(cid:19) and ϕ = qR : (A, Φ) →(cid:18) A R , Φ(cid:19) Theorem 3. (Fundamental Theorem of the C∞−Homomorphism) Let (A, Φ) be a C∞−ring and R ⊆ A × A be a C∞−congruence. For every C∞−ring (B, Ψ ) and for every C∞−homomorphism ϕ : (A, Φ) → (B, Ψ ) such that R ⊆ ker(ϕ), that is, such that: (a, a′) ∈ R ⇒ ϕ(a) = ϕ(a′), there is a unique C∞−homomorphism: such that the following diagram commutes: , Φ(cid:19) → (B, Ψ ) eϕ :(cid:18) A R / (B, Ψ ) 7♥ ♥ ♥ ϕ ♥ ♥ eϕ ♥ ♥ (A, Φ) q (cid:18) A R , Φ(cid:19) the quotient A R that is, such that eϕ ◦ q = ϕ, where Φ is the canonical C∞−structure induced on Proof. Define: → B A R a 7→ ϕ(a) eϕ : and note that eϕ defines a function, since given (a, a′) ∈ R, i.e., such that a = a′, we have ϕ(a) = ϕ(a′), so eϕ(a) = ϕ(a) = ϕ(a′) = eϕ(a′). As functions, we have eϕ ◦ q = ϕ. R(cid:19)n (cid:18) A Given n ∈ N and f ∈ C∞(Rn, R), the following diagram commutes, eϕn Bn Φ(f ) A R eϕ Ψ (f ) / B / 7 / / / 18 J. C. Berni and H. L. Mariano since for any (a1, · · · , an) ∈(cid:18) A R(cid:19)n we have: eϕ ◦ Φ(f )(a1, · · · , an) = eϕ(Φ(f )(a1, · · · , an)) := ϕ(Φ(f )(a1, · · · , an)) and since ϕ is a C∞−homomorphism, we have: ϕ(Φ(f )(a1, · · · , an)) = Ψ (f )(ϕ(a1), · · · , ϕ(an)) thus: On the other hand, we have: eϕ ◦ Φ(f )(a1, · · · , an) = Ψ (f )(ϕ(a1), · · · , ϕ(an)). so Ψ (f ) ◦eϕn(a1, · · · , an) = Ψ (f )(eϕ(a1), · · · ,eϕ(an)) = Ψ (f )(ϕ(a1), · · · , ϕ(an)) (∀(a1, · · · , an) ∈(cid:18) A R(cid:19)n and eϕ is a C∞−homomorphism. )(eϕ ◦ Φ(f )(a1, · · · , an) = Ψ (f ) ◦eϕn(a1, · · · , an)), Thus we have the following equation of C∞−homomorphisms: Since q is surjective, it follows that eϕ is unique. The following result is straightforward: eϕ ◦ q = ϕ. Proposition 7. Let (A, Φ) and (B, Ψ ) be two C∞−rings and let ϕ : (A, Φ) → (B, Ψ ) be a C∞−homomorphism. The ordered pair: where: (ϕ[A], Ψ ′) Ψ ′ :Sn∈N C∞(Rn, R) → Sn∈N Func (ϕ[A]n, ϕ[A]) 7→ Ψ (f ) ↾ϕ[A]n: ϕ[A]n → ϕ[A] (Rn f → R) is a C∞−subring of (B, Ψ ), called the homomorphic image of A by ϕ. A Universal Algebraic Survey of C∞−Rings 19 Corollary 2. Let (A, Φ) and (B, Ψ ) be two C∞−rings and let ϕ : (A, Φ) → (B, Ψ ) be a C∞−homomorphism. As we have noticed in Proposition 7, (ϕ[A], Ψ ′) is a C∞−subring of (B, Ψ ). Under these circumstances, there is a unique C∞−isomorphism: eϕ :(cid:18) A ker(ϕ) , Φ(cid:19) → (ϕ[A], Ψ ′) such that the following diagram commutes: ϕ / (ϕ[A], Ψ ′) 6❧❧❧❧❧❧❧ eϕ (A, Φ) q (cid:18) A ker(ϕ) , Φ(cid:19) A the quotient that is, such that eϕ ◦ q = ϕ, where Φ is the canonical C∞−structure induced on Proof. Applying the previous result to R = ker(ϕ) yields the existence of a unique C∞−homomorphism such that the diagram commutes. We need only to ker(ϕ) (a, a′) ∈ ker(ϕ) and a = a′. prove that eϕ is bijective. Given a, a′ such that eϕ(a) = eϕ(a′), by definition we have ϕ(a) = ϕ(a′), so eϕ(a) = ϕ(a) = y, so ϕ is surjective. Also, given any y ∈ ϕ[A], there is some a ∈ A such that ϕ(a) = y. Thus, 2.4 Directed Colimits of C∞−Rings The following result is going to be used to construct directed colimits of C∞−rings. Lemma 1. Let (A, Φ) be a C∞−ring. The ordered pair: (A × {α}, Φ × idα) where: Φ × idα :Sn∈N C∞(Rn, R) → (Rn f → R) 7→ Φ(f ) × idα : is a C∞−ring and: Sn∈N Func ((A × {α})n, A × {α}) (A × {α})n → A × {α} ((a1, α), · · · , (an, α)) 7→ (Φ(f )(a1, · · · , an), α) / 6 20 J. C. Berni and H. L. Mariano π1 : A × {α} → A 7→ a (a, α) is a C∞−isomorphism, that is: (A, Φ) ∼=π1 (A × {α}, Φ × idα) Proof. It is clear that π1 is a bijection, so it suffices to prove it is a C∞−homomorphism such that its inverse: ι1 : A → A × {α} a 7→ (a, α) is a C∞−homomorphism. Let n ∈ N and f ∈ C∞(Rn, R). Given ((a1, α), · · · , (an, α)) ∈ (A × {α})n, we have: π1◦((Φ×idα)(f ))((a1, α), · · · , (an, α)) = π1(Φ(f )(a1, · · · , an), α) = Φ(f )(a1, · · · , an) and Φ(f )◦πn 1 ((a1, α), · · · , (an, α)) = Φ(f )(π1(a1, α), · · · , π1(an, α)) = Φ(f )(a1, · · · , an). Thus, the following diagram commutes: (A × {α})n (Φ×idα)(f ) A × {α} πn 1 π1 An Φ(f ) / A and π1 is a C∞−homomorphism. Also, for any (a1, · · · , an) ∈ An we have: (Φ×idα)◦ιn(a1, · · · , an) = (Φ×idα)((a1, α), · · · , (an, α)) = (Φ(f )(a1, · · · , an), α) and ι ◦ Φ(f )(a1, · · · , an) = (Φ(f )(a1, · · · , an), α) so the following diagram commutes: An Φ(f ) A ιn ι (A × {α})n (Φ×idα) / A × {α} and ι is a C∞−homomorphism, inverse to π1. / / / / / / A Universal Algebraic Survey of C∞−Rings 21 Now we are going to describe the construction of directed colimits of directed families of C∞−rings. Theorem 4. Let (I, ≤) be a directed set and let ((Aα, Φα), µαβ)α,β∈I be a di- rected system. There is an object (A, Φ) in C∞Rng such that: Proof. Let: (A, Φ) ∼= lim −→ α∈I (Aα, Φα) A′ := [α∈I Aα × {α} and consider the following equivalence relation: R = {((a, α), (b, β)) ∈ A′ × A′(∃γ ∈ I)(γ ≥ α, β)(µαγ (a) = µβγ(b))} Take A = A′ R and define the following C∞−structure on A: Φ :Sn∈N C∞(Rn, R) →Sn∈N Func (An, A) 7→ (An Φ(f ) → A) (Rn f → R) where Φ(f ) : An → A is defined as follows: given ((a1, α1), · · · , (an, αn)) ∈ An, we have {α1, · · · , αn} ⊆ I, and since (I, ≤) is directed, there is some α ∈ I such that: (∀i ∈ {1, · · · , n})(αi ≤ α) We define: Φ(f )((a1, α1), · · · , (an, αn)) := (Φα(f )(µα1α(a1), · · · , µαnα(an)), α) We are going to show that this definition is independent of the choice of α. Let β ≥ α1, · · · , αn so we have: Φ(f )((a1, α1), · · · , (an, αn)) = (Φβ (f )(µα1β(a1), · · · , µαnβ(an)), β) We need to show that there is some γ ∈ I such that γ ≥ α, β and: µαγ(Φα(f )(µα1α(ai)), · · · , µαnα(an))) = µβγ(Φβ(f )(µα1β(ai)), · · · , µαnβ(an))) Choose γ ≥ α, β. Then: (∀i ∈ {1, · · · , n})(µβγ ◦ µαiβ = µαβ ◦ µαiα) 22 J. C. Berni and H. L. Mariano Since µαγ : (Aα, Φα) → (Aγ, Φγ) is a C∞−homomorphism, we have the following commutative diagram: µn αγ An α Φα(f ) An γ Φγ (f ) Aα µαγ / Aγ so µαγ(Φα(f )(µα1α(a1)), · · · , Φα(f )(µαnα(an))) = = Φγ(f )(µαγ(µα1α(a1)), · · · , µαγ(µαnα(an))) and since: we have: (∀i ∈ {1, · · · , n})(µαγ ◦ µαiα = µαiγ) µαγ(Φα(f )(µα1α(a1)), · · · , Φα(f )(µαnα(an))) = Φγ(f )(µα1γ(a1), · · · , µαnγ(an)) (1) Also, since µβγ : (Aβ, Φβ) → (Aγ, Φγ) is a C∞−homomorphism, we have the following commutative diagram: µn βγ An β Φβ (f ) An γ Φγ (f ) Aβ µβγ / Aγ so µβγ(Φβ (f )(µα1β(a1)), · · · , Φβ(f )(µαnβ(an))) = = Φγ(f )(µβγ(µα1β(a1)), · · · , µβγ(µαnβ(an))) and since: we have: (∀i ∈ {1, · · · , n})(µβγ ◦ µαiβ = µαiγ) µβγ(Φβ (f )(µα1β(a1)), · · · , Φβ(f )(µαnβ(an))) = Φγ(f )(µα1γ(a1), · · · , µαnγ(an)) (2) / / / / / / A Universal Algebraic Survey of C∞−Rings 23 Comparing (1) and (2), we get: µαγ(Φα(f )(µα1α(a1)), · · · , Φα(f )(µαnα(an))) = = µβγ(Φβ(f )(µα1β(a1)), · · · , Φβ(f )(µαnβ(an))) so (Φα(µα1α(a1), · · · , µαnα(an)), α) = (Φβ (µα1β(a1), · · · , µαnβ(an)), β) and Φ(f ) does not depend on the choice of the index. The definition of Φ(f ) does not depend on the choice of the representing elements, (ai, αi) + R, either5. Hence Φ is a C∞−structure on A and this shows that (A, Φ) is a C∞−ring. For each α ∈ I we define: ηα : Aα × {α} → A (a, α) 7→ (a, α) Claim: ηα : Aα ×{α} → A is a C∞−homomorphism between (Aα ×{α}, Φα× idα) and (A, Φ). Let n ∈ N and f ∈ C∞(Rn, R). Given ((a1, α), · · · , (an, α)) ∈ (Aα × {α})n. We have: Φ(f ) ◦ ηn α((a1, α), · · · , (an, α)) := (Φα × idα)(f )(ηα(a1, α), · · · , ηα(an, α)) = = (Φα(f )(a1, · · · , an), α) and ηα ◦ (Φα × idα)(f )((a1, α), · · · , (an, α)) = ηα(Φα(f )(a1, · · · , an), α) = = (Φα(f )(a1, · · · , an), α) so the following diagram commutes: (Aα × {α})n (Φα×idα)(f ) Aα × {α} ηn α ηα An Φ(f ) / A 5 the proof of this fact is analogous to the one we just made. / / / 24 J. C. Berni and H. L. Mariano Now we take, for each α ∈ I the following C∞−homomorphism: where ια : Aα → Aα × {α} is the C∞−isomorphism described in Lemma 1. λα := ηα ◦ ια : Aα → A Claim: (∀α ∈ I)(∀β ∈ I)(α ≤ β → λβ ◦ µαβ = λα), that is, the following diagram commutes: A ηα 6♠♠♠♠♠♠♠♠♠♠♠♠♠♠♠ Aα × {α} h◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ ηβ Aβ × {β} ια Aα µαβ ιβ / Aβ Indeed, given a ∈ Aα, we have: λβ ◦ µαβ(a) = ηβ ◦ ιβ(µαβ(a)) = ηβ(µαβ(a), β) = (µαβ(a), β) and λα(a) = ηα(ια(a)) = ηα(a, α) = (a, α) Given any γ ≥ α, β, we have µαγ = µβγ ◦ µαβ, so: µαγ(a) = µβγ(µαβ(a)) and ((a, α), (µαβ (a), β)) ∈ R, hence λβ ◦ µαβ(a) = λα(a), and the diagram com- mutes. Now we need only to show that ((A, Φ), λα) has the universal property of the colimit. Let (B, Ψ ) be a C∞−ring and let ζα : Aα → B be a family of C∞−homo- morphisms such that for every α, β ∈ I with α ≤ β we have ζβ ◦ µαβ = ζα, that is, the diagram: B ζα 7♦♦♦♦♦♦♦♦♦♦♦♦♦♦ µαβ Aα g❖❖❖❖❖❖❖❖❖❖❖❖❖❖ ζβ / Aβ commutes. We are going to show that there is a unique C∞−homomorphism θ : A → B such that the following diagram commutes: 6 h O O / O O 7 / g A Universal Algebraic Survey of C∞−Rings 25 ζα B ∃!θ A 7♦♦♦♦♦♦♦♦♦♦♦♦♦♦ λα µαβ Aα ζβ g❖❖❖❖❖❖❖❖❖❖❖❖❖❖ λβ / Aβ Indeed, given any (a, α) ∈ A, we have (a, α) + R = λα(a) = λβ(µαβ (a)), so we define: θ : A → B (a, α) 7→ ζα(a) which is a function, since (ζα)α∈I is a commutative co-cone. Claim: θ : (A, Φ) → (B, Ψ ) is a C∞−homomorphism. Let n ∈ N and f ∈ C∞(Rn, R). Given ((a1, α1), · · · , (an, αn)) ∈ An, we have, by definition, Φ(f )((a1, α1), · · · , (an, αn)) = (Φα(f )(a1, · · · , an), α) for every α ∈ I such that α ≥ αi, for every i ∈ {1, · · · , n}. Thus: θ ◦ Φ(f )((a1, α1), · · · , (an, αn)) = θ((Φα(f )(µα1α(a1), · · · , µαnα(an)), α)) = = ζα(Φα(f )(µα1α(a1), · · · , µαnα(an))) On the other hand, Ψ (f ) ◦ θn((a1, α1) + R, · · · , (an, αn) + R) = Ψ (f )(ζα1 (a1), · · · , ζαn (an)) = = Ψ (f )(ζα(µα1α(a1)), · · · , ζα(µαnα(an))) Since for every α ∈ I, ζα : Aα → B is a C∞−homomorphism, it follows that: ζα(Φα(f )(µα1α(a1), · · · , µαnα(an))) = Ψ (f )(ζα(µα1α(a1)), · · · , ζα(µαnα(an))) so θ is a C∞−homomorphism. Now, given a ∈ Aα, we have: so the required diagram commutes. θ(λα(a)) = θ((a, α)) = ζα(a) Since A =Sα∈I λα[Aα], θ is uniquely determined, and the result follows. O O ✤ ✤ ✤ 7 1 1 / g m m 26 J. C. Berni and H. L. Mariano Theorem 5. Given any small category J and any diagram: D : J → C∞Rng (α h→ β) 7→ (Aα, Φα) D(h) → (Aβ , Φβ) there is a C∞−ring (A, Φ) such that: (A, Φ) ∼= lim ←− α∈I D(α) Proof. Consider the product C∞−ring: Ai, Φ(I)! Yi∈I and take: A = {(aα)α∈I ∈ Yα∈I Aα(∀α, β ∈ I)(∀α h→ β)(D(h)(aα) = aβ)} ⊆ Yα∈I Aα together with the C∞−subring structure Φ′ of Φ(I). We have ((A, Φ′), πα ↾A: A → Aα) ∼= lim (Aα, Φα) ←−α∈I Remark 4. Let Σ = Sn∈N C∞(Rn, R) and let X = {x1, x2, · · · , xn, · · · } be a denumerable set of variables, so F (Σ, X) will denote the algebra of terms of this language Σ. A class of ordered pairs will be simply a subset S ⊆ F (Σ, X) × F (Σ, X). In our case, these pairs are given by the axioms, so S con- sists of the following: • For any n ∈ R, i ≤ n and a (smooth) projection map pi : Rn → R we have: (pi(x1, · · · , xi, · · · , xn), xi) ∈ S • for every f, g1, · · · , gn ∈ C∞(Rm, R) and h ∈ C∞(Rn, R) such that f = h ◦ (g1, · · · , gn), we have (h(g1(x1, · · · , xm), · · · , gn(x1, · · · , xm)), f (x1, · · · , xm)) ∈ S Remark 5. The class of C∞−rings is a model of an equational theory, thus it is a variety of algebras. However, if we were not given this information, noting that the category of C∞−rings is closed under products, subalgebras and ho- momorphic images, the HSP Birkhoff 's Theorem would lead us to the same conclusion, that is, that the class C∞Rng is a variety of algebras, and by the previous remark, C∞Rng = V (S), the variety of algebras defined by S. In particular, we have some classical results. We list some of them: A Universal Algebraic Survey of C∞−Rings 27 • for every set X there is a free C∞−ring determined by X; • any C∞−ring is a homomorphic image of some free C∞−ring; • a C∞−homomorphism is monic if, and only if, it is an injective map; • any indexed set of C∞−rings, {(Aα, Φα)α ∈ I}, has a coproduct in C∞Rng. We end this section by proving that the (variety) of all C∞−rings is a reflec- tive subcategory of C∞Str. Theorem 6. The inclusion functor ι : C∞Rng ֒→ C∞Str has a left adjoint L : C∞Str → C∞Rng: given by M 7→ M/θM where θM is the least C∞-congruence of M such that M/θM ∈ Obj(C∞Rng). Moreover, the unit of the adjunction L ⊣ ι has components (qM )M∈Obj(C∞Str), where qM : M ։ M/θM is the quotient homomorphism. Proof. Let (M, µ) be any C∞−structure and let M denote the underlying set of (M, µ). Consider ΓM = {θ ⊆ M × M ; it is a congruence relation and M/θ ∈ Obj(C∞Rng)}. Γ is non-empty, since θ = M × M is a congruence relation and M/θ = {∗} ∼= C∞(∅) ∈ Obj(C∞Rng). Let θM =T ΓM . We will show first that θM ∈ ΓM . Since θM is a C∞-congruence in M , it remains to check that M/θM ∈ Obj(C∞Rng). Consider the diagonal C∞-morphism: δM : M →Qθ∈ΓM M/θ m 7→ ([m]θ)θ∈ΓM We have (m, n) ∈ ker(δM ) ⇔ ([m]θ)θ∈ΓM = ([n]θ)θ∈ΓM ⇔ (∀ θ ∈ ΓM )([m]θ = [n]θ) ⇔ (∀ θ ∈ ΓM )(mθn) ⇔ mθM n. Thus, by the homomorphism theorem for C∞Str, there is a unique C∞- M/θ such the diagram below commutes monomorphism ¯δM : M/θM ֌Qθ∈ΓM δM M/θ Qθ∈ΓM 8qqqqqqqqqq ¯δM M qM M/θM Since C∞Rng is closed under products, we have thatQθ∈ΓM We also have that C∞Rng is closed under substructures and isomorphisms, then M/θ ∈ Obj(C∞Rng). / / 8 28 J. C. Berni and H. L. Mariano M/θM is Obj(C∞Rng). Denote by L(M ) := M/θM . We show that qM : M ։ ι(L(M )) satisfies the universal property relatively to C∞-homomorphisms f : M → ι(N ), with N ∈ Obj(C∞Rng). Thus, we obtain an injective C∞-morphism ¯f : M/ ker(f ) ֌ ι(N ). Since C∞Rng is closed under substructures and isomorphisms, we have that M/ ker(f ) ∈ Obj(C∞Rng). Hence ker(f ) ∈ ΓM and θM ⊆ ker(f ). Then, again by the homo- morphism theorem, there is a unique homomorphism f : M/θM → N such that the following diagram commutes M qM/ ι(L(M )) #●●●●●●●●● f ι( f ) ι(N ) 3 Free C∞−Rings Let E be any set. We are going to describe the free C∞−ring determined by E. For any set E (finite or infinite), we set: RE := {f : E → Rf is a function} and Func (RE, R) = {g : RE → Rg is a function}. Given any two finite subsets of E, Ei, Ej ⊆ E, whenever Ei ⊆ Ej we have the restriction map: µi,j : REj ։ REi f 7→ f ◦ ıi,j : Ei → R where ıi,j : Ei ֒→ Ej x 7→ x is the inclusion map of Ei into Ej , and ιi : Ei ֒→ E x 7→ x is the inclusion map of Ei into E. / # A Universal Algebraic Survey of C∞−Rings 29 We also have the injective pullback (defined by "composition") associated with µi,j, namely: bµi,j : Func (REi, R) ֌ Func (REj , R) / R ) 7→ ( REj g◦µi,j / ( REi / R ) g Analogously that for each finite subset Ei of E we have the injective map ιEi,E : Func (REi, R) → Func (RE, R) given by ιEi,E : Func (REi , R) → ( REi f / R ) 7→ ef : RE → Func (RE, R) R g 7→ f (g ◦ ιi : REi → R) Given f ∈ Func (REi , R), on the one hand we have: where ιEj ,E ◦dµi,j(f ) = ιEj ,E(f ◦ µi,j) = ^f ◦ µi,j, ^f ◦ µi,j : RE → R g 7→ f ◦ µi,j(g ◦ ιj) = f ◦ (g ◦ ιj ◦ ıi,j) = f (g ◦ ιi) and on the other hand, ef : RE → R g 7→ f (g ◦ ιi) so (∀f ∈ Func (REi , R))(ιEj ,E ◦dµi,j(f ) = ιEi,E(f )). Thus, for every Ei, Ej such that Ei ⊆ Ej, the following diagram commutes: Func (REi , R) bµi,j / Func (REj , R) (3) ιEi ,E Func (RE, R) ιEj ,E that is, Consider the following commutative diagram: ιEj ,E ◦bµi,j = ιEi,E. Func (REi , R) bµi,j / Func (REj , R) ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ bµi t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ bµj lim −→E′⊆f E Func (RE′ , R) / / + + / / / s s * * / / / t t t 30 J. C. Berni and H. L. Mariano which is the colimit diagram of the directed system (Func (REi , R),bµi,j). Recall that, concretely we have: Func (RE′ lim −→ E′⊆f E , R) = SEi⊆f E Func (REi, R) × {Ei} ∼ where (fi, Ei) ∼ (fj, Ej) ⇐⇒ (∃Ek ⊆f E)(Ei ⊆ Ek)(Ej ⊆ Ek)(bµi,k(fi) =bµj,k(fj)) Given the cone (3), the universal property of the colimit yields a unique function u : lim −→ E′⊆f E Func (RE′ , R) → Func (RE, R) such that the following diagram commutes: Func (REi , R) bµi,j / Func (REj , R) ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ bµi t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ bµj lim −→E′⊆f E Func (RE′ , R) ιEi,E ∃!u ιEj ,E Func (RE, R) Now we claim that u is an injective function. Given any [(fi, Ei)], [(fj, Ej )] ∈ lim −→E′⊆f E Func (RE′ , R) such that: u([(fi, Ei)]) = u([(fj, Ej)]), and since bµi(fi) = [(fi, Ei)] and bµj(fj) = [(fj, Ej )], this is equivalent to: ιEi,E(fi) = ιEj ,E(fj), so: i.e., (∀g ∈ RE)(efi(g) = efj(g)) (∀g ∈ RE)(fi(g ◦ ιi) = fj(g ◦ ιj )) (4) Now, given any finite Ei and Ej, we can take Ek ⊆ E such that Ei ⊆ Ek and Ej ⊆ Ek, namely Ek = Ei ∪ Ej. We are going to show that bµi,k(fi) =bµj,k(fj). , , / / / t r r A Universal Algebraic Survey of C∞−Rings 31 Given any g : Ei ∪ Ej → R consider: g : E → R x 7→(g(x), if x ∈ Ek 0 otherwise and note that g ◦ ιi = g ◦ ıik and g ◦ ιj = g ◦ ıjk. Thus we have: bµi,k(fi)(g) = fi(g ◦ ıik) = fi(g ◦ ιi) Since g is arbitrary, we have: (4) = fj(g ◦ ιj ) = fj(g ◦ ıjk) =bµj,k(fj)(g) and so u is injective. bµi,k(fi) =bµj,k(fj) [(fi, Ei)] = [(fj, Ej )], Therefore we can identify lim −→E′⊆f E Func (RE′ , R) with a subset of Func (RE, R), namely: u" lim −→ E′⊆f E Func (RE′ , R)# . In this sense, we say that lim −→E′⊆f E Func (RE′ f : RE → R for which there are some finite E′ ⊆ E and some f ′ : RE′ u([(f ′, E′)]) = f . As an "abus de langage", one says that lim is the set of all functions f : RE′ → R for some E′ ⊆f E. −→E′⊆f E , R) consists of all functions → R with , R) Func (RE′ Now we proceed to describe the free C∞−ring determined by an arbitrary set E. C∞(REj ) Given two finite subsets of E, Ei, Ej ⊆ E, we set bµi,j ↾C∞(REi ): C∞(REi ) ֌ bµi,j ↾C∞(REi ): C∞(REi ) → to be the restriction of the map bµi,j to C∞(REi ), so the following rectangle f 7→bg(f ◦ ıij : REi → R) → R) 7→ bg ◦ µi,j : REj → (REi bg C∞(REj ) R commutes 32 J. C. Berni and H. L. Mariano C∞(REi) ιi bµi,j ↾ C∞ (REi ,R) C∞(REj ) ιj Func (REi , R) / bµi,j / Func (REj , R) and consider the following colimit diagram in Set: ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ ℓi lim −→E′⊆f E bµi,j ↾ C∞ (REi ) / C∞ (REj ) t❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ℓj C∞ (RE′ ) C∞ (RE′ lim −→ E′⊆f E ) = SEi⊆f E C∞(REi) × {Ei} ∼ C∞ (REi) Recall that: where (fi, Ei) ∼ (fj, Ej ) ⇐⇒ ⇐⇒ (∃Ek ⊆f E)(Ei ⊆ Ek)(Ej ⊆ Ek)(cid:16)bµi,k ↾C∞(REi ) (fi) =bµj,k ↾C∞(REj ) (fj)(cid:17) so: ℓi : C∞(REi ) → lim −→E′⊆f E C∞ (RE′ ) (REi fi→ R) 7→ [(fi, Ei)] C∞−homomorphism. We now claim that bµi,j ↾C∞(REi ): (C∞(REi), ΦEi ) → (C∞(REj ), ΦEj ) is a Given any f ∈ C∞(Rn, R), we are going to show that the following diagram commutes: C∞(REi)n n C∞ (REi ) bµi,j ↾ / C∞(REj )n ΦEi (f ) C∞(REi ) ΦEj (f ) / C∞(REj ) bµi,j ↾ C∞(REi ) Let (bg1, · · · ,bgn) ∈ C∞(REi )n. On the one hand we have: / / _ _ / / * * / / / t t t / / A Universal Algebraic Survey of C∞−Rings 33 ΦEj (f )◦bµi,j ↾C∞(REi ) n(bg1, · · · ,bgn) = ΦEj (f )(bµi,j ↾C∞(REi ) (bg1), · · · ,bµi,j ↾C∞(REi ) (bgn))) = = ΦEj (f )(bg1 ◦ µij , · · · ,bgn ◦ µi,j) and on the other hand we have: bµi,j ↾C∞(REi ) ◦ΦEi(f )(bg1, · · · ,bgn) =bµi,j ↾C∞(REi ) ( \f ◦ (g1, · · · , gn)) = = ( \f ◦ (g1, · · · , gn)) ◦ µi,j Given any h ∈ REj , we have, on the one side: ΦEj (f )(bg1 ◦ µij, · · · ,bgn ◦ µi,j)(h) = ΦEj (f )(bg1(h ◦ ıij), · · · ,bgn(h ◦ ıi,j)) = = ΦEj (f )(bg1, · · · ,bgn)(h ◦ ıij) = \f ◦ (g1, · · · , gn)(h ◦ ıij) and on the other hand: = [ \f ◦ (g1, · · · , gn) ◦ µij ](h) = \f ◦ (g1, · · · , gn)(h ◦ ıij) [bµij ↾C∞(REi ) ◦ΦEi(f )(bg1, · · · ,bgn)](h) = [bµij ↾C∞(REi ) ◦( \f ◦ (g1, · · · , gn))](h) = so bµij ↾C∞(REi ) is a C∞−homomorphism. We have, thus, the directed system ((C∞(REi), ΦEi ),bµij ↾C∞(REi )) of C∞−rings, and the following commutative diagrams (for any Ei, Ej ⊆f E) in C∞Rng: (C∞ (REi), ΦEi ) bµi,j ↾ C∞ (REi ) / (C∞ (REj ), ΦEj ) ❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱ ℓi t❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤❤ ℓj ), ΦE′ ) (C∞ (RE′ lim −→E′⊆f E Note that each ℓi : (C∞ (REi), ΦEi ) → lim −→E′⊆f E (C∞ (RE′ ), ΦE′ ) is a C∞−homomorphism. Let us examine this colimit more closely. We denote: C∞(RE) = lim −→ E′⊆f E C∞ (RE′ ) and describe the C∞−structure on C∞(RE), * * / / / t t t 34 J. C. Berni and H. L. Mariano ΦE :Sn∈N C∞(Rn, R) →Sn∈N Func (C∞(RE)n, C∞(RE)) 7→ ΦE(f ) : C∞(RE)n → C∞(RE) (Rn f → R) where: ΦE(f ) : C∞(RE)n → C∞(RE) and Ek = ∪n i=1Ei. We have thus: ([(bf1, E1)], · · · , [(bfn, En)]) 7→ [(f ◦ (bµ1k ↾C∞(RE1 ) (bf1), · · · ,bµnk ↾C∞(REn ) (bfn)), Ek)] f ◦ (bµ1k ↾C∞(RE1 ) (bf1), · · · ,bµnk ↾C∞(REn ) (bfn)) : REk → R, which belongs to C∞(REk ), sincebµ1k ↾C∞(RE1 ) (bf1), · · · ,bµnk ↾C∞(REn ) (bfn) ∈ C∞(REk ) and f ∈ C∞(Rn). Thus we have the C∞−ring: (C∞(RE), ΦE) Let E be any set, and write ιE′ 6❧❧❧❧❧❧❧❧❧❧❧❧❧ E′ lim −→E′⊆f E E′ ιE′ E′′ ιE′′ h❘❘❘❘❘❘❘❘❘❘❘❘❘ / E′′ For every finite subset E′ ⊆f E, we have the free C∞−ring: E′ : E′ → U (C∞(RE′ ), ΦE′ ). so we can form the commutative cone: C∞(RE) ℓE′ ◦E′ 6♠♠♠♠♠♠♠♠♠♠♠♠♠♠ ιE′′ E′ E′ ℓE′′ ◦E′′ h◗◗◗◗◗◗◗◗◗◗◗◗◗◗ / E′′ The universal property of lim −→E′⊆f E C∞(RE) such that the following prism commutes in Set: E′ yields a unique function E : lim E′ → −→E′⊆f E 6 / h 6 / h A Universal Algebraic Survey of C∞−Rings 35 E′ $❏❏❏❏❏❏❏❏❏❏❏❏❏ ιE′′ E′ ιE′ E′′ E′ lim −→E′⊆f E :ttttttttttt ιE′′ E′ ∃!E E′′ ℓE′ C∞(RE′ ) C∞(RE) $❏❏❏❏❏❏❏❏❏❏❏ C∞ (RE′ ) bµE′E′′ ↾ :ttttttttttt ℓE′′ C∞(RE′′ ) In Set we have: lim −→ E′⊆f E E′ ∼= [E′⊆f E E′ = E so we have the function E : E → C∞(RE). Remark 6. Given a set E, for every finite subset E′ ⊆f E, with ♯E′ = n, there is a bijection ωn : {1, · · · , n} → E′ and a corresponding C∞−isomorphism: Also, given E′′ ⊆f E and a bijection ωm : {1, · · · , m} → E′′, with E′ ⊆ E′′ (so with n ≤ m) we have the composite: ), ΦE′ ) → (C∞(Rn), Ω). cωn : (C∞(RE′ −1 ◦bµE′E′′ ↾C∞(RE′ ) ◦cωn : C∞(Rn) → C∞(Rm) bµnm := cωm so the following rectangle commutes: C∞(RE′ ) bµE′ E′′ ↾ C∞(RE′ ) / C∞(RE′′ ) cωn C∞(Rn) dωm / C∞(Rm) dµnm Considering Ei ⊆ Ej and the inclusion maps ιi : C∞(REi ) ֒→ Func (REi ), we have the following commutative diagram: / / $ ✤ ✤ ✤ ✤ ✤ ✤ ✤ ✤ ✤ ✤ ✤ ✤ ✤ : / / $ : / O O / O O 36 J. C. Berni and H. L. Mariano C∞(REi) ιi Func (REi , R) bµi,j ↾ C∞ (REi ) bµi,j C∞(REj ) ιj / Func (REj , R) ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ bµi t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ bµj lim −→E′⊆f E Func (RE′ , R) By the universal property of lim C∞(REi, R), in the category Set, given −→Ei⊆f E the cone: (bµi ◦ ιi : C∞(REi , R) → lim −→ E′⊆f E C∞(RE′ Func (RE′ , R)), , R) → lim −→E′⊆f E Func (RE′ , R) there is a unique function v : lim −→E′⊆f E such that the following prism commutes: C∞(REi , R) / bµi,j ↾ C∞ (REi ,R) C∞(REj , R) $❏❏❏❏❏❏❏❏❏❏❏ ℓi ιi lim −→E′⊆f E ℓj zttttttttttt C∞(RE′ , R) ιj Func (REi, R) / $❏❏❏❏❏❏❏❏❏❏❏ bµi lim −→E′⊆f E v Func (REj , R) bµj zttttttttttt Func (RE′ , R) where ℓi : C∞(REi , R) → lim −→E′⊆E C∞(RE′ , R) are the canonic colimit arrows. We claim that v = ι : lim −→E′⊆f E C∞(RE′ , R) ֒→ lim −→E′⊆f E Func (RE′ , R) (the C∞(RE′ , R), there is some Ek ⊆f E and some fk ∈ inclusion). Given f ∈ lim −→E′⊆f E C∞(REk , R) such that: We have: f = [(fk, Ek)] = ℓk(fk) v(f ) = v(ℓk(fk)) =bµk ◦ ιk(f ) =bµk(f ) = [(fk, Ek)] = f / / _ _ / / / t / / / $ _ _ z / / / $ z A Universal Algebraic Survey of C∞−Rings 37 So we have lim −→E′⊆f E C∞(RE′ , R) ⊆ lim −→E′⊆f E Func (RE′ , R) ֌ Func (RE, R). Hence, we can regard lim −→E′⊆f E C∞(RE′ , R) as (an actual) subset of Func (RE′ , R), lim −→ E′⊆f E and consider, as in [22], (C∞(RE), ΦE) as the ring of functions RE → R which smoothly depend on finitely many variables only. The results given in the previous section assure the existence of some con- structions within the category of C∞−rings, such as quotients, products, coprod- ucts and so on. As we shall see, the category of C∞−rings holds a strong relation with rings of the form C∞(Rn). In this section we are going to describe concretely such constructions in C∞Rng. Our definition of C∞−ring yields a forgetful functor: U : C∞Rng (A, Φ) ((A, Φ) ϕ → (B, Ψ )) 7→ (A → Set A 7→ U(ϕ) → B) We are going to show that this functor has a left adjoint, the "free C∞−ring", that we shall denote by L : Set → C∞Rng. Before we do it, we need the follow- ing: Remark 7. Given any m ∈ N, we note that the set C∞(Rn) may be endowed with a C∞−structure: Ω :Sn∈N C∞(Rn, R) → (Rn f → R) 7→ Sn∈N Func (C∞(Rm)n, C∞(Rm)) Ω(f ) = f ◦ − : C∞(Rm)n → C∞(Rm) (h1, · · · , hn) 7→ f ◦ (h1, · · · , hn) so it is easy to see that it can be made into a C∞−ring (C∞(Rm), Ω). From now, when dealing with this"canonical" C∞−structure, we shall omit the symbol Ω, writting C∞(Rm) instead of (C∞(Rm), Ω). In fact, we are going to show that for finite sets X with ♯X = m, for in- stance, the C∞−ring given in the previous remark is (up to isomorphism) the free C∞−ring on m generators. 38 J. C. Berni and H. L. Mariano Indeed, given any finite set X with ♯X = m ∈ N, we define: together with the canonical C∞−structure Ω given in Remark 7. L(X) := C∞(Rm) We have the following: Proposition 8. Let U : C∞Rng → Set, (A, Φ) 7→ A , be the forgetful functor. The pair (n, (C∞(Rn), Ω)), where: n : {1, · · · , n} → U (C∞(Rn), Ω) 7→ πi : Rn → R , i is the free C∞-ring with n generators, which are the projections: πi : Rn → R (x1, · · · , xi, · · · , xn) 7→ xi Proof. (Cf. Proposition 1.1 of [25].) Given any C∞−ring (A, Φ) and any function α : {1, 2, · · · , n} → U (A, Φ), we (A, Φ) such that the following diagram commutes: are going to show that there is a unique C∞−homomorphismeα : (C∞(Rn), Ω) → {1, 2, · · · , n} n U (C∞(Rn), Ω) ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ α U(eα) U (A, Φ) Given α(1), α(2), · · · , α(n) ∈ A, define eα : C∞(Rn) → f A 7→ Φ(f )(α(1), α(2), · · · , α(n)) for every i ∈ {1, 2, · · · , n}: Note that such a function satisfies (∀i ∈ {1, 2, · · · , n})(eα(n(i)) = α(i)), since eα(n(i)) =eα(πi) = Φ(πi)(α(1), · · · , α(i), · · · , α(n)) = Next thing we show is that eα is a C∞−homomorphism. Let f : Rm → R be any smooth function. We claim that the following diagram = πi(α(1), · · · , α(i), · · · , α(n)) = α(i). commutes: / / A Universal Algebraic Survey of C∞−Rings 39 C∞(Rn)m Ω(f ) C∞(Rn) eαm eα Am Φ(f ) / A Let (ϕ1, · · · , ϕm) ∈ C∞(Rn)m. On the one hand we have: Φ(f ) ◦eαm(ϕ1, · · · , ϕm) = Φ(f ) ◦ (eα(ϕ1), · · · ,eα(ϕm)) = Φ(f )(Φ(ϕ1)(α(1), α(2), · · · , α(n)), · · · , Φ(ϕm)(α(1), α(2), · · · , α(n))) On the other hand we have: eα ◦ C∞(Rn)(f )(ϕ1, · · · , ϕm) =eα(f ◦ (ϕ1, · · · , ϕm)) = Since (A, Φ) is a C∞−ring, we have also: = Φ(f ◦ (ϕ1, · · · , ϕm))(α(1), α(2), · · · , α(n)) Φ(f ◦ (ϕ1, · · · , ϕm)) = Φ(f ) ◦ (Φ(ϕ1), · · · , Φ(ϕm)), hence: Φ(f ) ◦eαm(ϕ1, · · · , ϕm) =eα ◦ Ω(f )(ϕ1, · · · , ϕm), and eα is indeed a C∞−homomorphism. For the uniqueness of eα, suppose Ψ : (C∞(Rn), Ω) → (A, Φ) is a C∞−homo- morphism such that (∀i ∈ {1, 2, · · · , n})(Ψ (n(i)) = α(i)). Since Ψ is a C∞−homo- morphism, in particular the following diagram commutes: C∞(Rn)n Ω(f ) C∞(Rn) Ψ n Ψ An Φ(f ) / A Note that for any f ∈ C∞(Rn), f ◦ (π1, · · · , πn) = f , and since the diagram above commutes, we have: Φ(f )(Ψ (π1), · · · , Ψ (πn)) = Ψ (f ◦ (π1, · · · , πn)) = Ψ (f ) so Ψ (f ) = Φ(f )(Ψ (π1), · · · , Ψ (πn)) = Thus it is proved that (C∞(Rn), Ω) is (up to isomorphism) the free C∞−ring = Φ(f )(Ψ (n(1)), · · · , Ψ (n(n))) = Φ(f )(α(1), · · · , α(n)) =eα(f ). on n generators. / / / / / / 40 J. C. Berni and H. L. Mariano Given a finite set X with ♯X = n, consider a bijection ω : {1, · · · , n} → X and denote, for any i ∈ {1, · · · , n}, xi = ω(i). We define: RX = {f : X → Rf is a function}. Due to the bijection ω, any smooth function g ∈ C∞(Rn, R) can be inter- preted as: so we define: RX → f → R) 7→ g(f (x1), · · · , f (xn)) R (X bg : Note that C∞ (RX ) has a natural C∞−structure given by: C∞(RX ) := {bg : RX → Rg ∈ C∞(Rn, R)} ΦX :Sn∈N C∞ (Rn, R) →Sn∈N Func (C∞(RX )n, C∞(RX )) 7→ ΦX (f ) : C∞(RX )n → C∞(RX ) (Rn f → R) where: ΦX (f ) : C∞(RX )n → C∞(RX ) (bg1, · · · ,cgn) 7→ \f ◦ (g1, · · · , gn) : RX → R which is well-defined since the map g 7→bg is a bijection for a fixed ω. Thus we have the following C∞−ring: Note that given any x ∈ X, the evaluation map: (C∞(RX ), ΦX ) evx : RX → R f 7→ f (x) belongs to C∞(RX ). In fact, given x ∈ X = {x1, · · · , xn}, there is i ∈ {1, · · · , n} such that x = xi. coordinate. Claim: evxi = bπi, where πi : Rn → R is the (smooth) projection on the i−th bπi(f ), so evx ∈ C∞(RX ). We have, thus, the following function: We have, in fact, for any f ∈ RX , evxi(f ) = f (xi) = πi(f (x1), · · · , f (xi), · · · , f (xn)) = A Universal Algebraic Survey of C∞−Rings 41 X : X → U (C∞(RX ), ΦX ) evx : RX → R x 7→ f 7→ f (x) Proposition 9. Let X be a finite set with ♯X = n and consider: RX = {f : X → Rf is a function} together with some bijection ω : {1, · · · , n} → X, ω(i) = xi ∈ X. Under those circumstances, the following function is a C∞−isomorphism: bω : (C∞(RX ), ΦX ) → (C∞(Rn), Ω) bg : RX → R f 7→bg(f ) Proof. First we show that bω is a C∞−homomorphism. 7→ g : Rn → R mutes: Let m ∈ N and h ∈ C∞(Rm, R). We claim that the following diagram com- C∞(RX )m ΦX (h) C∞(RX ) bωm bω C∞(Rn)m Ω(h) / C∞(Rn) Given (bg1, · · · ,cgm) ∈ C∞(RX )m, we have: and bω(ΦX (h)(bg1, · · · ,cgm)) =bω( \h ◦ (g1, · · · , gm)) = h ◦ (g1, · · · , gm) Ω(h)(bω(bg1), · · · ,bω(cgm)) = Ω(h) ◦ (g1, · · · , gm) = h ◦ (g1, · · · , gm) so bω is a C∞−homomorphism. bδ : (C∞(Rm), Ω) 7→ f 7→ g(f (x1), · · · , f (xn)) (C∞(RX ), ΦX ) We claim that: (Rm g → R) 7→ bg : RX → R is a C∞−homomorphism. In fact, for every n ∈ N, h ∈ C∞(Rn, R) the follow- ing diagram commutes: / / / 42 J. C. Berni and H. L. Mariano C∞(Rm)n Ω(h) C∞(Rm) bδn bδ C∞(RX )n ΦX (h) / C∞(RX ) since given (g1, · · · , gn) ∈ C∞(Rm)n we have: and ΦX (h)(bδ(g1), · · · ,bδ(gm)) = ΦX (h) ◦ (bg1, · · · ,cgn) = \h ◦ (g1, · · · , gn) bδ(Ω(h)(g1, · · · , gn)) =bδ(h ◦ (g1, · · · , gn)) = \h ◦ (g1, · · · , gn), sobδ is a C∞−homomorphism. Given g ∈ C∞(Rm), we have: so so Also, givenbg ∈ C∞(RX ), we have: bω(bδ(g)) =bω(bg) = g bω ◦bδ = idC∞(Rm). bδ(bω(bg)) =bδ(g) =bg, bδ ◦bω = idC∞(RX ) As an immediate consequence of Propositions 9 and 8, we have: Corollary 3. For any finite set X, (X , (C∞(RX ), ΦX )) is the free C∞−ring defined by X. Proof. Let ω : {1, · · · , n} → X be a bijection,bω : (C∞(RX ), ΦX ) → (C∞(Rn), Ω) be the C∞−isomorphism induced by ω, as given in Proposition 9, (A, Φ) be any C∞−ring and let f : X → U (A, Φ) be any function. Note that the following diagram commutes: {1, · · · , n} ω X n X U (C∞(Rn), Ω) , U(bω−1) / U (C∞(RX ), ΦX ) / / / / / / A Universal Algebraic Survey of C∞−Rings 43 X ◦ ω(i). since given any i ∈ {1, · · · , n},bω−1(n)(i) =bω−1(πi) = bπi = evxi = X (xi) = We have the following diagram: {1, · · · , n} ω X n X U (C∞(Rn), Ω) U(bω−1) U (C∞(RX ), ΦX ) ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ f U (A, Φ) Since (n, (C∞(Rn), Ω)) is free, given the function f ◦ ω : {1, · · · , n} → U (A, Φ), there is a unique C∞−homomorphism ]f ◦ ω : (C∞(Rn), Ω) → (A, Φ) such that the following diagram commutes: {1, · · · , n} n U (C∞(Rn), Ω) ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ f ◦ω U(]f ◦ω) U (A, Φ) We have, thus, the following commutative diagram: {1, · · · , n} ω X n X U (C∞(Rn), Ω) U(bω−1) U (C∞(RX ), ΦX ) ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ f U (A, Φ) U(]f ◦ω) (A, Φ), and the following diagram commutes: It now suffices to take the C∞−homomorphism ef := (]f ◦ ω)◦bω : (C∞(RX ), ΦX ) → X X / U (C∞(RX ), ΦX ) )❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙ f U(ef ) U (A, Φ) determined by X. By construction such a ef is unique, so (X , (C∞(RX ), ΦX )) is the free C∞−ring / / / / / / * / / v v / / * / ) 44 J. C. Berni and H. L. Mariano Now we turn to the definition of the free C∞−ring determined by an arbi- trary set. Let E be any set, and consider: (C∞(RE), ΦE) = lim −→ E′⊆f E (C∞(RE′ ), ΦE′ ), where "⊆f " stands for "is a finite subset of", together with the unique arrow E : E = SE′⊆f E E′ → C∞(RE) such that for every E′ ⊆f E the following diagram commutes: E E′ E / C∞(RE) ℓE′ / C∞(RE′ ) E′ Proposition 10. Let E be any set. The pair (E, (C∞(RE), ΦE)) is the free C∞−ring determined by E. Proof. Let (A, Φ) be any C∞−ring and let f : E → A be any function. We decompose the set E as E = [E′⊆f E E′ = lim −→ E′⊆f E E and take the following cone: A f ◦ιE′ 7♦♦♦♦♦♦♦♦♦♦♦♦♦♦ ιE′ E′′ E′ g❖❖❖❖❖❖❖❖❖❖❖❖❖❖ f ◦ιE′′ / E′′ Since for every finite E′ ⊆ E, (E′ , C∞(RE′ )) is the free C∞−ring determined by E′, given the function f ◦ ιE′ : E′ → A, there is a unique C∞−homomorphism cfE′ : (C∞(RE′ ), ΦE′ ) → (A, Φ) such that the following diagram commutes: E′ E′ (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ f ◦ιE′ C∞(RE′ ) U(bfE′ ) U (A, Φ) Note that the following diagram commutes: / ? O O / O O 7 / g / / ( A Universal Algebraic Survey of C∞−Rings 45 (C∞(RE′ ), ΦE′ ) bµE′E′′ ↾ C∞ (RE′ ) (C∞(RE′′ ), ΦE′′ ) )❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙❙ bfE′ u❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ bfE′′ (A, Φ) In fact, the following rectangle commutes: E′ E′ C∞(RE′ ) ιE′′ E′ U( \µE′ E′′ ↾ C∞ (RE′ ) ) E′′ E′′ / C∞(RE′′ ) so: E′′ ◦ ιE′′ E′ = U (bµE′E′′ ↾C∞(RE′ )) ◦ E′ ), ΦE′ ) → (A, Φ) is the unique C∞−homomorphism such (5) that: and bfE′ : (C∞(RE′ commutes, i.e., E′ E′ C∞(RE′ ) (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ f ◦ιE′ U(bfE′ ) U (A, Φ) U (bfE′) ◦ E′ = f ◦ ιE′ is such that: We are going to show that bfE′′ ◦bµE′E′′ ↾C∞(RE′ ): (C∞(RE′ E′ E′ C∞(RE′ ) (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ f ◦ιE′ U(bfE′′ ◦bµE′E′′ ↾ C∞ (RE′ ) ) U (A, Φ) (6) ), ΦE′ ) → (A, Φ) commutes. By uniqueness it will follow that: bfE′′ ◦bµE′E′′ ↾C∞(RE′ )= bfE′. Composing (5) with U (bfE′′ ) by the left yields: E′ = U (bfE′′) ◦ U (bµE′E′′ ↾C∞(RE′ )) ◦ E′ U (bfE′′) ◦ E′′ ◦ ιE′′ / / ) u / / / / / ( / / ( 46 J. C. Berni and H. L. Mariano i.e., (f ◦ ιE′′ ) ◦ ιE′′ We know that ιE′′ ◦ ιE′′ U (bfE′′) ◦ E′′ ◦ ιE′′ Since U (bfE′′ ) ◦ E′′ = f ◦ ιE′′ , we have: E′ = U (bfE′′ ◦bµE′E′′ ↾C∞(RE′ )) ◦ E′ E′ = U (bfE′′ ◦bµE′E′′ ↾C∞(RE′ )) ◦ E′ f ◦ ιE′ = U (bfE′′ ◦bµE′E′′ ↾C∞(RE′ )) ◦ E′ , C∞−homomorphism bf : (C∞(RE), ΦE) → (A, Φ) such that: and the diagram commutes. E′ = ιE′ , so we get: By the universal property of the colimit (C∞(RE), ΦE), there is a unique (C∞(RE′ ), ΦE′ ) bµE′E′′ ↾ C∞ (RE′ ) (C∞(RE′′ ), ΦE′′ ) ❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯❯ ℓE′′ t✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐✐ ℓE′′ (C∞(RE), ΦE) bfE′ bf (A, Φ) bfE′′ Now we need only to show that U (bf ) ◦ E = f . Given any x ∈ E, there is some finite E′ ⊆f E such that x ∈ E′. We have the following commutative diagram: E′ E′ / U (C∞(RE′ ), ΦE′ ) ℓE′ U (C∞(RE), ΦE) ❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱ U(bfE′ ) U(bf ) U (A, Φ) so: Since ℓE′ ◦ E′ = E ◦ ιE′ and U (bfE′ ) ◦ E′ = f ◦ ιE′ , we get: U (bf ) ◦ ℓE′ ◦ E′ = U (bfE′ ) ◦ E′ U (bf ) ◦ E ◦ ιE′ = f ◦ ιE′ / / * . . t p p / / / * A Universal Algebraic Survey of C∞−Rings 47 thus: U (bf ) ◦ E(x) = U (bf ) ◦ E ◦ ιE′ (x) = f (ιE′ (x)) = f (x) and the diagram commutes. The uniqueness of bf comes from its construction. forgetful functor U : C∞Rng → Set. In the following proposition, we present a description of a left adjoint to the Proposition 11. The functions: L0 : Obj (Set) → Obj (C∞Rng) 7→ (C∞ (RX ), ΦX ) X and L1 : Mor (Set) → Mor (C∞Rng) (X f → Y ) 7→ (C∞(RX ), ΦX ) ef → (C∞(RY ), ΦY ) where ef : L0(X) → L0(Y ) is the unique C∞−homomorphism given by the universal property of the free C∞−ring X : X → C∞(RX ): given the function Y ◦ f : X → U (C∞(RY )), there is a unique C∞−homomorphism such that the following diagram commutes: X X U (C∞(RX )) (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ Y ◦f U(ef ) U (C∞(RY )) define a functor L : Set → C∞Rng which is left adjoint to the forgetful functor U : C∞Rng → Set. Proof. We claim that L1(idX ) = id(C∞(RX ),ΦX ). On the one hand, gidX is the only C∞−homomorphism such that: U (C∞(RX )) X X (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ X ◦idX U(gidX ) U (C∞(RX )) commutes. / / ( ✤ ✤ ✤ / / ( ✤ ✤ ✤ 48 J. C. Berni and H. L. Mariano On the other hand, since id(C∞(RX ),ΦX ) is a C∞−homomorphism which makes the following diagram to commute: X it follows that: X U (C∞(RX )) (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ X U(id(C∞ (RX ),ΦX )) U (C∞(RX )) Now, let X, Y, Z be sets and f : X → Y and g : Y → Z be functions. We claim that: L1(idX ) =gidX = id(C∞(RX ),ΦX ) L1(g ◦ f ) = L1(g) ◦ L1(f ) C∞−homomorphism such that: By definition, L1(f ) = ef : (C∞(RX ), ΦX ) → (C∞(RY ), ΦY ) is the only X X U (C∞(RX )) (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ Y ◦f U(ef ) U (C∞(RY )) commutes, that is, such that U (ef ) ◦ X = Y ◦ f . Also, by definition, L1(g) = eg : (C∞(RY ), ΦY ) → (C∞(RZ ), ΦZ ) is the only C∞−homomorphism such that: U (C∞(RY )) Y Y (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ Z ◦g U(eg) U (C∞(RZ)) commutes, that is, such that U (eg) ◦ Y = Z ◦ g. Finally, ]g ◦ f : (C∞(RX ), ΦX ) → (C∞(RZ ), ΦZ ) is the unique C∞−homo- morphism such that: X X U (C∞(RX )) (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ Z ◦(g◦f ) U(]g◦f ) U (C∞(RZ )) commutes, that is, such that U (]g ◦ f ) ◦ X = Z ◦ (g ◦ f ). / / ( ✤ ✤ ✤ / / ( ✤ ✤ ✤ / / ( ✤ ✤ ✤ / / ( ✤ ✤ ✤ A Universal Algebraic Survey of C∞−Rings 49 this property. We are going to show that eg ◦ ef : (C∞(RX ), ΦX ) → (C∞(RZ ), ΦZ ) also has In fact, U (ef ) ◦ X = Y ◦ f ⇒ U (eg) ◦ (U (ef ) ◦ X ) = (U (eg) ◦ Y ) ◦ f ⇒ ⇒ (U (eg) ◦ U (ef )) ◦ X = (Z ◦ g) ◦ f and thus By uniqueness it follows that: U (eg ◦ ef ) ◦ X = Z ◦ (g ◦ f ). L1(g ◦ f ) = L1(g) ◦ L1(f ), so L = (L0, L1) defines a functor. We claim that: φX,(A,Φ) : HomSet (X, A) → HomC∞Rng ((L(X), ΦX ), (A, Φ)) (X ϕ → A) 7→ (L(X), ΦX ) eϕ → (A, Φ) is a natural bijection, so L ⊣ U . where eϕ is the unique C∞−homomorphism such that U (eϕ) ◦ X = ϕ φX,(A,Φ) is surjective. Given any C∞−homomorphism eϕ : (L(X), ΦX ) → (A, Φ), taking ϕ := U (eϕ) ◦ X : X → A we have L(ϕ) = eϕ. Given ϕ, ψ : X → A such that φX,(A,Φ)(ϕ) = eϕ = eψ = φX,(A,Φ)(ψ), we have: U (eϕ) = U (eψ) ϕ = U (eϕ) ◦ X = U (eψ) ◦ X = ψ and so φX,(A,Φ) is a bijection. Let (A, Φ) be a C∞−ring and let X ⊆ A. Given ιA X : X ֒→ A, there is a X : (L(X), ΦX ) → (A, Φ) such that the following unique C∞−homomorphism fιA diagram commutes: X ηX (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ ιA X U (L(X)) X ) U(fιA U (A, Φ) (7) / / ( 50 J. C. Berni and H. L. Mariano We claim that if hXi = A, then fιA Indeed, X is surjective. X = ιA X [X] = im(ιA X ) and since U (fιA X ) ◦ ηX = ιA X , we have: X )) Since the diagram (7) commutes, on the other hand, thus im(ιA X ) ◦ ηX ). im(U (fιA X ) = im(U (fιA X ) ◦ ηX ) ⊆ im (U (fιA X ⊆ im (U (fιA hXi ⊆ him (U (fιA Since X generates A and him (U (fιA X ))i =fιA X ))i = im (fιA X )) and X [A] it follows that:: X ) is a C∞−subring of (A, Φ), hXi = A ⊆ im(fιA X is surjective. X ) ⊆ A X ) = A and ιA so im (fιA In particular, taking X = A yields ιA A = idA, and since εA = φA,(A,Φ)(idA) = ^id(A,Φ), we have: A ηA (◗◗◗◗◗◗◗◗◗◗◗◗◗◗◗ idA U (L(A)) U(εA) U (A, Φ) so im (εA) = (A, Φ), and εA is surjective. Now, given any C∞−ring (A, Φ) we have the surjective morphism: εA : L(U (A, Φ)) ։ (A, Φ). We have seen that since εA is a C∞−homomorphism, ker(εA) is a C∞−congru- ence. By the Fundamental Theorem of the C∞−Isomorphism we have: (A, Φ) ∼= L(A) ker(εA) / / ( A Universal Algebraic Survey of C∞−Rings 51 4 Other Constructions In this section we describe further results and constructions involving C∞−rings. 4.1 Ring-Theoretic Ideals and C∞−Congruences Our next goal is to classify the congruences of any C∞−ring. We shall see that they are classified by their ring-theoretic ideals. We begin with the finitely generated case. An interesting result, which is a consequence of Hadamard's Lemma, The- orem 7, is the description of the ideals of finitely generated C∞−rings in terms of C∞−congruences. For the reader's benefit, we state and prove the following result: Theorem 7. (Hadamard's Lemma)For every smooth function f ∈ C∞(Rn) there are smooth functions g1, · · · , gn ∈ C∞(R2n) such that ∀(x1, · · · , xn), (y1, · · · , yn) ∈ Rn f (x1, · · · , xn) − f (y1, · · · , yn) = nXi=1 (xi − yi) · gi(x1, · · · , xn, y1, · · · , yn) Proof. Holding (v1, · · · , vn) = v ∈ Rn fixed, we define: h(t) = f (y + t · (x − y)). We have: f (x) − f (y) =Z 1 0 h′(t)dt =Z 1 0 ∂f ∂vi nXi=1 (y + t · (x − y))) · (xi − yi)dt, where the second equality uses the chain rule. The result follows by putting: gi(x, y) =Z 1 0 ∂f ∂vi (y + t · (x − y))dt Proposition 12. Given a finitely generated C∞−ring (A, Φ), considering the forgetful functor given in Remark 1, eU : C∞Rng → CRing, we have that if I is a subset of A that is an ideal (in the ordinary ring-theoretic sense) in eU (A, Φ), then bI = {(a, b) ∈ A × Aa − b ∈ I} is a C∞−congruence in A. 52 J. C. Berni and H. L. Mariano Proof. (Cf. p. 18 of [24]) Let n ∈ N and f ∈ C∞(Rn, R). ai − bi ∈ I, then: We have to show that if for every i ∈ {1, · · · , n} we have (ai, bi) ∈bI, that is, Φ(2)(f )((a1, b1), · · · , (an, bn)) = (Φ(f )(a1, · · · , an), Φ(f )(b1, · · · , bn)) ∈bI, that is, Φ(f )(a1, · · · , an) − Φ(f )(b1, · · · , bn) ∈ I. Suppose that (a1, b1), · · · , (an, bn) ∈ A × A are such that ai − bi ∈ I for every i ∈ {1, · · · , n}. By Hadamard's Lemma, there are smooth functions g1, · · · , gn : Rn×R → R such that for all x = (x1, · · · , xn) and y = (y1, · · · , yn) ∈ Rn, f (x) − f (y) = Φ preserves this equation, i.e., (xi − yi) · gi(x, y). nXi=1 Φ(f )(a1, · · · , an) − Φ(f )(b1, · · · , bn) = (ai − bi) ·Φ(gi)(a1, · · · , an, b1, · · · , bn) ∈ I, = nXi=1 and the result follows. ∈I {z } The following result tells us that the ideals of the C∞−rings classify its con- gruences. Proposition 13. Given any finitely generated C∞−ring (A, Φ), let Cong (A, Φ) denote the set of all the C∞−congruences in A and let I(A, Φ) denote the set of all ideals of A. The following map is a bijection: ψA : Cong (A, Φ) → I(A, Φ) R 7→ {x ∈ A(x, 0) ∈ R} whose inverse is given by: ϕA : I(A, Φ) → Cong (A, Φ) I 7→ {(x, y) ∈ A × Ax − y ∈ I} Proof. Proposition 12 assures that ϕA maps ideals of (A, Φ) to C∞−congruences. A Universal Algebraic Survey of C∞−Rings 53 Also, given a C∞−congruence R in A, we have that ψA(R) is an ideal. Claim: {x ∈ A(x, 0) ∈ R} is closed under Φ(+) and Φ(−), since for any x, y ∈ A such that (x, 0), (y, 0) ∈ R, we have Φ(2)(+)((x, 0), (y, 0)) = (Φ(+)(x, y), Φ(+)(0, 0)) = (x + y, 0) ∈ R, for R is a C∞−congruence and (R, Φ(2)′ ) is a C∞−subring of (A × A, Φ(2)). Also, given x ∈ A such that (x, 0) ∈ R, since (R, Φ(2)′ ) is a C∞−subring of (A × A, Φ(2)), we have Φ(2)′ (−)(x, 0) = (−x, 0) ∈ R. Finally, given x ∈ A such that (x, 0) ∈ R and any y ∈ A, we have both ) is a (x, 0), (y, y) ∈ R, and since R is a C∞−congruence, and thus (R, Φ(2)′ C∞−subring of (A × A, Φ(2)), we have: Φ(2)′ (·)((x, 0), (y, y)) = (x · y, y · 0) = (x · y, 0) ∈ R so x · y ∈ {x ∈ A(x, 0) ∈ R}. Hence {x ∈ A(x, 0) ∈ R} is an ideal in (A, Φ). Now we are going to show that ψA and ϕA are inverse functions. Given R ∈ Cong (A, Φ), we have: ϕA(ψA(R)) = ϕA({x ∈ A(x, 0) ∈ R}) = {(x, y) ∈ A × A(x − y, 0) ∈ R} = R so ϕA ◦ ψA = idCong (A,Φ). Given I ∈ I (A), we have: ψA(ϕA(I)) = ψA({(x, y) ∈ A × Ax − y ∈ I}) = {z ∈ Az − 0 ∈ I} = I, so ψA ◦ ϕA = idI (A,Φ). For any C∞−ring (A, Φ) we have the function: ψA : Cong (A, Φ) → I(A, Φ) R 7→ {a ∈ A(a, Φ(0)) ∈ R} and whenever (A, Φ) is a finitely generated C∞−ring, we have seen in Propo- sition 13 that ψA has ϕA as inverse - which is a consequence of Hadammard's Lemma. In order to show that ψA is a bijection for any C∞−ring (A, Φ), first we decompose it as a directed colimit of its finitely generated C∞−subrings (cf. Theorem 8): 54 J. C. Berni and H. L. Mariano (A, Φ) ∼= lim −→ (Ai,Φi)⊆f.g.(A,Φ) (Ai, Φ ↾Ai) and then we use Proposition 13 to obtain a bijection: lim ←− (Ai,Φi)⊆f.g.(A,Φ) I(Ai, Φi) → lim ←− (Ai,Φi)⊆f.g.(A,Φ) Cong (Ai, Φi) such that for every (Ai, Φi) ⊆f .g. (A, Φ) the following diagram commutes: eϕ : lim ←−(Ai,Φi)⊆f.g.(A,Φ) I(Ai, Φi) I(Ai, Φi) eϕ ϕAi / lim ←−(Ai,Φi)⊆f.g.(A,Φ) Cong (Ai, Φi) / Cong(Ai, Φi) where the vertical downward arrows are the canonical arrows of the projec- tive limit. Finally we show that there is a bijective correspondence, α, between lim ←− (Ai,Φi)⊆f.g.(A,Φ) I(Ai, Φi) −→(Ai,Φi)⊆f.g.(A,Φ) lim ←−(Ai,Φi)⊆f.g.(A,Φ) −→(Ai,Φi)⊆f.g.(A,Φ) and I(cid:16)lim the bijections α, β,eϕ and ψ are complete lattices isomorphisms. (Ai, Φi)(cid:17), and a bijective correspondence, β, between (Ai, Φi)(cid:17). In fact, Cong (Ai, Φi) and Cong(cid:16)lim By composing these bijections we prove that the congruences of (C∞(RE), ΦE) are classified by the ring-theoretic ideals of (C∞(RE), ΦE). Theorem 8. Let (A, Φ) be any C∞−ring. There is a directed system of C∞−rings, ((Ai, Φi), αij ), where each (Ai, Φi) is a finitely generated C∞−ring and each αij : (Ai, Φi) → (Aj , Φj) is a monomorphism such that: (A, Φ) ∼= lim −→ (Ai, Φi) Proof. Let Pf be the set of all finite subsets of A, on which we consider the partial order defined by inclusion, i.e., (∀i ∈ I)(∀j ∈ I)(i (cid:22) j ↔ i ⊆ j). Let (Ai, Φi) := hii denote the intersection of all C∞−subrings (B, ΦB) of (A, Φ) which contain i, i.e., (Ai, Φi) = (hii, Φi) = \i⊂B (B, Φ′) / / A Universal Algebraic Survey of C∞−Rings 55 where Φi = Φ′ is the C∞−structure described in Proposition 1 Note that I = {hii = (Ai, Φi)i ∈ Pf }, partially ordered by the inclusion, is a directed set, since it is not empty and given any hii and hji we can take, for instance, k = i ∪ j ∈ Pf such that hii ⊆ hki and hji ⊆ hki. By Proposition 2, and using its notation, (hii, Φ′) [i∈Pf is a C∞−subring of (A, Φ). By construction, each (Ai, Φi) is a finitely gener- ated C∞−subring of (A, Φ). We clearly have the following set-theoretic equality: hence the equality Φ′ = Φ, so: A = U (A, Φ) = U[i∈Pf (A, Φ) = [i∈Pf (hii, Φ′) = [i∈Pf hii, (hii, Φ′) For each i ∈ I we define Ai = hii and take D(i) := (Ai, Φi). Whenever i ⊆ j, we have the inclusion D(i ⊆ j) := αij : (hii, Φi) ֒→ (hji, Φj ) which makes the following triangle commute: (hii, Φi) αij / (hji, Φj ) &◆◆◆◆◆◆◆◆◆◆◆ αi x♣♣♣♣♣♣♣♣♣♣♣ αj (cid:0)∪i∈Pf Ai, Φ′(cid:1) where αi : (hii, Φi) → (A, Φ) and αj : (hji, Φj ) → (A, Φ) are the C∞−homo- morphisms of inclusion. Consider the following diagram, where Pf is viewed as a category: D : Pf → C∞Rng i → j 7→ (hii, Φi) αij→ (hji, Φj ) We claim that: D(i) ∼= (A, Φ) lim −→ i∈Pf Let (B, Ψ ) be any C∞−ring and hi : (hii, Φi) → (B, Ψ ) and hj : (hji, Φj ) → (B, Ψ ) be any two C∞−homomorphisms. We are going to show that there exists & / x 56 J. C. Berni and H. L. Mariano a unique C∞−homomorphism h : (∪i∈Pf Ai, Φ) → (B, Ψ ) such that the following diagram commutes: (hii, Φi) αij / (hji, Φj ) &▼▼▼▼▼▼▼▼▼▼▼ αi x♣♣♣♣♣♣♣♣♣♣ αj (cid:0)∪i∈Pf Ai, Φ(cid:1) ∃!h hi hj (B, Ψ ) We now define the function which underlies h in the following way: given a ∈ ∪i∈Pf Ai there is some i0 ∈ Pf such that a ∈ i0. For any j0 ∈ Pf such that a ∈ j0, since I is directed, there is some k0 ∈ Pf such that k0 ⊇ i0, j0, so we have a ∈ k0 and the following commutative diagram: hi0i αi0 k0 / "❊❊❊❊❊❊❊❊ hi0 hk0i αj0 k0 hj0i hk0 ②②②②②②②② hj0 B so: hi0 (a) = hk0 (αi0k0 (a)) = hk0 (a) = hk0 (αj0k0 (a)) = hj0 (a). Hence, given a ∈ ∪i∈Pf Ai, choose any i0 ∈ Pf such that a ∈ i0 and define h(a) = hi0 (a) (up to here h is merely a function, not a C∞−homomorphism). Now we prove that h is a C∞−homomorphism. Let f ∈ C∞(Rn, R) be any n−ary function symbol. We must show that the following diagram commutes: (cid:0)∪i∈Pf Ai(cid:1)n h(n) Bn Φ(f ) Ψ (f ) / ∪i∈Pf Ai h / B that is, (∀a1)(∀a2) · · · (∀an)(Ψ (f )(h(a1), h(a2), · · · , h(an))) = h(Φ(f )(a1, a2, · · · , an)). Let i1, i2, · · · , in ∈ Pf be such that a1 ∈ hi1i, a2 ∈ hi2i, · · · , an ∈ hini, and take j ∈ Pf such that j ⊇ i1, · · · , in, so a1, a2, · · · , an ∈ hji + + & / x s s / " o o / / A Universal Algebraic Survey of C∞−Rings 57 For any ℓ ∈ {1, 2, · · · , n}, hi(aℓ) = hj(aℓ), so h(a1) = hj(a1), h(a2) = hj(a2), · · · , h(an) = hj(an). We have: Ψ (f )(h(a1), h(a2), · · · , h(an)) = Ψ (f )(hj(a1), hj(a2), · · · , hj(an)) (1) = = hj(Φj(f )(a1, a2, · · · , an)) (2) = hj(Φ(f )(a1, a2, · · · , an)) (3) = h(Φi(f )(a1, a2, · · · , an)) where (1) is due to the fact that hj : (hii, Φi) → (B, Ψ ) is a C∞−homomorphism, (2) is due to the fact that (hji, Φj ) is a C∞−subring of (∪i∈Pf Ai, Φ) and (3) oc- curs by the very definition of h. The uniqueness is granted by the property that h must satisfy as a function. Under those circumstances, so (Ai, Φi) −→ i∈I [i∈Pf (A, Φ) =[i∈Pf hii, Φ ∼= lim hii, Φ ∼= lim −→ i∈Pf (Ai, Φi) Remark 8. Let ((Ai, Φi)i∈I , αij : (Ai, Φi) → (Aj, Φj)) be an inductive directed system of C∞−rings, and let (Jℓ)ℓ∈I ∈ lim I(Aℓ, Φℓ). ←−ℓ∈I For every i ∈ I, we have the maps: αi ∗ : I(cid:16)lim bα : I(cid:16)lim −→i∈I J −→i∈I J (Ai, Φi)(cid:17) → I(Ai, Φi) (Ai, Φi)(cid:17) → I(Ai, Φi) 7→ α⊣ 7→ (α⊣ i [J])i∈I i [J] and the following limit diagram: lim ←−i∈I α∗ i u❥❥❥❥❥❥❥❥❥❥❥❥❥❥ I(Ai, Φi) , α∗ j )❚❚❚❚❚❚❚❚❚❚❚❚❚❚ I(Aj , Φj) α∗ ij I(Ai, Φi) where: αij ∗ : I(Aj , Φj) → I(Ai, Φi) ⊣[Jj] 7→ αij Jj u ) o o 58 J. C. Berni and H. L. Mariano We note, first, that α maps ideals of lim In fact, given any ideal J ∈ I(cid:16)lim i (J))i∈I ∈Yi∈I C∞−homomorphism, it follows that for every i ∈ I, α∗ (Ai, Φi), so α(J) = (α∗ −→i∈I −→i∈I is compatible, since: (Ai, Φi) to an element of lim (Ai, Φi)(cid:17), since for every i ∈ I, αi is a i (J) = αi ←−i∈I ⊣[J] is an ideal of i (J))i∈I I(Ai, Φi). Moreover, the family (α∗ I(Ai, Φi). (∀i ∈ I)(∀j ∈ I)(i (cid:22) j)(αj ◦ αij = αi ⇒ α∗ i = αij ∗ ◦ α∗ j ) (∀i ∈ I)(∀j ∈ I)(i (cid:22) j)(αi ∗(J) = (αij ∗ ◦ α∗ j )(J) = αij ∗(α∗ j (J)) and so α(J) = (α∗ i (J))i∈I ∈ lim ←− i∈I I(Ai, Φi). Proposition 14. Let (I, (cid:22)) be a directed partially ordered set and {(Ai, Φi), αij : (Ai, Φi) → (Aj, Φj )}i,j∈I be a directed inductive system of C∞−rings and C∞−homomorphisms. For every i ∈ I, we have the map: αi ∗ : I(cid:16)lim −→i∈I J 7→ α⊣ (Ai, Φi)(cid:17) → I(Ai, Φi) I(Ai, Φi), there is a unique α : I(cid:16)lim i [J] −→i∈I (Ai, Φi)(cid:17) → By the universal property of lim ←−i∈I lim ←−i∈I I(Ai, Φi) such that for every i ∈ I the following diagram commutes: I(cid:16)lim −→i∈I ∃!α (Ai, Φi)(cid:17) α∗ i lim ←−i∈I I(Ai, Φi) ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ I(Ai ,Φi ) πi↾lim←−i∈I I(Ai, Φi) that is, such that αi We have, thus: ∗ = πi ↾lim←−i∈I I(Ai,Φi) ◦α. α : I(cid:16)lim −→i∈I J (Ai, Φi)(cid:17) → lim ←−i∈I 7→ (α∗ I(Ai, Φi) i (J))i∈I A Universal Algebraic Survey of C∞−Rings 59 For any (Ji)i∈I ∈ lim ←−i∈I and the map: (Ai, Φi) −→i∈I I(Ai, Φi), Si∈I αi[Ji] is an ideal of lim I(Ai, Φi) → I(cid:16)lim (Ai, Φi)(cid:17) 7→ Si∈I αi[Ji] −→i∈I (Ji)i∈I α′ : lim ←−i∈I is an inverse for α, so α is a bijection. Proof. In Set, we have: U lim −→ i∈I (Ai, Φi)! = Si∈I Ai × {i} ∼ where: and (ai, i) ∼ (aj , j) ⇐⇒ (∃k ∈ I)(i, j (cid:22) k)(αik(ai) = αjk(aj)) U lim ←− i∈I I(Ai, Φi)! =((Jℓ)ℓ∈I ∈Yi∈I I(Ai, Φi)(∀i, j ∈ I)(i (cid:22) j)(αij ∗(Jj) = Ji)) Given the following colimit diagram: (Ai, Φi) lim −→i∈I αi 5❦❦❦❦❦❦❦❦❦❦❦❦❦❦ αij αj i❙❙❙❙❙❙❙❙❙❙❙❙❙❙ / (Aj, Φj) (Ai, Φi) with: αi : (Ai, Φi) → lim (Ai, Φi) −→i∈I ai 7→ [(ai, i)] consider the following limit diagram: lim ←−i∈I α∗ i u❥❥❥❥❥❥❥❥❥❥❥❥❥❥ I(Ai, Φi) , α∗ j )❚❚❚❚❚❚❚❚❚❚❚❚❚❚ I(Aj , Φj) α∗ ij I(Ai, Φi) where: αij ∗ : I(Aj , Φj) → I(Ai, Φi) ⊣[Jj] 7→ αij Jj 5 / i u ) o o 60 J. C. Berni and H. L. Mariano Given any (Ji)i∈I , we are going to show that α′((Ji)i∈I ) ∈ I(cid:16)lim (Ai, Φi)(cid:17). (Ai, Φi) is directed, given x, y ∈ Si∈I αi[Ji], there are i, j ∈ I, xi ∈ Ji and yj ∈ Jj such that αi(xi) = [(xi, i)] = x and αj(yj) = [(yj, j)] = y. Also, since the colimit is directed, there is k ∈ I with i, j ≤ k such that: −→i∈I Since lim −→i∈I αik(xi) ∈ αik[Ji], αjk(yj) ∈ αjk[Jj], so αk(αik(xi)), αk(αjk(yj)) ∈ αk[Jk] Since αk[Jk] is an ideal in αk[Ak], it follows that: αk(αik(xi)) − αk(αjk(yj)) ∈ αk[Jk] and x − y = αk(αik(xi)) − αk(αjk(yj)) ∈ αk[Jk] ⊆[i∈I αi[Ji]. αi[Ji] there are some i ∈ I, xi ∈ Ji such that x = αi(xi) = (Ai, Φi), there are some j ∈ I and some aj ∈ Aj Given x ∈ lim −→i∈I [(xi, i)], and given a ∈ lim such that a = αj(aj) = [(aj, j)]. −→i∈I Since the colimit is directed, there is some k ∈ I with i, j ≤ k such that: αik(xi) ∈ αik[Ji], αjk(aj) ∈ αjk[Aj] so and αi(xi) = αk(αik(xi)) ∈ αi[αik[Ji]] = αk[Jk] αj(aj) = αk(αjk(aj)) ∈ αj[αjk[Jj]] = αk[Jk] Since αk[Jk] is an ideal of αk[Ak], it follows that: αk(αik(xi)) · αk(αjk(aj)) ∈ αk[Jk]. Thus, x · a = αi(xi) · αj (aj) = αk(αik(xi)) · αk(αjk(aj)) = = αk(αik(xi)) · αk(αjk(aj )) ∈ αk[Jk] ⊆[i∈I αi[Ji] so α′ maps elements of lim ←−i∈I I(Ai, Φi) to ideals of lim −→i∈I (Ai, Φi). Claim: α ◦ α′ = idlim←−i∈I Given (Jℓ)ℓ∈I ∈ lim ←−i∈I so A Universal Algebraic Survey of C∞−Rings 61 I(Ai,Φi). I(Ai, Φi), we have: α′((Jℓ)ℓ∈I ) :=[ℓ∈I αℓ[Jℓ], α(α′((Jℓ)ℓ∈I )) = α [ℓ∈I αℓ[Jℓ]! := αi ⊣"[ℓ∈I αℓ[Jℓ]#!i∈I . We are going to show that for every i ∈ I, Given i ∈ I, on the one hand we have: αi ∗(α′((Jℓ)ℓ∈I )) = Ji Ji ⊆ αi ⊣[αi[Ji]] ⊆ αi ⊣"[ℓ∈I αℓ[Jℓ]# = αi ∗(α′((Jℓ)ℓ∈I )) =[ℓ∈I αi ⊣[αℓ[Jℓ]] On the other hand, given ai ∈ αi ⊣(cid:2)Sℓ∈I αℓ[Jℓ](cid:3), that is, ai ∈ Ai such that αi(ai) ∈ Sℓ∈I αℓ[Jℓ], there is some j ∈ I and some bj ∈ Jj ∗(α′((Jℓ)ℓ∈I )) = αi such that: αi(ai) = αj(bj). Since the system is directed, there is some k ∈ I, i, j (cid:22) k such that αik(ai) = αjk(bj). By compatibility we have αjk ⊣[Jk] = αjk ∗(Jk) = Jj, so in particular α⊣ jk[Jk] ⊇ Jj and αjk(bj) ∈ αjk[Jj] ⊆ αjk[αjk ⊣[Jk]] ⊆ Jk. Thus, taking ck = αjk(bj), we have: ck = αjk(bj) ∈ α⊣ jk[Jj] ⊆ Jk, that is, For any k′ ≥ k we have: ck ∈ Jk. so αik′ (ai) = αkk′ (ck) ∈ Jk′ ai ∈ αik′ ⊣[Jk′ ] = Ji. 62 J. C. Berni and H. L. Mariano Hence: and (∀i ∈ I) [ℓ∈I ⊣[αℓ[Jℓ]] = Ji! αi α(α′((Ji)i∈I )) = (Ji)i∈I . Since (Ji)i∈I is arbitrary, we have: α ◦ α′ = idlim←−i∈I I(Ai,Φi). On the other hand, given J ∈ I(cid:16)lim −→i∈I (Ai, Φi)(cid:17), we have α(J) = (αi and: ∗(J))i∈I α′(α(J)) = α′((αi ∗(J))i∈I ) =[i∈I αi[αi ∗(J)]. On the one hand, since for every i ∈ I, αi[αi ∗(J)] = αi[αi ⊣[J]] ⊆ J, we have α′(α(J)) ⊆ J. Recall that given J ∈ I(cid:16)lim −→i∈I (Ai, Φi)(cid:17), we have: (Ai, Φi) =[i∈I αi[Ai], J ⊆ lim −→ i∈I so given x ∈ J ⊆Si∈I αi[Ai], there are some i0 ∈ I and some xi0 ∈ αi0 ⊣[J] such that: ∗(J) = and since xi0 ∈ α∗ i0 (J), we have [(xi0 , i0)] ∈ αi0 [αi0 x = [(xi0 , i0)] ∗(J)] ⊆ Si∈I αi[α∗ i (J)], αi[α∗ i (J)] = α′((αi ∗(J))i∈I ) = α′(α(J)) x ∈[i∈I and: αi0 so α′(α(J)) = J. Since J is an arbitrary ideal of lim ←−i∈I I(Ai, Φi), it follows that: α′ ◦ α = idI(cid:16)lim−→i∈I (Ai,Φi)(cid:17). so α is a bijection whose inverse is α′. A Universal Algebraic Survey of C∞−Rings 63 The proof of the following proposition is similar to the proof we just made, so we are going to omit it. Proposition 15. Let (I, (cid:22)) be a directed partially ordered set and {(Ai, Φi), αij : (Ai, Φi) → (Aj, Φj)}i,j∈I be a directed inductive system of C∞−rings and C∞−homomorphisms. The following function is a bijection: β : Cong(cid:16)lim −→i∈I R (Ai, Φi)(cid:17) → lim ←−i∈I Cong(Ai, Φi) 7→ ((αi × αi)⊣(R))i∈I whose inverse is given by: β′ : lim ←−i∈I Cong(Ai, Φi) → Cong(cid:16)lim (Ri)i∈I 7→ (Ai, Φi)(cid:17) Ri −→i∈I lim −→i∈I The following result extends Proposition 13 in the sense that it shows us, with details, that the congruences of any free C∞−ring are classified by their ring-theoretic ideals (in the finitely generated case it follows from Hadamard's lemma, and this case is used here). Lemma 2. The congruences of (C∞(RE), ΦE), the free C∞−ring determined by the set E, are classified by their ring-theoretic ideals. Proof. By Proposition 12, for any finite E the result holds. Suppose E is any set (not necessarily finite). We have, by definition, (C∞(RE), ΦE) ∼= lim −→ E′⊆f E (C∞(RE′ , R), ΦE′ ) where ΦE is the C∞−structure of the colimit. The family of bijections {ψE′ : Cong(C∞(RE′ ), ΦE′ ) → I(C∞(RE′ ), ΦE′ )} always exist, so it yields a coherent bijection: Cong(C∞(RE′ lim ←− E′⊆f E ), ΦE′ ) → lim ←− E′⊆f E I(C∞(RE′ ), ΦE′ ). We have the following commutative diagram: eψ : Cong (C∞(RE′ ), ΦE′ ) lim ←−E′⊆f E Cong(C∞(RE′ ), ΦE′ ) α Cong(C∞(RE), ΦE) ψE′ eψ ψE / I(C∞(RE′ ), ΦE′ ) lim ←−E′⊆f E I(C∞(RE′ ), ΦE′ ) β / I(C∞(RE), ΦE) / / / O O O O O O / O O 64 J. C. Berni and H. L. Mariano where the two upper vertical arrows are the canonical arrows of the projec- tive limit, α is given in Proposition 14 and β is given in Proposition 15. I(C∞(RE), ΦE) is a bijection. Since α, β and eψ are bijections, it follows that ψE : Cong(C∞(RE), ΦE) → The following lemma is a well-known result of Universal Algebra applied to C∞−rings: Lemma 3. Let (A, Φ) be a C∞−ring and let R ∈ Cong (A, Φ). Given the quo- tient C∞−homomorphism: qR : (A, Φ) →(cid:18) A R , Φ(cid:19) 7→ x + R x we have the bijection: Cong(cid:18) A , Φ(cid:19) R R × A 7→ {(qR(s), qR(t)) ∈ A R (s, t) ∈ S} (qR)∗ : {S ∈ Cong (A, Φ)R ⊆ S} → S whose inverse is given by: (qR)∗ : Cong(cid:18) A R S′ , Φ(cid:19) → {S ∈ Cong(A, Φ)R ⊆ S} (qR × qR)⊣[S′] 7→ As a consequence of the above lemma, we have: Lemma 4. Let (A, Φ) be a C∞−ring and R ∈ Cong (A, Φ), and suppose that ψA : Cong (A, Φ) → I(A, Φ) is a bijection with an inverse, ϕA : I(A, Φ) → Cong (A, Φ). Under those circumstances, the quotient C∞−homomorphism: induces a pair of inverse bijections: R , Φ(cid:19) qR : (A, Φ) →(cid:18) A , Φ(cid:19) → {I ′ ∈ I(A, Φ)ψA(R) ⊆ I ′} 7→ qR[J] (qR)+ : I(cid:18) A R J and (pR)− : {I ′ ∈ I(A, Φ)ψA(R) ⊆ I ′} → I(cid:18) A R , Φ(cid:19) 7→ (qR)⊣[J ′] J ′ A Universal Algebraic Survey of C∞−Rings 65 Proof. The proof is a direct consequence of Lemma 3 and a diagram chase. Proposition 16. Let (A, Φ) and (B, Ψ ) be two C∞−rings and let h : (A, Φ) → (B, Ψ ) be a surjective C∞−homomorphism. The following functions are bijec- tions: h∗ : Cong (B, Ψ ) → {S ∈ Cong (A, Φ) ker(h) ⊆ S} R 7→ (h × h)⊣[R] h− : I(A, Φ) → {I ′ ∈ I(B, Ψ ) ϕA(ker(h)) ⊆ I ′} J 7→ h⊣[J] Proof. By the Theorem of the C∞−Isomorphism, we have the C∞−isomorphism: A h : ( , Φ) → (B, Ψ ) ker(h) a + ker(h) 7→ h(a) so α : Cong(cid:18) A ker(h) S′ , Φ(cid:19) → Cong(B, Ψ ) 7→ {(h(s), h(s′)) (s, s′) ∈ S′} is a bijection whose inverse is given by: β : Cong (B, Ψ ) → R −1 7→ {(h −1 (r), h Cong(cid:18) A ker(h) , Φ(cid:19) (r′))(r, r′) ∈ R} = {(a + ker(h), a′ + ker(h))(h(a), h(a′)) ∈ R} Taking R = ker(h) in the Lemma 3 yields a bijection (qker(h))∗ : Cong(cid:18) A ker(h)(cid:19) → {S ∈ Cong (A) ker(h) ⊆ S}. Now it suffices to prove that h∗ = (qker(h))∗ ◦ β. Since h is an isomorphism, note that β = h h∗ = q∗ ker(h) ◦ h ∗ , as we claimed. ∗ , and as h = h ◦ qker(h), we get The proof for h− is analogous, since h− = (qker(h))− ◦ h − . The following proposition gives us a description of C∞−rings via generators and relations. Proposition 17. Let (A, Φ) be any C∞−ring. The C∞−congruences of (A, Φ) are classified by the ring-theoretic ideals of (A, Φ). 66 J. C. Berni and H. L. Mariano Proof. Given any C∞−ring (A, Φ), we have the following isomorphism: εA : C∞(RU(A)) ker(εA) ∼=→ A so we can write: L(A) ker(εA) By Lemma 2, we have a bijection: A ∼= = C∞(RA) ker(εA) . ψA : Cong (C∞(RA), ΦA) → I(C∞(RA), ΦA) Note that given any R ∈ Cong (C∞(RA), ΦA) such that ker(εA) ⊆ R, we have ψA(ker(εA)) = {g ∈ C∞(RA)(g, 0) ∈ ker(εA)} ⊆ {g ∈ C∞(RA)(g, 0) ∈ R} = ψA(R), so: ψA[{R ∈ Cong (C∞(RA), ΦA) ker(εA) ⊆ R}] ⊆ {I ′ ∈ I(C∞(RA), ΦA) ψA(ker(εA)) ⊆ I ′}. Also, let I ′ ∈ I(C∞(RA), ΦA) be such that ψA(ker(εA)) = {g ∈ C∞(RA) (g, ΦA(0)) ∈ ker(εA)} ⊆ I ′. Given (g1, g2) ∈ ker(εA), we have (g1 − g2, ΦA(0)) ∈ ker(εA), and since {g ∈ C∞(RA) (g, ΦA(0)) ∈ ker(εA)} ⊆ I ′, g1 − g2 ∈ I ′, so ker(εA) ⊆ ϕA(I ′) and : ϕA[{I ′ ∈ I(C∞(RA), ΦA) ψA(ker(εA)) ⊆ I ′}] ⊆ {R ∈ Cong (C∞(RA), ΦA) ker(εA) ⊆ R}. We have, thus the functions: A : {R ∈ Cong (C∞(RA), ΦA) ker(εA) ⊆ R} → {I ′ ∈ I(C∞(RA), ΦA) ψA[R] ⊆ I ′} ψ′ R 7→ ϕA(R) = {g ∈ C∞(RA)(g, φA(0)) ∈ R} that is, ψ′ A = ψA ↾{R∈Cong (C∞(RA),ΦA)ker(εA)⊆R}, and: A : {I ′ ∈ I(C∞(RA), ΦA) ψA[R] ⊆ I ′} → ϕ′ {R ∈ Cong (C∞(RA), ΦA) ker(εA) ⊆ R} I ′ 7→ ϕA(I ′) = {(g1, g2) ∈ C∞(RA) × C∞(RA) g1 − g2 ∈ I ′} where ϕ′ A = ϕA ↾{I ′∈I(C∞(RA),ΦA)ϕA[R]⊆I ′} . Since ϕ′ A and ψ′ A are inverse bijections, we have the following commutative diagram: Cong (C∞(RA), ΦA) ı {R ∈ Cong(C∞(RA), ΦA) ker(εA) ⊆ R} ψA ψ′ A / I(C∞(RA), Φ)  / {I ′ ∈ I(C∞(RA), ΦA) ψA(ker(εA)) ⊆ I ′} / ? O O / ? O O A Universal Algebraic Survey of C∞−Rings 67 where ı and  are the ordinary inclusions. We also have the bijection given in the Lemma 3: (qker(εA))∗ : Cong(cid:18) C∞(RA) ker(εA) , Φ(cid:19) → {R ∈ Cong (A, Φ) ker(εA) ⊆ R} and the bijection given in lemmas Lemma 4 and 2: (qker(εA))− : I(cid:18) C∞(RA) ker(εA) , Φ(cid:19) → {I ′ ∈ I(C∞(RA), Φ)ϕA(ker(εA)) ⊆ I ′}. Hence, we have the following bijection: ψ C∞ (RA) ker(εA ) : Cong(cid:18) C∞(RA) ker(εA) , Φ(cid:19) → I(cid:18) C∞(RA) ker(εA) , Φ(cid:19) , since the following diagram commutes: Cong (C∞(RU(A)), Φ) {R ∈ Cong(C∞(RA), ΦA) ker(εA) ⊆ R} (qker(εA))∗ Cong(cid:18) C∞(RA) ker(εA) , Φ(cid:19) ψU (A) ψ′ A ψ C∞(RU(A)) ker(εA) / I(C∞(RU(A)), Φ) / {I ′ ∈ I(C∞(RA), ΦA) ϕA(ker(εA)) ⊆ I ′} (qker(εA ))− I(cid:18) C∞(RA) ker(εA) , Φ(cid:19) Since εA : C∞(RU(A)) ker(εA) and Cong(A, Φ) I(A, Φ) ∼=→ A is an isomorphism, we have the bijections: (εA)∗ (εA)− ker(εA) , Φ(cid:19) → Cong(cid:18) C∞(RA) → I(cid:18) C∞(RA) , Φ(cid:19) ker(εA) so we have the bijection: ψ(A,Φ) : Cong (A, Φ) → I(A, Φ), since the following diagram commutes: / ? O O / ? O O O O / / 3 3 O O k k 68 J. C. Berni and H. L. Mariano Cong(cid:18) C∞(RA) ker(εA) , Φ(cid:19) ψ C∞ (RU (A) ) ker(εA ) θ / I(cid:18) C∞(RA) ker(εA) , Φ(cid:19) (εA)− Cong (A, Φ) ψ(A,Φ) / I(A, Φ) Given any C∞−ring (A, Φ), there is a ring-theoretical ideal I = ψA(ker(εA)) such that: (A, Φ) ∼=(cid:18) C∞(RA) I , Φ(cid:19) , that is, every C∞−ring is the quotient of a free C∞−ring by some of its ring- theoretic ideals. We say that any C∞−ring is given by generators and relations. Remark 9. Let (A, Φ) be a C∞−ring. The set Cong (A, Φ) is partially ordered by inclusion. Also, given {Rii ∈ I} ⊆ Cong (A, Φ), we have: Ri ∈ Cong (A, Φ), \i∈I so we can define: V : P(Cong(A, Φ)) → {Ri i ∈ I} Cong 7→Ti∈I {Ri i ∈ I} . Also, given {Ri i ∈ I} ⊆ Cong (A, Φ), we define: {Ri i ∈ I} W : P(Cong(A, Φ)) → so (Cong (A, Φ),V,W) is a complete lattice. 7→T{R ∈ Cong(A, Φ) Si∈I {Ri i ∈ I} ⊆ R} , Cong Note that I(A, Φ), partially ordered by inclusion, also has a structure of com- plete lattice, since it the set of ring-theoretic ideals of (A, Φ(+), Φ(·), Φ(−), Φ(0), Φ(1)). We have constructed, in Proposition 17, a bijection: ϕ(A,Φ) : Cong (A, Φ) → I(A, Φ). ϕ(A,Φ) : Cong (A, Φ) → I(A, Φ) R 7→ {g ∈ A(g, 0) ∈ R} We claim that ϕ(A,Φ) : Cong(A, Φ) → I(A, Φ) is an isomorphism of lattices. / O O / A Universal Algebraic Survey of C∞−Rings 69 Given {Rii ∈ I} ⊆ Cong (A, Φ), it is easy to see that: ψ(A,Φ)(cid:16)^{Rii ∈ I}(cid:17) =^ ψ(A,Φ)(Ri), so ψ(A,Φ) is a homomorphism of lattices. Also, given R, S ∈ Cong (A, Φ) such that R ⊆ S, we have: ψ(A,Φ)(R) ⊆ ψ(A,Φ)(S). Given I ′ ⊇ ψ(A,Φ)(R), since ψ′ A is surjective, there is some S ∈ Cong (A, Φ) with ψ′ A(S) = I ′. Also, since ψ′ A is injective, such an S is unique. Now, so ψ(A,Φ)(R) ⊆ ψ(A,Φ)(S), R = ψ⊣ (A,Φ)[ψ(A,Φ)(R)] ⊆ ψ(A,Φ) ⊣[ψ(A,Φ)(S)] = S Since ψ(A,Φ) is bijective, it follows that ψ(A,Φ) is an isomorphism of complete lattices. Moreover, both lattices are algebraic lattices, whose compact elements are the finitely generated congruences or ideals. The following result relates the ideals of a product of C∞−rings with the ideals of its factors. Proposition 18. Let A and B be two C∞−rings and let I(A) be the set of all ideals of A and I(B) be the set of all ideals of B. Every ideal of the product A × B has the form a × b, where a is an ideal of A and b is an ideal of B, so we have the following bijection: Φ : I(A) × I(B) → I(A × B) (a, b) 7→ a × b Proof. Let p1 : A × B → A and p2 : A × B → B be the canonical projections of the product: A × B p1 v♠♠♠♠♠♠♠♠♠♠♠♠♠♠ A p2 (◗◗◗◗◗◗◗◗◗◗◗◗◗◗ B and let I be any ideal of A × B. We have: p1[I] = {a ∈ A(∃b ∈ B)((a, b) ∈ I)} = {a ∈ A(a, 0) ∈ I} p2[I] = {b ∈ B(∃a ∈ A)((a, b) ∈ I)} = {b ∈ B(0, b) ∈ I} We claim that p1[I] × p2[I] = I. v ( 70 J. C. Berni and H. L. Mariano Given (a, b) ∈ I, then p1(a, b) = a ∈ p1[I] and p2(a, b) = b ∈ p2[I], so (a, b) ∈ p1[I] × p2[I]. Conversely, given (a, b) ∈ p1[I] × p2[I], a ∈ p1[I] and b ∈ p2[I], so (a, 0) ∈ I and (0, b) ∈ I, so (a, b) = (a, 0) + (0, b) ∈ I. Since for every a, ideal of A and for every b ideal of B, a × b is an ideal of A × B, the following map is a bijection: Φ : I(A) × I(B) → I(A × B) (a, b) 7→ a × b In the following proposition we are going to describe how to calculate any limit and any directed colimit, making use of the forgetful functor U : C∞Rng → Set. Proposition 19. In the category C∞Rng of the C∞−rings, all the limits and all filtered colimits exist and are created by the forgetful functor U : C∞Rng → Set. Proof. Cf. p. 7 of [24] By a general argument, it can be shown that the category C∞Rng has all small colimits. In particular, coequalizers of pairs of C∞−homomorphisms, f, g : (A, Φ) → (B, Ψ ), are given by quotients: (A, Φ) f g / (B, Ψ ) qI /(cid:18) B I , Ψ(cid:19) where I = h{(f (a), g(a))a ∈ A}i. In order to describe all small colimits, it is enough to construct coproducts, and since C∞Rng has filtered colimits, it suffices to construct only finite coprod- ucts. Also, since R ∼= C∞({∗}) is the initial C∞−ring, it is enough, by induction, to describe binary coproducts in C∞Rng. 4.2 The C∞−Coproduct In this subsection we describe the coproduct in the category C∞Rng, which E. Dubuc calls "the C∞−tensor product", and we call "the C∞−coproduct". First we give its categorial definition, then we define the binary C∞−coproduct of free, finitely generated and finally the binary C∞−coproduct of arbitrary C∞−rings. Then we give a description of the C∞−coproduct of an arbitrary family of arbitrary C∞−rings. / / / / A Universal Algebraic Survey of C∞−Rings 71 Definition 13. Let (A, Φ) and (B, Ψ ) be two C∞−rings. We will denote the underlying set of the coproduct of (A, Φ) and (B, Ψ ) by A ⊗∞ B, and its corre- sponding canonical arrows by ιA and ιB: A B ιA A ⊗∞ B #❍❍❍❍❍❍❍❍❍ ;✈✈✈✈✈✈✈✈✈ ιB In order to describe concretely the coproduct in C∞Rng, first we compute the coproduct of two free C∞−rings with m and n generators. Since m = {0, · · · , m − 1}, n = {0, · · · , n − 1}, m ⊔ n ∼= m + n and the functor L : Set → C∞Rng preserves coproducts (since it is a left adjoint functor), we have: C∞(Rm) ⊗∞ C∞(Rn) ∼= C∞(Rm × Rn) ∼= C∞(Rm+n) Now, given ideals I ⊂ C∞(Rm) and J ⊂ C∞(Rn), then: C∞(Rm) I ⊗∞ C∞(Rn) J ∼= C∞(Rm × Rn) (I, J) , where (I, J) = hf ◦π1, g ◦π2(f ∈ I)&(g ∈ J)i, where π1 : Rm ×Rn → Rm and π2 : Rm × Rn → Rn are the projections on the first and the second coordinates. Now, given any two C∞−rings, (A, Φ) and (B, Ψ ), we describe concretely their coproduct. First we write (A, Φ) and (B, Ψ ) as colimits of their finitely generated C∞−sub- rings: and (A, Φ) ∼= lim −→ i∈I (Ai, Φi) (B, Ψ ) ∼= lim −→ j∈I (Bj, Ψj) Then, observing that colimits commute with coproducts, we have: A ⊗∞ B ∼= lim −→ i∈I j∈J Ai ⊗∞ Bj Now let {(Ai, Φi)i ∈ I} be any set of C∞−rings. As mentioned in Remark 5 of Subsection 2.4 , such a family has a coproduct in C∞Rng. This coproduct is given by the colimit: # ; 72 J. C. Berni and H. L. Mariano O∞ i∈I Ai = lim −→ I ′⊆finIO∞ i∈I ′ Ai 4.3 Addition of Variables: The C∞−Ring of Polynomials As an application of the construction given above, we can describe the process of "adding a set S of variables to a C∞−ring (A, Φ)". The construction is given as follows: Let (A, Φ) be any C∞−ring and let S be any set. Consider L(S) = C∞(RS), the free C∞−ring on the set S of generators, together with its canonical map, S : S → C∞(RS). If we denote by: A C∞(RS) ιA ❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚❚ 5❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥❥ ιC∞(RS ) A ⊗∞ C∞(RS) the coproduct of A and C∞(RS), define: xs := ιC∞(RS )(S(s)). We thus define: A{xss ∈ S} := A ⊗∞ C∞(RS). We have a natural bijection: C∞(RS) → A ⊗∞ C∞(RS) S → U (A ⊗∞ C∞(RS)) References 1. P. Arndt, H.L. Mariano, The von Neumann-regular Hull of (preordered) rings and quadratic forms, South American Journal of Logic, Vol. X, n. X, pp. 1-43, 2016. 2. J. L Bell, A. B. Slomson, Models and Ultraproducts - an Introduction Dover Publications, Inc. Mineola, New York, 2006. ISBN-10: 0-486-44979-3 3. J.C. Berni,H.L. Mariano, Topics on Smooth Commutative Algebra, in preparation 4. J.C. Berni. Alguns Aspectos Alg´ebricos e L´ogicos dos An´eis C∞ (Some Algebraic and Logical Apects of C∞−Rings, in English). PhD thesis, Instituto de Matem´atica e Estat´ıstica (IME), Universidade de Sao Paulo (USP), Sao Paulo, 2018. 5. J.C. Berni,H.L. Mariano, Classifying Toposes for Some Theories of C∞−Rings, to appear in South American Journal of Logic, ISSN 2446-6719, 2019. 5 A Universal Algebraic Survey of C∞−Rings 73 6. J.C. Berni,H.L. Mariano, Von Neumann Regular C∞−Rings and Applications (in preparation). 7. F. Borceux, Handbook of Categorical Algebra, 2, Categories and Structures, Cam- bridge University Press, 1994. 8. F. Borceux, Handbook of Categorical Algebra, 3, Categories of Sheaves, Cambridge University Press, 1994. 9. D. Borisov, K. Kremnizer Beyond perturbation 1: de Rham spaces https://arxiv.org/pdf/1701.06278.pdf 10. M. Bunge, F. Gago, A.M. San Luiz, Synthetic Differential Topology, Cambridge University Press, 2017. 11. B.A. Davey, H. A. Priestley Introduction to Lattices and Order Cambridge University Press, 1992 12. D. van Dalen, Logic and Structure, Springer Verlag, 2012. 13. E.J. Dubuc, C∞−Schemes, American Journal of Mathematics, Vol. 103, No. 4, Aug. 1981, pp. 683-690. 14. J. Dugundji, Topology, Allyn and Bacon, Inc. Boston 15. D. Joyce, Algebraic Geometry over C∞−Rings, volume 7 of Memoirs of the Amer- ican Mathematical Society, arXiv: 1001.0023, 2016. 16. D. Joyce, An introduction to C∞-schemes and C∞-algebraic geometry, arXiv:1104.4951v2[math.DG], 16 Nov 2012. 17. A. Kock, Synthetic Differential Geometry, Cambridge University Press, 2006. 18. A. Kock, A simple axiomatic for differentiation, Math. Scand. 40 (1977), 183-193. 19. S. MacLane, I. Moerdijk, Sheaves in Geometry and Logic, A first introduction to Topos Theory, Springer Verlag, 1992. 20. M. Makkai, G.E. Reyes, First Order Categorical Logic - Model-Theoretical Methods in the Theory of Topoi and Related Categories, Springer Verlag, 2008. 21. I. Moerdijk, G. Reyes, Rings of Smooth Functions and Their Localizations I, Jour- nal of Algebra, 99, 324-336, 1986. 22. I. Moerdijk, G. Reyes, N.v. Que, Rings of Smooth Functions and Their Localiza- tions II, Mathematical Logic and Theoretical Computer Science, Lecture Notes in Pure and Applied Mathematics 106, 277-300, 1987. 23. A. Kock (editor), Category theoretic Methods in Geometry, Various Publications Series n35, Aarhus, 1983. 24. I. Moerdijk, G. Reyes, Models for Smooth Infinitesimal Analysis, Springer Verlag, 1991. 25. I. Moerdijk, G. Reyes, Cohomology theories in synthetic differential geometry, in ([23]), 1-67. 26. H. Schubert, Categories, Springer Verlag, 1972.
1503.05529	2	1503	2015-04-13T11:02:23	Some special features of Cayley algebras, and $G_2$, in low characteristics	[ "math.RA", "math.RT" ]	Some features of Cayley algebras (or algebras of octonions) and their Lie algebras of derivations over fields of low characteristic are presented. More specifically, over fields of characteristic $7$, explicit embeddings of any twisted form of the Witt algebra into the simple split Lie algebra of type $G_2$ are given. Over fields of characteristic $3$, even though the Lie algebra of derivations of a Cayley algebra is not simple, it is shown that still two Cayley algebras are isomorphic if and only if their Lie algebras of derivations are isomorphic. Finally, over fields of characteristic $2$, it is shown that the Lie algebra of derivations of any Cayley algebra is always isomorphic to the projective special linear Lie algebra of degree four. The twisted forms of this latter algebra are described too.	math.RA	math	SOME SPECIAL FEATURES OF CAYLEY ALGEBRAS, AND G2, IN LOW CHARACTERISTICS ALONSO CASTILLO-RAMIREZ⋆ AND ALBERTO ELDUQUE∗∗ Abstract. Some features of Cayley algebras (or algebras of octonions) and their Lie algebras of derivations over fields of low characteristic are presented. More specifically, over fields of characteristic 7, explicit embeddings of any twisted form of the Witt algebra into the simple split Lie algebra of type G2 are given. Over fields of characteristic 3, even though the Lie algebra of derivations of a Cayley algebra is not simple, it is shown that still two Cayley algebras are isomorphic if and only if their Lie algebras of derivations are isomorphic. Finally, over fields of characteristic 2, it is shown that the Lie algebra of derivations of any Cayley algebra is always isomorphic to the projective special linear Lie algebra of degree four. The twisted forms of this latter algebra are described too. 1. Introduction Let F be an arbitrary field. Cayley algebras (or algebras of octonions) over F constitute a well-known class of nonassociative algebras (see, e.g. [KMRT98, Chapter VIII] and references therein). They are unital nonassociative algebras C of dimension eight over F, endowed with a nonsingualr quadratic multiplicative form (the norm) q : C → F. Hence q(xy) = q(x)q(y) for any x, y ∈ C, and the polar form bq(x, y) := q(x + y) − q(x) − q(y) is a nondegenerate bilinear form. Any element in a Cayley algebra C satisfies the degree 2 equation: x2 − bq(x, 1)x + q(x)1 = 0. (1) Besides, the map x 7→ ¯x := bq(x, 1)1 − x is an involution (i.e., an antiautomorphism of order 2) and the trace t(x) := bq(x, 1) and norm q(x) are given by t(x)1 = x + ¯x, q(x)1 = x¯x = ¯xx for any x ∈ C. Two Cayley algebras C1 and C2, with respective norms q1 and q2, are isomorphic if and only if the norms q1 and q2 are isometric. If the characteristic of F is not 2, then C = F1 ⊕ C0, where C0 is the subspace of trace zero elements (i.e., the subspace orthogonal to F1 relative to bq). For x, y ∈ C0, (1) shows that xy + yx = −bq(x, y)1, while t([x, y]) = [x, y] + [x, y] = [x, y] + [¯y, ¯x] = 0, so [x, y] := xy − yx ∈ C0. In particular, xy = − 1 2 bq(x, y)1 + 1 2 [x, y], (2) so the projection of xy in C0 is 1 e.g. [EK13, Theorem 4.23]): 2 [x, y]. Moreover, the following relation holds (see, so the multiplication in C and its norm are determined by the bracket in C0. [[x, y], y] = 2bq(x, y)y − 2bq(y, y)x, (3) 2010 Mathematics Subject Classification. Primary 17A75; Secondary 17B60, 17B25. Key words and phrases. Octonions; Cayley algebra; Derivations; G2. ⋆ Supported by the 150th Anniversary Postdoctoral Mobility Grant (PMG14-15 01) of the London Mathematical Society. ∗∗ Supported by the Spanish Ministerio de Econom´ıa y Competitividad -- Fondo Europeo de Desarrollo Regional (FEDER) MTM2013-45588-C3-2-P, and by the Diputaci´on General de Arag´on -- Fondo Social Europeo (Grupo de Investigaci´on de ´Algebra). 1 2 A. CASTILLO-RAMIREZ AND A. ELDUQUE The Lie algebra of derivations of a Cayley algebra C is defined by Der(C) := {d ∈ gl(C) : d(xy) = d(x)y + xd(y) ∀x, y ∈ C}. In general, if M is a module for a Lie algebra L, we say that a bilinear product · : M × M → M is L-invariant if L acts on (M, ·) by derivations: x(u · v) = x(u) · v + u · x(u), for any x ∈ L, u, v ∈ M . If the norm q of a Cayley algebra C is isotropic (i.e., there exists 0 6= x ∈ C with q(x) = 0), then C is unique up to isomorphism. In this case, the Cayley algebra C is said to be split, and it has a good basis {p1, p2, u1, u2, u3, v1, v2, v3} with multiplication given in Table 1. We denote by Cs the split Cayley algebra. p1 p1 0 0 0 0 v1 v2 v3 p1 p2 u1 u2 u3 v1 v2 v3 p2 0 p2 u1 u1 0 u2 u2 0 u3 u3 0 v1 0 v1 0 u1 u2 −v3 u3 v2 −v1 v3 −v2 −p1 0 0 v2 0 v2 0 −p1 v3 0 v3 0 0 0 −p1 v1 0 0 0 0 −p2 0 0 0 −p2 0 0 0 0 0 −p2 u3 −u2 0 u1 0 −u3 u2 −u1 Table 1. Multiplication table in a good basis of the split Cayley algebra. Given a finite-dimensional simple Lie algebra g of type Xr over the complex numbers, and a Chevalley basis B, let gZ be the Z-span of B (a Lie algebra over Z). The Lie algebra gF := gZ ⊗Z F is the Chevalley algebra of type Xr. In particular, the Chevalley algebra of type G2 is isomorphic to Der(Cs) (see, e.g. [EK13, §4.4]). For any Cayley algebra C, the Lie algebra Der(C) is a twisted form of the Chevalley algebra Der(Cs). (Recall that, if A and B are algebras over F, then A is a twisted form of B whenever A ⊗F Falg ∼= B ⊗F Falg, for an algebraic closure Falg of F.) If the characteristic of F is neither 2 nor 3, then the Chevalley algebra of type G2 is simple; this is the split simple Lie algebra of type G2. Moreover, two Cayley algebras C1 and C2 are isomorphic if and only if their Lie algebras of derivations are isomorphic (see [Sel67, Theorem IV.4.1] or [EK13, Theorem 4.35]). The goal of this paper is to show some surprising features of Cayley algebras over fields of characteristic 7, 3 and 2. Section 2 studies the case of characteristic 7 and is divided in three subsections. In Section 2.1, we review a construction of Cs due to Dixmier [Dix84] in terms of transvectants which was originally done in characteristic 0, but it is valid in any characteristic p ≥ 7. In this construction, Cs appears as the direct sum of the trivial one-dimensional module and the restricted irreducible seven-dimensional module V6 for the simple Lie algebra sl2(F), which embeds into Der(Cs) as its principal sl2 subalgebra. When the characteristic is 7, this action of sl2(F) by derivations on Cs may be naturally extended to an action by derivations of the Witt algebra W1 := Der(cid:0)F[X]/(X 7)(cid:1), explaining the fact, first proved in [Pre83, Lemma 13] (see also [HSpr]), that W1 embeds into the split simple Lie algebra of type G2. In Section 2.2, we show that, when F is algebraically closed, V6 is the unique non-trivial non-adjoint restricted irreducible module for W1 with a nonzero invariant product. SOME FEATURES OF CAYLEY ALGEBRAS, AND G2, IN LOW CHARACTERISTICS 3 Then, in Section 2.3 we prove that, in characteristic 7 and even when the ground field is not algebraically closed, all the twisted forms of the Witt algebra embed into Der(Cs), and any two embeddings of the same twisted form are conjugate by an automorphism. Section 3 is devoted to the case of characteristic 3. In this situation, it is known that the Chevalley algebra of type G2 is not simple, but it contains an ideal isomor- phic to the projective special linear Lie algebra psl3(F). We review this situation and prove that it is still valid that two Cayley algebras are isomorphic if and only if their Lie algebras of derivations are isomorphic. Finally, in Section 4, we prove that the Lie algebra of derivations Der(C) of any Cayley algebra C over a field F of characteristic 2 is always isomorphic to the projective special linear Lie algebra psl4(F). A proof of this fact when C = Cs appears in [EK13, Corollary 4.32]. Hence, in this case, it is plainly false that two Cayley algebras are isomorphic if and only if their Lie algebras of derivations are isomorphic. We show that the isomorphism classes of twisted forms of psl4(F), which is here the Chevalley algebra of type G2, are in bijection with the isomorphism classes of central simple associative algebras of degree 6 endowed with a symplectic involution. 2. Characteristic 7 2.1. Dixmier's construction. Let F[x, y] be the polynomial algebra in two vari- ables over a field F of characteristic 0. The general linear Lie algebra gl2(F) acts by derivations on F[x, y] preserving the degree of each polynomial. For any n ≥ 0, denote by Vn the subspace of homogeneous polynomials of degree n in F[x, y]. For any i, j, q ≥ 0 with q ≤ i, j, consider the q-transvectant Vi × Vj → Vi+j−2q given by (f, g)q := (i − q)! (j − q)! i! j! (cid:18) ∂qf ∂xq for f ∈ Vi and g ∈ Vj . ∂qg ∂yq −(cid:18)q ∂qg ∂x∂yq−1 ∂xq−1∂y 1(cid:19) ∂qf 2(cid:19) ∂qf +(cid:18)q ∂xq−2∂y2 ∂qg ∂x2∂yq−2 − + · · ·(cid:19) , It turns out that the split Cayley algebra Cs is isomorphic to the algebra defined on F1 ⊕ V6 with multiplication given by 1 20 (α1 + f )(β1 + g) :=(cid:0)αβ − (f, g)6(cid:1)1 +(cid:0)αg + βf + (f, g)3(cid:1), (4) for any α, β ∈ F and f, g ∈ V6 (see [Dix84, 3.6 Proposition]); equipped with this product, the subspace V6 becomes the subspace of trace zero elements in Cs. The existence of this isomorphism is based on the following identity given in [Dix84, 3.5 Lemme]: for any f, g ∈ V6. ((f, g)3, g)3 = 1 20(cid:16)(f, g)6g − (g, g)6f(cid:17), (5) . Consider the following endomorphisms of V6: ∂ e−1 := x , e0 := − − y 1 2 (cid:18)x ∂ ∂x ∂ ∂y(cid:12)(cid:12)(cid:12)(cid:12)V6 ∂y(cid:19)(cid:12)(cid:12)(cid:12)(cid:12)V6 , e1 := − y ∂ ∂x(cid:12)(cid:12)(cid:12)(cid:12)V6 A direct calculation gives [ei, ej] = (j − i)ei+j, for i, j ∈ {−1, 0, 1}, so these endo- morphisms span a subalgebra of gl(V6) isomorphic to sl2(F) that acts by derivations on V6. This construction also works when the characteristic of F is p ≥ 7, and, more- over, the map f ⊗ g 7→ (f, g)3 gives the only, up to scalars, linear map V6 ⊗ V6 → V6 invariant under the action of sl2(F). 4 A. CASTILLO-RAMIREZ AND A. ELDUQUE Now assume that the characteristic of F is 7. First, we will find a simpler formula describing the 3-transvectant ( , )3 : V6 × V6 → V6. For 0 ≤ i ≤ 6, denote mi := x6−iyi ∈ V6. Taking the indices modulo 7, the action of sl2(F) on V6 is given by the following formulas: e−1(mi) = imi−1 for 1 ≤ i ≤ 6, and e−1(m0) = 0, e0(mi) = (i + 4)mi for any 0 ≤ i ≤ 6, e1(mi) = (i + 1)mi+1 for 0 ≤ i ≤ 5, and e1(m6) = 0. For i, j in the prime subfield F7 of F, define the element c(i, j) by (6) (7) c(i, j) = 2(j − i)(4i + j − 1)(4j + i − 1) (cid:0)∈ F7 ⊆ F(cid:1). It is clear that c(i, j) = −c(j, i) for any i, j ∈ F7, and c(0, 6) = 1. A straightforward computation gives, for any i, j, k ∈ F7: (i + j + 4k + 1)c(i, j) = (i + 4k + 4)c(i + k, j) + (j + 4k + 4)c(i, j + k). (8) Therefore, defining mi · mj := c(i, j)mi+j−3 (9) (with indices modulo 7), we have ek(mi · mj) = ek(mi) · mj + mi · ek(mj), for any k ∈ {−1, 0, 1}, 0 ≤ i, j ≤ 6. This means that the product in (9) is sl2(F)- invariant; hence, by uniqueness and since (m0, m6)3 = m3 = m0 · m6, we conclude that (10) for any 0 ≤ i, j ≤ 6. The multiplication table of V6 with this product is given in Table 2. mi · mj = (mi, mj)3 · m0 0 0 0 m0 m1 m2 m3 m4 −3m1 m5 3m2 m6 −m3 m0 m1 0 0 −3m0 −m1 0 m3 3m4 m2 0 3m0 0 −m2 m3 0 m3 −m0 m1 m2 0 m4 3m1 0 −m3 m4 0 −m4 −m5 −3m6 −3m5 m6 0 m6 m3 −3m2 −m3 −3m4 3m5 −m6 0 m5 m5 3m6 0 0 0 0 0 Table 2. Multiplication table of (V6, ·). Remark 1. The above arguments show that the split Cayley algebra Cs is iso- morphic to the algebra defined on F1 ⊕ V6 with multiplication given by (4), or equivalently, by (α1 + f )(β1 + g) :=(cid:0)αβ + (f, g)6(cid:1)1 +(cid:0)αg + βf + f · g(cid:1). An explicit isomorphism between F1 ⊕ V6 and Cs, in terms of a good basis of Cs as in Table 1, is given by: 1 ↓ m0 ↓ p1 + p2 −3v3   m1 m2 ↓ ↓ 3u1 −p1 + p2 −3v1 −3v2 3u2 m4 ↓ m5 ↓ m3 ↓ m6 ↓ 3u3   . SOME FEATURES OF CAYLEY ALGEBRAS, AND G2, IN LOW CHARACTERISTICS 5 Now, for any k ∈ {2, . . . , 5}, define the endomorphism ek of V6 by ek(mi) =((i + 4k + 4)mi+k 0 if i + k ≤ 6, otherwise. (11) We will show that ek is a derivation of (V6, ·) for any k = {−1, 0, . . . , 5}. Take 0 ≤ i, j ≤ 6. If i + j + k − 3 ≤ 6, i + k ≤ 6, and j + k ≤ 6, then ek(mi · mj) = (i + j + 4k + 1)c(i, j)mi+j+k−3 =(cid:16)(i + 4k + 4)c(i + k, j) + (j + 4k + 4)c(i, j + k)(cid:17)mi+j+k−3 = ek(mi) · mj + mi · ek(mj). (12) If i + j + k − 3 > 6, then ek(mi · mj) = 0 = ek(mi) · mj = mi · ek(mj). Finally, if i + j + k − 3 ≤ 6 and i + k > 6 (the same happens if j + k > 6), then ek(mi) = 0 and c(i + k, j) is one of c(0, 0), c(0, 1), c(0, 2) or c(1, 1), but all these are equal to 0, so (12) applies. Because of (4) and (5), we may extend the action of ek to F1 ⊕ V6 ∼= Cs by means of ek(1) = 0, obtaining that ek is a derivation of the split Cayley algebra Cs. Furthermore, one checks at once that [ei, ej] = (j − i)ei+j (13) for any i, j ∈ {−1, 0, . . . , 5}, with ei = 0 when i is outside {−1, 0, . . . , 5}. Thus, the span of {ei : −1 ≤ i ≤ 5} in Der(V6, ·) is isomorphic to the Witt algebra ∂x gives an explicit isomorphism of W1 := Der(cid:0)F[X]/(X 7)(cid:1); the map ei ↔ xi+1 ∂ these Lie algebras, where x denotes the class of X modulo (X 7). Since Der(Cs) is the split simple Lie algebra of type G2, the above arguments provide an elementary proof of the next result. Theorem 2 ([Pre83]). If the characteristic of the ground field F is 7, then the Witt algebra W1 embeds as a subalgebra of the simple split Lie algebra of type G2. Remark 3. A specific embedding of W1 into Der(Cs) may be given in terms of a good basis of Cs. Given x, y ∈ Cs, the derivation of Cs defined by Dx,y := ad[x,y] + [adx, ady] (where adx(y) := [x, y]) is called the inner derivation induced by x and y. Then, the assignment e−1 7→ Dp1−p2,u1 + Du2,v1 , e0 7→ 2Du3,v3 + 3Du2,v2, e1 7→ Dp1−p2,v1 + Du1,v2 , e2 7→ 3Du1,u3 = 2Dp1−p2,v2, e3 7→ −Dv1,v2 = 4Dp1−p2,u3, e4 7→ 3Dv1,u3 , e5 7→ 5Dv2,u3 , gives an explicit embedding of W1 into Der(Cs). 2.2. Invariant bilinear products on modules for W1. It turns out that V6 is a very special module for the Witt algebra in characteristic 7 (see Theorem 4 below). Assume for this section that F is an algebraically closed field of characteristic p ≥ 5. Let L be a finite-dimensional restricted Lie algebra over F with p-mapping denoted by [p]. Given any character χ ∈ L∗, there is a finite-dimensional algebra u(L, χ), called the reduced enveloping algebra of L associated with χ, that is a quotient of the universal enveloping algebra of L and whose irreducible modules coincide precisely with the irreducible modules for L with character χ. We say that V is a restricted module for L if there is a representation ρ : L → gl(V ) that is a morphism of restricted Lie algebras, i.e., ρ(x[p]) = ρ(x)p, for any x ∈ L. When χ = 0, it turns out that the irreducible modules for u(L, χ) coincide precisely with the restricted irreducible modules for L. 6 A. CASTILLO-RAMIREZ AND A. ELDUQUE bracket as in (13). It is well known that W has a p-mapping given by e[p] i Let W = W1 be the Witt algebra over F with basis {ei : −1 ≤ i ≤ p − 2} and Lie := δi,0ei. For each i ∈ {−1, 0, ..., p − 2}, define W(i) = hei, ..., ep−2i. Consider the p- dimensional Verma modules L(λ) = u(W, 0) ⊗u(W(0) ,0) kλ, where λ ∈ {0, 1, ..., p − 1}, and kλ is the one-dimensional module for W(0) on which W(1) acts trivially and e0 acts by multiplication by λ. By [Nak92, Lemma 2.2.1]), L(λ) has a basis {m0, m1, ..., mp−1} on which the action of W is given by ek(mj) = (j + (λ + 1)(k + 1))mk+j , where mj = 0 for j outside {0, ..., p − 1}. It was established in [Cha41] (see also [Str77, FN98]) that any restricted irre- ducible module for W is isomorphic to one of the following: (1) The trivial one-dimensional module. (2) The (p − 1)-dimensional quotient L(p − 1)/hm0i. (3) The p-dimensional Verma module L(λ), with λ ∈ {1, ..., p − 2}. In the next result the invariant bilinear products L × L → L (not to be con- fused with invariant bilinear forms!) on irreducible restricted modules for the Witt algebra are determined. Theorem 4. Let W = W1 be the Witt algebra over an algebraically closed field F of characteristic p ≥ 5. Then: (1) If p 6= 7, there is no non-trivial non-adjoint restricted irreducible module for W with a nonzero invariant bilinear product. (2) If p = 7, there is a unique non-trivial non-adjoint restricted irreducible module for W with a nonzero invariant bilinear product. Up to isomorphism and scaling of the product, this unique module is V6 with product given in (9). (V6 is a module for W by means of (6) and (11).) Proof. We will use the above classification of restricted irreducible modules for W and follow several steps: (i) Let λ ∈ {0, 1, ..., p − 1} and suppose that L(λ) is equipped with ·, a W - invariant bilinear product. We claim that m0 ·mp−1 = µmλ, for some scalar µ ∈ F that determines the invariant product. Indeed, first observe that e0 (m0 · mp−1) = (λ + 1) m0 · mp−1 + (p − 1 + (λ + 1)) m0 · mp−1 = (2λ + 1) m0 · mp−1. Therefore, m0 · mp−1 is an eigenvector of the action of e0 with eigenvalue 2λ + 1. As mλ is also an eigenvector of the action of e0 with eigenvalue 2λ + 1, and the action of e0 is diagonal with p distinct eigenvalues, we must have that m0 · mp−1 is a scalar multiple of mλ. Furthermore, m0 ⊗ mp−1 generates the W -module L(λ)⊗L(λ), so this scalar determines the invariant product. (ii) For any p ≥ 5, we will show that if · is a W -invariant bilinear product on the irreducible module L(p − 1)/hm0i, then it must be the zero product. Denote by ¯x the class of an element x ∈ L(p − 1) modulo hm0i; then, e0 (m1 · mp−1) = m1 · mp−1 + (p − 1)m1 · mp−1 = 0. As all the eigenvalues of the action of e0 on L(p − 1)/hm0i are nonzero, this implies that m1 · mp−1 = 0. Now, using the W -invariance, it is easy to show that · must be the zero product. (iii) As L(p − 2) is the adjoint module for W , it obviously has a W -invariant bilinear product. Hence, we exclude this case from further observations. SOME FEATURES OF CAYLEY ALGEBRAS, AND G2, IN LOW CHARACTERISTICS 7 (iv) Let p ≥ 5 and λ ∈ {1, ..., p − 3}. We will show that if L (λ) has a nonzero W -invariant bilinear product ·, then p = 7 and λ = 3. By step (i), we may assume that m0 · mp−1 = mλ. For k ∈ {1, ..., p − 2}, (λ + 1)(k + 1)mk · mp−1 = ek(m0) · mp−1 = ek(m0 · mp−1) = ek(mλ) = (λ + (λ + 1)(k + 1))mλ+k, so mk · mp−1 = (λ + 1) (k + 1) + λ (λ + 1) (k + 1) mλ+k, (14) for k ∈ {1, ..., p − 2}. On the other hand, e1(e1(mλ)) = e1(e1(m0) · mp−1, and this gives 3 (3λ + 2) (λ + 1) mλ+2 = 2 (λ + 1) (2λ + 3) m2 · mp−1. (15) If 2λ + 3 = 0, (15) gives 3λ + 2 = 0, and this implies p = 5 and λ = 1. But then, using (14) we get e1(m2 · m4) = 2e1(m3) = 4m4, while at the same time e1 (m2 · m4) = (e1m3) · m4 + m3 · (e1m4) = m3 · m4 = −2m4, which is a contradiction. Hence, we assume for the rest of the proof 2λ + 3 6= 0. By (15), we have m2 · m6 = 3 (3λ + 2) 2 (2λ + 3) mλ+2. Comparing this with (14) with k = 2 we obtain 3 (3λ + 2) 2 (2λ + 3) = 4λ + 3 3 (λ + 1) , so λ (11λ + 9) = 0. If p = 11, the above relation implies that λ = 0, which is a contradiction with our choice of λ, so no invariant bilinear product exists for p = 11. For the rest of the proof, assume that p 6= 11 and hence λ = − 9 11 . Now, from e1 (e1 (e1(mλ))) = e1 (e1 (e1(m0) · mp−1)), we obtain (3λ + 2) (3λ + 3) (3λ + 4) mλ+3 = (2λ + 2) (2λ + 3) (2λ + 4) m3 · mp−1. Thus, 3 (3λ + 2) (3λ + 4) 4 (2λ + 3) (λ + 2) Comparing this with (14) for k = 3 we obtain m3 · mp−1 = mλ+3. Hence, 3 (3λ + 2) (3λ + 4) (2λ + 3) (λ + 2) = 5λ + 4 (λ + 1) . 17λ2 + 38λ + 20 = 0. Substituting λ = − 9 or p = 5. no nonzero invariant bilinear product exists in this case. λ = − 9 11 in this relation we obtain that p 35, so either p = 7 11 = 1, and it was shown above that If p = 7, then (cid:3) If p = 5, then λ = − 9 11 = 3. This completes the proof. Remark 5. Let V and U be two irreducible modules for a Lie algebra L over an arbitrary field F, and let Falg be an algebraic closure of F. Suppose that V ⊗F Falg and U ⊗F Falg are isomorphic as modules for L ⊗F Falg; then, V and U are isomorphic as modules for L. Indeed, if V ⊗F Falg and U ⊗F Falg are isomorphic, then HomL(V, U ) ⊗F Falg ∼= HomL⊗FFalg(V ⊗F Falg, U ⊗F Falg) 6= 0, so HomL(V, U ) 6= 0. The result follows since, by irreducibility, any nonzero L-module homomorphism 8 A. CASTILLO-RAMIREZ AND A. ELDUQUE from V to U is an L-module isomorphism. In particular, this implies that, even when the ground field is not algebraically closed, Theorem 4 applies to modules for the Witt algebra that are absolutely irreducible (i.e., they remain irreducible after extending scalars to an algebraic closure). 2.3. Embeddings of W1 and its twisted forms into G2. It is shown in [Pre83, HSpr] that, over an algebraically closed field of characteristic 7, the simple Lie algebra of type G2 contains a unique conjugacy class of subalgebras isomorphic to the Witt algebra. With our results above, a different proof may be given, valid for not necessarily algebraically closed fields. Theorem 6. Let F be an arbitrary field of characteristic 7. Then any two subal- gebras S1 and S2 of Der(Cs) isomorphic to the Witt algebra are conjugate: there is an automorphism ϕ of Cs such that S2 = ϕS1ϕ−1. Proof. Let S := span {ei : −1 ≤ i ≤ 5} be the subalgebra of g := Der(Cs) isomor- phic to the Witt algebra given by equations (6) and (11), and let S be an arbitrary subalgebra of g isomorphic to the Witt algebra. Take a basis {ei : −1 ≤ i ≤ 5} of S with [ei, ej] = (j − i)ei+j. We will follow several steps: (i) S is a maximal subalgebra of g. Proof. Since the Witt algebra does not admit nonsingular invariant bilinear forms (see, e.g. [Far86, Theorem 4.2]), the restriction of the Killing form κ of g to S is trivial; hence, by dimension count, g/ S is isomorphic, as a module for S, to the dual of the adjoint module for S. In particular, g/ S is an irreducible module for S, and this shows the maximality of S. (cid:3) (ii) The representation of S on C0 s, the subspace of trace zero elements of Cs, is restricted and absolutely irreducible. Proof. As S is an ideal in its p-closure in g, step (i) implies that S is a restricted subalgebra of g, and hence, the corresponding representation of S on C0 s is restricted. In order to prove the second part, we may assume that F is algebraically closed. Recall that the possible dimensions of restricted irreducible rep- resentations of the Witt algebra are 1, 6 and 7. Suppose that C0 s has a one-dimensional trivial S-submodule X. The space X may be either non- degenerate or totally isotropic with respect to the symmetric bilinear form bq of Cs. If X = Fx is nondegenerate, then q(x) 6= 0 and bq(x, 1) = 0. By (1), F1⊕X is a two-dimensional composition subalgebra of Cs, so it is isomorphic to Fp1 ⊕ Fp2 (with pi as in Table 1). Now, by [SV00, Corollary 1.7.3], the isomorphism F1 ⊕ Fx ∼= Fp1 ⊕ Fp2 may be extended to an automorphism of Cs. As S annihilates F1 ⊕ Fx, and the subalgebra of the derivations that annihilate Fp1 ⊕ Fp2 is isomorphic to sl3(F) ([EK13, Proposition 4.29]), we obtain that the Witt algebra embeds into sl3(F), which is impossible. If X is totally isotropic, then X ⊆ X ⊥, and X ⊥/X is a 5-dimensional module for S. As this cannot be irreducible, we deduce that all the compo- s are one-dimensional. Hence, the representation ρ of S sition factors of C0 on C0 s is nilpotent. Since ad(e0+ei) is diagonalizable for i 6= 0, −1 ≤ i ≤ 5, then adp s is restricted, then ρ(e0 + ei)p = ρ(e0 + ei), and the nilpotency implies that ρ(e0 + ei) = 0, for any i 6= 0, −1 ≤ i ≤ 5. This is a contradiction. (e0+ei) = ad(e0+ei). As the representation of S on C0 SOME FEATURES OF CAYLEY ALGEBRAS, AND G2, IN LOW CHARACTERISTICS 9 Finally, if Y is a 6-dimensional irreducible S-submodule for C0 s, then it must be nondegenerate with respect to bq. This implies that Y ⊥ is a one- dimensional nondegenerate submodule, and we may use the above argument with X = Y ⊥. (cid:3) (iii) Note that C0 s is not the adjoint module for S because of the existence of the invariant bilinear form bq on C0 s. (iv) The previous steps together with Theorem 4 and Remark 5 imply that, even when F is not algebraically closed, there is a unique possibility, up to s as a module for S, and a unique, up to scalars, nonzero isomorphism, for C0 S-invariant product on C0 s. Therefore, there is a basis { mi : 0 ≤ i ≤ 6} of C0 s with ek( mi) =((i + 4k + 4) mi+k 0 if i + k ≥ 6, otherwise, and 1 2 [ mi mj] = c(i, j) mi+j−3 for −1 ≤ k ≤ 5 and 0 ≤ i, j ≤ 6. Since the multiplication on a Cayley algebra is determined by the bracket of trace zero elements, the linear map ϕ that takes 1 to 1 and mi to mi, for i ∈ {0, . . . , 6}, is an automorphism of Cs such that S = ϕSϕ−1. (cid:3) To finish this section, note that the Witt algebra W over a field F of characteristic 7 is equal to the Lie algebra F[X]/(X 7) = F[Z]/(Z 7 − 1), where Z = X + 1. Denote by z the class of Z modulo (Z 7 − 1) = ((Z − 1)7). The elements fi = zi+1 ∂ ∂z , for i ∈ {−1, 0, . . . 5}, form a basis of W with [fi, fj] = (j − i)fi+j, (16) where, contrary to (13), the indices are taken modulo 7. Now we may define an action of W on V6 by changing slightly the definition in (11): fk(mi) = (i + 4k + 4)mi+k, (17) for −1 ≤ k ≤ 5 and 0 ≤ i ≤ 6, with indices taken modulo 7. This gives another representation of W on V6: fr(fs(mi)) − fs(fr(mi)) =(cid:16)(i + s + 4r + 4)(i + 4s + 4) − (i + r + 4s + 4)(i + 4r + 4)(cid:17)mi+s+r = (s − r)(i + 4(r + s) + 4)mi+s+r = [fr, fs](mi). Equation (8) proves that this is a representation by derivations, so W embeds in Der(V6, ·), and hence on Der(Cs) as well. This embedding is different from the one obtained through (11), but Theorem 6 shows that they are conjugate. We may even go a step further. For any 0 6= α ∈ F, consider the Lie algebra W α = Der(cid:0)F[Y ]/(Y 7 − α)(cid:1), and denote by y the class of Y modulo (Y 7 − α). For ∂y in W α, so { fi : −1 ≤ i ≤ 5} any natural number i, consider the element fi = yi+1 ∂ is a basis of W α, and fi+7 = αfi for any i. Define mi+7 := αmi. For 0 ≤ i, j ≤ 6, c(i, j) = 0 if i + j − 3 > 6 or i + j − 3 < 0, so if we modify (17) as follows: fk(mi) = (i + 4k + 4)mi+j−3, (18) 10 A. CASTILLO-RAMIREZ AND A. ELDUQUE for any i, k (but now fk+7 = α fk and mi+7 = αmi). The same computations as above show that W α embeds in Der(Cs). The twisted forms of the Witt algebra are precisely the algebras W α; if 0 6= α ∈ F7, W α is isomorphic to the Witt algebra, while if α ∈ F \ F7, W α is the Lie algebra of derivations of the purely inseparable field extension F[Y ]/(Y 7 −α). Two algebras W α and W β are isomorphic if and only if so are the algebras F[Y ]/(Y 7 − α) and F[Y ]/(Y 7 − β). (See [AS69] or [Wat71].) Therefore, Theorems 2 and 6 may be extended as follows: Theorem 7. Over a field of characteristic 7, all the twisted forms of the Witt algebra embed in the Lie algebra of derivations of the split Cayley algebra. Moreover, any two embeddings of the same twisted form of the Witt algebra in Der(Cs) are conjugate. The last part of this theorem follows by the same arguments as in the proof of Theorem 6. Remark 8. If C is a non split Cayley algebra (and hence it is a division alge- bra, since q is anisotropic), then Der(C) contains no nonzero nilpotent derivation. Indeed, if d ∈ Der(C) is nilpotent, then ker dC0 is a subspace of C0 and, since q is anisotropic, C0 = ker dC0 ⊕ (ker dC0 )⊥. Besides, since q is invariant under the action of d (because of (1)), d leaves (ker dC0 )⊥ invariant; this is a contradiction be- cause the nilpotency of d implies that any nonzero invariant subspace has nontrivial intersection with the kernel. Therefore, Der(C) cannot contain subalgebras isomorphic to twisted forms of the Witt algebra, because these algebras contain nilpotent elements. (Recall that the s is restricted, and hence so is the action on C0 of action of the Witt algebra on C0 any subalgebra of Der(C) isomorphic to a twisted form of the Witt algebra.) 3. Characteristic 3 Let C be a Cayley algebra over a field F of characteristic p 6= 2. Then, the subspace of trace zero elements C0 is closed under the commutator [ , ], and it satisfies (3). The anticommutative algebra (cid:0)C0, [ , ](cid:1) is a central simple Malcev algebra. If p 6= 2, 3, any central simple non Lie Malcev algebra is isomorphic to one of these. However, if p = 3, then C0 is a simple Lie algebra; more precisely, it is a twisted form of the projective special linear Lie algebra psl3(F), and any such twisted form is obtained, up to isomorphism, in this way. (See [AEMN02] or [EK13, Theorem 4.26].) Denote by Aut(A) the affine group scheme of automorphisms of a finite-dimensional algebra A. Equations (2) and (3) show that the restriction map Aut(C) −→ Aut(C0) f 7→ f C0, (19) gives an isomorphism of group schemes. (The reader may consult [KMRT98, Chap- ter VI] for the basic facts of affine group schemes.) For the rest of this section, assume that C is a Cayley algebra over a field F of characteristic 3, and write g := Der(C). Any derivation d ∈ g satisfies d(1) = 0 and d(C0) ⊆ C0, and hence, because of (2) and (3), we may identify g with Der(C0). Since C0 is a Lie algebra, adC0 is an ideal of g: the ideal of inner derivations. In fact, adC0 is the only proper ideal of g, and the quotient g/adC0 is again isomorphic to C0 ∼= adC0 (see [AEMN02]). In the split case, adC0 s is the ideal of the Chevalley algebra of type G2 generated by the root spaces corresponding to the short roots (see [Ste68, p. 156]). SOME FEATURES OF CAYLEY ALGEBRAS, AND G2, IN LOW CHARACTERISTICS 11 Lemma 9. Any derivation of g := Der(C) is inner. Proof. Let d ∈ Der(g) and let i = adC0 be the unique proper ideal of g. The ideal i is simple, so i = [i, i], and hence d(i) ⊆ [d(i), i] ⊆ i, so di is a derivation of i = adC0 ∼= C0. Since any derivation δ of C0 extends to a derivation of C by means of δ(1) = 0, it follows that there exists a δ ∈ g such that di = adδi. That is, d(f ) = [δ, f ] for any f ∈ i = adC0 , so the derivation d = d − adδ satisfies d(i) = 0. But if d is a derivation of g such that d(i) = 0, then [ d(g), i] ⊆ d(cid:0)[g, i](cid:1)+[g, d(i)] = 0, so d(g) is contained in the centralizer of i in g, which is trivial. In particular, the derivation d = d − adδ above is trivial, and hence d = adδ is inner. (cid:3) Theorem 10. Let C be a Cayley algebra over a field F of characteristic 3, and let g := Der(C) be its Lie algebra of derivations. The the adjoint map Ad : Aut(C) −→ Aut(g) f 7→ ϕ(f ) : d 7→ f df −1, is an isomorphism of affine group schemes. Proof. Let Falg be an algebraic closure of F. The group homomorphism AdFalg : Aut(CFalg) → Der(CFalg ) is injective, where CR = C⊗FR, for any unital commutative and associative algebra R over F. This is because f adxf −1 = adf (x) for any x ∈ C0 and f ∈ Aut(CFalg), so AdFalg(f ) = id implies f C0 = id; hence, f = id. Falg Falg As any ϕ ∈ Aut(cid:0)gFalg(cid:1) preserves the only proper ideal iFalg = adC0 tomorphism ϕ induces an automorphism f of C0 , which extends to an automor- phism of CFalg also denoted by f . Then ϕAd(f )−1iFalg = id, and, as in the proof above, a simple argument gives ϕAd(f )−1 = id, so ϕ = Ad(f ). This shows that AdFalg is a bijection. , the au- Falg Falg But it also shows that dim Aut(g) = dim Aut(C) and, since this latter group scheme is smooth (this follows from [SV00, Proposition 2.2.3], see also [EK13, Proof of Theorem 4.35]), we obtain dim Aut(g) = dim Der(C) = dim Der(g) by Lemma 9. Therefore, Aut(g) is smooth. Since the differential map d(Ad) : Der(C) → Der(g), δ 7→ adδ, is injective, (cid:3) [KMRT98, (22.5)] shows that Ad is an isomorphism. Denote by Isom(Cayley), Isom(G2), and Isom( ¯A2), the sets of isomorphism classes of Cayley algebras, twisted forms of the Chevalley algebra of type G2, and twisted forms of psl3(F), respectively. Theorem 10 and (19) immediately give the following consequence, where [A] denotes the isomorphism class of the algebra A. Corollary 11. The maps [C] 7→ [Der(C)] and [C] 7→ [C0] give bijections Isom(Cayley) → Isom(G2) and Isom(Cayley) → Isom( ¯A2), respectively. 4. Characteristic 2 In this section, assume that the characteristic of the ground field F is 2. In [EK13, Corollary 4.32] it is proved that the Chevalley algebra of type G2 (i.e., the Lie algebra Der(Cs)), is isomorphic to the projective special linear Lie algebra psl4(F). Here we extend this result for the Lie algebra of derivations of any Cayley algebra over F. Let V be a finite-dimensional vector space over F, and let b be a nondegenerate alternating (i.e., b(u, u) = 0 for any u ∈ V ) bilinear form of V . Denote by sp(V, b) the corresponding symplectic Lie algebra: sp(V, b) = {f ∈ gl(V ) : b(f (u), v) + b(u, f (v)) = 0 ∀u, v ∈ V }. 12 A. CASTILLO-RAMIREZ AND A. ELDUQUE In particular, sp2n(F) denotes the symplectic Lie algebra sp(F2n, bs), where bs is the alternating bilinear form with coordinate matrix (cid:18) 0 In In 0(cid:19) in the canonical basis. (Notice that, as char(F) = 2, there is no need of minus signs.) We identify the elements of sp2n(F) with their coordinate matrices in the canonical basis. A matrix in Mn(F) is called alternating if it has the form a + at for some a ∈ Mn(F), where at denotes the transpose of a. These are the coordinate matrices of the alternating bilinear forms. Lemma 12. The second derived power of the symplectic Lie algebra on a vec- tor space of dimension 6 is isomorphic to the projective special linear Lie algebra psl4(F): sp6(F)(2) ∼= psl4(F). Proof. For any natural number n we have b A direct computation gives sp2n(F) =(cid:26)(cid:18)a c at(cid:19) : a, b, c ∈ Mn(F), bt = b, ct = c(cid:27) . c at(cid:19) : a, b, c ∈ Mn(F), b and c alternating(cid:27) , c at(cid:19) : a, b, c ∈ Mn(F), a ∈ sln(F), b and c alternating(cid:27) . sp2n(F)(1) =(cid:26)(cid:18)a sp2n(F)(2) =(cid:26)(cid:18)a The dimension of sp2n(F)(2) is then n2 − 1 + 2(cid:0)n V2 V as a module for sl(V ). Fix a nonzero linear isomorphism det :V4 V → F and Let V be a four-dimensional vector space, and consider the second exterior power 2(cid:1) = 2n2 − n − 1. define the nondegenerate alternating bilinear form b b b :V2 V ×V2 V −→ F (u1 ∧ u2, v1 ∧ v2) 7→ det(u1 ∧ u2 ∧ v1 ∧ v2). with kernel FIV (where IV denotes the identity map on V ), so Φ induces an injection Φ : sl(V ) → sp(V2 V , b) ∼= sp6(F), Then, the action of sl(V ) on V2 V gives a Lie algebra homomorphism psl(V ) ֒→ sp(V2 V, b). But psl(V ) is simple of dimension 14, so, in particular, psl(V )(2) = psl(V ), and the dimension of sp(V2 V, b)(2) ∼= sp6(F)(2) is 2×32−3−1 = 14. Therefore, Φ induces an isomorphism psl(V ) ∼= sp(V2 V, b)(2), as required. (cid:3) We will need some extra notation. As above, let (V, b) be a finite-dimensional vector space endowed with a nondegenerate alternating bilinear form. Denote by gsp(V, b) the Lie algebra of the group of similarities (i.e., the general symplectic Lie algebra): gsp(V, b) = {f ∈ gl(V ) : ∃λ ∈ F such that b(f (u), v) + b(u, f (v)) = λb(u, v) ∀u, v ∈ V } , and by pgsp(V, b) the projective general symplectic Lie algebra (i.e., the quotient of gsp(V, b) by the one-dimensional ideal generated by IV ). In particular, after choosing a basis, we get the Lie algebras gsp2n(F) and pgsp2n(F). Corollary 13. The Lie algebra of derivations of psl4(F) is isomorphic to the pro- jective general symplectic Lie algebra pgsp6(F): Der(cid:0)psl4(F)(cid:1) ∼= pgsp6(F). SOME FEATURES OF CAYLEY ALGEBRAS, AND G2, IN LOW CHARACTERISTICS 13 Proof. Note that we have the decomposition 0 and one may easily check that gsp6(F)(1) = sp6(F). gsp6(F) = sp6(F) ⊕ F(cid:18)I3 0 0(cid:19) , For any A ∈ gsp6(F), adA : B 7→ [A, B] leaves invariant gsp6(F)(3) = sp6(F)(2), which is isomorphic to psl4(F) by Lemma 12, so we obtain a Lie algebra homomor- phism Φ : gsp6(F) −→ Der(cid:0)sp6(F)(2)(cid:1)(cid:16)∼= Der(cid:0)psl4(F)(cid:1)(cid:17), A 7→ adAsp6(F)(2) . The kernel of Φ is the centralizer in gsp6(F) of sp6(F)(2), which is FI6, so Φ in- and (as it may be calculated in GAP as in [AMMpr]) this is also the dimension of (cid:3) duces an injection pgsp6(F) → Der(cid:0)psl4(F)(cid:1). The dimension of pgsp6(F) is 21, Der(cid:0)psl4(F)(cid:1). The result follows. dimension of Der(cid:0)psl4(F)(cid:1) is not required; we just need the bound In fact, in order to prove the previous result, the exact computation of the For completeness, let us provide an elementary proof of this fact. dim Der(cid:0)psl4(F)(cid:1) ≤ 21. Lemma 14. dim Der(cid:0)psl4(F)(cid:1) ≤ 21. Proof. For 1 ≤ i, j ≤ 4, let Eij be the matrix in gl4(F) with 1 in the (i, j) entry and 0's elsewhere. Denote by ¯A the class of a matrix A ∈ sl4(F) in psl4(F). Then, sl4(F) is graded by Z3, with degree(E12) = (1, 0, 0) = − degree(E21), degree(E23) = (0, 1, 0) = − degree(E32), degree(E34) = (0, 0, 1) = − degree(E43). As the identity matrix I4 is homogeneous of degree (0, 0, 0), this induces a grading by Z3 on g = psl4(F): Γ : g = Mα∈Z3 gα, with support Supp Γ := {α ∈ Z3 : gα 6= 0} = {(0, 0, 0), ±(1, 0, 0), ±(0, 1, 0), ±(0, 0, 1), ±(1, 1, 0), ±(0, 1, 1), ±(1, 1, 1)}. Let h = span(cid:8)H1 := E11 + E22, H2 := E22 + E33(cid:9) = g(0,0,0) be the 'diagonal' sub- algebra of g. Then, the decomposition in eigenspaces for the adjoint action of h is g = h ⊕ g1 ⊕ g2 ⊕ g3, with g1 = span(cid:8) ¯E12, ¯E21, ¯E34, ¯E43(cid:9) = g(1,0,0) ⊕ g(−1,0,0) ⊕ g(0,0,1) ⊕ g(0,0,−1), g2 = span(cid:8) ¯E23, ¯E32, ¯E14, ¯E41(cid:9) = g(0,1,0) ⊕ g(0,−1,0) ⊕ g(1,1,1) ⊕ g(−1,−1,−1), g3 = span(cid:8) ¯E13, ¯E31, ¯E24, ¯E42(cid:9) = g(1,1,0) ⊕ g(−1,−1,0) ⊕ g(0,1,1) ⊕ g(0,−1,−1). As g is Z3-graded, so is Der(g). Several steps are required now: (i) dim Der(g)(0,0,0) ≤ 3, and d(h) = 0 for any d ∈ Der(g)(0,0,0). Proof. Any d ∈ Der(g)(0,0,0) preserves the one-dimensional spaces gα, for α ∈ Supp Γ \ {(0, 0, 0)}. Then d and adh commute, so d(h) = 0. Also d( ¯E12) = λ ¯E12 and d( ¯E23) = µ ¯E23 for some λ, µ ∈ F. From d(h) = 0, we obtain that d( ¯E21) = −λ ¯E21 and d( ¯E32) = −µ ¯E32. Hence d := d − 14 A. CASTILLO-RAMIREZ AND A. ELDUQUE adµH2+λH1 annihilates ¯E12 and ¯E23 (and ¯E21 and ¯E32). Since the elements ¯E12, ¯E21, ¯E23, ¯E32, ¯E34 and ¯E43 generate g, it follows that d is determined by the value d(E34). We conclude that dim Der(g)(0,0,0) − dim adh ≤ 1. (cid:3) (ii) Der(g) = adg + {d ∈ Der(g) : d(h) = 0}. Proof. We already have Der(g)(0,0,0) ⊆ {d ∈ Der(g) : d(h) = 0}, and it is clear that d ∈ Der(g)α, with α ∈ Z3 \ Supp Γ, implies d(h) ⊆ gα = 0. On the other hand, if d ∈ Der(g)α, with α ∈ Supp Γ \ {(0, 0, 0)}, then the restriction of d to h defines a linear map β : h → F by d(H) = β(H) ¯Ers, where ¯Ers is the basic element in the one-dimensional space gα. But for any H, H ′ ∈ h we get: 0 = d([H, H ′]) = [d(H), H ′] + [H, d(H ′)] = β(H)[ ¯Ers, H ′] + β(H ′)[H, ¯Ers] =(cid:0)−β(H)γ(H ′) + β(H ′)γ(H)(cid:1) ¯Ers, where γ : h → F is the nonzero linear form such that [H, ¯Ers] = γ(H) ¯Ers for any H ∈ h. Hence β is a scalar multiple of γ, and if β = νγ with ν ∈ F, then (d − νad ¯Ers)(h) = 0. (This argument is similar to the one in [EK12, Proposition 8.1].) (cid:3) (iii) Now note that ¯E12, ¯E23, ¯E34 and ¯E41 generate g. Let 0 6= d ∈ Der(g)α, with α ∈ Z3 \ {(0, 0, 0)}, and d(h) = 0. Then d(gi) ⊆ gi, for i = 1, 2, 3, and d( ¯E12) ∈ g1 so either d( ¯E12) = 0 or α ∈ {(−2, 0, 0), (−1, 0, 1), (−1, 0, −1)}, d( ¯E23) ∈ g2 so either d( ¯E23) = 0 or α ∈ {(0, −2, 0), (1, 0, 1), (−1, −2, −1)}, d( ¯E34) ∈ g1 so either d( ¯E34) = 0 or α ∈ {(0, 0, −2), (1, 0, −1), (−1, 0, 1)}, d( ¯E41) ∈ g2 so either d( ¯E41) = 0 or α ∈ {(2, 2, 2), (1, 2, 1), (1, 0, −1)}. But Der(g)α = 0 for α ∈ {(−2, 0, 0), (0, −2, 0), (0, 0, −2), (2, 2, 2)}, be- cause any d in one of these homogeneous spaces annihilates the generators ¯E21, ¯E32, ¯E43 and ¯E14. Hence, our homogeneous d belongs to Der(g)α with α ∈ X := {±(1, 0, 1), ±(1, 0, −1), ±(1, 2, 1)}. Now if, for instance, d ∈ Der(g)(−1,0,−1), then d annihilates ¯E23 ∈ g(0,1,0) (because (0, 1, 0) + (−1, 0, −1) 6∈ Supp Γ), ¯E41 ∈ g(−1,−1,−1), and ¯E24 ∈ g(1,1,0). Moreover, since ¯E34 = [[ ¯E31, ¯E12], ¯E24], it follows that d is determined by the value d( ¯E12). Thus, we get dim Der(g)(−1,0,−1) ≤ 1. A similar argument applies to the other possibilities, so dim Der(g)α ≤ 1 for any α ∈ X. (iv) Therefore, we conclude that dim Der(g) − dim g =(cid:16)dim Der(g)(0,0,0) − dim h(cid:17) +Xα∈X so that dim Der(g) ≤ dim g + 7 = 21. dim Der(g)α ≤ 7, (cid:3) Our next result extends [EK13, Corollary 4.32]. Theorem 15. Let C be a Cayley algebra over a field F of characteristic 2. The Lie algebra of derivations Der(C) is isomorphic to the projective special linear Lie algebra psl4(F). (Independently of the isomorphism class of C!) Proof. Recall that Der(C) is a 14-dimensional simple Lie algebra, a twisted form of Der(Cs), which is the Chevalley algebra of type G2. Any d ∈ Der(C) leaves the norm q invariant, annihilates the unity 1 and preserves C0, the subspace of trace zero elements. Since the characteristic is 2, the unity 1 is in C0, so d induces an element d in the symplectic Lie algebra sp(cid:0)C0/F1, bq(cid:1) ∼= sp6(F), SOME FEATURES OF CAYLEY ALGEBRAS, AND G2, IN LOW CHARACTERISTICS 15 where bq is the nondegenerate alternating bilinear form on C0 induced by bq. (Note that C0 = {x ∈ C : bq(x, 1) = 0}, so bq is nondegenerate.) Therefore, we have a homomorphism of Lie algebras Φ : Der(C) −→ sp(cid:0)C0/F1, bq(cid:1), 7→ d. d The simplicity of Der(C) implies that Φ is injective, and hence Φ(cid:0)Der(C)(cid:1) = Φ(cid:0)Der(C)(2)(cid:1) ⊆ sp(cid:0)C0/F1, bq(cid:1)(2) ∼= sp6(F)(2). By dimension count, the image of Φ is sp(cid:0)C0/F1, bq(cid:1)(2) sp6(F)(2), and hence to psl4(F) by Lemma 12. , which is isomorphic to (cid:3) The previous theorem shows that, in characteristic 2, it is no longer true that two Cayley algebras are isomorphic if and only if their Lie algebras of derivations are isomorphic. Remark 16. Given an irreducible root system of type Xr and its corresponding Chevalley algebra g over a field F, the quotient g/Z(g) (where Z(g) is the center of g) is usually called the classical Lie algebra of type Xr. In particular, in characteristic 2, Theorem 15 implies that the classical Lie algebras of type A3 and G2 coincide. Write A := M6(F), and let σ be the symplectic involution (attached to the standard alternating form bs), such that, for any X ∈ A, σ(X) is the adjoint relative to bs. Theorem 17. The affine group scheme of automorphisms of Der(Cs) is isomorphic to the affine group scheme of automorphisms of the algebra with involution (A, σ). Over an algebraic closure Falg, the group Aut(psl4(Falg) is the adjoint Chevalley group of type C3 (see [Hog82]), and hence it is isomorphic to projective general symplectic group PGSp6(Falg) ∼= Aut(cid:0)A, σ(cid:1). However, Theorem 17 considers arbi- trary fields, and we shall give an explicit isomorphism of schemes in its proof. Note that this result over Falg shows that Aut(psl4(F)) is connected, and Corollary 4 shows that it is smooth. Proof of Theorem 17. As in the proof of Theorem 15, we may identify Der(Cs) ∼= psl4(F) with the Lie algebra Skew(A, σ)(2), where Skew(A, σ) := {x ∈ A : σ(x) = x} (as the characteristic is two!) Consider the morphism of affine group schemes ϕ : Aut(A, σ) −→ Aut(cid:16)Skew(A, σ)(2)(cid:17), f : AR → AR 7→ f Skew(AR,σR)(2) , where R is a unital, commutative and associative F-algebra, AR = A⊗FR, and σR = σ ⊗ id is the induced involution in AR, so Skew(AR, σR)(2) = Skew(A, σ)(2) ⊗F R. The group homomorphism on points in an algebraic closure Falg ϕFalg : Aut(AFalg ) → Aut(cid:0)Skew(AFalg , σFalg)(2)(cid:1) is injective, because Skew(AFalg , σFalg)(2) = sp6(Falg)(2) generates AFalg = M6(Falg) as an algebra. Also, the differential dϕ is an isomorphism: it is the isomorphism Φ in the proof of Corollary 4. Therefore, ϕ is a closed imbedding ([KMRT98, (22.2)]). But both schemes are smooth, connected, and of the same dimension, so ϕ is an isomorphism. (cid:3) 16 A. CASTILLO-RAMIREZ AND A. ELDUQUE Since Aut(psl4(F) is smooth, the set of isomorphism classes of twisted forms of psl4(F) is in bijection with H 1(F, Aut(psl4(F)) (see [Wat79, Chapters 17 and 18]), and also the set of isomorphism clases of central simple associative algebras of degree 6 endowed with a symplectic involution is in bijection with H 1(F, Aut(M6(F), σ) = H 1(F, PGSp6(F)). Our last result is then a direct consequence of Theorem 17. Corollary 18. Let F be a field of characteristic 2. The map that sends any central simple associative algebra of degree 6 over F endowed with a symplectic involution (B, τ ) to the Lie algebra Skew(B, τ )(2) gives a bijection between the set of isomor- phism classes of such pairs (B, τ ) to the set of twisted forms over F of the Lie algebra psl4(F). Recall that psl4(F) is (isomorphic to) the Chevalley algebra of type G2, so this corollary gives the twisted forms of the classical simple Lie algebras of type G2 in characteristic 2. References [AEMN02] Bjerregaard, P. Alberca-Bjerregaard, A. Elduque, C. Mart´ın-Gonz´alez, and F.J. Navarro-M´arquez, On the Cartan-Jacobson theorem, J. Algebra 250 (2002), no. 2, 397 -- 407. [AMMpr] P. Alberca-Bjerregaard, D. Mart´ın-Barquero, C. Mart´ın-Gonz´alez, Inner derivations [AS69] [Cha41] [Dix84] [EK12] [EK13] [Far86] [FN98] [HSpr] [Hog82] of exceptional Lie algebras in prime characteristic, arXiv:1402.2212 [math.RA]. H.P. Allen and M.E. Sweedler, A theory of linear descent based upon Hopf algebraic techniques, J. Algebra 12 (1969), 242 -- 294. H.J. Chang, Uber Wittsche Lie-Ringe (German), Abh. Math. Sem. Hansischen Univ. 14 (1941) 151 -- 184. J. Dixmier, Certaines algebres non associatives simples definies par la transvection des formes binaires (French), J. Reine Angew. Math. 346 (1984) 110 -- 128. A. Elduque and M. Kochetov, Gradings on the exceptional Lie algebras F4 and G2 revisited, Rev. Mat. Iberoam. 28 (2012), no. 3, 773 -- 813. A. Elduque and M. Kochetov, Gradings on simple Lie algebras. Mathematical Surveys and Monographs 189, American Mathematical Society, Providence, RI, 2013. R. Farnsteiner, The associative forms of the graded Cartan type Lie algebras', Trans. Amer. Mat. Soc. 295 (1986) 417 -- 427. J. Feldvoss and D.K. Nakano, Representation Theory of the Witt Algebra, J. Algebra 203 (1998) 447 -- 469. S. Herpel and D.I. Stewart, Maximal subalgebras of Cartan type in the exceptional Lie algebras, arXiv:1409.3861 [math.RA]. G.M.D. Hogeweij, Almost-classical Lie algebras. II, Nederl. Akad. Wetensch. Indag. Math. 44 (1982), no. 4, 453 -- 460. [Pre83] [Nak92] [KMRT98] M.A. Knus, A. Merkurjev, M. Rost, and J.P. Tignol, The Book of Involutions, Amer- ican Mathematical Society Colloquium Publications 44. American Mathematical So- ciety, Providence, R.I., 1998. D.K. Nakano, Projective modules over Lie algebras of Cartan type, Mem. Amer. Math. Soc. 98 (470) (1992). A.A. Premet, Algebraic groups associated with Lie p-algebras of Cartan type (Rus- sian), Mat. Sb. (N.S.) 122(164) (1983), no. 1, 82 -- 96. English translation: Math. USSR-Sb. 50 (1985), no. 1, 85 -- 97. G.B. Seligman, Modular Lie algebras. Ergebnisse der Mathematik und ihrer Grenzge- biete, Band 40 Springer-Verlag, New York 1967. T.A. Springer and F.D. Veldkamp, Octonions, Jordan algebras and exceptional groups, Springer Monographs in Mathematics, Springer-Verlag, Berlin, (2000). R. Steinberg, Lectures on Chevalley groups, Yale University, New Haven, Conn., 1968. Notes prepared by John Faulkner and Robert Wilson. H. Strade, Representations of the Witt Algebra, J. Algebra 49 (1977) 595 -- 605. [Str77] [Wat71] W.C. Waterhouse, Automorphism schemes and forms of Witt Lie algebras, J. Algebra [Sel67] [SV00] [Ste68] 17 (1971), 34 -- 40. [Wat79] W.C. Waterhouse, Introduction to affine group schemes, Graduate Texts in Mathe- matics 66, Springer-Verlag, New York, 1979. SOME FEATURES OF CAYLEY ALGEBRAS, AND G2, IN LOW CHARACTERISTICS 17 School of Engineering and Computing Sciences, Durham University, South Road, Durham, DH1 3LE, United Kingdom E-mail address: [email protected] Departamento de Matem´aticas e Instituto Universitario de Matem´aticas y Aplica- ciones, Universidad de Zaragoza, 50009 Zaragoza, Spain E-mail address: [email protected]
1810.09866	2	1810	2019-12-31T07:03:45	Leavitt path algebras of weighted Cayley graphs $C_n(S,w)$	[ "math.RA" ]	For a postive integer $n$ and a subset $S$ of $\mathbb{Z}_n$, let $\left\langle S\right\rangle =\mathbb{Z}_n$, and $w:S\rightarrow\mathbb{N}$ be a function. The weighted Cayley graph of the cyclic group $\mathbb{Z}_n$ with respect to $S$ and $w$ is denoted by $C_n(S,w)$. We give an explicit description of the Grothendieck group of the Leavitt path algebras of $C_n(S,w)$. We also give description of Leavitt path algebras of $C_n(S,w)$ in some special cases.	math.RA	math	LEAVITT PATH ALGEBRAS OF WEIGHTED CAYLEY GRAPHS CnpS, wq MOHAN.R Abstract. For a postive integer n and a subset S of Zn, let xSy " Zn, and w : S Ñ N be a function. The weighted Cayley graph of the cyclic group Zn with respect to S and w is denoted by CnpS, wq. We give an explicit description of the Grothendieck group of the Leavitt path algebras of CnpS, wq. We also give description of Leavitt path algebras of CnpS, wq in some special cases. 1. Introduction For a finite group G and a subset S Ď G, let the associated Cayley graph be denoted by CaypG, Sq. When the given group is Zn we write CnpSq :" CaypZn, Sq. Leavitt path algebras of Cayley graphs of the finite cyclic group Zn with respect to the subset S " t1, n ´ 1u were initially studied in [8]. It was shown that there are exactly four isomorphism classes represented by the collection tLpCnp1, n ´ 1qq n P Nu. In subsequent work, [5] contains the computation of the important integers K0pLpCnp1, jqqq and detpIn ´ At Cnp1,jqq, where Ap´q denotes the adjacency matrix of a directed graph, and K0p´q denotes the Grothendieck group of a ring. Also in [5] the collections of K-algebras were described upto isomorphism: The descriptions of all these algebras follow from an application of the powerful tool known as the (Restricted) Algebraic Kirchberg-Philips Theorem. tLpCnp1, jqq n P Nu for j " 0, 1, 2. Cnp1,jq to that of calculating the Smith Normal Form of a j j matrix pM n In [6] the study was extended and a method to compute the Grothendieck group of the Leavitt path algebra LpCnp1, jqq to the case where 0 ď j ď n ´ 1 and n ě 3 was derived. Specifically it was shown how to reduce the computation of the Smith Normal Form of the n n matrix In´ At j qt´ Ij. Further a description of K0pLpCnp1, jqq was also given. In this paper we generalize the work done in [6] to study LpCnpS, wqq, where S is any nonempty generating subset of Zn, w : S Ñ N is a map and CnpS, wq is the weighted Cay- ley graph. In section 2 we recall the background information required. In Section 3 we present a method to compute the Grothendieck group. Specifically we find the conditions to determine the sign of detpIn ´ At CnpS,wqq and also the cardinality of K0pLpCnpS, wqqq. Also we find a method to reduce the computation of the Smith Normal form of the n n matrix In ´ At to that of calculating the Smith Normal form of a square matrix of smaller size if 0 R S (Theo- rem 3.9). In Section 4 we use the method developed in Section 3 to study the following simple cases when xSy " Zn: Case 1 : S " 1 Case 2 : S " 2 Case 3 : S " n Moreover we recover the results studied in [8],[5], and [6] as special cases and get some new results. Among these new results, in particular, we show that LpKnq -- Lp1, nq where Kn is the unweighted complete n-graph (See 4.1 for definition) and Lp1, nq is the Leavitt algebra. We also show that the main result of [8] holds true if Cnp1, n´ 1q is replaced by Dn for every n P N, where Dn denotes the Cayley graph of Dihedral group with respect to usual generating set. CnpS,wq Key words and phrases. Leavitt path algebra, Weighted Cayley graph. 1 2 MOHAN.R 2. Preliminaries Notation 2.1. Throughout by K we mean a fixed field. N denotes the set of natural numbers, Z` denotes the set of non-negative integers, Z denotes the set of integers and Q denotes the set of rationals. S is the cardinality of the set S. 2.1. Leavitt path algebras and the Algebraic KP theorem. A graph E " pE0, E1, s, rq consists of two sets E0, E1 and functions s, r : E1 Ñ E0. The elements of E0 are called vertices and the elements of E1 are called edges. If e is an edge, then speq is called its source and rpeq its range. E is called finite if E0 and E1 are finite sets. A vertex v is called a source (resp. sink ) if r´1pvq " H (resp. if s´1pvq " H). A graph is called sink-free (resp. source-free) if it has no sinks (resp. no sources). A non-sink v P E0 is called regular is s´1pvq is finite. (v emits only finitely many edges). A path µ in a graph E is either a vertex in E or a finite sequence e1e2 . . . en of edges in E such that rpeiq " spei`1q for 1 ď i ď n ´ 1. A path µ " e1e2 . . . en for which n ď 1 and spe1q " rpenq " v is called a closed path based at v. A closed path µ " e1e2 . . . en based at v for which speiq ‰ spejq for any i ‰ j is called a cycle (based at v). A graph which contains no cycles is called acyclic. Let E " pE0, E1, s, rq be graph. The adjacency matrix AE of E is the E0 E0 matrix whose entries are given by AEpv, wq " te P E1 speq " v, rpeq " wu for any v, w P E0 Definition 2.2. Let E be a graph and H Ď E0. (1) H is hereditary if whenever v P H and w P E0 for which there exists a path µ such that (2) H is saturated if whenever v P E0 is regular such that trpeq e P E1, speq " vu Ď H, (3) E satisfies condition (L) if every cycle in E has an exit. spµq " v and rpµq " w, then w P H. then v P H. Definition 2.3. Let K be a field and let E " pE0, E1, r, sq be a graph. The Leavitt path algebra of E is the K-algebra presented by generators E0 \ E1 \ E1 where E1 :" te e P E1u and the following relations. (V) @ v, w P E0, vw " δvwv (E) @ e P E1, speqe " e " erpeq and rpeqe " east " espeq (CK1) @ e, f P E1, ef " δef rpeq (CK2) @ v P E0 if 0 ă s´1pvq ă 8 then v " řePs´1pvq where δij is Kronecker delta. We denote the Leavitt path algebra by LpEq if underlying field K is fixed. Remark 2.4. It is easy to see that LpEq is unital if and only if E0 is finite, in which case řvPE 0 acts as the unity. In this paper we focus only on finite graphs. ee v One of the motivations to define Leavitt path algebra is to study a natural abelian monoid associated to a given graph E called the graph monoid ME which we define below. Definition 2.5. Let E be a finite graph with vertex set E0 and adjacency matrix AE " pavwq. The graph monoid ME of E is the abelian monoid presented by generating set E0 and following relations: avww v " ÿwPE 0 Recall that for a unital K-algebra R, the V-monoid of R, denoted by VpRq, is the set of isomorphism classes of finitely generated projective left R-modules. We denote the elements of VpRq using brackets, for example, rRs P VpRq represents the isomorphism class of the left regular module RR. Then VpRq is an abelian monoid, with operation ', and zero element r0s, LEAVITT PATH ALGEBRAS OF WEIGHTED CAYLEY GRAPHS CnpS, wq 3 where 0 is the zero R-module. Also, the moniod pVpRq,'q is conical; that is the sum of any two nonzero elements of VpRq is nonzero, or rephrased, VpRq " VpRq ´ r0s is a semigroup under '. The group completion of VpRq is denoted by K0pRq and called the Grothendieck group of R. Theorem 2.6. [10, Theorem 3.5] As monoids, VpLpEqq -- ME and rLpEqs Ø řvPE 0rvs under this isomorphism. Definition 2.7. A unital K-algebra A is called purely infinite simple in case A is not a division ring, and A has the property that for every nonzero element x of A there exists b, c P A for which bxc " 1A. The finite graphs E for which the Leavitt path algebra LpEq is purely infinite simple have been explicitly described in [4]: Theorem 2.8. LpEq is purely infinite simple if and only if E is sink-free, satisfies Condition (L), and only hereditary and saturated subsets of E0 are H and E0. In other words, the graph E satisfies the following properties: every vertex in E connects to every cycle of E; every cycle in E has an exit; and E contains at least one cycle. It is shown in [9, Corollary 2.2], that if A is a unital purely infinite simple K-algebra, then the semigroup pVpAq,'q is in fact a group, and moreover, that VpAq -- K0pAq, the Grothendieck if VpLpEqq is a group of A. For unital Leavitt path algebras, the converse is true as well: group, then LpEq is purely infinite simple. (This converse is not true for general K-algebras.) Theorem 2.9. If LpEq is unital purely infinite simple then K0pLpEqq -- VpLpEqq -- M E. The following important theorem will be used to yield a number of key results in the following sections. Theorem 2.10 ((Restricted) Algebraic KP Theorem). [7, Corollary 2.7] Suppose E and F are finite graphs for which the Leavitt path algebras LpEq and LpFq are purely infinite simple. Suppose that there is an isomorphism ϕ : K0pLpEqq Ñ K0pLpFqq for which ϕprLpEqsq " rLpFqs, and suppose also that the two integers detpIE 0 ´ At Eq and detpIF 0 ´ At Fq have the same sign. Then LpEq -- LpFq as K-algebras. Example 2.11 (Leavitt algebras). For any integer m ě 2, Lp1, mq is the free associative K- algebra in 2m generators x1, x2, . . . , xm, y1, y2, . . . , ym, subject to the relations m yixj " δi,j1K and xiyi " 1K. ÿi"1 These algebras were first defined and investigated in [12] in the context of finding counter- examples for the invariant basis number problem, and formed the motivating examples for the It is easy to see that for m ą 2, if Rm is the more general notion of Leavitt path algebra. graph having one vertex and m loops, then LpRmq -- Lp1, mq. From Theorem 2.8 it follows that LpRmq is unital purely infinite simple and hence K0pLpRmqq -- M is the cyclic group Zm´1, where the regular module rLpRmqs in K0pLpRmqq corresponds to 1 in Zm´1. It is shown in [1] that LpRd Unital purely infinite simple Leavitt path algebras LpEq whose corresponding K0 groups are Eq ď 0 are relatively well-understood, and arise as matrix rings m having two cyclic and for which det(IE 0´At over the Leavitt algebras Lp1, mq, as follows. Let d ě 2, and consider the graph Rd vertices v1, v2; d ´ 1 edges from v1 to v2; and m loops at v2. mq is isomorphic to the matrix algebra MdpLp1, mqq. By standard Morita equivalence theory we have that K0pMdpLp1, mqqq -- K0pLp1, mqq. Moreover, the ele- ment rMdpLp1, mqqs of K0pMdpLp1, mqqq corresponds to the element d in Zm´1. In particular, the element rMm´1pLp1, mqqs of K0pMm´1pLp1, mqqq corresponds to m ´ 1 " 0 in Zm´1. Fi- nally, an easy computation yields that detpI2´ At mq " ´pm´ 1q ď 0 for all m, d. Therefore, by invoking the Algebraic KP Theorem, the previous discussion immediately yields the following. Rm Rd 4 MOHAN.R pd ´ 1q Rd m " v1 pmq v2 Proposition 2.12. Suppose that E is a graph for which LpEq is unital purely infinite simple. Suppose that M E is isomorphic to the cyclic group Zm´1, via an isomorphism which takes the element řvPE 0rvs of M Eq ď 0. Then LpEq -- MdpLp1, mqq. 2.1.1. Computation of Grothendieck group. E to the element d of Zm´1. Finally, suppose that detpIE 0 ´ At Let E be a finite directed graph for which E0 " n. We view In ´ At E both as a matrix, and E : Zn Ñ Zn, via left multiplication (viewing elements of Zn as a linear transformation In ´ At as column vectors). As discussed in [1, Section 3], we have Proposition 2.13. If LpEq is purely infinite simple, then M E -- K0pLpEqq -- Zn{ImpIn ´ At Eq " CokerpIn ´ At Eq. Under this isomorphism rvis ÞÑ ~bi ` ImpIn ´ At the ith coordinate and 0 elsewhere. Eq, where ~bi is the element of Zn which is 1 in Let M P MnpZq and view M as a linear transformation M : Zn Ñ Zn via left multiplication on columns. The cokernel of M is a finitely generated abelian group, having at most n summands; as such, by the invariant factors version of the Fundamental Theorem of Finitely Generated Abelian Groups, we have CokerpMq -- Zsl ' Zsl`1 ' ' Zsn, for some 1 ď l ď n, where either n " l and sn " 1 (i.e., Coker(M ) is trivial group), or there are (necessarily unique) nonnegative integers sl, sl`1, . . . , sn, for which the nonzero values sl, sl`1, . . . , sr satisfy sj ě 2 for 1 ď j ď r and sisi`1 for l ď i ď r´ 1, and sr`1 " " sn " 0. Coker(M ) is a finite group if and only if r " n. In case l ą 1, we define s1 " s2 " " sl´1 " 1. Clearly then we have since any additional direct summands are isomorphic to the trivial group Z1. CokerpMq -- Zs1 ' Zs2 ' ' Zsl ' ' Zsn, We note that if P, Q are invertible in MnpZq (hence their determinant is 1), then CokerpMq -- CokerpP M Qq. In other words, if N P MnpZq is a matrix which is constructed by performing any sequence of Z-elementary row (or column) operations starting with M , then CokerpMq -- CokerpNq as abelian groups. Definition 2.14. Let M P MnpZq, and suppose CokerpMq -- Zs1 ' Zs2 '' Zsn as described above. The Smith Normal Form of M ((SNF(M ) to be short)) is the n n diagonal matrix diagps1, s2, . . . , sr, 0, . . . , 0q. For any matrix M P MnpZq, the Smith Normal Form of M exists and is unique. If D P MnpZq is a diagonal matrix with entries d1, d2, . . . , dn then clearly CokerpDq -- Zd1 ' Zd2 ' ' Zdn. We also note the following. Proposition 2.15. Let M P MnpZq, and let S denote the Smith Normal Form of M . Suppose the diagonal entries of S are s1, s2, . . . , sn. Then CokerpMq -- Zs1 ' Zs2 ' ' Zsn. In particular, if there are no zero entries in the Smith Normal Form of M , then CokerpMq " s1s2 . . . sn " detpSq " detpMq. LEAVITT PATH ALGEBRAS OF WEIGHTED CAYLEY GRAPHS CnpS, wq 5 Proposition 2.15 yields the following Proposition 2.16. Let E be a finite graph with E0 " n and adjacency matrix AE. Suppose that LpEq is purely infinite simple. Let S be the Smith Normal Form of the matrix In ´ At E, with diagonal entries s1, s2, . . . , sn. Then K0pLpEqq -- Zs1 ' Zs2 ' ' Zsn. E yields Moreover, if K0pLpEqq is finite, then an analysis of the Smith Normal Form of the matrix In ´ At K0pLpEqq "detpIn ´ At Eq , Conversely, K0pLpEqq is infinite if and only if detpIn´At nullitypIn ´ At Eq " 0 and in this case rankpK0pLpEqq " Eq. We record the following theorem which will be used in computations of Smith Normal Forms in later sections Theorem 2.17 (Determinant Divisors Theorem). [14, Theorem II.9] Let M P MnpZq. Define α0 :" 1, and for each 1 ď i ď n, define the ith determinant divisor of M to be the integer αi :" the greatest common divisor of the set of all i i minors of M . Let s1, s2, . . . , sn denote the diagonal entries of the Smith Normal Form of M , and assume that each si is nonzero. Then si " αi αi´1 for each 1 ď i ď n. 2.2. Weighted Cayley graphs and circulant matrices. w, sw, rwq where E1 Let E be a graph and w : E1 Ñ N be a function, the weighted graph of E associated to w w :" te1 . . . ewpeq e P E1u, swpeiq " speq and is a new graph Ew " pE0, E1 rwpeiq " rpeq. Recall that given a group G, and a subset S Ď G, the associated Cayley graph CaypG, Sq is the directed graph EpG, Sq with vertex set tvg g P Gu, and in which there is an edge epg, hq from vg to vh in case there exists (a necessarily unique) s P S with h " gs in G. Thus, in CaypG, Sq, at every vertex vg, the number of edges emitted is S. The identity of G is in S if and only CaypG, Sq contains a loop at every vertex. Definition 2.18. Let G be a group, S Ď G and w : S Ñ N be a map. Then w induces a map (also denoted by w) from the set of edges of CaypG, Sq to N by epg, hq ÞÑ wpsq whenever h " gs. The weighted graph of CaypG, Sq associated to the map w is called the weighted Cayley graph (or w-Cayley graph) and denoted by CaypG, S, wq In particular CaypG, Sq is a special case of CaypG, S, wq when w is the constant map wpeq " 1 for every edge e. In this case we say CaypG, Sq is unweighted. Remark 2.19. CaypG, S, wq is strongly connected if and only if xSy " G. In particular, CaypxSy, S, wq is a connected component of CaypG, S, wq, where xSy is the subgroup generated by S. Notation 2.20. For a positive integer n, let G " Zn, and S be any non-empty subset of G. We denote the w-Cayley graph CaypG, S, wq simply by CnpS, wq. In other words, if S " ts1, s2, . . . sku then the w-Cayley graph CnpS, wq is the directed graph with vertex set tv0, v1, v2, . . . , vn´1u and edge set telpi, sjq 0 ď i ď n ´ 1, 1 ď j ď k, 1 ď l ď wpsjqu for which spelpi, sjqq " vi, and rpelpi, sjqq " vi`sj where indices are interpreted modulo n. Therefore CnpS, wq is a finite graph. Definition 2.21. For a positive integer n, let c " pc0, c1, . . . , cn´1q P Qn. Consider the shift operator T : Qn Ñ Qn, defined by Tpc0, c1, . . . , cn´1q " pcn´1, c0, . . . , cn´2q. The circulant matrix circpcq, associated with c is the n n matrix C whose kth row is T k´1pcq, for k " 1, 2, . . . .n. Thus C is of the form 6 MOHAN.R C " c0 cn´1 ... c1 c1 c0 ... c2 c2 c1 ... c3 . . . . . . . . . . . . cn´1 cn´2 ... c0 ‹‹‹‚ In other words, a circulant matrix is obtained by taking an arbitrary first row, and shifting it cyclically one position to the right in order to obtain successive rows. The pi, jq element of C is cj´i, where subscripts are taken modulo n. Note that ACnpS,wq is the n n matrix with the pi, jqth entry is wpsq if i` s " j modulo n, for some s P S, otherwise 0. Hence ACnpS,wq is a circulant matrix with non-negative integer entries. In the case of unweighted Cayley graph CnpSq, the adjacency matrix is binary circulant matrix. Definition 2.22. For c P Qn, let C " circpcq. The representer polynomial of C is defined to be the polynomial PCpxq " c0 ` c1x ` ` cn´1xn´1 P Qrxs. Lemma 2.23. Let C " circpcq be a circulant matrix. Then the eigenvalues of C equal PCpζ k c0 ` c1ζ k unity. Further nq " n , the primitive nth root of for k " 0, 1, . . . , n ´ 1, where ζn " e 2πi n ` ` cn´1ζ kpn´1q n detpCq " n´1 źl"0n´1 ÿk"0 cjζ lk n ¸ . For a proof of Lemma 2.23 we refer the reader to [11, Theorem 6]. Note that the nth cyclotomic polynomial, denoted by Φnpxq " ź1ďaăn gcdpa,nq"1 px ´ ζ a nq, is an element of Zrxs. Also, xn ´ 1 " śdn fpζnq " 0 for some fpxq P Zrxs implies Φnpxq divides fpxq. By applying Lemma 2.23 we get Lemma 2.24. Let C " circpcq. Then the following are equivalent. Φdpxq. Since Φnpxq is the minimal polynomial of ζn, (a) C is singular (b) PCpζ k (c) The polynomials PCpxq and xn ´ 1 are not relatively prime. nq " 0 for some k P Z. 3. Leavitt path algebras of CnpS, wq wpsq. Then the following are equivalent. Theorem 3.1. Let G be a finite group, S its generating set and w : S Ñ N a weight function. Let W " řsPS (1) LpCaypG, S, wq is purely infinite simple (2) W ě 2 (3) LpCaypG, S, wq does not have Invariant Basis Number Proof. p1q ñ p2q ð p3q. Let G " n. If W " 1, then S " 1. Setting S " tgu we have G is cyclic group generated by g. Hence CaypG, S, wq is the graph Cn which is cycle of length n , which does not satisfy condition L, which contradicts p1q. By [2, Theorem 3.8 and 3.10] LpCnq -- MnpKrx, x´1sq which has Invariant Basis Number. p2q ñ p1q. Let W ě 2. In CaypG, S, wq, the number of edges emitted at each vertex vg is W . So there is at least two edges emitted from each vertex. This also implies condition pLq. Since xSy " G, CaypG, S, wq is strongly connected. Hence for any vertex vg there is a non-trivial path connecting vg to v1 and vice versa. Therefore CaypG, S, wq contains a cycle and there is no non-trivial hereditary subset of vertices. LEAVITT PATH ALGEBRAS OF WEIGHTED CAYLEY GRAPHS CnpS, wq 7 p2q ñ p3q. For a finite graph E, LpEq has Invariant Basis Number if and only if for each pair of positive integers m and n, In MCaypG,S,wq, for each vg we have rvgs " řsPS wpsqrvgss " ÿsPS ÿgPG rvgs " ÿgPGÿsPS m ÿvPE 0rvs " n ÿvPE 0rvs in ME ñ m " n. wpsqrvgss and hence, wpsqrvgss " ÿsPS wpsq ÿgPG ÿgPG rvgss " W ÿgPG rvgss. Since G is a finite group we have G " tgs g P Gu and hence řgPGrvgss " řgPGrvgs. Hence we have řgPGrvgs " W řgPGrvgs. If W ě 2 then LpCaypG, S, wqq does not have Invariant Basis Number. (cid:3) Corollary 3.2. [13, Proposition 4.1, Theorem 4.2] Let G be a finite group, S its generating set. Then the following are equivalent. (1) LpCaypG, Sq is purely infinite simple (2) LpCaypG, Sq does not have Invariant Basis Number (3) S ě 2 Proof. In this case W " S. From now on we work with the following assumption. (cid:3) Let σ "řn´1 σ " wpsjq. Assumption 3.3. Let n P N, S " ts1, s2, . . . sku Ď Zn. s1 ă s2 ă ă sk. Further set W :" řsjPS Theorem 3.4. Let xSy " Zn and W ě 2. Then in the group M divides W ´ 1. Further if gcdpW ´ 1, nq " 1 then order of řn´1 Proof. Let S " ts1, s2, . . . , sku. Then in M rvis " ÿsjPS CnpS,wq, the order of řn´1 i"0 rvis CnpS,wq, we have the following relations i"0 rvis is W ´ 1. wpsjqrvi`sjs. i"0 rvis. Then using the defining relations in M CnpSq,w, we have n´1 n´1 Thus, in the group M ři"0rvi`sjs ři"0 řsjPS wpsjqn´1 wpsjqn´1 wpsjq¸ σ " W σ. wpsjqrvi`sjs¸ " řsjPS ři"0rvis " řsjPS ři"0rvis " " řsjPS CnpS,wq, we have pW ´ 1q σ " 0. This proves the first part of the theorem. CnpS,wq -- CokerpIn ´ At CnpS,wqq, and under the isomorphism rvis ÞÑ CnpS,wqq, where ~bi is the element of Zn which has 1 in the ith coordinate and 0 CnpS,wqq where ~v " p1, 1, . . . , 1qt. This is equivalent to ~u ´ At~u " d~v for some ~u " pu0, u1, . . . , un´1q P Zn, which in turn equivalent to Hence for a natural number d, dσ " 0 in M CnpS,wq if and only if d~v P ImpIn ´ At ~bi ` ImpIn ´ At elsewhere. By Theorem 2.13, M ul ´ ÿsjPS wpsjqun´sj`l " d 0 ď l ď n ´ 1 Adding all the above equations we get n´1 ÿl"0 ul ´ n´1 ÿl"0 ÿsjPS wpsjqun´sj`l " nd 8 MOHAN.R n´1 n´1 LHS " wpsjqun´sj `l un´sj`l n´1 ÿl"0 wpsjq wpsjq ‚ ÿl"0 ul ´ ÿsjPS ÿl"0 ul ´ ÿsjPS " " 1 ´ ÿsjPS ÿl"0 " p1 ´ Wq n´1 ul n´1 ÿl"0 ÿl"0 n´1 ul Thus W ´ 1 divides nd. If gcdpW ´ 1, nq " 1, then W ´ 1 divies d. In particular, when n´1 gcdpW ´ 1, nq " 1 order of ři"0rvis is W ´ 1. Assumption 3.5. In what follows, we always assume that xSy " Zn and W ě 2. As we noted in 2.23 for a circulant matrix C, n´1 (cid:3) detpCq " źl"0n´1 ÿj"0 cj ζ lj n¸ 2πi where ζn " e circulant. Also In ´ At n , the primitive nth root of unity. For CnpS, wq, the adjacency matrix ACnpS,wq is CnpS,wq is circulant (with integer entries). Let S " ts1, s2, . . . , sku. Then, detpIn ´ At CnpSqq " detpIn ´ ACnpSqq " n´1 źl"0 1 ´ ÿsjPS wpsjqζ lsj n ‚. Proposition 3.6. Let S0 :" tj P S j " 0 pmod 2qu, S1 :" tj P S j " 1 pmod 2qu, W0 :" řsj PS0 CnpS,wqq ą 0 if and only if n is even and 1 ` W1 ă W0. Proof. Let Ppxq " 1 ´řsj Ppζ l nq " 1´řsj for all l. Thus detpIn ´ At wpsjq, and W1 :" řsjPS1 CnpS,wq. Let zl " wpsjqζ lsj . It is easy to see that z0 " 1´řsjPS wpsjq " 1´ W ă 0 and zn´l " zl wpsjqxsj be the representer polynomial of In ´ At wpsjq. Then detpIn ´ At 2 ă 0. Since CnpSqq ą 0 if and only if n is even and z n wpsjq, z n wpsjq ` ÿj odd 2 " 1 ´ ÿj even 2 ă 0 iff 1 ` W1 ă W0. Thus z n Proposition 3.7. Let Ppxq P Zrxs be the representer polynomial associated with the circulant matrix In ´ At CnpS,wq. Then K0pLpCnpS, wqqq is infinite if and only if Ppxq and xn ´ 1 are relatively prime. (cid:3) Proof. Follows from Lemma 2.24 and Proposition 2.16 (cid:3) In order to compute the Grothendieck group of the Leavitt path algebra of CnpS, wq, we look at the generating relations for M CnpS,wq rvis " ÿsjPS wpsjqrvi`sjs. where 0 ď i ď n ´ 1, (subscripts are modulo n) and S " ts1, s2, . . . , sku, (sl ă sm for l ă m). Any statement about rv0s in M CnpS,wq, has an analogous statement for rvks for 0 ď k ď n ´ 1, by symmetry of relations. LEAVITT PATH ALGEBRAS OF WEIGHTED CAYLEY GRAPHS CnpS, wq 9 Definition 3.8. The companion matrix of the monic polynomial pptq " c0`c1t``cn´1tn´1` tn, the n n matrix defined as Tppq " 0 0 . . . 0 ´c0 1 0 . . . 0 ´c1 0 1 . . . 0 ´c2 ... ... 0 0 . . . 1 ´cn´1 . . . ... ... ‹‹‹‹‹‚ Tppq " 0 0 . . . 0 c0 1 0 . . . 0 c1 0 1 . . . 0 c2 ... ... ... 0 0 . . . 1 cn´1 . . . ... ‹‹‹‹‹‚ Let a linear recursive sequence is of the form un`k ´ cn´1un`k´1 ´ ´ c0un " 0 pn ě 0q, where c0, c1, . . . , cn´1 are constants. The characteristic polynomial of the above linear recursive sequence is defined as pptq " tn ´ cn´1tn´1 ´ ´ c1t ´ c0 whose companion matrix is This matrix generates the sequence in the sense that, `ak ak`1 . . . ak`n´1 Tppq "`ak`1 ak`2 . . . ak`n In particular, the pn, nqth entry of Tppqk is un`k´2. When 0 R S, from the linear recursive relation in M nomial ppS, w, tq " tsk ´řsjPS wpsjqtsk´sj . The companion matrix of ppS, w, tq is denoted by TCnpS,wq, is then the sk sk matrix CnpS,wq, we have the characteristic poly- TCnpS,wq " 0 0 . . . 0 1 0 . . . 0 0 1 . . . 0 ... ... . . . ... 0 0 . . . 1 ‹‹‹‹‹‹‹‚ c In M where c is the last column of TCnpS,wq which contains entry wpsjq at positions sk ´ sj ` 1 and 0 elsewhere. CnpS,wq, we observe that by writing the generating relations and then expanding the equation such that the subscripts are kept in increasing order, at ith step we get the coefficients to be the last column of T i CnpS,wq. The computation of the Smith Normal Form of In ´ At CnpS,wq is the key tool for determining the K0 of the Leavitt path algebra of CnpS, wq. We show that this computation reduces to calculating the Smith Normal Form of an sk sk matrix. Theorem 3.9. Let n P N, S " ts1, s2, . . . , sku Ĺ Zn such that xSy " Zn, 0 R S, and W ě 2. Then CokerpIn ´ At Proof. Since the Smith normal form of In ´ At cokernels are same and we only show that CokerpACnpS,wq ´ Inq -- CokerpT n simplicity we write B " ACnpS,wq ´ In and T " TCnpS,wq. First we observe that CnpS,wq and ACnpS,wq ´ In are the same, their CnpS,wq ´ Inq. For CnpS,wqq -- CokerpT n CnpS,wq ´ Inq. ´1 wpsjq 0 Bpq "$'& '% q " p q " p ` sj if if otherwise 10 MOHAN.R Let P be a psk skq lower triangular matrix given by 0 wpsjq 0 q ą p if if p ´ q " sk ´ sj otherwise Ppq "$'& '% and let Q be a psk skq upper triangular matrix given by Qpq " 0 ´1 wpsjq 0 q ă p q " p q ´ p " sj if if if otherwise $'''& '''% It is direct that P and Q are invertible. Let R " ´Q´1. Then a direct computation yields P R " T sk, and also QR " ´Isk . Let P 1 be the block matrix "P 0skpn´skq‰ and Q1 be the block matrix "0skpn´skq Q‰. The psk pn ´ skqq submatrix of B consisting of bottom s ´ k rows can be written as P 1 ` Q1. The first pn´ skq reduction steps of the Smith normal form will result in an pn´ skqpn´ skq identity submatrix in the upper left corner. On the bottom sk rows, the ith reduction step adds the ith column to the sum of wpsjq times pi` skqth columns, then zeros out the ith column. The matrix that accomplishes this reduction step is P ´1T P " r ... 1 0 0 . . . 0 0 0 1 0 . . . 0 0 ... ... 0 0 0 . . . 1 0 . . . ... ... ‹‹‹‹‹‚ where r is the first row contains entry wpsjq at positions sk and 0 elsewhere. After i reduction steps, the first psk skq submatrix with nonzero column vectors on the bottom sk rows will be Therefore the first pn ´ skq reduction steps of the Smith Normal Form will result in the following form. P pP ´1T Pqi " T iP. B „ Ipn´skq 0skpn´skq T n´skP ` Q 0pn´skqsk Since pT n´sk P ` QqR " T n ´ Isk , B „ Ipn´skq 0kpn´skq 0pn´skqsk T n ´ Isk Hence CokerpBq -- CokerpT n ´ Iskq. 4. Illustrations (cid:3) n´1 As illustrations of the above discussion we consider some simple cases when W ě 2 and ři"0rvis is the identity in M CnpS,wq, which recovers the examples obtained in [8],[5],and [6]. LEAVITT PATH ALGEBRAS OF WEIGHTED CAYLEY GRAPHS CnpS, wq 11 4.1. S " Zn. In this subsection we only look at the following two simple cases when in M CnpZn,wq n´1 ři"0rvis is the identity Definition 4.1. Let n, l be two positive integers. We define K plq n to be the graph with n vertices v0, v1, . . . vn´1, in which there is exactly one edge from vi to vj for each 0 ď i ‰ j ď n ´ 1 and l loops at each vertex. We call K plq Theorem 4.2. Let n ě 2 be a positive integer. n the complete n-graph with l loops. (1) LpK p1q (2) Let E be a finite graph such that LpEq is purely infinite simple. If K0pLpEqq -- Zn and n q -- Lp1, nq. rLpEqs is identity in K0pLpEqq, then LpEq -- LpK p2q n`1q. Proof. Let wl : S Ñ N be the weight function defined by wlp0q " l and wlpiq " 1 for 1 ď i ď n´1. Then it is direct that CnpZn, wlq -- K plq n q " ζ ljq " ´pn ´ 1q ă 0. detpIn ´ At n . 1. We note that p´1qp n´1 n´1 p1q K źl"0 ÿj"1 Also we have W ´ 1 " S ´ 1 " n ´ 1. So gcdpW ´ 1, nq " 1 and hence in M K . Also determinant divisors theorem yields that p1q n n´1 ři"0rvis is the identity SNFpIn ´ At n q " diagp1, 1, . . . , 1, n ´ 1q. n qq -- Zn´1. By Proposition 2.12, the result follows. p1q K Hence, K0pLpK p1q and rank(In ´ At 2. We note that pIn´At K K p2q n q " 1. Therefore if n ě 2, then n q is the nn matrix with every entry ´1. Hence detpIn´At p2q K p2q n q " 0 Also in K0pLpK p2q K0pLpK p2q ÿi"1rvis "2rv0s ` n qq -- Zn´1. ÿi"1rvis¸ ` n´1 n´1 n´1 n qq, ÿi"0rvis " rv0s ` n´1 σ " ři"0rvis is the identity in K0pLpK p2q Hence result. 4.2. S " 1. n´1 n´1 ÿi"1rvis " 2 ÿi"0rvis " 2σ. n qq. Applying Algebraic KP Theorem we have the (cid:3) Let S " tiu. Since xSy " Zn, gcdpi, nq " 1 and the weight function w : S Ñ N is given by n be the graph with n vertices v0, v1, . . . , vn and kn edges such that every wpiq " W . Let Dk vertex vi emit k edges to vi`1. We call Dk n an k-cycle of length n. Dk n " pkq vn´2 pkq vn´1 pkq v0 pkq v1 pkq v2 pkq 12 MOHAN.R It is easy to see that CnpS, wq -- DW n . The generating relations for M are given by DW n rvis " Wrvi`1s for 0 ď i ď n, where the subscripts are interpreted mod n. So for each 0 ď i ď n we have that rvis " Wrvi`1s " W 2rvi`2s " " W n´irvn´1s " W n`1´irv0s. In particular, each rvis is in the subgroup of M DW n i ď n ´ 1u generates M , we conclude that M We also observe that DW n DW n generated by rv0s. Since the set trvis 0 ď is cyclic, and rv0s is a generator. detpIn ´ At DW n q " p1 ´ W ζ lq " 1 ´ W n ă 0. n´1 źl"0 We conclude that K0pLpDW n qq " W n ´ 1. Thus we have K0pLpCnpS, wqq -- M DW n -- ZW n´1. Proposition 4.3. Let S " tiu, gcdpi, nq " 1, and gcdpW ´ 1, nq " 1, then LpCnpS, wqq -- MW n´1pLp1, W nqq. follows. i"0 rvis is the identity in the group M Proof. řn´1 Corollary 4.4. ([5], Proposition 3.4) Assume the hypothesis of Proposition 4.3 and W " 2, then LpCnpS, wqq -- M2n´1pLp1, 2nqq 4.3. S " 2. CnpS,wq. Hence by Proposition 2.12 the result (cid:3) Let S " ts1, s2u with s1 ă s2. Let a, b P N. We define wps1q :" a and wps2q :" b. Thus W " a ` b ě 2. Since xSy " Zn, it is sufficient to consider only the following subcases: (1) s1 " 0 and s2 " 1. (2) s1 " 1. (3) s1 and s2 divide n with 1 ă s1 ă s2, and gcdps1, s2q " 1. In what follows we consider these subcases separately. Lemma 4.5. In each of the above subcases if a " b " 1, then Proof. Since W ´ 1 " S ´ 1 " 1 in these subcases we have gcdpW ´ 1, nq " 1 and the result follows from Theorem 3.4. ři"0rvis is the identity in M CnpS,wq. (cid:3) n´1 Proposition 4.6. Let n, a, b P N be fixed. Let 0 ď s1 ă s2 ď n ´ 1. Consider the w-Cayley graph CnpS, wq where S " ts1, s2u and wps1q " a, wps2q " b. Then detpIn ´ At CnpS,wqq " 0 if and only if exactly one of the following occurs: (1) a " b " 1, n " 0 pmod 6q, s2 " 5s1 pmod 6q. (2) a " b ` 1, n is even, s1 is even, s2 is odd. (3) b " a ` 1, n is even, s1 is odd, s2 is even. Proof. Let ∆ " detpIn ´ At CnpS,wqq and zl " aζ ls1 ` bζ ls2. Since źl"0´1 ´ aζ ls1 ´ bζ ls2¯ ∆ " n´1 Then ∆ " 0 if and only if zl " 1 for some l. We observe that z0 " a ` b ą 1 and zn´l " zl. So we can write LEAVITT PATH ALGEBRAS OF WEIGHTED CAYLEY GRAPHS CnpS, wq 13 n´1 2 ∆ " p1 ´ z0q śl"1p1 ´ zlqp1 ´ zlq p1 ´ z0qp1 ´ ap´1qs1 ´ bp´1qs2q $'''& '''% n 2 ´1 śl"1 p1 ´ zlqp1 ´ zlq 2s is the integer part of n if n is odd if n is even Hence we can assume 1 ď l ď r n aζ ls1 ` bζ ls2 " 1. So 1 "aζ ls1 ` bζ ls2 ě a ´ b " a ´ b ě 0. Since a, b P N, only possibli- ties are a " b, a " b ` 1, or b " a ` 1. 2 . Further, zl " 1 implies 2s, where r n Case 1: a " b Let θ " 2πl n . Then zl " 1 if and only if apcos s1θ ` cos s2θq " 1 and apsin s1θ ` sin s2θq " 0. The second equation implies that s1θ " ´s2θ pmod 2qπ. Substituting back in first equation we get, 1 " apcosp´s2θq ` cos s2θq " 2a cos s2θ s2 " arccos 1 2a . ñ cos s2θ " 2 " π Thus n P N only if a " 1. Assuming a " 1, we have arccos 1 see that 5π 3 ñ 5n " 6s2l. π 3 ñ n " 6s2l, 2πls2 n " 1 2a ñ n " 3 or 5π 2πl n 2πls2 or 3 . Substituting back, we In either case, n " 0 pmod 6q. Also, s2θ " ´s1θ pmod 2qπ implies that for some integer m, ps2 ` s1q π 3 " 2πm ñ s2 ` s1 " 6m or ps2 ` s1q In either case, s2 ` s1 " 0 pmod 6q, or s2 " 5s1 pmod 6q. `´e zl " ωls1 ` ωls2 "´e 6 ¯s1 2πi 2πi 6 ¯´s1 " 1 Conversely, when a " 1, n " 0 pmod 6q and s2 " 5s1 pmod 6q, then letting l " 6 implies that 5π 3 " 2πm ñ 5ps2 ` s1q " 6m. Case 2: a " b ` 1 As in case 1, let θ " 2πl n . Then zl " 1 if and only if pb ` 1q cos s1θ ` b cos s2θ " 1 and pb ` 1q sin s1θ ` b sin s2θ " 0. b`1 sin s2θ¯. Substituting back in the first sin s2θ ` b cos s2θ " 1. The second equation implies that s1θ " arcsin´ ´b equation we get, pb ` 1q cosarcsin ´b b ` 1 Since cosparcsin xq " ?1 ´ x2, we have pb ` 1qd1 ´ b ab2 ` 2b ` 1 ´ b2 sin2 s2θ " 1 ´ b cos s2θ. sin s2θ2 Squaring both sides, b ` 1 Hence, ` b cos s2θ " 1. b2 cos2 s2θ ` 2b ` 1 " b2 cos2 s2θ ´ 2b cos s2θ ` 1 ñ cos s2θ " ´1. Therefore, s2θ " π pmod 2qπ. Substituting θ " 2πl n , we see that n is even. Also, s2θ " π pmod 2qπ implies that ps2 ´ 1qπ " 2πm for some integer m. So, s2 " 2m ` 1 or s2 is odd. Also 14 MOHAN.R b`1 sin s2θ¯ " arcsinp0q, s1π " 0 or π. pb ` 1q cos s1π ´ b " 1 ñ s1π " 0 since, s1θ " arcsin´ ´b pmod 2qπ. Hence s1 is even. Conversely, let n, s1 be even and s2 be odd then by taking l " n zl " pb`qωs1 l " pb ` 1qp´1qs1 ` bp´1qs2 " b ` 1 ´ b " 1. l ` bωs2 2 , we get Case 3: b " a ` 1 The proof is similar to that of case 2. Corollary 4.7. Assume the hypothesis of Proposition 4.6. Further assume that LpCnpS, wqq is unital purely infinite simple. Then K0pLpCnpS, wqqq is infinite abelian group if and only if one of the following holds: (cid:3) (1) a " b " 1, n " 0 pmod 6q, s2 " 5s1 pmod 6q. (2) a " b ` 1, n is even, s1 is even, s2 is odd. (3) b " a ` 1, n is even, s1 is odd, s2 is even. In which case rankpK0pLpCnpS, wqqqq " n ´ rankpIn ´ ACnpS,wqq. 4.3.1. Subcase 2.1: S " t0, 1u. Let F pa,bq n vertex vl, there are a loops and b edges getting emitted into vl`1 (subscripts are mod n). be the graph with n vertices v0, v1, . . . , vn´1 and ak ` bk edges such that at every F pa,bq n " paq paq pbq pbq pbq vn´1 v0 vn´2 v1 v2 pbq paq pbq pbq paq paq Then CnpS, wq -- F pa,bq n when S " t0, 1u. We note that n´1 detpIn ´ At F pa,bq n q " p1 ´ a ´ bζ lq " p1 ´ aqn ´ bn. źl"0 Lemma 4.8. Let n, a, b P N. Then detpIn ´ At F pa,bq n q ě 0 if and only if n is even and a ě b ` 1. Moreover, detpIn ´ At Proof. We refer to the proof of Proposition 4.6. We need to substitute s1 " 0, and s2 " 1. Since, q " 0 if and only if n is even and a " b ` 1. pa,bq n F n´1 2 p1 ´ z0q śl"1p1 ´ zlqp1 ´ zlq ∆ " n 2 ´1 $'''& '''% p1 ´ z0qp1 ´ ap´1qj ´ bp´1qkq śl"1 p1 ´ zlqp1 ´ zlq ∆ ě 0 if and only if n is even and a ě b ` 1, in which case 1 ´ z n follows that, detpIn ´ At q " 0 if and only if n is even and a " b ` 1. F pa,bq n 2 " 1 ´ a ` b ď 0. Also, it if n is odd if n is even LEAVITT PATH ALGEBRAS OF WEIGHTED CAYLEY GRAPHS CnpS, wq 15 (cid:3) We describe the Smith Normal Form of In ´ At F . pa,bq n Lemma 4.9. Suppose n P N. Let T be the n n circulant matrix whose first row is ~t " pp1 ´ aq,´b, 0, . . . , 0q.Let gcdp1 ´ a, bq " d. Then the Smith Normal Form SNFpTq " diagd, d, . . . , d, p1 ´ aqn ´ bn dn´1 Proof. In order to compute Smith Normal Form of T , we use at the determinant divisors theorem and look at i i minors of T for each 1 ď i ď n. Let αi be the gcd of the set of all i i minors of T and α0 " 1. Then SNFpTq " diag α1 α0 , α2 α1 αn´1 . , . . . , detpTq By the definition of T , it is easy to observe that αi " gcd`pa ´ 1qi, bi " gcdpa, bqi " di for 1 ď i ď n ´ 1 and detpTq " p1 ´ aqn ´ bn. Therefore d2 d , , 1 SNFpTq " diag d , . . . , p1 ´ aqn ´ bn " diagd, d, d, . . . , p1 ´ aqn ´ bn d3 d dn´1 dn´1 (cid:3) Theorem 4.10. Let n, a, b P N be fixed. Suppose S " t0, 1u Ă Zn, w : S Ñ N be defined as wp0q " a and wp1q " b. Let d " gcdpa ´ 1, bq. Then K0pLpCnpS, wqqq -- # pZdqn´1 ' Z pZdqn´1 ' Z p1´aqn´bn Proof. Follows from the above lemmas 4.8 and 4.9. dn´1 if a " b ` 1 and n is even otherwise (cid:3) Example 4.11. LpCnp0, 1qq -- Lp1, 2q. Proof. In Theorem 4.10 we take a " b " 1. Then gcdpa ´ 1, bq " 1. Hence detpIn ´ At Cnp0,1qq " ´1 ă 0 and K0pLpCnp0, 1qqq is trivial. By Proposition 2.12 we have LpCnp0, 1qq -- Lp1, 2q. (cid:3) The above example was observed in [5], Proposition 3.3. 4.3.2. Subcase 2.2: S " t1, ju with j ą 1. We note that by Proposition 3.6 detpIn ´ At b ą a ` 1. Also by Proposition 4.6 detpIn ´ At occurs: CnpS,wqq ą 0 if and only if n, j are even and CnpS,wqq " 0 if and only if one of the following (1) a " b " 1, n " 0 pmod 6q, j " 5 pmod 6q (2) b " a ` 1, n, j are even. In order to compute K0pLpCnpS, wqqq, we apply Theorem 3.9 and compute the Smith normal form of T n CnpS,wq´In. This procedure is performed for unweighted Cayley graph in [6]. However, the record an interesting example here. 16 MOHAN.R 4.3.3. Leavitt Path algebras of Cayley graphs of Dihedral groups. Let Dn be the dihedral group of order 2n. i.e. Dn " @r, s rn " s2 " e, rsr " sD. Let Dn denote the Cayley graph of Dn with respect to the generating subset S " tr, su. The following discussion is taken from [7]. A graph transformation is called standard if it is one of the following types: in-splitting, in-amalgamation, out-splitting, out-amalgamation, expansion, or contraction. For definitions the reader to referred to [3]. If E and F are graphs having no sources and no sinks, a flow equivalence from E to F is a sequence E " E0 Ñ E1 Ñ Ñ En " F of graphs and standard graph transformations which starts at E and ends at F . Proposition 4.12. [3, Corollary 6.3.13] Suppose E and F are finite graphs with no sources whose corresponding Leavitt path algebras are purely infinite simple. Then E is flow equivalent to F if and only if detpIE ´ AEq " detpIF ´ AFq and CokerpIE ´ AEq -- CokerpIF ´ AFq. Definition 4.13 (In-splitting). Let E " pE0, E1, r, sq be a directed graph. For each r´1pvq ‰ φ, partition the set r´1pvq into disjoint nonempty subsets E v mpvq where mpvq ě 1. If v is a source then set mpvq " 0. Let P denote the resulting partition of E1. We form the in-split graph ErpPq from E using the partition P as follows: 1 , . . . , E v ErpPq0 " tvi v P E0, 1 ď i ď mpvqu Y tv mpvq " 0u, ErpPq1 " tej e P E1, 1 ď j ď mpspeqqu Y te mpspeqq " 0u, and define rErpPq, sErpPq : ErpPq1 Ñ ErpPq0 by sErpPqpejq " speqj and sErpPqpeq " speq rErpPqpejq " rpeqi and sErpPqpeq " speqi where e P E rpeq i . We observe that Dn can be obtained from C n´1 by the standard operation in-splitting with respect to the partition P of the edge set of C n´1 that places each edge in its own singleton partition class. In [8] the collection of Leavitt path algebras tLpC n´1 q n P Nu is completely described and by Proposition 4.12 we have that the same description holds true if we replace C n´1 n with Dn for every n P N. Hence we have Theorem 4.14. For each n P N, detpIn ´ At Dnq ď 0. And n n n (1) If n " 1 or 5 pmod 6q then K0pLpDnqq -- t0u and LpDnq -- Lp1, 2q. (2) If n " 2 or 4 pmod 6q then K0pLpDnqq -- Z{3Z and LpDnq -- M3pLp1, 4qq. (3) If n " 3 pmod 6q then K0pLpDnqq -- pZ{2Zq2 (4) If n " 0 pmod 6q then K0pLpDnqq -- Z2 and LpDnq -- LpK p2q 3 q 4.3.4. Subcase 2.3: S " ts1, s2u where s1, s2 divide n, 1 ă s1 ă s2 and gcdps1, s2q " 1. By Proposition 4.6 and by Proposition 3.6, we have that detpIn ´ At CnpS,wqq " 0 if and only if one of the following occurs: (1) a " b " 1, n " 0 pmod 6q, d2 " 5d1 pmod 6q. (2) a " b ` 1, n, d1 are even, d2 is odd. (3) b " a ` 1, n, d2 are even, d1 is odd. and detpIn ´ At CnpS,wqq ą 0 if and only if one of the following occurs: (1) a ą b ` 1, n, d1 are even, d2 is odd. (2) b ą a ` 1, n, d2 are even, d1 is odd. CnpS,wq ´ In. In order to compute K0pLpCnpS, wqqq, we apply Theorem 3.9 and compute the Smith Normal form of T n We illustrate this when S " td1, d2u, where d1, d2 divides n, gcdpd1, d2q " 1 and a " b " 1. In this special case we have detpIn ´ At Cnpd1,d2qq " 0 if and only if n " 0 pmod 6q and d2 " 5d1 pmod 6q. In all other cases, we have detpIn ´ At nq. In order to compute K0pLpCnpd1, d2qqq, we apply Theorem 3.9 and compute the Smith normal form of T n nq ă 0. Define Hpd1,d2qpnq :" detpIn ´ At C k C k Cnpd1,d2q ´ In. LEAVITT PATH ALGEBRAS OF WEIGHTED CAYLEY GRAPHS CnpS, wq 17 For 1 ď j, k P N let us define a sequence Fpj,kq recursively as follows: Fpj,kqpnq " In M Cnpd1,d2q, we have 0 1 0 Fpj,kqpn ´ jq ` Fpj,kqpn ´ kq 1 ď n ď k ´ 2 if if n " k ´ 1 if n " k if n ě k ` 1 $'''& '''% rv0s " rvjs ` rvks " rv2js ` rvks ` rvj`ks " rv3js ` rvks ` rvj`ks ` rv2j`ks " . . . The coefficients appearing in the above equations are terms in the sequence Fpd1,d2q and corre- sponding TCnpd1,d2q is given by the following Lemma 4.15. For fixed d1, d2, let d2 ´ d1 " k. Let T " TCnpd1,d2q. Suppose Gpnq :" Fpd1,d2qpnq be the sequence defined above. Then for each n P N, T n " Gpn ´ 1q Gpn ´ 2q ... Gpnq Gpn ´ 1q ... . . . . . . Gpn ` d1 ´ 1q Gpn ` d1q ... ... Gpnq Gpn ` 1q Gpn ` d2 ´ 2q Gpn ` d2 ´ 3q ... . . . . . . Gpn ` d2 ` d1 ´ 2q . . . . . . Gpn ` d2 ´ 1q ... ‹‹‹‹‹‹‹‚ where the highlighted row is pk ` 1qth row. Proof. We prove the lemma by induction on n. We extend the definition of G to the negative integers as well. Then, T " ... ... 0 0 . . . 0 . . . 0 1 1 0 . . . 0 . . . 0 0 ... ... 0 0 . . . 1 . . . 0 1 ... ... 0 0 . . . 0 . . . 0 0 0 0 . . . 0 . . . 1 0 ... ... ... ... ‹‹‹‹‹‹‹‹‹‚ " Gp0q Gp´1q ... Gp1q Gp0q ... Gpd1q Gpd1 ` 1q ... ... . . . Gpk ´ 1q . . . Gpk ´ 2q ... . . . Gpd2 ´ 1q ... . . . . . . Gpd2 ´ 2q Gpd2 ´ 3q ... Gpd2 ´ 1q Gpd2 ´ 2q ... . . . Gpd2 ` d1 ´ 2q Gpd2 ` d1 ´ 1q ... ... Gp2q Gp1q Gp3q Gp2q . . . Gpk ` 1q . . . Gpkq . . . . . . Gpd2q Gpd2 ´ 1q Gpd2 ` 1q Gpd2q ‹‹‹‹‹‹‹‹‹‚ where highlighted column is kth column. 18 MOHAN.R Thus we have the statement true for n " 1. Now suppose . . . . . . Gpn ´ 2q Gpn ´ 3q Gpn ´ 1q Gpn ´ 2q ... ... ... ... Gpn ` d1 ´ 2q Gpn ` d1 ´ 1q Gpn ´ 1q Gpnq Gpn ` d2 ´ 3q Gpn ` d2 ´ 4q ... . . . . . . Gpn ` d2 ` d1 ´ 3q . . . . . . Gpn ` d2 ´ 2q ... ‹‹‹‹‹‹‹‚ T n´1 " Gpn ´ 2q Gpn ´ 3q ... Then, " ... . . . 0 0 . . . 1 1 0 . . . 0 ... ... 0 0 . . . 1 ... ... 0 0 . . . 0 . . . ... ‹‹‹‹‹‹‹‚ T n " T n´1T . . . Gpn ` k ´ 2q . . . Gpn ` k ´ 3q . . . . . . ... Gpn ` d2 ´ 3q Gpn ` d2 ´ 4q ... Gpn ` d1 ´ 2q ... . . . Gpn ` d2 ´ 2q ... . . . Gpn ` d2 ` d1 ´ 3q ... . . . Gpn ` k ´ 2q . . . Gpn ´ 1q Gpn ´ 1q Gpn ´ 2q ... ... " Gpnq Gpn ´ 1q Gpn ´ 2q Gpn ` d1 ´ 1q Gpn ` d1q Gpn ` d2 ´ 2q Gpn ´ 2q ` Gpn ` k ´ 2q Gpn ´ 3q ` Gpn ` k ´ 3q . . . . . . Gpn ` d1 ´ 2q ` Gpn ` d2 ´ 2q . . . . . . ... Gpn ´ 1q ` Gpn ` k ´ 2q . . . . . . Gpn ` d2 ´ 2q Gpn ` d2 ´ 3q ... ... ‹‹‹‹‹‹‹‚ . . . . . . Gpnq Gpn ´ 1q ... ... Gpn ` 1q Gpnq Gpn ´ 1q ... ... " ... ... Gpn ` d1 ´ 1q Gpn ` d1q Gpnq Gpn ` 1q . . . . . . Gpn ` d2 ` d1 ´ 2q . . . . . . Gpn ` d2 ´ 1q ... ‹‹‹‹‹‹‹‚ ‹‹‹‹‹‹‹‚ (cid:3) Using the determinant divisors theorem, the Smith normal form of T n reduced to SN FpT n Cnpd1,d2q ´ Ikq " α1pnq α2pnq α1pnq . . . αd2 pnq αd2´1pnq Cnpd1,d2q ´ Ik can be ‹‹‹‹‚ where αi is the greatest common divisor of the set of all i i minors of T n Example 4.16. Let n " 6, d1 " 2, d2 " 3. The corresponding Cayley graph is Cnpd1,d2q. v5 v0 C6p2, 3q " v4 v1 v3 v2 LEAVITT PATH ALGEBRAS OF WEIGHTED CAYLEY GRAPHS CnpS, wq 19 Corresponding companion matrix is given by and, T " T 6 ´ I3 " 0 1 0 ‚ 1 2 1 ‚ SN FpT 6 ´ I3q " 0 0 7 ‚ Hence, K0pLpC6p2, 3qqq -- Z7 and LpC6p2, 3qq -- Lp1, 8q. whose Smith normal form is given by 0 0 1 1 0 1 0 1 2 2 1 3 1 0 0 0 1 0 Acknowledgment The author would like to thank B.Sury for fruitful discussions during the preparation of this paper. The author is grateful to Aditya Challa for his valuable help with Python Programming Language. The author sincerely thanks Ramesh Sreekantan, Roozbeh Hazrat, Gene Abrams, and Crist´obal Gil Canto for their very useful comments towards improving the paper. The author gratefully acknowledges Department of Atomic Energy (National Board for Higher Mathematics), Government Of India for their financial support through Ph.D. Scholarship. References [1] G. Abrams, P. N. ´Anh, A. Louly, and E. Pardo. The classification question for Leavitt path algebras. J. Algebra, 320(5):1983 -- 2026, 2008. [2] G. Abrams, G. Aranda Pino, and M. Siles Molina. Locally finite Leavitt path algebras. Israel J. Math., 165:329 -- 348, 2008. [3] Gene Abrams, Pere Ara, and Mercedes Siles Molina. Leavitt path algebras, volume 2191 of Lecture Notes in Mathematics. Springer, London, 2017. [4] Gene Abrams and Gonzalo Aranda Pino. Purely infinite simple Leavitt path algebras. J. Pure Appl. Algebra, 207(3):553 -- 563, 2006. [5] Gene Abrams and Gonzalo Aranda Pino. The Leavitt path algebras of generalized Cayley graphs. Mediterr. J. Math., 13(1):1 -- 27, 2016. [6] Gene Abrams, Stefan Erickson, and Crist´obal Gil Canto. Leavitt path algebras of Cayley graphs C j n. Mediterr. J. Math., 15(5):Art. 197, 23, 2018. [7] Gene Abrams, Adel Louly, Enrique Pardo, and Christopher Smith. Flow invariants in the classification of Leavitt path algebras. J. Algebra, 333:202 -- 231, 2011. [8] Gene Abrams and Benjamin Schoonmaker. Leavitt path algebras of Cayley graphs arising from cyclic groups. In Noncommutative rings and their applications, volume 634 of Contemp. Math., pages 1 -- 10. Amer. Math. Soc., Providence, RI, 2015. [9] P. Ara, K. R. Goodearl, and E. Pardo. K0 of purely infinite simple regular rings. K-Theory, 26(1):69 -- 100, 2002. [10] P. Ara, M. A. Moreno, and E. Pardo. Nonstable K-theory for graph algebras. Algebr. Represent. Theory, 10(2):157 -- 178, 2007. [11] Irwin Kra and Santiago R. Simanca. On circulant matrices. Notices Amer. Math. Soc., 59(3):368 -- 377, 2012. [12] W. G. Leavitt. The module type of a ring. Trans. Amer. Math. Soc., 103:113 -- 130, 1962. [13] T. G. Nam and N. T. Phuc. The structure of Leavitt path algebras and the invariant basis number property. J. Pure Appl. Algebra, 223(11):4827 -- 4856, 2019. [14] Morris Newman. Integral matrices. Academic Press, New York-London, 1972. Pure and Applied Mathemat- ics, Vol. 45. Statistics and Mathematics Unit, Indian Statistical Institute Bangalore, India E-mail address: [email protected]
1102.2965	1	1102	2011-02-15T05:26:26	Simple archimedean dimension groups	[ "math.RA", "math.FA" ]	We answer a question of Goodearl, by constructing for every metrizable Choquet simplex, a dimension group that is simple and archimedean and whose trace space is the desired Choquet simplex.	math.RA	math	Simple archimedean dimension groups0 David Handelman1 Let (G, u) be an unperforated partially ordered abelian group with an order unit, u. It is simple if every nonzero positive element is an order unit. By [G, ], it is archimedean2 if and only if for g in G, τ (g) ≥ 0 for all pure traces τ implies g ∈ G+. Among unperforated groups, archimedean and simple represent properties that are maximal and minimal, respectively -- the former are those for which the weakest necessary condition (that the values at traces be nonnegative) implies positivity, while the latter are those for which the strongest sufficient condition (that positivity implies strict positivity on traces) is implied by positivity. So it is a little difficult to construct dimension groups that are both simple and archimedean, aside from subgroups of the reals. Question 4 of [G] asks whether every Choquet simplex can be the trace space a of a simple archimedean dimension group; we show this is the case for metrizable simplices, based on an interesting construction over the interval. The normalized trace space of (G, u) is denoted S(G, u); the latter's extremal boundary is denoted ∂eS(G, u). We use to denote the natural map G → Aff S(G, u), given by g(τ ) = τ (g). An alternative formulation of simple and archimedean (in the presence of an order unit) is the following. Suppose G admits an unperforated partial ordering such that the map G → Aff S(G, u) is an embedding; then the ordering is unique (the smallest ordering is strict ordering, the largest ordering is the pointwise one). The following is practically tautological. LEMMA 1 Let (G, u) be an unperforated partially ordered abelian group with order unit. Then G is simple and archimedean if and only for all g ∈ G \ {0}, inf τ ∈∂eS(G,u) τ (g) 6= 0. Proof. A standard facial argument shows that inf τ ∈∂eS(G,u) τ (g) > 0 implies inf τ ∈S(G,u) τ (g) > 0 (even though ∂eS(G, u) need not be compact) -- the condition implies that τ (g) ≥ 0 for all traces τ ; set F = {τ ∈ S(G, u) τ (g) = 0} -- it is easy to see that if nonempty, this is a closed face, hence has extreme points, which (since F is a face) are extreme with respect to S(G, u). Assume G is archimedean and simple; the former says that inf τ ∈∂eS(G,u) τ (g) = 0 entails g ∈ G+; simplicity implies g would be an order unit, hence the infimum would be strictly greater than zero. Conversely, suppose nonzero g satisfies τ (g) ≥ 0 for all pure τ . By hypothesis (the infimum is not zero, hence must be strictly positive), τ (g) > 0 for all pure τ , hence (by the first paragraph), τ (g) > 0 for all traces, whence g is an order unit and thus in the positive cone, so G is simple and • archimedean (simultaneously). Note that the criterion refers to all nonzero g, not just those in G+ (which would characterize simplicity). In particular, a simple archimedean group which is also an ordered real vector space must be the reals with the usual ordering (pick any nonzero g in G+, and let α = inf τ (g); 0 Working document. 1 Supported in part by a Discovery Grant from NSERC. 2 This uses the classical definition of archimedean: for elements g and h of G, ng ≤ h for all positive integers n implies −g ≥ 0. In the presence of an order unit, this is equivalent to the trace-determining condition in the text, and in particular shows that the map G → Aff S(G, u) is an embedding (an order-embedding, in fact) when G is archimedean. A much weaker definition -- irrelevant here -- is used by a large group of workers in real algebraic geometry, causing confusion. 1 then g − αu vanishes at an extreme point -- again, using the facial argument -- and is nonnegative everywhere; archimedeanness entails g ≥ αu, and simplicity then forces g = αu). An extreme version is the following. An unperforated partially ordered abelian group with order unit, (G, u), is extremely simple if for all g ∈ G \ {0}, and all pure traces τ , τ (g) 6= 0. To see that this implies the criterion of Lemma 1, we use the facial remark in the first paragraph of the argument -- if G is extremely simple and τ (g) ≥ 0 for all pure τ , the hypothesis ensures that τ (g) > 0 for all pure τ , and the facial argument entails g is an order unit, so the infimum is strictly positive. Obvious examples are subgroups of the reals with the relative ordering. There are others. Examples of extremely simple dimension groups with modestly interesting trace spaces. 1 Suppose αi (i = 1, 2, . . . , n) are real numbers such that each of {1, αi} is linearly independent over the rationals (that is, none of αi is rational). Let G be the subgroup of Rn spanned (as a group) by the standard basis vectors {ei} together with the element E = P αjej. Then G is a free abelian group of rank n + 1; equipped with the relative order inherited from Rn (i.e., G+ = (Rn)+ ∩ G, where Rn has the usual coordinatewise ordering), G is unperforated and it is easy to check that all the pure traces on G are given by the n coordinate evaluations, and the linear independence hypotheses ensure G is extremely simple. In this case, the pure trace space consists of n points. If we additionally insist that the set {1, α1, α2, . . . , αn} be linearly independent over the ra- tionals, then as is well known, G is a dense subgroup of Rn, and being simple, is thus a dimension group. 2 Suppose that K is an algebraic extension field of Q, and in addition K is formally real, i.e., if not all ki are zero, then P k2 i 6= 0 (or what amounts to the same thing, K admits a real embedding). Impose on K the sums of squares ordering (that is, K + consists of the set of sums of squares). As K is formally real, this is a proper cone, and from algebraicity, it follows and is easy to check that 1 is an order unit for K; moreover, inverses of positive elements are positive, and multiplication preserves the positive cone. It is known that K is a dimension group [H; old paper], i.e., satisfies interpolation (this is true for any formally real field, not just algebraic extensions of the rationals). For a partially ordered ring with 1 as order unit, the pure traces are multiplicative, in particular are ring homomorphisms. Since K is a field, none of these have nontrivial kernel, verifying extreme simplicity. In particular, K is a simple archimedean dimension group, which is also an ordered ring. The pure trace space can be interesting. Let K be QR, the subfield of the reals consisting of all elements algebraic over Q (QR is of index two in the algebraic closure of Q). Then the pure traces can be identified with the Galois automorphisms of QR, and in particular, the pure trace space is the infinite, nonatomic, and totally disconnected separable set (sometimes called the (or a) Cantor set, although von Neumann compactum would be at least as appropriate since he proved its uniqueness). • Extreme simplicity is drastic, as evidenced by the trivial Lemma 2. LEMMA 2 Suppose that (G, u) is a dimension group with order unit, and there exists a connnected subset U of ∂eS(G, u). Then there exists g in G and s in U such that g is not constant on U, and s(g) = 0. Proof. Given distinct v and w in U , there exists h in G such that h(v) 6= h(w). Since hU is continuous and U is connected, the range of hU is a nontrivial interval. Let q = a/b be a rational number (with a an integer and b a positive integer) in the interval. There exists s in U such that h(s) = q. Set g = bh − au; since h is nonconstant on U and u is the constant function 1, g is not constant on U . • 2 In particular, if (R, 1) is an unperforated partially ordered ring with 1 as order unit, and R is extremely simple, then the pure trace space (known to be compact, since the pure traces are exactly the multiplicative ones) must be totally disconnected, as in Example 2 above. We simply note that if ∂eS(R, 1) contained a connected subset, then by Lemma 2, there would be an element r together with a pure trace x such that x(r) = 0, and moreover archimedeanness entails r 6= 0, so r2 is not zero. Then r2 is nonnegative at every pure trace (since all such are multiplicative), hence by archimedeanness, r2 would belong to R+, and of course, x(r2) = 0, contradicting Lemma 1. This seems about the end of the road for extremely simple (dimension) groups. EXAMPLE 3 Simple archimedean dimension groups with the unit interval as pure trace space. Proof. Recall from Example 2 above, the maximal algebraic (over the rationals) subfield of the reals, QR. Let {αi}i∈N be a countably infinite set of real numbers such that the enlarged set {1} ∪ {αi}i∈N is linearly independent over QR -- e.g., if t is a transcendental number, we could take αi = ti, or we could simply insist that {αi} be algebraically independent over Q. Inside the real polynomial algebra R[x], define the elements, e0 = 1, ei = xi − αi (i = 1, 2, 3, . . . ), and define G ⊂ R[x] to be the rational span of {ej}j≥0. Equip R[x] with the strict ordering as a subgroup of C([0, 1], R), so that R[x] is a simple dimension group (since the image is dense), and put the relative ordering on G. Automatically, G is simple. Next, G is a rational vector space, so its closure with respect to the supremum norm -- equivalently the norm on R[x] with the strict ordering -- is a real vector space, and thus each xi is contained in the closure of G. Hence G is dense in C([0, 1], R) (since R[x] is), and in particular, its pure trace space is the same as that of R[x], namely [0, 1], and moreover, G is a dimension group. We show that G is archimedean by verifying the condition of Lemma 1. Pick (to begin with) an arbitrary nonzero element of G, g = q0 + Pn 1 (xi − αi)qi, where qi are rationals; since g is not zero, not all the rational coefficients are zero. If α is a real number that is algebraic over the rationals, then g(α) = 0 entails q0 + Pn 1 qi(αi − αi) = 0, yielding the equation, q0 + Pi=1 qiαi = P qiαi; the left side is algebraic (over the rationals), hence belongs to QR, while the right side is a rational- (hence a QR-) linear combination of {αi}. By our assumption, both sides must be zero, which forces q1 = q2 = · · · = 0, and this in turn forces q0 = 0, a contradiction. The conclusion is that if g is an element of G and not a constant, then it cannot have any zeros at algebraic points. Now suppose that nonzero g in G has minimum 0 as a function on the unit interval. Then inf g(α) = 0 for some α in the unit interval. By the preceding paragraph, α is not algebraic, so in particular, cannot be zero or one; thus it must be an interior point, and since g is nonnegative, α is the location of a minimum of g (as a continuous function on [0, 1]). Since g is a polynomial and α is an interior point of the interval, we must have g′(α) = 0. But this entails P qiiαi−1 = 0, which in turn entails either that α is algebraic, or that q1 = q2 = · · · = qn = 0, hence q0 = 0; either way, we reach a contradiction. Thus g in G \ {0} with g[0, 1] ≥ 0 forces g to have no zeros in [0, 1]. This verifies the criterion of Lemma 1. Thus G is a simple archimedean dimension group whose pure trace space is the unit interval. A sensitivity phenomenon occurs if we change the endpoints of the interval from [0, 1] to [a, b]. If both a and b are algebraic, then the same argument applies (since the putative functions cannot vanish at either endpoint, hence any zeros must be in the interior, whence the derivative argument works). On the other hand, if either a or b is of the form α1/k (where this makes sense, e.g., if k is even, then αk must be positive) for some k, then one of ±ek = ±(xk − αk) (an element of G) will vanish at one point of the interval while being strictly positive on the rest of it -- in particular, the so-constructed simple dimension group G will not be archimedean. • k 3 In this example, we can consider the ordered tensor product, H := G⊗ZR; as vector spaces, the obvious map H → R[x] is an isomorphism inducing an affine homeomorphism on their respective trace spaces; it is also positive. Since G and R are simple dimension groups, so is H, and it follows that the map is an order-isomorphism (of ordered vector spaces). However, we could just as well have imposed a different ordering on R[x] which yields exactly the same pure traces -- for example, the positive cone generated additively and multiplicatively by {R+, x, 1 − x} (Renault's example). This is a non-simple dimension group with pure trace space [0, 1]. The inclusion G ⊂ R[x] yields the same ordering on G (that is, with this latter ordering on R[x], the relative ordering on G is the same as the original), so that in this case the natural map H → R[x] is not an order isomorphism (the left is simple, the right isn't), although it is a vector space isomorphism which is also positive. Another candidate for the ordering on R[x] is pointwise on [0, 1], that is, make R[x] itself archimedean; the same remarks apply, except I cannot see whether it is a dimension group. The idea underlying Example 3 yields a complete answer (at least in the metrizable case -- without metrizability, there probably is an argument, but it looks like a lot of effort) to Goodearl's question. EXAMPLE ∞ For every metrizable Choquet simplex K, there exists a countable simple archimedean dimension group whose trace space is (affinely homeomorphic to) K. Proof. Let K be a metrizable Choquet simplex, and let u0 = 1, u1, . . . be a countable set of elements of A := Aff K (where 1 simply means the constant function) such that {ui} is linearly independent over the reals, and its real span is dense in A. We will construct out of this a rational vector subspace of A (parallel to the development of the polynomial example), G, satisfying the criterion of Lemma 1. Let s−(g) = inf τ ∈∂eK g(τ ) and s+(g) = supτ ∈∂eK g(τ ). The facial argument yields that the values are actually attained on ∂eK, i.e., there exist τ+ and τ− in ∂eK such that both equations τ±(g) = s±(g) hold. For a subgroup J of A, denote by s(J ) the subgroup of the reals generated by set of values of s−(g) as g varies over J ; since J is closed under multiplication by −1, this is the same as the group generated by the set of values of s+(g), which explains the lack of sign in s(J ). Obviously s(J ) is a subgroup of the reals; if J is a rational vector space, so is s(J ), and if J is countable, so is s(J ). It is not clear that the set of values of the s−(g) (running over J ) is a group, but fortunately all that matters for this example is cardinality of the group it generates. We proceed to define vi inductively, so that if Hi is the rational vector space span of {v0, v1, . . . , vi}, then RHi is the real span of {u0, u1, . . . , ui} for all i, and various other properties. Let v0 = u0 = 1; suppose we have v0, . . . , vk−1 with the following properties: (a) there exist nonzero real numbers λi such that for all 1 ≤ i ≤ k − 1, vi = ui − λi (this notation identifies λi with the corresponding constant function; to be pedantic, vi = ui − λi111 where 111 is the constant function with value 1); (b) On defining Hi as above, for 1 ≤ i ≤ k − 1, for all g ∈ Hi \ Hi−1, each of s±(g) is a nonzero rational multiple of λi modulo s(Hi−1 + Qui), and moreover, λiQ ∩ s(Hi−1 + Qui) = {0}. We will show the inductive process can be continued. Consider the countable rational vector subspace of the reals, s(Hk−1 + Quk); we may thus select nonzero real λk such that λkQ ∩ s(Hk−1 + Quk) = {0}, and set vk = uk − λk. Obviously the real span of {ui}k i=0. We observe that since the set {uj} is linearly independent over the reals, so is the set {v0, v1, . . . , vk}, and therefore it is linearly independent over the rationals. Then define Hk = Hk−1 + vkQ, and select g in Hk \ Hk−1. By linear independence, we have g = (uk − λk)qk + Pi≤k−1 qivi with qk 6= 0. Set g0 = g + qkλk; this is in Hk−1 + ukQ. i=0 coincides with the real span of {vi}k 4 Suppose s−(g) = α, so that g ≥ α (as functions on K), and thus g0 ≥ α + λkqk. There exists pure τ0 such that τ0(g) = s−(g); obviously, τ0(g0) = α + λkqk. Thus s−(g0) = α + λkqk, so the latter number is in s(Hk−1 + ukQ). Hence s−(g) = α = −λkqk + (α + λkqk), so belongs to the coset, −λkqk + s(Hk−1 + Qui), in R/s(Hk−1 + Quk); this is nonzero since qk and λk are not zero and λkQ ∩ s(Hk−1 + Quk) = {0}. The same computation with a couple of inequalities reversed shows that s+(g) ∈ −λkqk + s(Hk−1 + Quk). This completes the inductive process. Now set G to be the rational span of {vi}∞ 0 ; obviously, this is an increasing union of the Hk. Since G is a rational vector space, its closure (within A) is a real vector space, and therefore contains RG, which in turn contains the real span of {ui}, and is thus dense. Hence G is dense in A. Impose the strict ordering on G, so that G is a simple dimension group (we finally use the fact that K is a Choquet simplex). Now we verify that G satisfies the condition of Lemma 1. Select nonzero g in G. Since G is the union of the ascending chain of subgroups {Hi}, there exists k ≥ 0 such that g belongs to Hk \ Hk−1 (define H−1 = {0}). From the computation above, s−(g) belongs to a nontrivial coset of a rational subgroup of the reals, hence cannot be zero. Thus G is a simple archimedean dimension group, a dense subgroup of Aff K, and it is immediate that • its normalized trace space is K. Mathematics Department, University of Ottawa, Ottawa ON K1N 6N5, Canada; [email protected] 5
1901.08947	1	1901	2019-01-25T16:07:24	Local derivations on associative and Jordan matrix algebras	[ "math.RA", "math.OA" ]	In the present paper we prove that every additive (not necessarily homogenous) local inner derivation on the algebra of matrices over an arbitrary field is an inner derivation, and every local inner derivation on the ring of matrices over a finite ring generated by the identity element or the ring of integers is an inner derivation. We also prove that every additive local inner derivation on the Jordan algebra of symmetric matrices over an arbitrary field is a derivation, and every local inner derivation on the Jordan ring of symmetric matrices over a finite ring generated by the identity element or the ring of integers is a derivation.	math.RA	math	LOCAL DERIVATIONS ON ASSOCIATIVE AND JORDAN MATRIX ALGEBRAS SHAVKAT AYUPOV1,2 AND FARHODJON ARZIKULOV3 Abstract. In the present paper we prove that every additive (not necessarily homogenous) local inner derivation on the algebra of matrices over an arbitrary field is an inner derivation, and every local inner derivation on the ring of matrices over a finite ring generated by the identity element or the ring of integers is an inner derivation. We also prove that every additive local inner derivation on the Jordan algebra of symmetric matrices over an arbitrary field is a derivation, and every local inner derivation on the Jordan ring of symmetric matrices over a finite ring generated by the identity element or the ring of integers is a derivation. 1. Introduction The present paper is devoted to local derivations on associative and Jordan matrix algebras. Recall that a local derivation is defined as follows: given an algebra A, a linear map ∇ : A → A is called a local derivation if for every x ∈ A there exists a derivation D : A → A such that ∇(x) = D(x). In [14], R. Kadison introduces the concept of local derivation and proves that each continuous local derivation from a von Neumann algebra into its dual Banach bemodule is a derivation. B. Jonson [13] extends the above result by proving that every local derivation from a C-algebra into its Banach bimodule is a derivation. In particular, Johnson gives an automatic continuity result by proving that local derivations of a C-algebra A into a Banach A-bimodule X are continuous even if not assumed a priori to be so (cf. [13, Theorem 7.5]). Based on these results, many authors have studied local derivations on operator algebras, for example, see in [2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 15, 16, 17, 18, 19, 20, 22, 23]. In this paper we develop a pure algebraic approach to investigation of deriva- tions and local derivations on associative and Jordan algebras. Since we consider a sufficiently general case we restrict our attention only on inner derivations and local inner derivations. In section 2 we introduce and investigate a notion of additive local derivation on the algebra Mn(F ) of matrices over an arbitrary field F . It is proved that, given an arbitrary field F , every additive local inner derivation on the algebra Mn(F ) is an inner derivation, and every local inner derivation on the ring of matrices over a finite ring generated by the identity element or the ring of integers is an inner derivation, where a finite ring is a ring that has a finite number of elements. 2010 Mathematics Subject Classification. 16W25, 46L57, 47B47, 17C65. Key words and phrases. derivation, inner derivation, local inner derivation, associative alge- bra of matrices, Jordan algebra of matrices. 1 2 SH. AYUPOV AND F. ARZIKULOV Here we define an additive local inner derivation as follows: given an algebra A, an additive (not necessarily homogenous) map ∇ : A → A is called additive local inner derivation if for every x ∈ A there exists an inner derivation D : A → A such that ∇(x) = D(x). In section 3 additive local derivations on the Jordan algebra of symmetric matrices over an arbitrary field are introduced and studied. It is proved that every additive local inner derivation on the Jordan algebra Hn(F ) of n-dimensional symmetric matrices over an arbitrary field F is a derivation, and every local inner derivation on the Jordan ring of symmetric matrices over a finite ring generated by the identity element or the ring of integers is a derivation. For this propose we use a Jordan analogue of the algebraic approach to the investigation of additive local derivations applied to algebras of matrices over an arbitrary field developed in section 2. The method developed in this paper is sufficiently universal and can also be applied to Jordan and Lie algebras. Its corresponding modification has been used when we considered a similar problem for Lie algebras of skew- symmetric matrices over an arbitrary field [2]. It should be noted that the notions of local inner derivation and local spatial derivation in theorems 5.4, 5.5 and 6.1 in [2] can be replaced by the notions of additive local inner derivation and additive local spatial derivation respectively. 2. local derivations on associative algebras of matrices Let A be an algebra. A liner map D : A → A is called a derivation, if D(xy) = D(x)y + xD(y) for any two elements x, y ∈ A. An additive map ∇ : A → A is called additive local derivation, if for any element x ∈ A there exists a derivation D : A → A such that ∇(x) = D(x). Now let A be a non commutative (but associative) algebra. A derivation D on A is called an inner derivation, if there exists an element a ∈ A such that D(x) = ax − xa, x ∈ A. This derivation D we denote by Da, i.e. Da(x) = ax − xa. An additive (not necessarily homogenous) map ∇ : A → A is called additive local inner derivation, if for any element x ∈ A there exists an inner derivation Da such that ∇(x) = Da(x). Let F be a field, and let Mn(F ) be the matrix algebra over F , n > 1, i.e. consisting of matrices   a1,1 a1,2 a2,1 a2,2 ... ... an,1 an,2 · · · a1,n · · · a2,n ... . . . · · · an,n   , ai,j ∈ F , i, j = 1, 2, . . . , n. Let {ei,j}n i,j=1 be the set of matrix units in Mn(F ), i.e. ei,j is the matrix with components ai,j = 1 and ak,l = 0 if (i, j) 6= (k, l), where 1 is the identity element, 0 is the zero element of the field F , and a matrix a ∈ Mn(F ) is written as a = Pn k,l=1 ak,l, where ak,l = ek,kael,l for k, l = 1, 2, . . . , n. k,l=1 ak,lek,l, where ak,l ∈ F for k, l = 1, 2, . . . , n, or as a = Pn LOCAL DERIVATIONS ON MATRIX ALGEBRAS 3 First, let us prove some lemmas which will be used in the proof of Theorem 2.14. Throughout the section, F denotes an arbitrary field, Mn(F ) denotes the algebra of n× n matrices over F , n > 1. Let ∇ : Mn(F ) → Mn(F ) be an additive local inner derivation. Lemma 2.1. For arbitrary λ, µ ∈ F and each pair i, j of distinct indices there exists an element a ∈ Mn(F ) such that ∇(λei,i) = Da(λei,i), ∇(µei,j) = Da(µei,j). Proof. We have Let di,i, di,j, d ∈ Mn(F ) be elements such that ∇(λei,i) + ∇(µei,j) = ∇(λei,i + µei,j). ∇(λei,i) = di,iλei,i − λei,idi,i, ∇(µei,j) = di,jµei,j − µei,jdi,j, ∇(λei,i + µei,j) = d(λei,i + µei,j) − (λei,i + µei,j)d. Note that ei,idi,jej,jµei,j = µei,jei,idi,jej,j = 0. So we can take ei,idi,jej,j = ei,idi,iej,j, i.e. di,j ei,i[di,iλei,i−λei,idi,i+di,jµei,j −µei,jdi,j]ei,i = ei,i[d(λei,i+µei,j)−(λei,i +µei,j)d]ei,i i,j = di,j i,i . Since we have The equalities −µei,jdi,jei,i = −µei,jdei,i, i.e dj,i i,j = dj,i. ej,j[di,iλei,i−λei,idi,i+di,jµei,j −µei,jdi,i]ei,i = ej,j[d(λei,i+µei,j)−(λei,i+µei,j)d]ei,i imply that Hence ej,jdi,iei,i = ej,jdei,i, i.e dj,i i,i = dj,i. i,j = dj,i dj,i i,i. Let e = ei,i + ej,j. Then (1 − e)(Dd(λei,i + µei,j))ei,i = (1 − e)(Ddi,i(λei,i))ei,i + (1 − e)(Ddi,j (µei,j))ei,i and Also (1 − e)dei,i = (1 − e)di,iei,i. (1 − e)(Dd(λei,i + µei,j))ej,j = (1 − e)(Ddi,i(λei,i))ej,j + (1 − e)(Ddi,j (µei,j))ej,j and (1 − e)dei,j = (1 − e)di,jei,j, i.e. (1 − e)dei,i = (1 − e)di,jei,i. Hence (1 − e)di,iei,i = (1 − e)di,jei,i. At the same time, since ei,idi,j(1 − e)ei,j = ei,jei,idi,j(1 − e) = 0, we can take ei,idi,j(1 − e) = ei,idi,i(1 − e). Therefore ∇(λei,i) = ei,idi,iλei,i + ej,jdi,iλei,i + (1 − e)di,iλei,i − λei,idi,iej,j− λei,idi,iei,i − λei,idi,i(1 − e) = ej,jdi,jλei,i − λei,idi,jej,j + ei,idi,jλei,i + (1 − e)di,jλei,i− λei,idi,jei,i − λei,idi,j(1 − e) = 4 SH. AYUPOV AND F. ARZIKULOV since di,jλei,i − λei,idi,j = Ddi,j (λei,i). ei,idi,iλei,i − λei,idi,iei,i = ei,idi,jλei,i − λei,idi,jei,i = 0, di,j i,j = di,j i,i . This completes the proof. (cid:3) Similarly we can prove the following lemma. Lemma 2.2. For arbitrary λ, µ ∈ F and each pair i, j of distinct indices there exists an element a ∈ Mn(F ) such that ∇(λei,i) = Da(λei,i), ∇(µej,i) = Da(µej,i). Lemma 2.3. For arbitrary λ, µ ∈ F and each pair i, j of distinct indices there exists an element a ∈ Mn(F ) such that ∇(λei,j) = Da(λei,j), ∇(µej,i) = Da(µej,i). Proof. We have Let di,j, dj,i, d ∈ Mn(F ) be such elements that ∇(λei,j) + ∇(µej,i) = ∇(λei,j + µej,i) ∇(λei,j) = di,jλei,j − λei,jdi,j, ∇(µej,i) = dj,iµej,i − µej,idj,i, ∇(λei,j + µej,i) = d(λei,j + µej,i) − (λei,j + µej,i)d. Then di,jλei,j − λei,jdi,j + dj,iµej,i − µej,idj,i = d(λei,j + µej,i) − (λei,j + µej,i)d. Note that ei,idi,jej,jλei,j = λei,jei,idi,jej,j = ej,jdj,iei,iµej,i = µej,iej,jdj,iei,i = 0. So we can take From ei,idi,jej,j = ei,idj,iej,j, i.e. di,j ej,jdj,iei,i = ej,jdi,jei,i, i.e. dj,i i,j = di,j j,i, j,i = dj,i i,j. ei,i[di,jλei,j−λei,jdi,j+dj,iµej,i−µej,idj,i]ej,j = ei,i[d(λei,j+µej,i)−(λei,j+µej,i)d]ej,j we have that i.e. ei,idi,jλei,j − λei,jdi,jej,j = ei,idλei,j − λei,jdej,j, Similarly we have di,i i,j − dj,j i,j = di,i − dj,j. ej,jdj,iµej,i − µej,idj,iei,i = ej,jdµej,i − µej,idei,i, i.e. Therefore j,i − di,i dj,j j,i = dj,j − di,i. i,j − dj,j di,i i,j = di,i j,i − dj,j j,i . LOCAL DERIVATIONS ON MATRIX ALGEBRAS 5 Now, let e = ei,i + ej,j. Then (1 − e)di,jej,jei,j = ei,j(1 − e)di,jej,j = 0, ej,jdj,i(1 − e)ej,i = ej,iej,jdj,i(1 − e) = 0, (1 − e)dj,iei,iej,i = ej,i(1 − e)dj,iei,i = 0, ei,idi,j(1 − e)ei,j = ei,jei,idi,j(1 − e) = 0 So we may assume that (1 − e)di,jej,j = (1 − e)dj,iej,j, ej,jdj,i(1 − e) = ej,jdi,j(1 − e), (1 − e)dj,iei,i = (1 − e)di,jei,i, ei,idi,j(1 − e) = ei,idj,i(1 − e). So ∇(λei,j) = ei,idi,jλei,j + ej,jdi,jλei,j + (1 − e)di,jλei,j − λei,jdi,jej,j− λei,jdi,jei,i − λei,jdi,j(1 − e) = ei,idi,jλei,j − λei,jdi,jej,j + ej,jdi,jei,iλei,j − λei,jej,jdi,jei,i+ (1 − e)di,jλei,j − λei,jdi,j(1 − e) = ei,idj,iλei,j − λei,jdj,iej,j + ej,jdj,iei,iλei,j − λei,jej,jdj,iei,i+ (1 − e)dj,iλei,j − λei,jdj,i(1 − e) = dj,iλei,j − λei,jdj,i = Ddj,i(λei,j), since The proof is complete. di,j i,j = di,j j,i, dj,i i,j = dj,i j,i. (cid:3) Lemma 2.4. For arbitrary λ, µ, ν ∈ F and each pair i, j of distinct indices there exists an element a ∈ Mn(F ) such that ∇(λei,i) = Da(λei,i), ∇(µei,j) = Da(µei,j), ∇(νej,i) = Da(νej,i). Proof. By lemmas 2.1, 2.2 and 2.3 there exist a, b, c ∈ Mn(F ) such that ∇(λei,i) = Da(λei,i), ∇(µei,j) = Da(µei,j), ∇(λei,i) = Db(λei,i), ∇(νej,i) = Db(νej,i) ∇(µei,j) = Dc(µei,j), ∇(νej,i) = Dc(νej,i). Da(λei,i) = Db(λei,i), Da(µei,j) = Dc(µei,j) Db(νej,i) = Dc(νej,i). and We have and Further from (Da(λei,i))ej,j = (Db(λei,i))ej,j, ej,j(Da(λei,i)) = ej,j(Db(λei,i)) we obtain that Similarly, from ei,iaej,j = ei,ibej,j, ej,jaei,i = ej,jbei,i. ei,i(Da(µei,j))ej,j = ei,i(Dc(µei,j))ej,j, ej,j(Db(νej,i))ei,i = ej,j(Dc(νej,i))ei,i it follows that ei,iaei,j − ei,jaej,j = ei,icei,j − ei,jcej,j, ej,jbej,i − ej,ibei,i = ej,jcej,i − ej,icei,i 6 i.e. Hence Also SH. AYUPOV AND F. ARZIKULOV ai,i − aj,j = ci,i − cj,j, bj,j − bi,i = cj,j − ci,i. ai,i − aj,j = bj,j − bi,i. gives us ei,ia(1 − e) = ei,ib(1 − e), and by the equality (Da(λei,i))(1 − e) = (Db(λei,i))(1 − e) (1 − e)Db(νej,i) = (1 − e)Dc(νej,i) we have (1 − e)bej,i = (1 − e)cej,i. At the same time (1 − e)aej,jei,j = ei,j(1 − e)aej,j = 0, (1 − e)cej,jei,j = ei,j(1 − e)cej,j = 0. Hence we may assume (1 − e)bej,j = (1 − e)cej,j = (1 − e)aej,j. Therefore, Db(νej,i) = bνej,i − νej,ib = ei,ibνej,i + ej,jbνej,i + (1 − e)bνej,i − νej,ibei,i− νej,ibej,j − νej,ib(1 − e) = ei,iaνej,i + (bj,j − bi,i)νej,i − νej,iaej,j + (1 − e)aνej,i − νej,ia(1 − e) = ei,iaνej,i + (aj,j − ai,i)νej,i − νej,iaej,j + (1 − e)aνej,i − νej,ia(1 − e) = aνej,i − νej,ia = Da(νej,i). The proof is complete. (cid:3) Lemma 2.5. For arbitrary λ, µ, ν ρ ∈ F and each pair i, j of distinct indices there exists an element a ∈ Mn(F ) such that ∇(λei,i) = Da(λei,i), ∇(µei,j) = Da(µei,j), ∇(νej,i) = Da(νej,i), ∇(ρej,j) = Da(ρej,j). Proof. By lemma 2.4 there exist a, b ∈ Mn(F ) such that ∇(λei,i) = Da(λei,i), ∇(µei,j) = Da(µei,j), ∇(νej,i) = Da(νej,i) ∇(µei,j) = Db(µei,j), ∇(νej,i) = Db(νej,i), ∇(ρej,j) = Db(ρej,j). We have From Da(µei,j) = Db(µei,j), Da(νej,i) = Db(νej,i). ej,j(Da(µei,j)) = ej,j(Db(µei,j)), ei,i(Da(νej,i)) = ei,i(Db(νej,i)) it follows that respectively. Also, by the equalities ej,jaei,i = ej,jbei,i, ei,iaej,j = ei,ibej,j (Da(µei,j))(1 − e) = (Db(µei,j))(1 − e), (1 − e)(Da(νej,i)) = (1 − e)(Db(νej,i)) we have ej,ja(1 − e) = ej,ja(1 − e) and (1 − e)aej,j = (1 − e)bej,j respectively, where e = ei,i + ej,j. Therefore Db(ρej,j) = bρej,j − ρej,jb = LOCAL DERIVATIONS ON MATRIX ALGEBRAS 7 ei,ibρej,j + ej,jbρej,j + (1 − e)bρej,j − ρej,jbei,i − ρej,jbej,j − ρej,jb(1 − e) = ei,ibρej,j − ρej,jbei,i + (1 − e)bρej,j − ρej,jb(1 − e) = ei,iaρej,j − ρej,jaei,i + (1 − e)aρej,j − ρej,ja(1 − e) = ei,iaρej,j − ρej,jaei,i + ej,jaρej,j − ρej,jaej,j + (1 − e)aρej,j − ρej,ja(1 − e) = aρej,j − ρej,ja = Da(ρej,j). This completes the proof. (cid:3) Lemma 2.6. For arbitrary λ, µ ∈ F and each index i there exists an element a ∈ Mn(F ) such that ∇(λei,i) = Da(λei,i), ∇(µei,i) = Da(µei,i). Proof. By Lemma 2.4 for arbitrary ν, ρ ∈ F there exist a, b ∈ Mn(F ) such that ∇(λei,i) = Da(λei,i), ∇(νei,j) = Da(νei,j), ∇(ρej,i) = Da(ρej,i) ∇(µei,i) = Db(µei,i), ∇(νei,j) = Db(νei,j), ∇(ρej,i) = Db(ρej,i). We have Since Da(νei,j) = Db(νei,j), Da(ρej,i) = Db(ρej,i). ej,j(Da(νei,j)) = ej,j(Db(νei,j)), ei,i(Da(ρej,i)) = ei,i(Db(ρej,i)) we have that respectively. Let e = ei,i + ej,j. Then by the equalities ej,jaei,i = ej,jbei,i, ei,iaej,j = ei,ibej,j (1 − e)(Da(νei,j)) = (1 − e)(Db(νei,j)), (Da(ρej,i))(1 − e) = (Db(ρej,i))(1 − e) we have (1 − e)aei,i = (1 − e)bei,i and ei,ia(1 − e) = ei,ib(1 − e) respectively. Therefore Db(µei,i) = bµei,i − µei,ib = ei,ibµei,i + ej,jbµei,i + (1 − e)bµei,i − µei,ibei,i − µei,ibej,j − µei,ib(1 − e) = ej,jbµei,i − µei,ibej,j + (1 − e)bµei,i − µei,ib(1 − e) = ej,jaµei,i − µei,iaej,j + (1 − e)aµei,i − µei,ia(1 − e) = ej,jaµei,i − µei,iaej,j + ei,iaµei,i − µei,iaei,i + (1 − e)aµei,i − µei,ia(1 − e) = aµei,i − µei,ia = Da(µei,i). This completes the proof. (cid:3) Similarly we can prove the following lemma using the above lemmas 2.4 and 2.5. Lemma 2.7. For arbitrary λ, µ ∈ F and each pair i, j of distinct indices there exist elements a ∈ Mn(F ) such that ∇(λei,j) = Da(λei,j), ∇(µei,j) = Da(µei,j). Lemma 2.8. For arbitrary λ, µ ∈ F and each pair i, j of distinct indices one has ∇(λei,i) = λ∇(ei,i), ∇(µei,j) = µ∇(ei,j). 8 SH. AYUPOV AND F. ARZIKULOV Proof. By lemma 2.6 there exists an element a ∈ Mn(F ) such that ∇(λei,i) = Da(λei,i), ∇(ei,i) = Da(ei,i). Hence ∇(λei,i) = Da(λei,i) = λDa(ei,i) = λ∇(ei,i). The second equality is proved in a similarly way. (cid:3) Theorem 2.9. Let F be an arbitrary field, and let M2(F ) be the algebra of 2 × 2 matrices over F . Then any additive local inner derivation on the matrix algebra M2(F ) is an inner derivation. Proof. Let ∇ : M2(F ) → M2(F ) be an additive local inner derivation. By lemma 2.5 there exists a ∈ M2(F ) such that for any indices i, j from {1, 2}. ∇(ei,j) = Da(ei,j) Let x be an arbitrary element in M2(F ). Then x = P2 k,l=1 xk,lek,l and by lemma 2.8 ∇(x) = 2 X k,l=1 ∇(xk,lek,l) = 2 X k,l=1 xk,l∇(ek,l) = 2 X k,l=1 xk,lDa(ek,l) = Da(x). Hence ∇ is an inner derivation. The proof is complete. (cid:3) Lemma 2.10. Let ∇ : Mn(F ) → Mn(F ) be an additive local inner derivation. Then for any indices i, j, k, l, at least three of which are pairwise distinct, there exists a ∈ Mn(F ) such that ∇(ei,j) = Da(ei,j), ∇(ek,l) = Da(ek,l). Proof. Let i, j, k be pairwise distinct indices. We have ∇(λei,j) + ∇(µej,k) = ∇(λei,j + µej,k) Let di,j, dj,k, d ∈ Mn(F ) be such elements that ∇(λei,j) = Ddi,j (λei,j), ∇(µej,k) = Ddj,k (µej,k), ∇(λei,j + µej,k) = Dd(λei,j + µej,k). Ddi,j (λei,j) + Ddj,k (µej,k) = Dd(λei,j + µej,k). ej,jdj,kei,iµej,k = µej,kej,jdj,kei,i = 0. ej,jdj,kei,i = ej,jdi,jei,i, i.e. dj,i j,k = dj,i i,j. Then Note that So we can take From ei,i[Ddi,j (λei,j) + Ddj,k(µej,k)]ej,j = ei,i[Dd(λei,j + µej,k)]ej,j LOCAL DERIVATIONS ON MATRIX ALGEBRAS 9 it follows that i.e. ei,idi,jλei,j − λei,jdi,jej,j = ei,idλei,j − λei,jdej,j, di,i i,j − dj,j i,j = di,i − dj,j. Similarly we have ej,jdj,kµej,k − µej,kdj,kei,i = ej,jdµej,k − µej,kdei,i, i.e. dj,j j,k − di,i j,k = dj,j − di,i. Therefore di,i i,j − dj,j Now, let e = ei,i + ej,j. Then i,j = di,i j,k − dj,j j,k. ej,jdj,k(1 − e)ej,k = ej,kej,jdj,k(1 − e) = 0, (1 − e)dj,kei,iej,k = ej,k(1 − e)dj,kei,i = 0. So we may assume that ej,jdj,k(1 − e) = ej,jdi,j(1 − e), (1 − e)dj,kei,i = (1 − e)di,jei,i. So since ∇(λei,j) = ei,idi,jλei,j + ej,jdi,jλei,j + (1 − e)di,jλei,j − λei,jdi,jej,j− λei,jdi,jei,i − λei,jdi,j(1 − e) = ei,idi,jλei,j − λei,jdi,jej,j + ej,jdi,jei,iλei,j − λei,jej,jdi,jei,i+ (1 − e)di,jλei,j − λei,jdi,j(1 − e) = ei,idj,kλei,j − λei,jdj,kej,j + ej,jdj,kei,iλei,j − λei,jej,jdj,kei,i+ (1 − e)dj,kλei,j − λei,jdj,k(1 − e) = dj,kλei,j − λei,jdj,k = Ddj,k (λei,j), i,j = dj,i dj,i j,k. Let i, j, k, l be pairwise distinct indices and let di,j, dk,l be elements in Mn(F ) such that ∇(ei,j) = Ddi,j (ei,j), ∇(ek,l) = Ddk,l(ek,l). Put e = ei,i + ej,j, f = ek,k + el,l. Then we have (1 − f )dk,l(1 − f )ek,l = ek,l(1 − f )dk,l(1 − f ) = 0. So we may take (1 − f )dk,l(1 − f ) = (1 − f )di,j(1 − f ). Hence (ei,i + ej,j)di,j(ei,i + ej,j) = (ei,i + ej,j)dk,l(ei,i + ej,j) and Also we have ei,idi,jei,i = ei,idk,lei,i, ei,idi,jej,j = ei,idk,lej,j, ej,jdi,jei,i = ej,jdk,lei,i, ej,jdi,jej,j = ej,jdk,lej,j. (1 − e − el,l)dk,lei,iek,l = ek,l(1 − e − el,l)dk,lei,i = 0, ej,jdk,l(1 − e − ek,k)ek,l = ek,lej,jdk,l(1 − e − ek,k) = 0. 10 SH. AYUPOV AND F. ARZIKULOV So we may take (1 − e − el,l)dk,lei,i = (1 − e − el,l)di,jei,i, ej,jdk,l(1 − e − ek,k) = ej,jdi,j(1 − e − ek,k). Now, let d be an element in Mn(F ) such that ∇(ei,j + ek,l) = Dd(ei,j + ek,l). Then Dd(ei,j + ek,l) = Ddi,j (ei,j) + Ddk,l(ek,l) and by the equalities ei,i(Dd(ei,j + ek,l))ek,k = ei,i(Ddi,j (ei,j) + Ddk,l(ek,l))ek,k, ej,j(Dd(ei,j + ek,l))el,l = ej,j(Ddi,j (ei,j) + Ddk,l(ek,l))el,l we have ej,jdek,k = ej,jdi,jek,k, ej,jdek,k = ej,jdk,lek,k. Hence ej,jdi,jek,k = ej,jdk,lek,k. Similarly by the equalities ek,k(Dd(ei,j + ek,l))ei,i = ek,k(Ddi,j (ei,j) + Ddk,l(ek,l))ei,i, el,l(Dd(ei,j + ek,l))ej,j = el,l(Ddi,j (ei,j) + Ddk,l(ek,l))ej,j we have Therefore el,ldi,jei,i = el,ldk,lei,i. ∇(ei,j) = ei,idi,jei,j + ej,jdi,jei,j + (1 − e)di,jei,j − ei,jdi,jej,j− ei,jdi,jei,i − ei,jdi,j(1 − e) = ei,idi,jei,j − ei,jdi,jej,j + ej,jdi,jei,iei,j − ei,jej,jdi,jei,i+ (1 − e)di,jei,j − ei,jdi,j(1 − e) = ei,idk,lei,j − ei,jdk,lej,j + ej,jdk,lei,iei,j − ei,jej,jdk,lei,i+ (1 − e)dk,lei,j − ei,jdk,l(1 − e) = dk,lei,j − ei,jdk,l = Ddk,l(ei,j). Similarly we can prove that for any pairwise distinct indices i, j, k there exists a in Mn(F ) such that ∇(ei,j) = Da(ei,j), ∇(ek,k) = Da(ek,k). This completes the proof. (cid:3) Lemma 2.11. There exists a ∈ Mn(F ) such that ∇(ei,i+1) = Da(ei,i+1), i = 1, 2, . . . n − 1. LOCAL DERIVATIONS ON MATRIX ALGEBRAS 11 Proof. By lemma 2.10 there exists a ∈ Mn(F ) such that ∇(e1,2) = Da(e1,2), ∇(e2,3) = Da(e2,3). Suppose that for k there exists a ∈ Mn(F ) such that ∇(ei,i+1) = Da(ei,i+1), i = 1, 2, . . . k − 1. We prove that for k + 1 there exists b ∈ Mn(F ) such that ∇(ei,i+1) = Db(ei,i+1), i = 1, 2, . . . k. By lemma 2.10 there exists c(i, i+1) ∈ Mn(F ) such that ∇(ei,i+1) = Dc(i,i+1)(ei,i+1), ∇(ek,k+1) = Dc(i,i+1)(ek,k+1), where i = 1, 2, . . . k − 1. Since ek+1,k+1aek+1,k+1ei,j = ei,jek+1,k+1aek+1,k+1 = 0, i, j = 1, 2, . . . k we may put and bi,j = ai,j, if i ≤ k or j ≤ k, bk+1,k+1 = c(k, k + 1)k+1,k+1 − c(k, k + 1)k,k + bk,k, bi,j = c(k, k + 1)i,j, (i, j) 6= (k + 1, k + 1), i ≥ k + 1, j ≥ k + 1. (5.1) In this case we have ek,k∇(ek,k+1)ek+1,k+1 = ek,kDc(k,k+1)(ek,k+1)ek+1,k+1 = ek,kc(k, k + 1)ek,k+1ek+1,k+1 − ek,kek,k+1c(k, k + 1)ek+1,k+1 = ek,kbek,k+1ek+1,k+1 − ek,kek,k+1bek+1,k+1 = ek,kDb(ek,k+1)ek+1,k+1 by equalities (5.1). If i 6= k and j 6= k + 1 then ei,i∇(ek,k+1)ej,j = 0 = ei,iDb(ek,k+1)ej,j (5.2) If j > k + 1 then ek,k∇(ek,k+1)ej,j = ek,kDc(k,k+1)(ek,k+1)ej,j = ek,kc(k, k + 1)ek,k+1ej,j − ek,kek,k+1c(k, k + 1)ej,j = ek,kbek,k+1ej,j − ek,kek,k+1bej,j = ek,kDb(ek,k+1)ej,j by (5.1). From c(j, j + 1)ej,j+1 − ej,j+1c(j, j + 1) = aej,j+1 − ej,j+1a = bej,j+1 − ej,j+1b it follows that where j = 1, 2, . . . k. Therefore, if j < k + 1 then ek+1,k+1c(j, j + 1)ej,j = ek+1,k+1bej,j, (5.3) ek,k∇(ek,k+1)ej,j = ek,kDc(j,j+1)(ek,k+1)ej,j = ek,kc(j, j + 1)ek,k+1ej,j − ek,kek,k+1c(j, j + 1)ej,j = ek,kbek,k+1ej,j − ek,kek,k+1bej,j = ek,kDb(ek,k+1)ej,j by (5.3). Similarly ej,j∇(ek,k+1)ek+1,k+1 = ej,jDb(ek,k+1)ek+1,k+1, j = 1, 2, . . . n. So ∇(ek,k+1) = n X i,j=1 ei,i∇(ek,k+1)ej,j = n X i,j=1 ei,iDb(ek,k+1)ej,j = Db(ek,k+1). 12 SH. AYUPOV AND F. ARZIKULOV Hence by the induction we obtain that there exists a ∈ Mn(F ) such that ∇(ei,i+1) = Da(ei,i+1), i = 1, 2, . . . n − 1. (cid:3) Lemma 2.12. For any indices i, j there exists a ∈ Mn(F ) such that ∇( n−1 X k=1 ek,k+1) = Da( n−1 X k=1 ek,k+1), ∇(ei,j) = Da(ei,j). Proof. By lemma 2.11 there exists a ∈ Mn(F ) such that ∇(ei,i+1) = Da(ei,i+1), i = 1, 2, . . . n − 1. Fix i and j from {1, 2, . . . n} such that i ≤ j. We have ∇(Pj−1 k=i ek,k+1) + ∇(ej,i). There exist b, d ∈ Mn(F ) such that ∇(Pj−1 k=i ek,k+1 + ej,i) = k=i ek,k+1 + ∇(Pj−1 ej,i) = Dd(Pj−1 k=i ek,k+1 + ej,i), ∇(ej,i) = Dbej,i. Therefore Dd( j−1 X k=i ek,k+1 + ej,i) = Da( j−1 X k=i ek,k+1) + Db(ej,i). Hence di,i − di+1,i+1 = ai,i − ai+1,i+1, k = i, i + 1, i + 2, . . . , j, dj,j − di,i = bj,j − bi,i, and Now, if k 6= i and k 6= j then aj,j − ai,i = bj,j − bi,i. ek,kbek,kej,i = ej,iek,kbek,k = 0. So we may assume that ek,kbek,k = ek,kaek,k for every k such that k 6= i and k 6= j. By lemma 2.10 there exists c(i, i + 1) ∈ Mn(F ) such that ∇(ei,i+1) = Dc(i,i+1)(ei,i+1), ∇(ej,i) = Dc(i,i+1)(ej,i). We have Dc(i,i+1)(ei,i+1) = Da(ei,i+1), Dc(i,i+1)(ej,i) = Db(ej,i) and ej,jc(i, i + 1)ei,i+1 = ej,jaei,i+1, ej,jbei,iej,i = ej,iej,jbei,i = 0. So we may assume that ej,jbei,i = ej,jc(i, i + 1)ei,i. Hence ej,jbei,i = ej,jaei,i. Also, from Dc(i−1,i)(ei−1,i) = Da(ei−1,i), Dc(i−1,i)(ej,i) = Db(ej,i) it follows that ei−1,ic(i − 1, i)ej,j = ei−1,iaej,j, ei,ic(i − 1, i)ej,i = ei,ibej,i, and The remaining case is {k, l} 6= {i, j} for ek,l. 1) Suppose k = i, l 6= j, l 6= i. Take ei,ibej,j = ei,iaej,j. Dc(i−1,i)(ei−1,i) = Da(ei−1,i), Dc(i−1,i)(ej,i) = Db(ej,i) and we have ei−1,ic(i − 1, i)el,l = ei−1,iael,l, ej,ic(i − 1, i)el,l = ej,ibel,l, LOCAL DERIVATIONS ON MATRIX ALGEBRAS 13 and so ei,ibel,l = ei,iael,l. Now, we take Dc(i,i+1)(ei,i+1) = Da(ei,i+1), Dc(i,i+1)(ej,i) = Db(ej,i). Then el,lc(i, i + 1)ei,i+1 = el,laei,i+1, el,lbei,iej,i = ej,iel,lbei,i = 0. So we may assume el,lbei,i = el,laei,i. 2) Suppose k 6= i, l = j, k 6= j. Take Dc(j−1,j)(ej−1,j) = Da(ej−1,j), Dc(j−1,j)(ej,i) = Db(ej,i) and we have ej−1,jc(j − 1, j)ek,k = ej−1,jaek,k, ej,jbek,kej,i = ej,iej,jbek,k = 0. So we may assume that ej,jbek,k = ej,jaek,k. Take and we have Dc(j,j+1)(ej,j+1) = Da(ej,j+1), Dc(j,j+1)(ej,i) = Db(ej,i) ek,kc(j, j + 1)ej,j+1 = ek,kaej,j+1, ek,kc(j, j + 1)ej,i = ek,kbej,i, and so ek,kbej,j = ek,kaej,j. 3) Now, suppose k 6= i, j, l 6= i, j. Then ek,kbel,lej,i = ej,iek,kbel,l = 0. So we may assume that ek,kbel,l = ek,kael,l. Thus, for all {k, l} 6= {i, j} we have ek,kbel,l = ek,kael,l and, if {k, l} = {i, j} then bi,i − bj,j = ai,i − aj,j and bi,j = bi,j, bj,i = bj,i. Hence ∇(ej,i) = Db(ej,i) = Da(ej,i). Now, by the definition of additive local inner derivation we have that ∇(ei,j) = Dc(ei,j), ∇(ej,i) = Dc(ej,i) for some c in Mn(F ). Then Dc(ej,i) = Da(ej,i) and cj,j − ci,i = aj,j − ai,i, ei,icej,j = ei,iaej,j. Also we have ej,jcei,iej,i = ej,iej,jcei,i = 0. So we may assume that ej,jcei,i = ej,jaei,i. Now similar to the equality ∇(ej,i) = Da(ej,i) we prove that ∇(ei,j) = Dc(ei,j) = (cid:3) Da(ei,j). This completes the proof. Lemma 2.13. There exists a ∈ Mn(F ) such that ∇(ei,j) = Da(ei,j) for any indices i, j. Proof. By the previous lemmas we can repeat the proof of theorem 4 in [1] and get the statement of this lemma. The proof is complete. (cid:3) Theorem 2.14. Let F be an arbitrary field, and let Mn(F ) be the algebra of n × n matrices over F , n > 1. Then any additive local inner derivation on the algebra Mn(F ) is an inner derivation. Proof. Let ∇ : Mn(F ) → Mn(F ) be an additive local inner derivation. Then by lemma 2.13 there exists a ∈ Mn(F ) such that for any indices i, j ∇(ei,j) = Da(ei,j). 14 SH. AYUPOV AND F. ARZIKULOV Let x be an arbitrary element in Mn(F ). Then x = Pn 2.8 we have that k,l=1 xk,lek,l and by lemma ∇(x) = n X k,l=1 ∇(xk,lek,l) = n X k,l=1 xk,l∇(ek,l) = n X k,l=1 xk,lDa(ek,l) = Da(x). Thus ∇ is an inner derivation. The proof is complete. (cid:3) Let R be a ring. An additive map D : R → R is called a derivation, if D(xy) = D(x)y + xD(y) for any two elements x, y ∈ R. An additive map ∇ : R → R is called local derivation, if for any element x ∈ R there exists a derivation D : R → R such that ∇(x) = D(x). Now let R be a non commutative (but associative) ring. A derivation D on R is called an inner derivation, if there exists an element a ∈ R such that D(x) = ax − xa, x ∈ R. This derivation D we denote by Da, i.e. Da(x) = ax − xa. An additive map ∇ : R → R is called local inner derivation, if for any element x ∈ R there exists an inner derivation Da such that ∇(x) = Da(x). A finite ring is a ring that has a finite number of elements. A finite ring generated by the identity element is a finite ring, every element of which is a sum of some quantity of the identity element of this ring. By the proofs of the previous lemmas we have the following lemma. Lemma 2.15. Let ℜ be a finite ring generated by the identity element or the ring of integers, Mn(ℜ) be the ring of n × n matrices over ℜ, n > 1, and let ∇ : Mn(ℜ) → Mn(ℜ) be a local inner derivation. Then there exists a ∈ Mn(λℜ) such that for any indices i, j. ∇(ei,j) = Da(ei,j) By lemma 2.15 and by the definition of a local inner derivation we have the following theorem. Theorem 2.16. Let ℜ be a finite ring generated by the identity element or the ring of integers, Mn(ℜ) be the ring of n × n matrices over ℜ, n > 1. Then for every local inner derivation ∇ on the ring Mn(ℜ) there exists a matrix a ∈ Mn(ℜ) such that i.e. ∇ is a derivation. ∇(x) = Da(x), x ∈ Mn(ℜ), Let B be a subalgebra of an algebra A. A derivation D on B is said to be spatial, if D is implemented by an element in A, i.e. D(x) = ax − xa, x ∈ B, LOCAL DERIVATIONS ON MATRIX ALGEBRAS 15 for some a ∈ A. An additive local derivation ∇ on B is called additive local spatial derivation with respect to derivations implemented by an element in A, if for every element x ∈ B there exists an element a ∈ A such that ∇(x) = ax − xa. It should be noted that by the proofs of theorems 5.4, 5.5 and 6.1 in [2] the notions of local inner derivation and local spatial derivation in these theorems can be replaced by the notions of additive local inner derivation and additive local spatial derivation respectively. 3. Local derivations on Jordan algebras of symmetric matrices This section is devoted to derivations and local derivations of Jordan algebras. In this section the notations and terminology follow the paper [21] of H. Upmeier. Given subsets B and C of a Lie ring ℜ with bracket [·, ·], let [B, C] denote the subset of ℜ consisting of all finite sums of elements [b, c], where b ∈ B and c ∈ C. Consider a Jordan ring J and let m = {xM : x ∈ A}, where xM denotes the multiplication operator defined by (xM)y := x · y for all y ∈ A. Let aut(J ) denotes the Lie algebra of all derivations of J . The elements of the ideal int(J ) := [m, m] in aut(J ) are called inner derivations of the Jordan ring J (cf.[21]). Let R be an associative unital ring, and suppose 2 is invertible in R Then the set R with respect to the operations of addition and Jordan multiplication a · b = 1 2 (ab + ba), a, b ∈ R is a Jordan ring. This Jordan ring we will denote by (R, ·). Every inner derivation of (R, ·) is an inner derivation of R, and, conversely, every inner derivation of R is an inner derivation of (R, ·) [1]. Let ∇ be a local inner derivation of the Jordan ring (R, ·). Then for every element x ∈ R there is an inner derivation D of (R, ·) such that ∇(x) = D(x). But D is also an inner derivation of the associative ring R. Hence, ∇ is a local inner derivation of the associative ring R. Conversely, every local inner derivation of the associative ring R is a local inner derivation of the Jordan ring (R, ·). Now, let R be an involutive unital ring, and suppose 2 is invertible in R. Let Rsa be the set of all self-adjoint elements of the ring R. Then, it is known that (Rsa, ·) is a Jordan ring. Also, every inner derivation of the Jordan ring (Rsa, ·) is extended to an inner derivation of the ∗-ring R [1]. Such extension of derivations on a special Jordan algebra is considered in [21]. Concerning local inner derivation, till now it is not possible to obtain such extension without additional conditions. This problem shows the importance of the main result in the present section. Throughout of this section F is an arbitrary field with invertible 2, and Mn(F ) is the associative algebra of n × n matrices over F . In this case the set Hn(F ) = {   a1,1 a1,2 a2,1 a2,2 ... ... an,1 an,2 · · · a1,n · · · a2,n ... . . . · · · an,n   ∈ Mn(F ) : ai,j = aj,i, i, j = 1, 2, . . . , n} 16 SH. AYUPOV AND F. ARZIKULOV is a Jordan algebra with respect to the addition and the Jordan multiplication a · b = 1 2 (ab + ba), a, b ∈ Hn(F ). Let ¯ei,j = ei,j + ej,i and ¯ai,j = {ei,iaej,j} = (ei,ia)ej,j + ei,i(aej,j) for every a ∈ Hn(F ) and distinct i, j in {1, 2, . . . , n}. Lemma 3.1. Let D = Pm by a1, a2, . . . , am, b1, b2, . . . , bm ∈ Hn(F ). Then k=1 Dak,bk be an inner derivation on Hn(F ), generated m X k=1 [ak, bk]i,j = − m X k=1 [ak, bk]j,i, i, j = 1, 2, . . . , n, k=1[ak, bk] is a skew-symmetric matrix. i.e. Pm Proof. Indeed, let i, j be arbitrary indices in {1, 2, . . . , n}. Then for every k ∈ {1, 2, . . . , m} we have [ak, bk]i,j = n X l=1 ai,l k bl,j k − n X l=1 k al,j bi,l k , and ai,l k bl,j k al,j since ak and bk are symmetric matrices. k = al,i k , bi,l k bj,l k = bj,l k al,i k = bl,i k aj,l k = aj,l k bl,i k , l = 1, 2, . . . , n, Hence [ak, bk]i,j = n X l=1 ai,l k bl,j k − n X l=1 bi,l k al,j k = n X l=1 bj,l k al,i k − n X l=1 aj,l k bl,i k = −[ak, bk]j,i, k = 1, 2, . . . , m. This completes the proof. (cid:3) Let A be a Jordan algebra. An additive (not necessarily homogenous) map ∇ : A → A is called additive local inner derivation, if for any element x ∈ A there exists an inner derivation D such that ∇(x) = D(x). Lemma 3.2. Let Hn(F ) be the Jordan algebra of symmetric n × n matrices over F , n > 1. Let ∇ be an additive local inner derivation on Hn(F ). Then for arbitrary λ, µ ∈ F there exists an inner derivation D on Hn(F ) such that ∇(λei,i) = D(λei,i), ∇(µ¯ei,j) = D(µ¯ei,j). Proof. We have ∇(λei,i) + ∇(µ¯ei,j) = ∇(λei,i + µ¯ei,j). Let a1, a2, . . . , am, b1, b2, . . . , bm, c1, c2, . . . , cm, d1, d2, . . . , dm be elements in Hn(F ) such that ∇(λei,i) = m X k=1 Dak ,bk(λei,i), ∇(µ¯ei,j) = m X k=1 Dck,dk(µ¯ei,j). LOCAL DERIVATIONS ON MATRIX ALGEBRAS 17 We have m X k=1 m X k=1 Dak,bk(λei,i) = D 1 4 Pm k=1[ak,bk](λei,i), Dck,dk(µ¯ei,j) = D 1 4 Pm k=1[ck,dk](µ¯ei,j). Let p1, p2, . . . , pm, q1, q2, . . . , qm be elements in Hn(F ) such that ∇(λei,i + µ¯ei,j) = m X k=1 Dpk,qk(λei,i + µ¯ei,j). Then Let di,i = 1 ∇(λei,i + µ¯ei,j) = D 1 4 Pm k=1[ak, bk], di,j = 1 4 Pm 4 Pm k=1[pk,qk](λei,i + µ¯ei,j). k=1[ck, dk], d = 1 k=1[pk, qk]. Then, since 4 Pm ei,i[di,iλei,i−λei,idi,i+di,jµ¯ei,j −µ¯ei,jdi,j]ei,i = ei,i[d(λei,i+µ¯ei,j)−(λei,i +µ¯ei,j)d]ei,i we have di,j i,j − dj,i i,j = di,j − dj,i, i.e di,j i,j = di,j = dj,i i,j = dj,i by lemma 3.1. From the equality ei,i[di,iλei,i−λei,idi,i+di,jµ¯ei,j −µ¯ei,jdi,i]ej,j = ei,i[d(λei,i+µ¯ei,j)−(λei,i+µ¯ei,j)d]ej,j it follows that i,i + µdi,i and by lemma 3.1 we have di,i λdi,j i,j − µdj,j i,j = dj,j i,j = λdi,j + µdi,i − µdj,j i,j = di,i = dj,j = 0. Hence di,j i,i = di,j. Therefore di,j i,i = di,j i,j = dj,i i,i = dj,i i,j. Let e = ei,i + ej,j. Then as in the proof of lemma 2.1 we get (1 − e)dei,i = (1 − e)di,jei,i. Hence ei,id(1 − e) = ei,idi,j(1 − e), since di,j and d are skew-symmetric matrices. Therefore ∇(λei,i) = ei,idi,iλei,i + ej,jdi,iλei,i + (1 − e)di,iλei,i − λei,idi,iej,j− λei,idi,iei,i − λei,idi,i(1 − e) = ej,jdi,jλei,i − λei,idi,jej,j + ei,idi,jλei,i + (1 − e)di,jλei,i− λei,idi,jei,i − λei,idi,j(1 − e) = di,jλei,i − λei,idi,j = Ddi,j (λei,i). since ei,idi,iλei,i − λei,idi,iei,i = ei,idi,jλei,i − λei,idi,jei,i = 0, di,j i,j = di,j i,i . This completes the proof. (cid:3) 18 SH. AYUPOV AND F. ARZIKULOV Lemma 3.3. Let ∇ be an additive local inner derivation on Hn(F ). Then for arbitrary λ, µ ∈ F and each index i there exists an inner derivation D on Hn(F ) such that ∇(λei,i) = D(λei,i), ∇(µei,i) = D(µei,i). Proof. By Lemma 3.2 for arbitrary ν ∈ F there exist a1, a2, . . . , am, b1, b2, . . . , bm in Hn(F ) such that ∇(λei,i) = m X k=1 Dak,bk(λei,i), ∇(ν¯ei,j) = m X k=1 Dak ,bk(ν¯ei,j). Let a = 1 4 Pm k=1[ak, bk]. Then ∇(λei,i) = Da(λei,i), ∇(ν¯ei,j) = Da(ν¯ei,j). Similarly, for arbitrary ν ∈ F there exist b ∈ Mn(F ) such that ∇(µei,i) = Db(µei,i), ∇(ν¯ei,j) = Db(ν¯ei,j). We have From it follows that Da(ν¯ei,j) = Db(ν¯ei,j). ej,j(Da(ν¯ei,j)) = ej,j(Db(ν¯ei,j)) Let e = ei,i + ej,j. Then by the equalities ej,jaei,i = ej,jbei,i, ei,iaej,j = ei,ibej,j. (1 − e)(Da(ν¯ei,j)) = (1 − e)(Db(ν¯ei,j)), (Da(ν¯ej,i))(1 − e) = (Db(ν¯ej,i))(1 − e) we have (1 − e)aei,i = (1 − e)bei,i and ei,ia(1 − e) = ei,ib(1 − e) respectively. Therefore ei,ibµei,i + ej,jbµei,i + (1 − e)bµei,i − µei,ibei,i − µei,ibej,j − µei,ib(1 − e) = Db(µei,i) = bµei,i − µei,ib = ej,jbµei,i − µei,ibej,j + (1 − e)bµei,i − µei,ib(1 − e) = ej,jaµei,i − µei,iaej,j + (1 − e)aµei,i − µei,ia(1 − e) = ej,jaµei,i − µei,iaej,j + ei,iaµei,i − µei,iaei,i + (1 − e)aµei,i − µei,ia(1 − e) = aµei,i − µei,ia = Da(µei,i). This completes the proof. (cid:3) Similarly we can prove the following lemma using lemmas 2.4 and 2.5. Lemma 3.4. Let ∇ be an additive local inner derivation on Hn(F ). Then for arbitrary λ, µ ∈ F and each pair i, j of distinct indices there exists an inner derivation D on Hn(F ) such that ∇(λ¯ei,j) = D(λ¯ei,j), ∇(µ¯ei,j) = D(µ¯ei,j). Now we have the following Lemma 3.5. For arbitrary λ, µ ∈ F and each pair i, j of distinct indices we have that ∇(λei,i) = λ∇(ei,i), ∇(µ¯ei,j) = µ∇(¯ei,j). LOCAL DERIVATIONS ON MATRIX ALGEBRAS 19 Proof. By lemma 3.3 there exists an element a ∈ Mn(F ) such that ∇(λei,i) = Da(λei,i), ∇(ei,i) = Da(ei,i). Hence ∇(λei,i) = Da(λei,i) = λDa(ei,i) = λ∇(ei,i). The proof of the second equality is similar. (cid:3) Lemma 3.6. Let ∇ be an additive local inner derivation on Hn(F ). Then for any indices i, j, k, l, satisfying {i, j} 6= {k, l}, there exists an inner derivation D on Hn(F ) such that ∇(¯ei,j) = D(¯ei,j), ∇(¯ek,l) = D(¯ek,l). Proof. Let i, j, k be pairwise distinct indices. Then there exist inner derivation D1, D2 on Hn(F ) such that ∇(¯ei,j) = D1(¯ei,j), ∇(¯ej,j) = D1(¯ej,j), ∇(¯ej,k) = D2(¯ej,k), ∇(¯ej,j) = D2(¯ej,j). Hence there exist elements a, b ∈ Mn(F ) such that ∇(¯ei,j) = Da(¯ei,j), ∇(ej,j) = Da(ej,j), ∇(¯ej,k) = Db(¯ej,k), ∇(ej,j) = Db(ej,j). Da(ej,j) = Da(ej,j) So and (1 − ej,j)aej,j = (1 − ej,j)bej,j, ej,ja(1 − ej,j) = ej,jb(1 − ej,j). Since ej,jaej,j = ej,jbej,j = 0 we have aej,j = bej,j, ej,ja = ej,jb, ei,iaej,j = ei,ibej,j, ej,jaei,i = ej,jbei,i. Also we have Let d ∈ Mn(F ) be an element such that ∇(¯ei,j) + ∇(¯ej,k) = ∇(¯ei,j + ¯ej,k) ∇(¯ei,j + ¯ej,k) = Dd(¯ei,j + ¯ej,k). Then From and Da(¯ei,j) + Db(¯ej,k) = Dd(¯ei,j + ¯ej,k). ei,i[Da(¯ei,j) + Db(¯ej,k)]ej,j = ei,i[Dd(¯ei,j + ¯ej,k)]ej,j ej,j[Da(¯ei,j) + Db(¯ej,k)]ek,k = ej,j[Dd(¯ei,j + ¯ej,k)]ek,k it follows that ei,ibek,k = ei,idek,k and ei,iaek,k = ei,idek,k respectively. Hence ei,iaek,k = ei,ibek,k and ek,kaei,i = ek,kbei,i. Now, let e = ei,i + ej,j. Then (1 − e − ek,k)bei,i¯ej,k = ¯ej,k(1 − e − ek,k)bei,i = 0, ei,ib(1 − e − ek,k)¯ej,k = ¯ej,kei,ib(1 − e − ek,k) = 0. So we may assume that (1 − e − ek,k)bei,i = (1 − e − ek,k)aei,i, ei,ib(1 − e − ek,k) = ei,ia(1 − e − ek,k). 20 Hence SH. AYUPOV AND F. ARZIKULOV (1 − e)bei,i = (1 − e)aei,i, ei,ib(1 − e) = ei,ia(1 − e) since ei,iaek,k = ei,ibek,k and ek,kaei,i = ek,kbei,i. Therefore ∇(¯ei,j) = Da(¯ei,j) = a¯ei,j − ¯ei,ja = aej,i + (1 − e)aei,j + ei,iaei,j + ej,jaei,j− ei,ja − ej,ia(1 − e) − ej,iaei,i − ej,iaej,j = bej,i + (1 − e)bei,j + ei,ibei,j + ej,jbei,j− ei,jb − ej,ib(1 − e) − ej,ibei,i − ej,ibej,j = Db(¯ei,j). Let i, j, k, l be pairwise distinct indices and let a, b be elements in Mn(F ) such that and let e = ei,i + ej,j, f = ek,k + el,l. Then we have ∇(¯ei,j) = Da(¯ei,j), ∇(¯ek,l) = Db(¯ek,l) (1 − f )b(1 − f )¯ek,l = ¯ek,l(1 − f )b(1 − f ) = 0. So we may take (1 − f )b(1 − f ) = (1 − f )a(1 − f ). In particular, ei,ib(1 − f ) = ei,ia(1 − f ), (1 − f )bei,i = (1 − f )aei,i, ej,jb(1 − f ) = ej,ja(1 − f ), (1 − f )bej,j = (1 − f )aej,j. Then Dd(¯ei,j + ¯ek,l) = Da(¯ei,j) + Db(¯ek,l) and by the equalities ei,i(Dd(¯ei,j + ¯ek,l))ek,k = ei,i(Da(¯ei,j) + Db(¯ek,l))ek,k, ej,j(Dd(¯ei,j + ¯ek,l))el,l = ej,j(Da(¯ei,j) + Db(¯ek,l))el,l we have di,l − dj,k = bi,l − aj,k, dj,k − di,l = bj,k − ai,l, respectively. Hence bi,l − aj,k = ai,l − bj,k. Also, by the equalities ∇(¯ei,j − ¯ek,l) = ∇(¯ei,j) − ∇(¯ek,l), ei,i(Dd(¯ei,j − ¯ek,l))ek,k = ei,i(Da(¯ei,j) − Db(¯ek,l))ek,k, ej,j(Dd(¯ei,j − ¯ek,l))el,l = ej,j(Da(¯ei,j) − Db(¯ek,l))el,l we have di,l + dj,k = bi,l + aj,k, dj,k + di,l = bj,k + ai,l, respectively. Hence bi,l + aj,k = ai,l + bj,k. Therefore bi,l = ai,l, aj,k = bj,k and bl,i = al,i, ak,j = bk,j. Similarly we get Therefore bi,k = ai,k, aj,l = bj,l and bk,i = ak,i, al,j = bl,j. ∇(¯ei,j) = ek,ka¯ei,j + el,la¯ei,j + (1 − f )a¯ei,j − ¯ei,jael,l− ¯ei,jaek,k − ¯ei,ja(1 − f ) = ek,kb¯ei,j + el,lb¯ei,j + (1 − f )b¯ei,j − ¯ei,jbel,l− ¯ei,jbek,k − ¯ei,jb(1 − f ) = b¯ei,j − ¯ei,jb = Db(¯ei,j). LOCAL DERIVATIONS ON MATRIX ALGEBRAS This completes the proof. 21 (cid:3) Lemma 3.7. Let ∇ : Hn(F ) → Hn(F ) be an additive local inner derivation. Then there exists a skew-symmetric matrix a ∈ Mn(F ) such that for any indices i, j we have ∇(¯ei,j) = Da(¯ei,j). Proof. By the previous lemmas we can repeat the proof of lemma 14 in [1] and get the statement of the above lemma. The proof is complete. (cid:3) Theorem 3.8. Let Hn(F ) be the Jordan algebra of n×n symmetric matrices over F , n > 1. Then any additive local inner derivation on Hn(F ) is a derivation. Proof. Let ∇ : Hn(F ) → Hn(F ) be an additive local inner derivation. Then by lemma 3.7 there exists a skew-symmetric matrix a ∈ Mn(F ) such that for any indices i, j we have ∇(¯ei,j) = Da(¯ei,j). Let x be an arbitrary element in Hn(F ). Then x = Pn 3.5 k,l=1 xk,lek,l and by lemma n n ∇(x) = X k,l=1,k≤l ∇(xk,l¯ek,l) = X k,l=1,k≤l xk,l∇(¯ek,l) = n X k,l=1,k≤l xk,lDa(¯ek,l) = Da(x). Hence ∇ is a derivation on Hn(F ) (to be more precise, ∇ is a spatial derivation implemented by a skew-symmetric matrix a ∈ Mn(F )) . The proof is complete. (cid:3) Let J be a Jordan ring. An additive map ∇ : J → J is called local derivation, if for any element x ∈ J there exists a derivation D : J → J such that ∇(x) = D(x). An additive map ∇ : J → J is called local inner derivation, if for any element x ∈ J there exists an inner derivation D on the Jordan ring J such that ∇(x) = D(x). By the proofs of the previous lemmas of the present section we have the fol- lowing lemma. Lemma 3.9. Let ℜ be a finite ring generated by the identity element or the ring of integers, Hn(ℜ) be the Jordan ring of n × n symmetric matrices over ℜ, n > 1, and let ∇ : Hn(ℜ) → Hn(ℜ) be a local inner derivation. Then there exists a skew-symmetric matrix a ∈ Mn(λℜ) such that for any indices i, j we have ∇(¯ei,j) = Da(¯ei,j). By lemma 3.9 and by the definition of a local inner derivation we have the following theorem. Theorem 3.10. Let ℜ be a finite ring generated by the identity element or the ring of integers, Hn(ℜ) be the Jordan ring of n × n symmetric matrices over ℜ, 22 SH. AYUPOV AND F. ARZIKULOV n > 1. Then for every local inner derivation ∇ on the Jordan ring Hn(ℜ) there exists a skew-symmetric matrix a ∈ Mn(ℜ) such that i.e. ∇ is a derivation on the Jordan ring Hn(ℜ). ∇(x) = Da(x), x ∈ Hn(ℜ), References [1] Sh. Ayupov, F. Arzikulov, 2-Local derivations on associative and Jordan matrix rings over commutative rings, Linear Algebra Appl. 522 (2017) 28 -- 50. [2] Sh. Ayupov, F. Arzikulov, Description of 2-local and local derivations on some Lie rings of skew-adjoint matrices, Linear and Multilinear Algebra (2018), available from https://arxiv.org/abs/1803.06281. [3] S. Albeverio, SH. Ayupov, K. Kudaybergenov and B. Nurjanov, Local derivations on al- gebras of measurable operators, Commun. Contemp.Math. 13 (2011) 643 -- 657. [4] Sh. Ayupov, K. Kudaybergenov, Local derivations on measurable operators and commu- tativity, European Journal of Mathematics 2: 1023 (2016) 1023-1030. [5] Sh. Ayupov, K. Kudaybergenov, Local derivations on finite dimensional Lie algebras, Lin- ear Algebra Appl. 493 (2016) 381-398. [6] L. Chen, F. Lu, T. Wang, Local and 2-local Lie derivations of operator algebras on Banach spaces, Integr. Equ. Oper. Theory 77 (2013) 109-121. [7] R. Crist, Local derivations on operator algebras, J. Funct. Anal. 135 (1996) 72 -- 92. [8] D. Hadwin, J. Li, Local derivations and local automorphisms, J. Math. Anal. Appl. 290 (2004) 702-714. [9] D. Hadwin, J. Li, Local derivations and local automorphisms on some algebras, J. Operator Theory 60(1) (2008) 29 -- 44. [10] J. He, J. Li, G. An, W. Huang, Characterizations of 2-local derivations and local Lie derivations on some algebras, Sib. Math. J. 59 (2018) 912 -- 926 [11] J. He, J. Li, D. Zhao, Derivations, local and 2-local derivations on some algebras of oper- ators on Hilbert C∗-modules, Mediterr. J. Math. 14: 230 (2017) [12] W. Jing, Local derivations on reflexive algebras II, Proc. Amer. Math. Soc. 129 (2001) 1733-1737. [13] B. Johnson, Local derivations on C*-algebras are derivations, Trans. Amer. Math. Soc. 353 (2001) 313 -- 325. [14] R. Kadison, Local derivations, J. Algebra 130 (1990) 494 -- 509. [15] S. Kim, J. Kim, Local automorphisms and derivations on M n, Proc. Amer. Math. Soc. 132 (2004) 1389 -- 1392. [16] D. Larson, A. Sourour, Local derivations and local automorphisms, Proc. Sympos. Pure Math. 51 (1990) 187 -- 194. [17] J. Li, Z. Pan, Annihilator-preserving maps, multipliers and local derivations, Linear Alge- bra Appl. 432 (2010) 5-13 [18] D. Liu, J. Zhang, Local Lie derivations on certain operator algebras, Ann. Funct. Anal. 8 (2017) 270 -- 280. [19] D. Liu, J. Zhang, Local Lie derivations of factor von Neumann algebras, Linear Algebra Appl. 519 (2017) 208 -- 218 [20] Y. Pang, W. Yang, Derivations and local derivations on strongly double triangle subspace lattice algebras, Linear Multilinear Algebra 58 (2010) 855 -- 862. [21] H. Upmeier, Derivations on Jordan C∗-algebras, Math. Scand. 46 (1980) 251 -- 264. [22] J. Zhang, G. Ji, H. Cao, Local derivations of nest subalgebras of von Neumann algebras, Linear Algebra Appl. 392 (2004) 61-69. [23] J. Zhang, F. Pan, A. Yang, Local derivations on certain CSL algebras, Linear Algebra Appl. 413 (2006) 93-99. LOCAL DERIVATIONS ON MATRIX ALGEBRAS 23 1 V.I. Romanovskiy Institute of Mathematics, Uzbekistan Academy of Sciences, Tashkent, Uzbekistan. 2 National University of Uzbekistan, Tashkent, Uzbekistan. E-mail address: sh−[email protected] 3 Department of Mathematics, Andizhan State University, Andizhan, Uzbek- istan. E-mail address: [email protected]
1306.4006	1	1306	2013-06-17T20:21:04	A note on orthogonal Lie algebras in dimension 4 viewed as current Lie algebras	[ "math.RA" ]	Orthogonal Lie algebras in dimension 4 are identified as current Lie algebras, thus producing a natural decomposition for them over any field.	math.RA	math	A NOTE ON ORTHOGONAL LIE ALGEBRAS IN DIMENSION 4 VIEWED AS CURRENT LIE ALGEBRAS MARTIN CHAKTOURA AND FERNANDO SZECHTMAN Abstract. Orthogonal Lie algebras in dimension 4 are identified as current Lie algebras, thus producing a natural decomposition for them over any field. Let f : V × V → F be a non-degenerate symmetric bilinear form defined on a vector space V of dimension 4 over a field F . If char(F ) = 2 we further assume that f is not alternating. Let D be the discriminant of f relative to a basis of V . Let L(f ) be the subalgebra of gl(V ) associated to f , formed by all x ∈ gl(V ) that are skew-adjoint relative to f . Let M = [L(f ), L(f )], which coincides with L(f ) if and only if char(F ) 6= 2. In any case, M is a 6-dimensional orthogonal Lie algebra. According to [B], Chapter 1, §6, Exercise 26, if char(F ) 6= 2 then M is either the direct sum of two 3-dimensional simple ideals or simple, depending on whether D is a square in F or not. On the other hand, if char(F ) = 2 then either M = N ⋉ R or M is simple, depending, again, on whether D is a square in F or not; here N is a simple 3-dimensional simple subalgebra and R is the solvable radical of M . We may view these orthogonal Lie algebras as current Lie algebras, and in this way explain all cases described above in a uniform manner. Theorem. We have M ∼= [L(f W ), L(f W )] ⊗ F [X]/(X 2 − D), where W is an arbitrary 3-dimensional subspace of V such that f W is non-degenerate. Proof. It follows from [K], Theorems 4 and 20, that V admits an orthogonal basis B = {v1, v2, v3, v4}. Thus, the Gram matrix of f relative to B is diagonal with non-zero entries a, b, c, d and D = abcd. Let f1 = be12 − ae21, f2 = ce23 − be32, f3 = ce13 − ae31, h1 = ab(de34 − ce43), h2 = bc(de14 − ae41), h3 = ac(be42 − de24). Then f1, f2, f3, h1, h2, h3 is a basis of M , with multiplication table [f1, f2] = bf3, [f2, f3] = cf1, [f3, f1] = af2, [f1, h2] = bh3, [f2, h3] = ch1, [f3, h1] = ah2, [f2, h1] = −bh3, [f3, h2] = −ch1, [f1, h3] = −ah2, and [h1, h2] = Dbf3, [h2, h3] = Dcf1, [h3, h1] = Daf2. Thus M has the same multiplication table as [L(f W ), L(f W )] ⊗ F [X]/(X 2 − D), where W is the span by v1, v2, v3. By [K], Theorems 4 and 20, W can be replaced by any 3-dimensional subspace of V where the restriction of f is non-degenerate. (cid:3) 2000 Mathematics Subject Classification. 17B05. Key words and phrases. Classical Lie algebras, current Lie algebras. The second author was supported in part by an NSERC discovery grant. 1 2 MARTIN CHAKTOURA AND FERNANDO SZECHTMAN It is obvious from the Theorem that M decomposes exactly as prescribed above if D is a square. Suppose D is not a square. Then K = F [X]/(X 2 − D) is a quadratic field extension of F . Since [L(f W ), L(f W )] is a perfect 3-dimensional Lie algebra over F , it follows that [L(f W ), L(f W )] ⊗ K is a simple Lie algebra over K, and hence over F , as seen below. Thus, by the Theorem, M is a simple Lie algebra over F . Lemma. Suppose that F is a subfield of K. Then L is simple over F . [LP], Lemma 2.7) Let L be a simple Lie algebra over a field K. (cf. Proof. Let I be a non-zero ideal of L over F . The K-span of I is a non-zero ideal J of L over K, so J = L = [L, L] = [L, J] = [L, I] ⊆ I. (cid:3) A simple Lie algebra need not remain simple, or even semisimple, upon field extension. For instance, let L be an absolutely simple Lie algebra over an imperfect field F of characteristic p, and let s ∈ F \ F p and K = F [X]/(X p − s), which has an element t satisfying tp = s. Then P = L ⊗ K is simple over K and hence over F , but P ⊗ K ∼= L ⊗ K[X]/(X p − s) ∼= L ⊗ K[X]/(X − t)p is not semisimple over K. Acknowledgment. We are grateful to A. Pianzola for pointing out reference [LP], which replaced a more convoluted argument. References [B] N. Bourbaki, Lie groups and Lie algebras, Chapters 1-3, Springer-Verlag, Berlin, 1989. [K] I. Kaplansky, Linear algebra and geometry. A second course., Allyn and Bacon, Boston, 1969. [LP] M. Lau and A. Pianzola, Maximal Ideals and Representations of Twisted Forms of Algebras, to appear in Algebra and Number Theory. Department of Mathematics and Statistics, Univeristy of Regina, Canada E-mail address: martin [email protected] Department of Mathematics and Statistics, Univeristy of Regina, Canada E-mail address: [email protected]
1906.04519	1	1906	2019-06-11T12:14:31	Morphisms, direct sums and tensor products of K\"ahler-Poisson algebras	[ "math.RA" ]	In this paper we introduce the concept of morphisms of K\"ahler-Poisson algebras and study their algebraic properties. In particular, we find conditions, in terms of the metric, for two algebras to be isomorphic, and we introduce direct sums and tensor products of K\"ahler-Poisson algebras. We provide detailed examples to illustrate the novel concepts.	math.RA	math	MORPHISMS, DIRECT SUMS AND TENSOR PRODUCTS OF KÄHLER -- POISSON ALGEBRAS AHMED AL-SHUJARY Abstract. In this paper we introduce the concept of morphisms of Kähler-Poisson algebras and study their algebraic properties. In particular, we find conditions, in terms of the metric, for two algebras to be isomorphic, and we introduce direct sums and tensor products of Kähler-Poisson algebras. We provide detailed examples to illustrate the novel concepts. 1. Introduction The study of geometry via Poisson algebras goes back to two centuries ago through the works of Lagrange, Poisson and Lie. Poisson [16] invented his brackets as a tool for classical dynamics, Jacobi [10] realized the importance of these brackets and studied their algebraic properties, and Lie [15] began the study of their geometry. The study of geometry via Poisson algebras has experienced an amazing development since the 1980,s, starting with the foundational work of Weinstein [18] on Poisson manifolds. Since then many authors have studied the geometric and algebraic properties of symplectic and Poisson manifolds (see e.g. [5, 8, 9, 11, 14]). Later, Kontsevich [11] has shown that the classification of formal deformations of the algebra of functions for any manifold Σ is equivalent to the classification of formal families of Poisson structures on Σ. However, metric aspects of Poisson manifolds have not been investigated to the same extent. In [3] we began to study these metric aspects by defining Kähler-Poisson algebras as algebraic analogues of the algebra of functions on Kähler manifolds (the concept of Kähler manifolds was introduced by E. Kähler [12]). This study of metric aspects was motivated by the results in [1, 2], where many aspects of the differential geometry of embedded Riemannian manifolds Σ can be formulated in terms of the Poisson structure of the algebra of functions of Σ. In [3] we showed that "the Kähler -- Poisson condition", being the crucial identity in the definition of Kähler-Poisson algebras, allowed for an identification of geometric objects in the Poisson algebra which share crucial properties with their classical counterparts. For instance, we proved the existence of a unique Levi-Civita connection on the module generated by the inner derivations of the Kähler-Poisson algebra, and show that the curvature operator has all the classical symmetries. It is generally interesting to ask how many different (up to isomorphism) Kähler-Poisson structures do there exist on a given Poisson algebra? The aim of this paper is to explore further algebraic properties of Kähler-Poisson algebras. In particular, we find appropriate definitions of morphisms of Kähler-Poisson algebras. We illustrate with examples when two Kähler-Poisson algebras are isomorphic, for instance, we begin by taking algebras A and A′, where A is finitely generated algebra, A′ = A and we consider different set of generators for the Kähler-Poisson algebra structures of A and A′. We then use the concept of morphism to define subalgebras of Kähler-Poisson algebras. Again, we present examples to understand subalgebras of Kähler-Poisson algebras. Finally, we introduce direct sums and tensor products of Kähler-Poisson algebras together with their basic properties. 1 2 AHMED AL-SHUJARY 2. Kähler-Poisson algebras We begin this section by recalling the main object of our investigation. In [3] we introduce Kähler-Poisson algebras over the field K (either R or C). Let us consider a Poisson algebra (A, {., .}), over a field K and let {x1, ..., xm} be a set of distinguished elements of A. These elements play the role of functions providing an embedding into Rm for Kähler manifolds. Kähler manifolds and their properties have been extensively studied (see e.g. [5, 14, 18, 19]). Let us recall the definition of Kähler-Poisson algebras together with a few basic results. Definition 2.1. Let (A, {·, ·}) be a Poisson algebra over K and let x1, ..., xm ∈ A. Given a symmetric m × m matrix g = (gij) with entries gij ∈ A, for i, j = 1, ..., m, we say that the triple K = (A, g, {x1, ..., xm}) is a Kähler -- Poisson algebra if there exists η ∈ A such that (2.1) m Xi,j,k,l η{a, xi}gij{xj, xk}gkl{xl, b} = −{a, b} for all a, b ∈ A. We call equation (2.1) "the Kähler -- Poisson condition". Given a Kähler-Poisson algebra K = (A, g, {x1, ..., xm}), let g denote the A-module gener- ated by all inner derivations , i.e. g = {a1{c1, .} + ... + aN {cN , .} : ai, ci ∈ A and N ∈ N}. It is a standard fact that g is a Lie algebra over K with respect to the bracket [α, β](a) = α(β(a)) − β(α(a)), where α, β ∈ g and a ∈ A (see e.g.[7, 13]). It was shown in [3] that g is a projective module and that every Kähler -- Poisson algebra is a Lie-Rinehart algebra. More details for Lie-Rinehart algebra can be found in [9, 17]. Moreover, it follows that the matrix g defines a metric on g by for α, β ∈ g. Let us now introduce some notation for Kähler -- Poisson algebras. We set g(α, β) = α(xi)gij β(xj ), and for a ∈ A, as well as P ij ={xi, xj} P i(a)={xi, a} Dij = η{xi, xl}glk{xj, xk} Di(a) = η{xk, a}gkl{xl, xi}. Note that Dij = Dji. The metric g will be used to lower indices in analogy with differential geometry. E.g. P i j = P ikgkj Di j = Dikgkj Di = gijDj . In this notation, (2.1) can be stated as (2.2) Di(a)P i(b) = {a, b}. Furthermore, one immediately derives the following identities (2.3) Dij Pj(a) = P i(a), P ijDj(a) = P i(a) and Di jDjk = Djk. If we denote by P the matrix with entries P ij, the Kähler-Poisson condition (2.1) in Definition 2.1, can be written in matrix notation as ηPgPgP = −P, for an algebra A generated by {x1, ..., xm}. Let us also recall Example 6.1 in [3] (see also [4]). This example shows that any Poisson algebra generated by two elements can be endowed with a Kähler-Poisson algebra structure. Example 2.2. [3] Let A be a Poisson algebra generated by two elements x and y ∈ A and let 3 It is easy to check that for an arbitrary symmetric matrix with (det(g) 6= 0) P =(cid:18) 0 −{x, y} {x, y} 0 (cid:19). c g =(cid:18)a c b(cid:19) −a{x, y} −c{x, y}(cid:19) (c{x, y})2 − ab{x, y}2(cid:19) . b{x, y} 0 PgPg =(cid:18) c{x, y} −a{x, y} −c{x, y}(cid:19)(cid:18) c{x, y} b{x, y} =(cid:18)(c{x, y})2 − ab{x, y}2 0 one obtains Therefore 0 {x, y}((c{x, y})2 − ab{x, y}2) −{x, y}((c{x, y})2 − ab{x, y}2) 0 (cid:19) PgPgP =(cid:18) = −{x, y}2(ab − c2)P = −{x, y}2 det(g)P, giving η = ({x, y}2 det(g))−1, provided that the inverse exists. Thus, as long as {x, y}2 det(g) is not a zero-divisor, one may localize A to obtain a Kähler-Poisson algebra K = (A[({x, y}2 det(g))−1], g, {x, y}). 3. Homomorphisms of Kähler-Poisson algebras In this section we are interested in studying maps between Kähler-Poisson algebras. Given a Poisson algebra, we will investigate isomorphism classes of Kähler-Poisson algebra structures on the Poisson algebra. As the definition of a Kähler-Poisson algebra involves the choice of a set of distinguished elements, we will require a morphism to respect the subalgebra generated by these elements. To this end, we start by making the following definition: Definition 3.1. Given a Kähler-Poisson algebra K = (A, g, {x1, ..., xm}) on a Poisson algebra (A, {·, ·}), let Afin ⊆ A denote the Poisson subalgebra generated by {x1, ..., xm}. We now introduce morphisms of Kähler-Poisson algebras in the following way: Definition 3.2. Let K = (A, g, {x1, ..., xm}) and K′ = (A′, g′, {y1, ..., ym′}) be Kähler-Poisson ′, respectively. A (homo)morphism algebras together with their modules of derivations g and g of Kähler-Poisson algebras is a pair of maps (φ, ψ), with φ : A → A′ a Poisson algebra homomorphism and ψ : g → g ′ a Lie algebra homomorphism, such that (1) ψ(aα) = φ(a)ψ(α), (2) φ(α(a)) = ψ(α)(φ(a)), (3) φ(g(α, β)) = g′(ψ(α), ψ(β)), (4) φ(Afin) ⊆ A′ for all a ∈ A and α, β ∈ g. fin, Remark 3.3. Note that a morphism of Kähler-Poisson algebras is also a morphism of the underlying Lie-Rinehart algebras. 4 AHMED AL-SHUJARY Furthermore, in next Proposition we show that the composition of two Kähler-Poisson algebras morphisms is a morphism of Kähler-Poisson algebras as we expect and require. Proposition 3.4. Let K = (A, g, {x1, ..., xm}) and K′ = (A′, g′, {y1, ..., ym′}) be Kähler- ′ and let K′′ = Poisson algebras together with their corresponding modules of derivations g, g (A′′, g′′, {z1, ..., zm′′}) be a Kähler-Poisson algebra with its module of derivations g ′′. If (φ, ψ) : K → K′ and (φ′, ψ′) : K′ → K′′ are homomorphisms of Kähler-Poisson algebras then (φ′ ◦ φ, ψ′ ◦ ψ) : K → K′′ is a homomorphism of Kähler-Poisson algebras. Proof. Let φ = φ′ ◦ φ and ψ = ψ′ ◦ ψ, where φ : A → A′, φ′ : A′ → A′′, ψ : g → g ψ′ : g Kähler-Poisson algebras we show that ( φ, ψ) satisfies properties (1-4) in Definition 3.2. ′ and ′′. To prove that ( φ, ψ) is a homomorphism of ′′, giving φ : A → A′′ and ψ : g → g ′ → g (1) ψ(aα) = φ(a) ψ(α), for all α ∈ g and a ∈ A. ψ′(ψ(aα)) − φ′(φ(a))ψ′(ψ(α)) = ψ′(φ(a)ψ(α)) − φ′(φ(a))ψ′(ψ(α)) = φ′(φ(a))ψ′(ψ(α)) − φ′(φ(a))ψ′(ψ(α)) = 0. (2) φ(α(a)) = ψ(α)( φ(a)), for all α ∈ g and a ∈ A. φ(α(a)) − ψ(α)( φ(a)) = φ′(φ(α(a))) − ψ′(ψ(α))(φ′(φ(a))) = φ′(ψ(α)(φ(a))) − ψ′(ψ(α))(φ′(φ(a))) = ψ′(ψ(α)(φ′(φ(a))) − ψ′(ψ(α))(φ′(φ(a))) = 0. (3) φ(g(α, β)) = g′′( ψ(α), ψ(β)), for α, β ∈ A. φ(g(α, β)) − g′′( ψ(α), ψ(β)) = φ′(φ(g(α, β))) − g′′(ψ′(ψ(α)), ψ′(ψ(β))) = φ′(g′(ψ(α), ψ(β))) − g′′(ψ′(ψ(α)), ψ′(ψ(β))) = g′′(ψ′(ψ(α)), ψ′(ψ(β))) − g′′(ψ′(ψ(α)), ψ′(ψ(β))) = 0. (4) φ′(φ(Afin)) ⊆ A′′ fin) ⊆ A′′ Therefore, ( φ, ψ) is a homomorphism of Kähler-Poisson algebras. fin, since, φ(Afin) ⊆ A′ fin and φ′(A′ fin. (cid:3) Remark 3.5. Note that, the composition of Kähler-Poisson algebra morphism is associative and the identity (idA, idg) of K = (A, g, {x1, ..., xm}) with idA : A → A and idg : g → g is a morphism of Kähler-Poisson algebras. Remark 3.6. An isomorphism of Kähler-Poisson algebras is a morphism (φ, ψ) of Kähler- Poisson algebras such that φ is a Poisson algebra isomorphism and φ(Afin) = A′ fin. Observe that, ψ can always be constructed from φ. When two Kähler-Poisson algebras K and K′ are isomorphic, we write K ∼= K′. Before continuing, we need to introduce some notation. Let (A, {., .}) and (A′, {., .}′) be Poisson algebras and let xi ∈ A for i = 1, ..., m. If p ∈ A is a polynomial in {x1, ..., xm} then, using Leibniz rule, one computes (3.1) {p, a} = ∂p ∂xi {xi, a} where ∂p ∂xi denotes the formal derivative of the polynomial p with respect to the variable xi. Note that, in general, ∂p ∂xi is itself not well-defined in the algebra, since there might exist several different (but equivalent) representations of p as a polynomial in x1, ..., xm, and the formal derivative then yields several, possibly non-equivalent, elements of the algebra. However, the combination in (3.1) is always well-defined, and gives the same result for all representations of p. 5 Given a matrix M = (mij) over A, we set φ(M ) = (φ(mij )). Given a morphism (φ, ψ) : (A, g, {x1, ..., xm}) → (A′, g′, {y1, ..., ym′}), it will be convenient to introduce the notation Ai α = ∂φ(xi) ∂yα (keeping in mind that this is not well-defined by itself); recall that if (φ, ψ) is a morphism of Kähler-Poisson algebras, then φ(Afin) ⊆ A′ fin, ensuring that φ(xi) is indeed a polynomial in y1, ..., ym′. This notation allows us to write φ({xi, xj}) = {φ(xi), φ(xj )}′ = Ai α{yα, yβ}′Aj β in matrix notation: φ(P) = AP ′AT , where P = ({xi, xj}) and P ′ = ({yα, yβ}′). Given two Kähler-Poisson algebras K = (A, g, {x1, ..., xm}) and K′ = (A′, g′, {y1, ..., ym′}), we would like to understand when they are isomorphic. In the following, we shall consider a number of examples in order to explore when Kähler-Poisson algebras are isomorphic. From now on, we consider only finitely generated Poisson algebras that have been properly localized (cf. Example 2.2) to allow the construction of a Kähler-Poisson algebra. We begin by taking finitely generated algebras A and A′, where A = A′ and we consider different set of generators for the Kähler-Poisson algebra structures of A and A′. Let us start with the following example. Example 3.7. Let (A, {., .}) be a Poisson algebra generated by two elements x and y. From Example 2.2 we know that K = (A[({x, y}2 det(g))−1], g, {x, y}) is a Kähler-Poisson algebra for an arbitrary symmetric matrix Consider K′ = (A[({x, y}2 det(g))−1], h, {x + y, x − y}), with a symmetric matrix g =(cid:18)g11 g12 g12 g22(cid:19), det(g) 6= 0 with η =(cid:0){x, y}2 det(g)(cid:1)−1 h =(cid:18)h11 h12 . h12 h22(cid:19), det(h) 6= 0. {x − y, x − y}(cid:19) =(cid:18) 0 {x + y, x − y} −2{x, y} 2{x, y} 0 (cid:19) . We have P ′ =(cid:18){x + y, x + y} {x + y, x − y} It is easy to check that for the symmetric matrix h we obtain (3.2) P ′hP ′hP ′ = −4{x, y}2 det(h)P ′, giving η′ = (4{x, y}2 det(h))−1. For each metric g in the definition of K, we find a suitable matrix h such that K ∼= K′. From g = {a1{x, ·} + a2{y, ·} : a1, a2 ∈ A}, and g ′ = {a1{x + y, ·} + a2{x − y, ·} : a1, a2 ∈ A} = {a1{x, ·} + a1{y, ·} + a2{x, ·} − a2{y, ·} : a1, a2 ∈ A} = {(a1 + a2){x, ·} + (a1 − a2){y, ·} : a1, a2 ∈ A}, ′. Since we require that K ∼= K′, we define maps φ : A → A and ψ : g → g we see that g = g satisfying properties (1-4). We choose φ = id , ψ = id and find a suitable choice of matrix h yielding that (φ, ψ) is an isomorphism of Kähler Poisson algebras. Now, we check that properties (1-4) in Definition 3.2 are satisfied. (1) φ(a)ψ(α) = aα = ψ(aα) (2) ψ(α)(φ(a)) = α(a) = φ(α(a)) 6 AHMED AL-SHUJARY (3) To show that g(α, β) = h(α, β), we start from the left hand side φ(g(α, β)) = α(xi)gijβ(xj ) = α(x)g11β(x) + α(x)g12β(y) + α(y)g21β(x) + α(y)g22β(y). From the right hand side we get h(α, β) = α(xi)hij β(xj ) = α(x + y)h11β(x + y) + α(x + y)h12β(x − y) + α(x − y)h21β(x + y) + α(x − y)h22β(x − y) = α(x)h11β(x) + α(x)h11β(y) + α(y)h11β(x) + α(y)h11β(y) + α(x)h12β(x)+ − α(x)h12β(y) + α(y)h12β(x) − α(y)h12β(y) + α(x)h21β(x) + α(x)h21β(y) − α(y)h21β(x) − α(y)h21β(y) + α(x)h22β(x) − α(x)h22β(y) − α(y)h22β(x) + α(y)h22β(y) = α(x)h11β(x) + α(x)h11β(y) + α(y)h11β(x) + α(y)h11β(y) + 2α(x)h12β(x) − 2α(y)h12β(y) + α(x)h22β(x) − α(x)h22β(y) − α(y)h22β(x) + α(y)h22β(y). We conclude that (3.3) (3.4) (3.5) (3.6) α(x)β(x) : h11 + 2h12 + h22 = g11 α(x)β(y) : h11 − h22 = g12 α(y)β(x) : h11 − h22 = g21 α(y)β(y) : h11 − 2h12 + h22 = g22. From (3.6) we obtain, h11 = 2h12 − h22 + g22 and setting this in (3.3) we get 2h12 − h22 + g22 + 2h12 + h22 = g11 4h12 = g11 − g22 h12 = g11−g22 . 4 From (3.4) we get h22 = h11 − g12, which in (3.3) gives h11 + 2( g11−g22 4 ) + h11 − g12 = g11 2h11 + ( g11 −g22 4h11 + g11 − g22 = 2(g11 + g12) ) − g12 = g11 2 4h11 = g11 + 2g12 + g22 h11 = g11+2g12+g22 , 4 which implies that h22 = h11 − g12 = g11 + 2g12 + g22 4 − g12 = g11 − 2g12 + g22 4 . Now, the symmetric matrix h becomes giving det(h) = = h =(cid:18) g11+2g12+g22 g11−g22 4 4 g11−g22 4 g11−2g12+g22 4 (cid:19), 1 1 16h(g11 + 2g12 + g22)(g11 − 2g12 + g22) − (g11 − g22)2i 16h(g11)2 − 2g11g12 + g11g22 + 2g11g12 − 4(g12)2 + 2g12g22 + g11g22 − 2g12g22 + (g22)2 − (g11)2 − (g22)2 + 2g11g22i = 1 16 (4g11g22 − 4(g12)2) = 1 4 (g11g22 − (g12)2) = 1 4 det(g). Inserting det(h) in equation 3.2 we get P ′hP ′hP ′ = −4{x, y}2 1 4 det(g)P ′ = −{x, y}2 det(g)P ′, giving η′ = ({x, y}2 det(g))−1. Therefore η′ = η. We conclude that g(α, β) = h(α, β). 7 (4) It is easy to see that φ(Afin) ⊆ A′ fin. Hence, K ∼= K′ if h =(cid:18) g11+2g12+g22 g11−g22 4 4 g11−g22 4 g11−2g12+g22 4 (cid:19) . Note that the above example extends to the case where we choose more (dependent) gen- erators of the algebra, giving many possible presentations of the same Kähler-Poisson algebra. Next, let us explore the case when we choose a different number of generators for the same algebra. Example 3.8. Let (A, {·, ·}) be a Poisson algebra generated by two elements x and y. Consider the Kähler-Poisson algebra K = (A[({x, y}2 det(g))−1], g, {x, y}), with an arbitrary symmetric matrix g =(cid:18)g11 g12 g12 g22(cid:19), det(g) 6= 0 and η =(cid:0){x, y}2 det(g)(cid:1)−1 . Let K′ = (A[({x, y}2 det(g))−1], h, {x, y, x}) with λ 0 0 −λ 0 −λ 0 λ P ′ =  1 4 g11 1 2 g12 1 4 g11 1 2 g12 g22 1 2 g12 h =  0  ,  , 1 4 g11 1 2 g12 1 4 g11 where λ = {x, y}. It is easy to check that for the symmetric matrix h one obtains P ′hP ′hP ′ = −{x, y}2 det(g)P ′, giving η′ = ({x, y}2 det(g))−1. We conclude that η′ = η. Now, we show that K ∼= K′. From and g = {a1{x, ·} + a2{y, ·} : a1, a2 ∈ A} g ′ = {a1{x, ·} + a2{y, ·} + a3{x, ·} : a1, a2, a3 ∈ A} = {(a1 + a3){x, ·} + a2{y, ·} : a1, a2, a3 ∈ A}, ′. We define maps φ : A → A and ψ : g → g by choosing φ = id and ψ = id, we see that g = g and we find a suitable choice of matrix h such that (φ, ψ) is an isomorphism of Kähler Poisson algebras. We again check properties (1-4) in Definition 3.2 are satisfied. (1) φ(a)ψ(α) = aα = ψ(aα). (2) ψ(α)(φ(a)) = α(a) = φ(α(a)). (3) To show that φ(g(α, β)) = h(ψ(α), ψ(β)), we start from the right hand side h(ψ(α), ψ(β)) = h(α, β) = α(xi)hij β(xj ) = α(x)h11β(x) + α(x)h12β(y) + α(y)h13β(x) + α(y)h21β(x) + α(y)h22β(y) + α(y)h23β(x) + α(x)h31β(x) + α(x)h32β(y) + α(x)h33β(x) 8 AHMED AL-SHUJARY = 1 4 + α(x)g11β(x) + 1 2 α(x)g12β(y) + 1 4 α(x)g11β(x) + 1 2 α(y)g12β(x) + α(y)g22β(y) 1 2 α(y)g12β(x) + 1 4 α(x)g11β(x) + 1 2 α(x)g12β(y) + 1 4 α(x)g11β(x) (3.7) = α(x)g11β(x) + α(x)g12β(y) + α(y)g12β(x) + α(y)g22β(y). From the left hand side we get φ(g(α, β)) = α(xi)gijβ(xj ) = α(x)g11β(x) + α(x)g12β(y) + α(y)g21β(x) + α(y)g22β(y), and we conclude that φ(g(α, β)) = h(ψ(α), ψ(β)). (4) It is easy to see that φ(Afin) ⊆ A′ fin. Therefore, K ∼= K′ for the matrix h with entries as in (3.7). The above examples show that two Kähler-Poisson algebras can be isomorphic when con- sidering different set of generators of the same finitely generated Poisson algebra. 3.1. Properties of isomorphisms. We are interested in studying properties of more gen- eral isomorphisms for Kähler-Poisson algebras. Let Let K = (A, g, {x1, ..., xm}) and K′ = (A′, g′, {y1, ..., ym′}) be Kähler-Poisson algebras, and assume that there exists a Poisson alge- bra isomorphism φ : (A, {., .}) → (A′, {., .}′). When does there exist a map ψ : g → g ′ such that (φ, ψ) is an isomorphism of Kähler-Poisson algebras?. The following result provides an answer to this question. Proposition 3.9. Let K = (A, g, {x1, ..., xm}) and K′ = (A′, g′, {y1, ..., ym′}) be Kähler- Poisson algebras. Then K and K′ are isomorphic if and only if there exists a Poisson algebra isomorphism φ : A → A′ such that φ(Afin) = A′ fin, and (3.8) P ′g′P ′ = P ′AT φ(g)AP ′, where Ai α = ∂φ(xi) ∂yα and (P ′)αβ = {yα, yβ}′. Proof. Assume that K ∼= K′. Then we need to show that P ′g′P ′ = P ′AT φ(g)AP ′. Let us start by computing φ(P ij ) φ(P ij ) = φ({xi, xj}) = {φ(xi), φ(xj )}′ = ∂φ(xi) ∂yα {yα, φ(xi)}′ = ∂φ(xi) ∂yα {yα, yβ}′ ∂φ(xj ) ∂yβ = Ai α(P ′)αβAj β. Now, in order to prove (3.8), we let γi = {xi, .} and γj = {xj, .} then φ(g(γi, γj)) = φ({xi, xk}gkl{xj, xl}) = φ(P ikgklP jl). From (2) in Definition 3.2 it follows that ψ(γi)(b) = φ({xi, φ−1(b)}) = {φ(xi), b}′, which implies that, ψ(γi) = {φ(xi), ·}′ and in the same way we get ψ(γj) = {φ(xj ), ·}′. Furthermore, and φ(P ikgklP jl) = {φ(xi), yα}′Ak αφ(gkl){φ(xj ), yβ}′Al β, g′(ψ(γi), ψ(γj )) = {φ(xi), yα}′g′ αβ{φ(xj), yβ}′. From (3) in Definition 3.2 one obtains (3.9) {φ(xi), yα}′(Ak αφ(gkl)Al β − g′ αβ){φ(xj ), yβ}′ = 0. Note that if {φ(xi), yβ}Cβ = 0, for Cβ ∈ A′, then 9 {yα, yβ}′Cβ = φ({φ−1(yα), φ−1(yα)})Cβ = φ(cid:16) ∂φ−1(yα) ∂xi {xi, φ−1(yβ)}(cid:17)Cβ = φ(cid:16) ∂φ−1(yα) ∂xi (cid:17)φ(cid:16){xi, φ−1(yβ)}(cid:17)Cβ = φ(cid:16) ∂φ−1(yα) ∂xi (cid:17){φ(xi), yβ}′Cβ = 0. Therefore, equation (3.9) yields {yγ, yα}′(Ak αφ(gkl)Al β − g′ αβ){φ(xj ), yβ}′ = 0, and furthermore {yγ, yα}′(Ak αφ(gkl)Al β − g′ αβ){yδ, yβ}′ = 0. In matrix notation this becomes (3.10) P ′Aφ(g)AT P ′ = P ′g′P ′. To prove the converse, we assume that φ : A → A′ (φ(a) = a′) is an isomorphism such that fin and (3.10) holds. First, we need to define the map ψ. Since φ is an isomorphism φ(Afin) = A′ of Poisson algebras, we may define ψ as: which clearly fulfills: ψ(γi)(a′) := φ(α(φ−1(a′))), φ(γi(a)) = ψ(γi)(φ(a)). With a slight abuse of notation, let us show that ψ(α) ∈ g and β = bi{bi, .} as inner derivations in g one obtains ′ for α ∈ g. Writing α = ai{bi, .} ψ(α)(a′) = φ(α(φ−1(a′))) = φ(ai{bi, φ−1(a′)}) = φ(ai){φ(bi), a′}′, which implies that ψ(α) = φ(ai){φ(bi), .}′ ∈ g ′. Secondly, we show that ψ is a Lie algebra isomorphism: g ′. Similarly, one obtains ψ(β) = φ(bi){φ(bi), .}′ ∈ I) [ψ(α), ψ(β)](a′) = ψ(α)(cid:0)ψ(β)(a′)(cid:1) − ψ(β)(cid:0)ψ(α)(a′)(cid:1) = ψ(α)(cid:0)φ(β(φ−1(a′)))(cid:1) − ψ(β)(cid:0)φ(α(φ−1(a′)))(cid:1) = φ(cid:0)α(β(φ−1(a′)))(cid:1) − φ(cid:0)β(α(φ−1(a′)))(cid:1) = φ(cid:0)α(β(φ−1(a′))) − β(α(φ−1(a′)))(cid:1) = φ(cid:0)[α, β](φ−1(a′)(cid:1) = ψ([α, β])(a′). II) ψ(α)(a′) + ψ(β)(a′) = φ(cid:0)α(φ−1(a′))(cid:1) + φ(cid:0)β(φ−1(a′))(cid:1) = φ(cid:0)α(φ−1(a′) + β(φ−1(a′))(cid:1) = φ(cid:0)(α + β)(φ−1(a′))(cid:1) = ψ(α + β)(a′). Therefore ψ is a Lie algebra homomorphism. Now, we show that the map ψ−1 defined by ψ−1(α′)(a) := φ−1(α′(φ(a))), 10 AHMED AL-SHUJARY for all a ∈ A and α′ ∈ g ′ is indeed the inverse of ψ: ψ−1(cid:0)ψ(α)(cid:1)(a) = φ−1(cid:16)ψ(α)(cid:0)φ(a)(cid:1)(cid:17) = φ−1(cid:16)φ(cid:0)α(φ−1φ(a))(cid:1)(cid:17) = α(a) and ψ(cid:0)ψ−1(α′)(cid:1)(a′) = φ(cid:16)ψ−1(α′)(cid:0)φ−1(a′)(cid:1)(cid:17) = φ(cid:16)φ−1(cid:0)α′(φφ−1(a′))(cid:1)(cid:17) = α′(a′). Finally, we show that (φ, ψ) is a morphism of Kähler-Poisson algebras. (1) φ(a)ψ(α)(a′) = φ(a)φ(α(φ−1(a′))) = φ(aα(φ−1(a′))) = ψ(aα)(a′). (2) ψ(α)(φ(a)) = φ(α(φ−1(φ(a)))) = φ(α(a)). (3) For α = αi{xi, .} and β = βi{xi, .} one gets φ(g(α, β)) = φ(αi{xi, xk}gklβj{xj, xl}) = φ(αi)φ({xi, xk})φ(gkl)φ(βj )φ({xj , xl}) = φ(αi)φ(P ik)φ(gkl)φ(βj )φ(P jl) = φ(αi)(AP ′AT )ikφ(gkl)φ(βj )(AP ′AT )jl = φ(αi)φ(βj )(cid:0)AP ′AT φ(g)A(−P ′)AT(cid:1)ij = −φ(αi)φ(βj )(cid:0)AP ′AT φ(g)AP ′AT(cid:1)ij . P ′g′P ′ = P ′AT φ(g)AP ′, By using one obtains φ(g(α, β)) = −φ(αi)φ(βj )(AP ′g′P ′AT )ij = −φ(αi)φ(βj )Ai α{yα, yβ}g′ βγ{yγ, yδ}Aj δ = −φ(αi)φ(βj ){φ(xi), yβ}g′ = g′(ψ(α), ψ(β)). βγ{yγ, φ(xj )} = ψ(α)(yβ)g′ βγ ψ(β)(yγ) Note that (4) is true by assumption. (cid:3) As an illustration, Proposition 3.9 gives us another method to do the calculations of Example 3.7. We have seen in Example 3.7 that two Kähler-Poisson algebras K = (A, g, {x1, ..., xm}) and K′ = (A, g′, {y1, ..., ym′}), where A is a finitely generated algebra, can be isomorphic when considering different sets of generators for the Kähler-Poisson algebra structures of A and A′. Example 3.10. (Continuation of 3.7) Proposition 3.9 tells us that if we set h = AT φ(g)A, then P ′hP ′ = P ′AT φ(g)AP ′ implying that K ∼= K′. Let y1 = x + y and y2 = x − y. Hence x = 1 ∂yα , with φ = id: 2 (y1 + y2) and y = 1 1 = ∂x1 ∂y1 = ∂y 2 (y1 − y2). We compute the matrix Ai ∂y2 = ∂( 1 ∂y1 = ∂( 1 ∂y1 = ∂( 1 ∂y2 = ∂y Therefore the matrix A becomes 2 = ∂x1 1 = ∂x2 2 , A1 2 , A2 1 = ∂x2 2 (y1−y2)) 2 (y1+y2)) = 1 = 1 A1 A2 ∂y1 ∂y1 α = ∂xi 2 (y1+y2)) ∂y2 = 1 2 , 2 (y1−y2)) ∂y2 = ∂( 1 ∂y2 = − 1 2 . 1 2 2(cid:19) 2 − 1 2 1 A =(cid:18) 1 2(cid:19) =(cid:18) 1 1 2 2 (g11 + g21) 1 2 (g11 − g21) g22 g22(cid:19)(cid:18) 1 2 1 2 − 1 1 1 4 (g12 + g22) 4 (g12 − g22) 1 4 (g11 − g22) 4 (g11 + g21) − 1 4 (g11 − g21) − 1 4 (g11 − 2g12 + g22)(cid:19) , 1 4 (g12 + g22) 4 (g12 − g22)(cid:19) 1 2 (g12 + g22) 1 2 (g12 − g22)(cid:19)(cid:18) 1 2 1 1 2 2(cid:19) 2 − 1 and 2 1 h =(cid:18) 1 =(cid:18) 1 =(cid:18) 1 1 1 2 2(cid:19)(cid:18)g11 g21 2 − 1 4 (g11 + g21) + 1 4 (g11 − g21) + 1 4 (g11 + 2g12 + g22) 1 4 (g11 − g22) which agrees with the result in Example 3.7. Also φ(Afin) = φ(A′ we can now use Proposition 3.9 to conclude that K ∼= K′. fin), since φ = idA. Therefore 11 In the above examples of isomorphism K ∼= K′, we noted that η = η′. The next proposition shows that this is indeed true in the generic case. Proposition 3.11. Let K = (A, g, {x1, ..., xm}) and K′ = (A′, g′, {y1, ..., ym′}) be Kähler- Poisson algebras and let (φ, ψ) : K → K′ be an isomorphism of Kähler-Poisson algebras. If ηPgPgP = −P and η′P ′g′P ′g′P ′ = −P ′ then (φ(η) − η′)P ′ = 0. Proof. From Proposition 3.9 we have P ′g′P ′ = P ′AT φ(g)AP ′. Starting from ηPgPgP = −P and using that φ(P) = AP ′AT and P ′g′P ′ = P ′AT φ(g)AP ′, one has −φ(P) = φ(η)φ(PgPgP). Multiplying both sides by η′, where η′P ′g′P ′g′P ′ = −P ′, we obtain −η′φ(P) = φ(η)φ(PgPgP)η′ = φ(η)AP ′AT φ(g)AP ′AT φ(g)AP ′AT η′ = φ(η)AP ′g′P ′AT φ(g)AP ′AT η′ = φ(η)Aη′P ′g′P ′g′P ′AT = −φ(η)AP ′AT . Hence, one obtains η′AP ′AT = φ(η)AP ′AT , which implies that(η′ − φ(η))P ′ = 0, using the same argument as in the proof of Proposition 3.9 (cid:3) Thus, if at least one of (P ′)αβ = {yα, yβ} is not a zero divisor, then Proposition 3.11 implies that φ(η) = η′. 4. Subalgebras of Kähler-Poisson algebras As shown in Section 3, a morphism of Kähler-Poisson algebras is a pair of maps (φ, ψ), with φ : A → A′ a Poisson algebra homomorphism and ψ : g → g ′ a Lie algebra homomorphism, satisfying certain conditions. In this section, we present some algebraic properties of Kähler- Poisson algebras. In particular, we introduce subalgebras of Kähler-Poisson algebras, which we define by using the concept of morphisms. Definition 4.1. Let K = (A, g, {x1, ..., xm}) and K′ = (A′, g′, {y1, ..., ym′}) be Kähler-Poisson ′, respectively. K is algebras together with their corresponding modules of derivations g and g a Kähler-Poisson subalgebra of K′ if: (A) A is a Poisson subalgebra of A′, (B) (id A, id g) is a homomorphism of Kähler-Poisson algebras,where id A and id g denotes the identity maps restricted to A and g, respectively. Given two Kähler-Poisson algebras K = (A, g, {x1, ..., xm}) and K′ = (A′, g′, {y1, ..., ym′}), we illustrate with examples when K is a Kähler-Poisson subalgebra of K′, where A is a proper Poisson subalgebra of a finitely generated algebra A′. We consider different set of generators for the Kähler-Poisson algebra structures of A and A′. Note that the property (B) in Definition 4.1 determines the metric g on K. Example 4.2. Let K = (A, g, {x, y}) and K′ = (A′, g′, {x, y, z}) be Kähler-Poisson algebras, where A is a subalgebra of A′ generated by {x, y} and A′ generated by {x, y, z}. Let {x, y} = p(x, y) and {x, z} = {y, z} = 0. Since A is a Poisson subalgebra of A′, to show that K is a Kähler-Poisson subalgebra of K′, we only need to show that (id A, id g) satisfies the properties in Definition 3.2. This fact determines the metric g on K. (1) ψ(aα) = aα = φ(a)ψ(α). (2) φ(α(a)) = α(a) = ψ(α)(φ(a)). (3) We see that φ(g(α, β)) = g′(ψ(α), ψ(β)). φ(g(α, β)) = α(xi)gijβ(xj ) 2 Xi,j=1 12 AHMED AL-SHUJARY = α(x1)g11β(x1) + α(x1)g12β(x2) + α(x2)g21β(x1) + α(x2)g22β(x2) = α(x)g11β(x) + α(x)g12β(y) + α(y)g21β(x) + α(y)g22β(y), and 3 g′(ψ(α), ψ(β)) = g′(α, β) = α(xi)g′ ijβ(xj ) Xi,j=1 = α(x1)g′ 11β(x1) + α(x1)g′ 12β(x2) + α(x1)g′ 13β(x3) + α(x2)g′ 21β(x1) + α(x2)g′ + α(x3)g′ 22β(x2) + α(x2)g′ 33β(x3) 23β(x3) + α(x3)g′ 31β(x1) + α(x3)g′ 32β(x2) = α(x)g′ + α(y)g′ 11β(x) + α(x)g′ 23β(z)α(z)g′ 11β(x) + α(x)g′ 12β(y) + α(x)g′ 31β(x) + α(z)g′ 12β(y) + α(y)g′ 13β(z) + α(y)g′ 32β(y) + α(z)g′ 21β(x) + α(y)g′ 21β(x) + α(y)g′ 33β(z) 22β(y), = α(x)g′ 22β(y) since, α = ai{xi, .} = a1{x, .} + a2{y, .} ∈ g, we get α(z) = a1{x, z} + a2{y, z} = 0. Similarly, β(z) = 0. By comparing both sides φ(g(α, β)) and g′(ψ(α), ψ(β)), we conclude that g11 = g′ 21 and g22 = g′ (4) By construction we have that φ(Afin) ⊆ A′ 12 ,g21 = g′ 11 ,g12 = g′ 22. Therefore, φ(g(α, β)) = g′(ψ(α), ψ(β)). fin. Therefore, K is a Kähler-Poisson subalgebra of K′ and g obtained from g′ as above. Let us take another example with different number of generators. Example 4.3. Let K = (A, g, {x, y}) and K′ = (A′, g′, {x, y, z, w}) be Kähler-Poisson algebras, where A is a subalgebra of A′ generated by {x, y} and A′ generated by {x, y, z, w}. Let {x, y} = p(x, y), {z, w} = q(z, w) and {x, z} = {x, w} = {y, z} = {y, w} = 0. To show that K is a Kähler-Poisson subalgebra of K′ we only need to show that (id A, id g) satisfies the properties in Definition 3.2. This fact determines the metric g on K. (1) ψ(aα) = aα = φ(a)ψ(α). (2) φ(α(a)) = α(a) = ψ(α)(φ(a)). (3) We see that φ(g(α, β)) = g′(ψ(α), ψ(β)), the left hand side is 2 φ(g(α, β)) = α(xi)gijβ(xj ) Xi,j=1 = α(x1)g11β(x1) + α(x1)g12β(x2) + α(x2)g21β(x1) + α(x2)g22β(x2) = α(x)g11β(x) + α(x)g12β(y) + α(y)g21β(x) + α(y)g22β(y), and the right hand side is 4 g′(ψ(α), ψ(β)) = g′(α, β) = α(xi)g′ ijβ(xj ) Xi,j=1 = α(x1)g′ 11β(x1) + α(x1)g′ 12β(x2) + α(x1)g′ 13β(x3) + α(x1)g′ 14β(x4) + α(x2)g′ + α(x3)g′ + α(x4)g′ 21β(x1) + α(x2)g′ 31β(x1) + α(x3)g′ 41β(x1) + α(x4)g′ 22β(x2) + α(x2)g′ 32β(x2) + α(x3)g′ 42β(x2) + α(x4)g′ 23β(x3) + α(x2)g′ 33β(x3) + α(x3)g′ 43β(x3) + α(x4)g′ 24β(x4) 34β(x4) 44β(x4) = α(x)g′ 11β(x) + α(x)g′ 12β(y) + α(x)g′ 13β(z) + α(x)g′ 14β(w) + α(y)g′ 21β(x) + α(y)g′ 22β(y) + α(y)g′ 23β(z) + α(y)g′ 24β(w) 13 + α(z)g′ + α(w)g′ 31β(x) + α(z)g′ 41β(x) + α(w)g′ 32β(y) + α(z)g′ 42β(y) + α(w)g′ 33β(z) + α(z)g′ 34β(w) 43β(z) + α(w)g′ 44β(w) = α(x)g′ 11β(x) + α(x)g′ 12β(y) + α(y)g′ 21β(x) + α(y)g′ 22β(y), since, α = ai{xi, .} = a1{x, .} + a2{y, .} ∈ g, we get α(z) = a1{x, z} + a2{y, z} = 0. Similarly, α(w) = β(z) = β(w) = 0. By comparing both sides φ(g(α, β)) and g′(ψ(α), ψ(β)), 22. Therefore, φ(g(α, β)) = 21 and g22 = g′ 11 ,g12 = g′ 12 ,g21 = g′ we conclude that g11 = g′ g′(ψ(α), ψ(β)). (4) By construction we have that φ(Afin) ⊆ A′ fin. Therefore, K is a Kähler-Poisson subalgebra of K′ and g obtained from g′ as above. Above we have given examples of when K = (A, g, {x1, ..., xm}) is a Kähler-Poisson subal- gebra of K′ = (A′, g′, {y1, ..., ym′}), where A is a Poisson subalgebra of A′. Next proposition shows that, in general a morphism (φ, ψ) : (K, g) → (K′, g ′) induces a Kähler-Poisson subalge- bra of K′, denoted by Im(φ, ψ). Proposition 4.4. Let K = (A, g, {x1, ..., xm}) and K′ = (A′, g′, {y1, ..., ym′}) be Kähler- Poisson algebras, and let (φ, ψ) : K → K′ be a homomorphism of Kähler-Poisson algebras. If yJ ∈ φ(A) for J = 1, ..., m′ then Im(φ, ψ) = (φ(A), g, {φ(x1), φ(x2), ..., φ(xm)}) is a Kähler- Poisson subalgebra of K′, where (4.1) gkl = φ(ηPkm){φ(xm), yJ }′g′ JM φ(ηPln){φ(xn), yM }′. Proof. First, we show that g is symmetric. gkl = φ(ηPkm){φ(xm), yJ }′g′ = φ(ηPln){φ(xn), yM }′g′ = φ(ηPln){φ(xn), yM }′g′ JM φ(ηPln){φ(xn), yM }′ JM φ(ηPkm){φ(xm), yM }′ M J φ(ηPkm){φ(xm), yM }′, since, g′ JM = g′ M J we obtain gkl = φ(ηPlm){φ(xm), yJ }′g′ JM φ(ηPkn){φ(xn), yM }′ = glk. As φ(A) is a subalgebra of A′, to show that Im(φ, ψ) is a subalgebra of K′, we need to show that (id φ(A), id g) is a morphism of Kähler-Poisson algebras, where g is module of derivations of Im(φ, ψ). Now, let α = αi{φ(xi), .}′ and β = βj{φ(xj), .}′ be arbitrary elements of g as usual, we have (1) ψ(aα) = aα = φ(a)ψ(α), for all α ∈ g and a ∈ φ(A). (2) φ(α(a)) = α(a) = ψ(α)(φ(a)), for all α ∈ g and a ∈ φ(A). (3) φ(g(α, β)) = g′(ψ(α), ψ(β)) = g′(α, β), for all α, β ∈ g and a ∈ φ(A). (4.2) φ(g(α, β)) − g′(α, β) =αi{φ(xi), φ(xk)}′gklβj{φ(xj ), φ(xl)}′ JM βj{φ(xj ), yM }′. − αi{φ(xi), yJ }′g′ Since yJ ∈ φ(A) and for each J, we can find yJ ∈ A such that φ(yJ ) = yJ . We can write gkl = φ(ηPkm){φ(xm), φ(yJ )}′g′ JM φ(ηPln){φ(xn), φ(yM )}′. We compute φ(g(α, β)) = φ(cid:16)ηαi{xi, xk}gkm{xm, xp}gpq{xq, yJ }(cid:17)g′ JM φ(cid:16)ηβj{xj, xl}glm{xm, xr}grs{xs, yM }(cid:17) = αi{φ(xi), yJ }′g′ JM βj{φ(xj), yM }′, by (4.2) we get φ(g(α, β)) = g′(α, β). (4) φ(A)fin = φ(Afin) ⊆ A′ fin, since, φ(A)fin is generated by {φ(x1), ..., φ(xm)} Therefore, (idφ(A), idg) is a morphism of Kähler-Poisson algebras, and Im(φ, ψ) is a subalgebra of K′. (cid:3) 14 AHMED AL-SHUJARY 5. Direct sums and tensor products of Kähler-Poisson algebras Let K = (A, g, {x1, ..., xm}) and K′ = (A′, g′, {y1, ..., ym′}) be Kähler-Poisson algebras. We have seen in Section 4, when K is Kähler-Poisson subalgebra of K′. In this section, we are interested in defining operations of Kähler-Poisson algebras. In particular, we introduce direct sums and tensor products of Kähler-Poisson algebras and study properties of these operations. We prove that, direct sums and tensor products of two Kähler-Poisson algebras are Kähler- Poisson algebras and we show that K and K′ are subalgebras of the direct sum K ⊕ K′. Let us first recall some basic facts of direct sums and tensor products of Poisson algebras. Firstly, we recall that the tensor product A ⊗ A′ of two Poisson algebras A and A′ is a Poisson algebra with Poisson products (5.1) {a1 ⊗ a2, b1 ⊗ b2} = {a1, b1} ⊗ a2b2 + a1b1 ⊗ {a2, b2}, for a1, b1 ∈ A and a2, b2 ∈ A′ (see [6]). Secondly, the direct sum of two Poisson algebras A ⊕ A′ is a Poisson algebra with Poisson product. (5.2) {(a1, a2), (b1, b2)} = ({a1, b1}, {a2, b2}) for a1, b1 ∈ A and a2, b2 ∈ A′. In next proposition, we see that direct sums of Kähler-Poisson algebras are Kähler-Poisson algebra. Proposition 5.1. Let K = (A, g, {x1, ..., xm}) and K′ = (A′, g′, {y1, ..., ym′}) be Kähler- Poisson algebras, and set K ⊕ K′ = (A ⊕ A′, g, {z1, ..., zm+m′ }) where zI =((xI , 0) (0, yI−m) if I ∈ {1, ..., m} if I ∈ {m + 1, ..., m + m′}, and (gIJ , 0) (0, g′ (0, 0) I−m,J −m) if I, J ∈ {1, ..., m} if I, J ∈ {m + 1, ..., m + m′} otherwise. Then K ⊕ K′ is a Kähler-Poisson algebra. Proof. Since K = (A, g, {x1, ..., xm}) and K′ = (A′, g′, {y1, ..., ym′}) are Kähler-Poisson alge- bras then there exists η ∈ A and η′ ∈ A′ such that m η{a1, xi}gij{xj, xk}gkl{xl, b1} = −{a1, b1} η′{a2, yα}′g′ αβ{yβ, yγ}′g′ γδ{yδ, b2}′ = −{a2, b2}′. where a1, b1 ∈ A and a2, b2 ∈ A′. We would like to show that K⊕K′ satisfies the Kähler-Poisson condition (2.1); that is m+m′ (5.5) (η, η′){(a1, a2), zI }gIJ{zJ , zK}gKL{zL, (b1, b2)} = −{(a1, a2), (b1, b2)} XI,J,K,L for a1, b1 ∈ A and a2, b2 ∈ A′. gIJ =  Xi,j,k,l=1 Xα,β,γ,δ=1 m′ (5.3) (5.4) Starting from the left hand side m+m′ 15 (η,η′){(a1, a2), zI }gIJ{zJ , zK}gKL{zL, (b1, b2)} XI,J,K,L (η, η′){(a1, a2), zi}giJ {zJ , zK}gKL{zL, (b1, b2)} (η, η′){(a1, a2), zα+m}gα+m,J {zJ , zK}gKL{zL, (b1, b2)} (η, η′){(a1, a2), zi}gij{zj, zK}gKL{zL, (b1, b2)} = + = + m+m′ m m′ m+m′ m+m′ XJ,K,L=1 XJ,K,L=1 XK,L=1 XK,L=1 Xi=1 Xα=1 Xi,j=1 Xα,β=1 m+m′ m′ m (η, η′){(a1, a2), zα+m}gα+m,β+m{zβ+m, zK}gKL{zL, (b1, b2)}, since giK = 0 when K > m, and gm+α,K = 0 when K ≤ m. Moreover, since {zi, zK} = 0 and {zα+m, zL} = 0 if 1 ≤ i ≤ m, 1 ≤ α ≤ m′, K > m and L ≤ m, then (η,η′){(a1, a2), zI }gIJ{zJ , zK}gKL{zL, (b1, b2)} (η, η′){(a1, a2), zi}gij{zj, zk}gkl{zl, (b1, b2)} (η, η′){(a1, a2), zα+m}gα+m,β+m{zβ+m, zγ+m}gγ+m,δ+m{zδ+m, (b1, b2)}. m+m′ XI,J,K,L = + and m+m′ XI,J,K,L m Xi,j,k,l=1 Xα,β,γ,δ=1 m′ = + = + = m m m′ Xi,j,k,l=1 Xα,β,γ,δ=1 Xi,j,k,l=1 Xα,β,γ,δ=1 Xi,j,k,l=1 m′ m (η,η′){(a1, a2), zI}gIJ {zJ , zK}gKL{zL, (b1, b2)} (η, η′){(a1, a2), (xi, 0)}(gij, 0){(xj, 0), (xk, 0)}(gkl, 0){(xl, 0), (b1, b2)} (η, η′){(a1, a2), (0, yα)}(0, g′ αβ{(0, yβ), (0, yγ)}(0, g′ γδ){(0, yδ), (b1, b2)} (η, η′)({a1, xi}, 0)(gij, 0)({xj, xk}, 0)(gkl, 0)({xl, b1}, 0) (η, η′)(0, {a2, yα}′)(0, g′ αβ)(0, {yβ, yγ}′)(0, g′ γδ)(0, {yδ, b2}′) (η{a1, xi}gij{xj, xk}gkl{xl, b1}, 0) + (0, η′{a2, yα}′g′ αβ{yβ, yγ}′g′ γδ{yδ, b2}′) = (−{a1, b1}, 0) + (0, −{a2, b2}′) = −({a1, b1}, {a2, b2}′) = −{(a1, a2), (b1, b2)}, m′ Xα,β,γ,δ=1 16 AHMED AL-SHUJARY since K and K′ are Kähler-Poisson algebras. Therefore, K ⊕ K′ is a Kähler-Poisson algebra. (cid:3) Next, given two Kähler-Poisson algebras K and K′, we show that they are Kähler-Poisson subalgebras of the direct sum K ⊕ K′. Remark 5.2. The module of inner derivations of A ⊕ A′ can be written as: g ⊕ g ′ = {(α, β) : α ∈ g and β ∈ g ′} since given α ∈ g ⊕ g ′ α(c, d) = m+m′ XI=1 with (aI , bI){zI , (c, d)} = (aI {xI , c}, 0) + (0, bI{yI−m, d}) m m+m′ XI=1 XI=m+1 = (α(c), β(d)) = (α, β)(c, d), m m′ α = aI {xI , ·} and β = bI+m{yI, ·}. XI=1 XI=1 Proposition 5.3. Let K = (A, g, {x1, ..., xm}) and K′ = (A′, g′, {y1, ..., ym′}) be Kähler- Poisson algebras, then K and K′ are Kähler-Poisson subalgebras of K ⊕ K′. To prove this proposition for K and K′, we first show the following results: Lemma 5.4. Let K = (A, g, {x1, ..., xm}) and K′ = (A′, g′, {y1, ..., ym′}) be Kähler-Poisson algebras, and let A = {(a, 0) : a ∈ A}. Then K = ( A, g, {(x1, 0), ..., (xm, 0)}) is a Kähler- Poisson subalgebra of K ⊕ K′, where gij =(cid:0)(gij , 0)(cid:1). Proof. Firstly, A is a Poisson subalgebra of A⊕A′. Denote the derivation module g = {(α, 0) ∈ ′. Secondly, we show that (id A, idg) is a morphism of g ⊕ g Kähler-Poisson algebras. Let φ = id A and ψ = idg, then by Definition 3.2 we get ′ : α ∈ g} as a submodule of g ⊕ g (1) ψ((a, 0)(α, 0)) = ψ(aα, 0) = (aα, 0) and φ((a, 0))ψ((α, 0)) = (a, 0)(α, 0) = (aα, 0). (2) φ(α(a, 0)) = φ(aα, 0) = (aα, 0) and ψ(α, 0)(φ(a, 0)) = (α, 0)(a, 0) = (aα, 0). (3) We see that φ(g((α, 0), (β, 0))) = g(ψ(α, 0), ψ(β, 0)). φ(g((α, 0), (β, 0))) = = m m Xi,j=1 Xi,j=1 ((α, 0)(xi, 0)(gij, 0)(β, 0)(xj , 0)) (α(xi)gijβ(xj ), 0), and g(ψ(α, 0), ψ(β, 0)) = g((α, 0), (β, 0)) = (α, 0)(xI , 0)gIJ (β, 0)(xJ , 0). Since gIJ is a diagonal block matrix, we get m g(ψ(α, 0), ψ(β, 0)) = + XI,J=1 XI,J=m+1 m+m′ (α, 0)(xI , 0)(gIJ , 0)(β, 0)(xJ , 0) (α, 0)(0, yI)(0, g′ I−m,J −m)(β, 0)(0, yJ ) since (α, 0)(0, yJ ) = 0, we get g(ψ(α, 0), ψ(β, 0)) = m Xi,j=1 (α(xi), 0)(gij, 0)(β(xj ), 0) = (α(xi)gijβ(xj ), 0). m Xi,j=1 (4) By construction, we have φ( Afin) ⊆ (A ⊕ A′)fin. Therefore, K is a Kähler-Poisson subalgebra of K ⊕ K′. 17 (cid:3) Lemma 5.5. Let K = (A, g, {x1, ..., xm}) and K = ( A, g, {(x1, 0), ..., (xm, 0)}) be Kähler- Poisson algebras as in Lemma 5.4, then K is isomorphic to K. Proof. We see that there is a morphism φ : A → A and ψ : g → g defined by φ(c) = (c, 0) (φ is a Poisson algebra isomorphism, see Remark 3.6), for all c ∈ A and let α = ai{xi, .} we define ψ(α) = (ai, 0){(xi, 0), .} and ψ(β) = (bi, 0){(xi, 0), .} for all α, β ∈ g, we get (1) φ(c)ψ(α) = (c, 0)(ai, 0){(xi, 0), .} = (cai, 0){(xi, 0), .} = ψ(cα). (2) ψ(α)(φ(a)) = (ai, 0){(xi, 0), (c, 0)} = (ai, 0)({xi, c}, 0) = (ai{xi, c}, 0) = φ(ai{xi, c}) = φ(α(c)). (3) We see that φ(g(α, β)) = g(ψ(α), ψ(β)) since φ(g(α, β)) = m Xi,j=1 (α(xi)gijβ(xj ), 0), and g(ψ(α), ψ(β)) = g((al, 0){(xl, 0), .}, (bk, 0){(xk, 0), .}) = (al, 0){(xl, 0), (xi, 0)}gij(bk, 0){(xk, 0), (xj, 0)} = (al, 0)({xl, xi}, 0)(gij, 0)(bk, 0)({xk, xj}, 0) = (al{xl, xi}, 0)(gij, 0)(bk{xk, xj}, 0) = (α(xi), 0)(gij, 0)(β(xj ), 0) = (α(xi)gij β(xj ), 0). (4) By construction, we have φ(Afin) = Afin. Therefore, K isomorphic to K. (cid:3) Proof of Proposition 5.3. To prove Proposition 5.3 we need the above results. Lemma 5.4 says that K = ( A, g, {(x1, 0), ..., (xm, 0)}) is a Kähler-Poisson subalgebra of K ⊕ K′ and Lemma 5.5 says that K is isomorphic to K. Therefore, we can consider K to be a Kähler-Poisson subalgebra of K ⊕ K′. (cid:3) Observe that, in the same way, K′ is a Kähler-Poisson subalgebra of K ⊕ K′. The next proposi- tion shows that the tensor product of Kähler-Poisson algebras is Kähler-Poisson algebra under certain conditions. Proposition 5.6. Let K = (A, g, {x1, ..., xm}) and K′ = (A′, g′, {y1, ..., ym′}) with η{a1, xi}gij{xj, xk}gkl{xl, b1} = −{a1, b1} and m Pi,j,k,l=1 Pα,β,γ,δ=1 m′ η′{a2, yα}′g′ αβ{yβ, yγ}′g′ γδ{yδ, b2}′ = −{a2, b2}′ 18 AHMED AL-SHUJARY be Kähler-Poisson algebras, and assume that there exist ρ ∈ A and ρ′ ∈ A′ such that ρ2 = η and ρ′2 = η′. Set K ⊗ K′ = (A ⊗ A′, g, {z1, ..., zm+m′ }) where zI =(xI ⊗ 1 1 ⊗ yI−m if I ∈ {1, ..., m} if I ∈ {m + 1, ..., m + m′}, and ρgIJ ⊗ 1 1 ⊗ ρ′g′ 0 I−m,J −m if I, J ∈ {1, ..., m} if I, J ∈ {m + 1, ..., m + m′} otherwise. gIJ =  Then K ⊗ K′ is a Kähler-Poisson algebra. Proof. Let us show that K ⊗ K′ satisfies the Kähler-Poisson condition: m+m′ (5.6) {a1 ⊗ a2, zI}gIJ{zJ , zK}gKL{zL, b1 ⊗ b2} = −{a1 ⊗ a2, b1 ⊗ b2} XI,J,K,L for a1, b1 ∈ A and a2, b2 ∈ A′. Starting from the left hand side m+m′ {a1 ⊗ a2, zI}gIJ {zJ , zK}gKL{zL, b1 ⊗ b2} = m+m′ m′ m+m′ XI,J,K,L XJ,K,L=1 XJ,K,L=1 XK,L=1 XK,L=1 Xi=1 Xα=1 Xi,j=1 Xα,β=1 m+m′ m+m′ m m′ + = + = + m Xi,j,k,l=1 Xα,β,γ,δ=1 m′ m {a1 ⊗ a2, zi}giJ {zJ , zK}gKL{zL, b1 ⊗ b2} {a1 ⊗ a2, zα+m}gα+m,J{zJ , zK}gKL{zL, b1 ⊗ b2} {a1 ⊗ a2, zi}gij{zj, zK}gKL{zL, b1 ⊗ b2} {a1 ⊗ a2, zα+m}gα+m,β+m{zβ+m, zK}gKL{zL, b1 ⊗ b2} since giK = 0 when K > m, and gm+α,K = 0 when K ≤ m. Moreover, since {zi, zK} = {xi ⊗ 1, 1 ⊗ yK−m} = 0 and {zα+m, zL} = {1 ⊗ yα, xL ⊗ 1} = 0 if 1 ≤ i ≤ m, 1 ≤ α ≤ m′, K > m and L ≤ m, then m+m′ {a1 ⊗ a2, zI }gIJ{zJ , zK}gKL{zL, b1 ⊗ b2} XI,J,K,L {a1 ⊗ a2, zi}gij{zj, zk}gkl{zl, b1 ⊗ b2} {a1 ⊗ a2, zα+m}gα+m,β+m{zβ+m, zγ+m}gγ+m,δ+m{zδ+m, b1 ⊗ b2}, and m+m′ 19 {a1 ⊗ a2, zI}gIJ {zJ , zK}gKL{zL, b1 ⊗ b2} XI,J,K,L {a1 ⊗ a2, xi ⊗ 1}ρgij ⊗ 1{xj ⊗ 1, xk ⊗ 1}ρgkl ⊗ 1{xl ⊗ 1, b1 ⊗ b2} {a1 ⊗ a2, 1 ⊗ yα}1 ⊗ ρ′g′ αβ{1 ⊗ yβ, 1 ⊗ yγ}1 ⊗ ρ′g′ γδ{1 ⊗ yδ, b1 ⊗ b2} ({a1, xi} ⊗ a2)(ρgij ⊗ 1)({xj, xk} ⊗ 1)(ρgkl ⊗ 1)({xl, b1} ⊗ b2) = + = + = + = + m m m′ m′ Xi,j,k,l=1 Xα,β,γ,δ=1 Xi,j,k,l=1 Xα,β,γ,δ=1 Xi,j,k,l=1 Xα,β,γ,δ=1 Xi,j,k,l=1 Xα,β,γ,δ=1 m′ m′ m m (a1 ⊗ {a2, yα}′)(1 ⊗ ρ′g′ αβ)(1 ⊗ {yβ, yγ}′)(1 ⊗ ρ′g′ γδ)(b1 ⊗ {yδ, b2}′) ρ2{a1, xi}gij{xj, xk}gkl{xl, b1} ⊗ a2b2 a1b1 ⊗ ρ′2{a2, yα}′g′ αβ{yβ, yγ}′g′ γδ{yδ, b2}′ (η{a1, xi}gij{xj, xk}gkl{xl, b1} ⊗ a2b2) (a1b1 ⊗ η′{a2, yα}′g′ αβ{yβ, yγ}′g′ γδ{yδ, b2}′), since ρ2 = η and ρ′2 = η′. Finally, since K and K′ are Kähler-Poisson algebras we get m+m′ {a1 ⊗ a2, zI }gIJ{zJ , zK}gKL{zL, b1 ⊗ b2} = −{a1, b1} ⊗ a2b2 − a1b1 ⊗ {a2, b2}′ XI,J,K,L = −{a1 ⊗ a2, b1 ⊗ b2}. Therefore, K ⊗ K′ is a Kähler-Poisson algebra. (cid:3) 6. Summary In this paper, we have introduced the concept of morphism of Kähler-Poisson algebras. We have recalled a few results from [3], in order to motivate and understand the concept of morphism of Kähler-Poisson algebras. We have studied properties of isomorphisms for Kähler-Poisson algebras and we illustrates with examples when two Kähler-Poisson algebras are isomorphic. We have used the concept of morphism to define subalgebras of Kähler-Poisson algebras and we have presented examples when K is a Kähler-Poisson subalgebra of K′, where A is a proper Poisson subalgebra of a finitely generated algebra A′. Finally, in Section 5, we have introduced direct sums and tensor products of Kähler-Poisson algebras and properties of these operations. 20 AHMED AL-SHUJARY There are many open questions that one would like to investigate in future work. For instance, is there a natural way to study the moduli spaces of Poisson algebras; i.e. how many (non-isomorphic) Kähler-Poisson structures does there exist on a given Poisson algebra? Acknowledgements I would like to thank my supervisor J. Arnlind for fruitful discussions and helpful comments. I would also like to thank my co-supervisor M. Izquierdo for ideas and discussions. The results in Section 3 are a part of my licentiate thesis [4]. References [1] J. Arnlind and G. Huisken. Pseudo-Riemannian geometry in terms of multi-linear brack- ets. Lett. Math. Phys., 104(12):1507-1521, 2014. [2] J. Arnlind, J. Hoppe, and G. Huisken. Multi-linear formulation of differential geometry and matrix regu- larizations. J. Differential Geom., 91(1):1-39, 2012. [3] J. Arnlind. A. Al-Shujary. Kähler-Poisson Algebras. J. Geometry and Physics., 136:156-172, 2019. [4] A. Al-Shujary. Kähler-Poisson Algebras. Linköping Studies in Science and Technology. No.1813, 2018. [5] J.-L. Brylinski. A differential complex for Poisson manifolds. J. Differential Geom., 28(1):93-114, 1988. [6] K. Christian. Quantum groups. Graduate texts in mathematics; 155: 1995. [7] S. Helgason. Differential geometry, Lie groups, and symmetric spaces, volume 34 of Graduate Studies in Mathematics. American Mathematical Society, Providence, RI, 2001. [8] J. Huebschmann. Poisson cohomology and quantization. J. Reine Angew. Math., 408:57- 113, 1990. [9] J. Huebschmann. Extensions of Lie-Rinehart algebras and the Chern-Weil construction. In Higher homo- topy structures in topology and mathematical physics, volume 227 of Contemp. Math., Amer. Math. Soc., Providence, RI, pages 145-176, 1999. [10] C.G.J. Jacobi, Vorlesungen über Dynamik, gehalten an der Universität zu Königsberg im Wintersemester 1842-1843 und nach einem von C.W. Borchardt ausgearbeiteten Hefte, herausgegeben von A. Clebsch, zweite revidirte Ausgabe,l884 (Chelsea, New York, 1969). [11] M. Kontsevich. Deformation quantization of Poisson manifolds. Lett. Math. Phys., 66(3):157-216, 2003. [12] E. Kähler, Uber eine bemerkenswerte Hermitesche Metrik, Abh. Math. Sem. Hamburg Univ., 173-186, 1933. [13] J. Lee. Manifolds and Differential Geometry. Graduate studies in mathematics.v107: 2009. [14] A. Lichnerowicz. Les vari´et´es de Poisson et leurs alg´ebres de Lie associ´ees. J. Differential Geometry, 12(2):253-300, 1977. [15] S. Lie, Theorie der Transformationsgruppen, Zweiter Abschnitt, unter Mitwirkung von Prof. Dr. Friedrich Engel (Teubner, Leipzig, 1890). [16] S.-D. Poisson, Sur la variation des constantes arbitraires dans les questions de m´ecanique, J. Ecole Poly- technique 8 (15):266-344, 1809. [17] G. S. Rinehart. Differential forms on general commutative algebras. Trans. Amer. Math. Soc., 108:195-222, 1963. [18] A. Weinstein. The local structure of Poisson manifolds. J. Differential Geom., 18(3):523- 557, 1983. [19] R.O. Wells jr., Differential analysis on complex manifolds, Springer-Verlag, New York-Heidelberg- Berlin,1980. Dept. of Mathematics, Linköping Univeristy, 58183 Linköping, Sweden. E-mail address: [email protected]
1002.3944	1	1002	2010-02-21T01:01:58	Hom-Maltsev, Hom-alternative, and Hom-Jordan algebras	[ "math.RA", "math-ph", "math-ph" ]	Hom-Maltsev(-admissible) algebras are defined, and it is shown that Hom-alternative algebras are Hom-Maltsev-admissible. With a new definition of a Hom-Jordan algebra, it is shown that Hom-alternative algebras are Hom-Jordan-admissible. Hom-type generalizations of some well-known identities in alternative algebras, including the Moufang identities, are obtained.	math.RA	math	HOM-MALTSEV, HOM-ALTERNATIVE, AND HOM-JORDAN ALGEBRAS DONALD YAU Abstract. Hom-Maltsev(-admissible) algebras are defined, and it is shown that Hom-alternative algebras are Hom-Maltsev-admissible. With a new definition of a Hom-Jordan algebra, it is shown that Hom-alternative algebras are Hom-Jordan-admissible. Hom-type generalizations of some well- known identities in alternative algebras, including the Moufang identities, are obtained. 1. Introduction Maltsev algebras were introduced by Maltsev [19], who called these objects Moufang-Lie algebras. A Maltsev algebra is a non-associative algebra A with an anti-symmetric multiplication [−, −] that satisfies the Maltsev identity J(x, y, [x, z]) = [J(x, y, z), x] (1.0.1) for all x, y, z ∈ A, where J(x, y, z) = [[x, y], z]+[[z, x], y]+[[y, z], x] is the Jacobian. In particular, Lie algebras are examples of Maltsev algebras. Maltsev algebras play an important role in the geometry of smooth loops. Just as the tangent algebra of a Lie group is a Lie algebra, the tangent algebra of a locally analytic Moufang loop is a Maltsev algebra [13, 14, 19, 22, 25]. The reader is referred to [9, 21, 23] for discussions about the relationships between Maltsev algebras, exceptional Lie algebras, and physics. Closely related to Maltsev algebras are alternative algebras. An alternative algebra is an algebra whose associator is an alternating function. In particular, all associative algebras are alternative, but there are plenty of non-associative alternative algebras, such as the octonions. Roughly speaking, alternative algebras are related to Maltsev algebras as associative algebras are related to Lie algebras. Indeed, as Maltsev observed in [19], every alternative algebra A is Maltsev-admissible, i.e., the commutator algebra A− is a Maltsev algebra. There are many Maltsev-admissible algebras that are not alternative; see, e.g., [21]. The reader is referred to [29] for applications of alternative algebras to projective geometry, buildings, and algebraic groups. Instead of the commutator, the anti-commutator also gives rise to interesting structures. A Jordan algebra is a commutative algebra that satisfies the Jordan identity (x2y)x = x2(yx). (1.0.2) Starting with an alternative algebra A, it is known that the Jordan product x ∗ y = 1 2 (xy + yx) Date: September 3, 2018. 2000 Mathematics Subject Classification. 17A20, 17C50, 17D05, 17D10, 17D25. Key words and phrases. Hom-Maltsev algebras, Hom-Maltsev-admissible algebras, Hom-alternative algebras, Hom-Moufang identities, Hom-Jordan algebras, Hom-Jordan-admissible algebras. 1 2 DONALD YAU gives a Jordan algebra A+ = (A, ∗). In other words, alternative algebras are Jordan-admissible. The reader is referred to [9, 12, 23, 28] for discussions about the important roles of Jordan algebras in physics, especially quantum mechanics. The purpose of this paper is to study Hom-type generalizations of Maltsev(-admissible) algebras, alternative algebras, and Jordan(-admissible) algebras. The reader is referred to the survey arti- cle [16] for discussions about other Hom-type algebras and to [32]-[40] for Hom-type analogues of Novikov algebras, quantum groups, and the Yang-Baxter equations. Roughly speaking, a Hom-type generalization of a kind of algebras is defined by twisting the defining identities by a self-map, called the twisting map. When the twisting map is the identity map, one recovers the original kind of algebras. Below is a description of the rest of this paper. In section 2 we define Hom-Maltsev algebras and prove two construction results, Theorems 2.11 and 2.13. Hom-Maltsev algebras include Maltsev algebras and Hom-Lie algebras as examples. The- orem 2.11 says that the class of Hom-Maltsev algebras is closed under the process of taking derived Hom-algebras (Definition 2.9), in which the structure maps are suitably twisted by the twisting map. Theorem 2.13 says that a Maltsev algebra (A, [−, −]) can be twisted into a Hom-Maltsev algebra Aα = (A, [−, −]α = α ◦ [−, −], α) along any algebra self-map α of A. In Examples 2.14 and 2.15, we show that, using Theorem 2.13 with different algebra self-maps, it is possible to twist a Maltsev algebra into a non-Hom-Lie Hom-Maltsev algebra, a Hom-Lie algebra, or a Lie algebra. The Hom-type analogue of an alternative algebra is called a Hom-alternative algebra, in which the Hom-associator (2.5.1) is alternating. Hom-alternative algebras were introduced by Makhlouf in [15]. In section 3 we show that Hom-alternative algebras are Hom-Maltsev-admissible (The- orem 3.8). That is, the commutator Hom-algebra (Definition 3.5) of a Hom-alternative algebra is a Hom-Maltsev algebra, generalizing the fact that alternative algebras are Maltsev-admissible. Hom-Lie-admissible algebras [17] and Maltsev-admissible algebras are obvious examples of Hom- Maltsev-admissible algebras. The proof of the Hom-Maltsev-admissibility of Hom-alternative alge- bras involves the Hom-type analogues of certain identities that hold in alternative algebras and of the Bruck-Kleinfeld function (Definition 3.11). In Example 3.19, starting with the octonions, we construct (non-Hom-Lie) Hom-Maltsev algebras using Theorem 3.8. In section 4 we consider the class of Hom-Maltsev-admissible algebras. In Proposition 4.3 we give several characterizations of Hom-Maltsev-admissible algebras that are also Hom-flexible [17]. In Theorems 4.4 and 4.5 we prove construction results for Hom-flexible and Hom-Maltsev-admissible algebras. Hom-alternative algebras are Hom-flexible [15], so by Theorem 3.8 Hom-alternative al- gebras are both Hom-flexible and Hom-Maltsev-admissible. In Examples 4.6, 4.7, and 4.8, we construct Hom-flexible, Hom-Maltsev-admissible algebras that are not Hom-alternative, not Hom- Lie-admissible, and not Maltsev-admissible. In section 5 we study Hom-Jordan(-admissible) algebras, which are the Hom-type generalizations of Jordan(-admissible) algebras. The first definition of a Hom-Jordan algebra was given by Makhlouf in [15]. Hom-alternative algebras are not Hom-Jordan-admissible under that definition. We intro- duce a different definition of a Hom-Jordan algebra and show that Hom-alternative algebras are Hom-Jordan-admissible under this new definition (Theorem 5.6). In other words, the plus Hom- algebra (Definition 5.1) of any Hom-alternative algebra is a Hom-Jordan algebra, generalizing the Jordan-admissibility of alternative algebras. Construction results analogous to Theorems 2.11 and HOM-MALTSEV, HOM-ALTERNATIVE, AND HOM-JORDAN ALGEBRAS 3 2.13 are proved for Hom-Jordan(-admissible) algebras (Theorems 5.8 and 5.9). In Example 5.10, we construct (non-Jordan) Hom-Jordan algebras using the 27-dimensional exceptional simple Jordan algebra of 3 × 3 Hermitian octonionic matrices. In section 6 we provide further properties for Hom-alternative algebras. First we observe that a Hom-algebra is Hom-associative if and only if it is both Hom-alternative and Hom-Lie-admissible (Proposition 6.1). Then we show that the class of Hom-alternative algebras is closed under taking derived Hom-algebras (Proposition 6.2). In Propositions 6.3 and 6.5 we provide further properties of the Hom-Bruck-Kleinfeld function in Hom-alternative algebras. It is well-known that the Moufang identities (6.7.4) hold in alternative algebras. In Theorem 6.8 we show that there are Hom-type generalizations of the Moufang identities in Hom-alternative algebras. 2. Hom-Maltsev algebras In this section we define Hom-Maltsev algebras and study their general properties. Other char- acterizations of the Hom-Maltsev identity are given (Proposition 2.8). We prove some construction results for Hom-Maltsev algebras (Theorems 2.11 and 2.13). Using Theorem 2.13, we demonstrate in Examples 2.14 and 2.15 that it is possible to twist a Maltsev algebra into a non-Hom-Lie Hom- Maltsev algebra or a Hom-Lie (or even Lie) algebra using different algebra morphisms. 2.1. Conventions. Throughout the rest of this paper, we work over a fixed field k of characteristic 0. Modules, tensor products, linearity, and Hom are all meant over k. If f : V → V is a linear self-map on a vector space V , then f n : V → V denotes the composition f ◦ · · · ◦ f of n copies of f , with f 0 = Id. For a map µ : V ⊗2 → V , we sometimes write µ(a, b) as ab for a, b ∈ V . If W is another vector space, then τ : V ⊗ W ∼= W ⊗ V denotes the twist isomorphism, τ (v ⊗ w) = w ⊗ v. More generally, we do not distinguish between a permutation θ on n letters and its induced linear isomorphism θ : V ⊗n → V ⊗n given by θ(v1 ⊗ · · · ⊗ vn) = vθ(1) ⊗ · · · ⊗ vθ(n). Let us give the definitions regarding Hom-algebras. Definition 2.2. By a Hom-algebra we mean a triple (A, µ, α) in which A is a k-module, µ : A⊗2 → A is a bilinear map (the multiplication), and α : A → A is a linear map (the twisting map) such that α ◦ µ = µ ◦ α⊗2 (multiplicativity). A Hom-algebra (A, µ, α) is usually denoted simply by A. A morphism f : A → B of Hom-algebras is a linear map f of the underlying k-modules such that f ◦ αA = αB ◦ f and µB ◦ f ⊗2 = f ◦ µA. Remark 2.3. If (A, µ) is a not-necessarily associative algebra in the usual sense, we also regard it as the Hom-algebra (A, µ, Id) with identity twisting map. This defines a fully faithful embedding from the category of algebras into the category of Hom-algebras. With this convention, the notion of a morphism between Hom-algebras with identity twisting maps reduces to the usual notion of an algebra morphism. Remark 2.4. The multiplicativity of the twisting map α is built into our definition of a Hom- algebra. Some authors (see, e.g., [15, 16, 17]) do not make this assumption. We chose to impose multiplicativity because many of our results depend on it and all of our concrete examples of Hom- Maltsev(-admissible), Hom-alternative, and Hom-Jordan(-admissible) algebras have this property. The algebraic structures studied in this paper are all defined using the Hom-versions of the associator and the Jacobian, which we now define. 4 DONALD YAU Definition 2.5. Let (A, µ, α) be a Hom-algebra. (1) The Hom-associator of A [17] is the trilinear map asA : A⊗3 → A defined as asA = µ ◦ (µ ⊗ α − α ⊗ µ). (2) The Hom-Jacobian of A [17] is the trilinear map JA : A⊗3 → A defined as JA = µ ◦ (µ ⊗ α) ◦ (Id + σ + σ2), where σ : A⊗3 → A⊗3 is the cyclic permutation σ(x ⊗ y ⊗ z) = z ⊗ x ⊗ y. (2.5.1) (2.5.2) If there is only one Hom-algebra under consideration, we will sometimes omit the subscript in the Hom-associator and the Hom-Jacobian Note that when (A, µ) is an algebra (with α = Id), its Hom-associator and Hom-Jacobian coincide with its usual associator and Jacobian, respectively. Since Hom-Maltsev algebras generalize Hom-Lie algebras (as we will see shortly), which in turn generalize Lie algebras, we use the bracket notation [−, −] to denote their multiplications. Definition 2.6. (1) A Hom-Lie algebra [10, 17] is a Hom-algebra (A, [−, −], α) such that [−, −] is anti-symmetric (i.e., [−, −] ◦ (Id + τ ) = 0) and that the Hom-Jacobi identity JA = 0 (2.6.1) is satisfied, where JA is the Hom-Jacobian of A (2.5.2). (2) A Hom-Maltsev algebra is a Hom-algebra (A, [−, −], α) such that [−, −] is anti-symmetric and that the Hom-Maltsev identity JA(α(x), α(y), [x, z]) = [JA(x, y, z), α2(x)] (2.6.2) is satisfied for all x, y, z ∈ A. Observe that when α = Id, the Hom-Jacobi identity reduces to the usual Jacobi identity [[x, y], z] + [[z, x], y] + [[y, z], x] = 0 for all x, y, z ∈ A. Likewise, when α = Id, by the anti-symmetry of [−, −], the Hom-Maltsev identity reduces to the Maltsev identity (1.0.1) or equivalently, [[x, y], [x, z]] = [[[x, y], z], x] + [[[y, z], x], x] + [[[z, x], x], y] (2.6.3) for all x, y, z ∈ A. Example 2.7. A Lie (resp., Maltsev [19]) algebra (A, [−, −]) is a Hom-Lie (resp., Hom-Maltsev) algebra with α = Id, since the Hom-Jacobi identity (2.6.1) (resp., the Hom-Maltsev identity (2.6.2)) reduces to the usual Jacobi (resp., Maltsev) identity when α = Id. Moreover, every Hom-Lie algebra is also a Hom-Maltsev algebra because the Hom-Jacobi identity JA = 0 clearly implies the Hom-Maltsev identity. (cid:3) Before we give more examples of Hom-Maltsev algebras, let us give some other characterizations of the Hom-Maltsev identity. Proposition 2.8. Let (A, [−, −], α) be a Hom-algebra with [−, −] anti-symmetric. Then the follow- ing statements are equivalent. (1) A is a Hom-Maltsev algebra, i.e., the Hom-Maltsev identity (2.6.2) holds. HOM-MALTSEV, HOM-ALTERNATIVE, AND HOM-JORDAN ALGEBRAS 5 (2) The condition J(α(w), α(y), [x, z]) + J(α(x), α(y), [w, z]) = [J(w, y, z), α2(x)] + [J(x, y, z), α2(w)] holds for all w, x, y, z ∈ A. (3) The condition α([[x, y], [x, z]]) = [[[x, y], α(z)], α2(x)] + [[[y, z], α(x)], α2(x)] + [[[z, x], α(x)], α2(y)] holds for all x, y, z ∈ A. (4) The condition α([[w, y], [x, z]]) + α([[x, y], [w, z]]) = [[[w, y], α(z)], α2(x)] + [[[x, y], α(z)], α2(w)] + [[[y, z], α(w)], α2(x)] + [[[y, z], α(x)], α2(w)] + [[[z, w], α(x)], α2(y)] + [[[z, x], α(w)], α2(y)] holds for all w, x, y, z ∈ A. (2.8.1) (2.8.2) (2.8.3) Proof. The equivalence between the Hom-Maltsev identity (2.6.2) and (2.8.1) follows from lineariza- tion: To get the latter, replace x with w + x in the former. Conversely, (2.8.1) yields (2.6.2) by setting w = x. Similarly, linearization implies the equivalence between (2.8.2) and (2.8.3). To prove the equivalence between the Hom-Maltsev identity and (2.8.2), observe that the left-hand side of the Hom-Maltsev identity (2.6.2) is: J(α(x), α(y), [x, z]) = [[α(x), α(y)], α([x, z])] + [[[x, z], α(x)], α2(y)] + [[α(y), [x, z]], α2(x)] = α([[x, y], [x, z]]) − [[[z, x], α(x)], α2(y)] + [[[z, x], α(y)], α2(x)]. In the last equality above, we used the multiplicativity of α and the anti-symmetry of [−, −]. Like- wise, the right-hand side of the Hom-Maltsev identity (2.6.2) is: [J(x, y, z), α2(x)] = [[[x, y], α(z)], α2(x)] + [[[z, x], α(y)], α2(x)] + [[[y, z], α(x)], α2(x)]. Since the summand [[[z, x], α(y)], α2(x)] appears on both sides of (2.6.2), the above calculation and a rearrangement of terms imply the equivalence between the Hom-Maltsev identity and (2.8.2). (cid:3) To state our next result, we need the following definition. Definition 2.9. Let (A, µ, α) be a Hom-algebra and n ≥ 0. Define the nth derived Hom-algebra of A by ). Note that A0 = A, A1 = (A, µ(1) = α ◦ µ, α2), and An+1 = (An)1. An = (A, µ(n) = α2n −1 ◦ µ, α2n The following elementary observations are used in the next result. Lemma 2.10. Let (A, µ, α) be a Hom-algebra. Then we have and for all n ≥ 0 JA ◦ α⊗3 = α ◦ JA JAn = α2(2n−1) ◦ JA (2.10.1) (2.10.2) 6 DONALD YAU Proof. The condition (2.10.1) holds because α⊗3 commutes with the cyclic permutation σ and α is multiplicative. For (2.10.2), observe that: µ(n) ◦ (µ(n) ⊗ α2n ) = α2n−1 ◦ µ ◦ ((α2n −1 ◦ µ) ⊗ α2n ) = α2n−1 ◦ µ ◦ (α2n−1)⊗2 ◦ (µ ⊗ α) = α2(2n−1) ◦ µ ◦ (µ ⊗ α), (2.10.3) where the last equality follows from the multiplicativity of α with respect to µ. Pre-composing (2.10.3) with the cyclic sum (Id + σ + σ2), we obtain (2.10.2). (cid:3) The following result shows that the category of Hom-Maltsev algebras is closed under taking derived Hom-algebras. Theorem 2.11. Let (A, [−, −], α) be a Hom-Maltsev algebra. Then the nth derived Hom-algebra is also a Hom-Maltsev algebra for each n ≥ 0. An = (A, [−, −](n) = α2n−1 ◦ [−, −], α2n ) Proof. Since A0 = A, A1 = (A, [−, −](1) = α ◦ [−, −], α2), and An+1 = (An)1, by an induction argument it suffices to prove the case n = 1. To show that A1 is a Hom-Maltsev algebra, first note that [−, −](1) is anti-symmetric because [−, −] is anti-symmetric and α is linear. Since α2 is multiplicative with respect to [−, −](1), it remains to show the Hom-Maltsev identity (2.6.2) for A1. For x, y, z ∈ A, we compute as follows: JA1 (α2(x), α2(y), [x, z](1)) = JA1 (α2(x), α2(y), α([x, z])) = α2 (cid:0)JA(α2(x), α2(y), α([x, z]))(cid:1) = α3 (JA(α(x), α(y), [x, z])) = α3 (cid:0)[JA(x, y, z), α2(x)](cid:1) = [α2(JA(x, y, z)), (α2)2(x)](1) = [JA1 (x, y, z), (α2)2(x)](1) (by (2.10.2)) (by (2.10.1)) (by (2.6.2) in A) (by multiplicativity of α) (by (2.10.2)). This establishes the Hom-Maltsev identity in A1 and finishes the proof. (cid:3) To state our next result, we need the following definition. Definition 2.12. Let (A, µ) be any algebra and α : A → A be an algebra morphism. Define the Hom-algebra induced by α as where µα = α ◦ µ. Aα = (A, µα, α), The next result shows that given a Maltsev algebra and an algebra morphism, the induced Hom- algebra is a Hom-Maltsev algebra. The Hom-Maltsev algebras constructed using this twisting result are generally not Maltsev algebras, as we will see in the examples later. Such a twisting result was first used by the author in [31] (Theorem 2.3) on G-associative algebras (where G is a subgroup of the symmetric group on three letters), which include associative, Lie, pre-Lie, and Lie-admissible algebras as examples. That result has since been employed and extended by various authors; see [2] (Theorem 2.7), [3] (Theorems 1.7 and 2.6), [7] (Section 2), [8] (Proposition 1), [15] (Theorems 2.1 and 3.5), [16], and [18]. HOM-MALTSEV, HOM-ALTERNATIVE, AND HOM-JORDAN ALGEBRAS 7 Theorem 2.13. Let (A, [−, −]) be a Maltsev algebra and α : A → A be an algebra morphism. Then the Hom-algebra Aα = (A, [−, −]α = α ◦ [−, −], α) induced by α is a Hom-Maltsev algebra. Proof. Since [−, −]α is clearly anti-symmetric, it remains to prove the Hom-Maltsev identity (2.6.2) for Aα. Here we regard A as the Hom-Maltsev algebra (A, [−, −], Id) with identity twisting map. For any algebra (A, [−, −]) (Maltsev or otherwise), by the multiplicativity of α with respect to [−, −], we have [−, −]α ◦ ([−, −]α ⊗ α) = α2 ◦ [−, −] ◦ ([−, −] ⊗ Id) and [−, −] ◦ ([−, −] ⊗ Id) ◦ α⊗3 = α ◦ [−, −] ◦ ([−, −] ⊗ Id). Pre-composing these identities with the cyclic sum (Id + σ + σ2) (2.5.2) (which commutes with α⊗3), we obtain and JAα = α2 ◦ JA JA ◦ α⊗3 = α ◦ JA. (2.13.1) (2.13.2) To prove the Hom-Maltsev identity for Aα, we compute as follows: JAα(α(x), α(y), [x, z]α) = α2 (JA(α(x), α(y), [α(x), α(z)])) = α2 ([JA(α(x), α(y), α(z)), α(x)]) = [α(JA(α(x), α(y), α(z)), α2(x)]α = [α2(JA(x, y, z)), α2(x)]α = [JAα(x, y, z), α2(x)]α (by (2.13.1)) (by (2.6.2) in A) (by multiplicativity of α) (by (2.13.2)) (by (2.13.1)). This shows that the Hom-Maltsev identity holds in Aα. (cid:3) We now discuss examples of Hom-Maltsev algebras that can be constructed using Theorem 2.13. Example 2.14. There is a four-dimensional non-Lie Maltsev algebra (A, [−, −]) [26] (Example 3.1) with basis {e1, e2, e3, e4} and multiplication table: [−, −] e1 e2 e3 e4 e1 0 e2 e3 −2e4 −e4 e2 e3 −e2 −e3 2e4 0 0 0 0 e4 e4 0 0 0 We want to apply Theorem 2.13 to this Maltsev algebra. Using suitable algebra morphisms, we can twist the Maltsev algebra A into non-Hom-Lie, non-Maltsev Hom-Maltsev algebras or (Hom-)Lie algebras. With a bit of computation, one can check that one class of algebra morphisms α1 : A → A is given by α1(e1) = e1 + a3e3 + a4e4, α1(e2) = b2e2 + b3e3 + a3b2e4, α1(e3) = ce3, α1(e4) = b2ce4, 8 DONALD YAU where a3, a4, b2, b3, and c are arbitrary scalars in k. By Theorem 2.13 there is a Hom-Maltsev algebra with multiplication table: Aα1 = (A, [−, −]α1 = α1 ◦ [−, −], α1) [−, −]α1 e1 e2 e3 e4 e1 0 α1(e2) e2 e3 −α1(e2) −ce3 2b2ce4 0 ce3 −2b2ce4 −b2ce4 0 0 0 e4 b2ce4 0 0 0 Note that Aα1 is in general not a Hom-Lie algebra, i.e., JAα1 6= 0. Indeed, we have Combining this with (2.13.1) we obtain JA(e1, e2, e3) = −6e4. JAα1 1(JA(e1, e2, e3)) (e1, e2, e3) = α2 = α2 = −6(b2c)2e4, 1(−6e4) which is not equal to 0 in general. Also, (A, [−, −]α1 ) is not a Maltsev algebra in general. Indeed, let J ′ denote the usual Jacobian of (A, [−, −]α1), i.e., J ′(x, y, z) = [[x, y]α, z]α + [[z, x]α, y]α + [[y, z]α, x]α, (2.14.1) where α = α1. To see that the Maltsev identity (1.0.1) does not hold in (A, [−, −]α1), it suffices to show that J ′(e1, e2, [e1, e3]α) 6= [J ′(e1, e2, e3), e1]α. (2.14.2) Indeed, we have J ′(e1, e2, e3) = [[e1, e2]α, e3]α + [[e3, e1]α, e2]α + [[e2, e3]α, e1]α = −2b2c(b2 + c + b2c)e4. This implies that, on the one hand, J ′(e1, e2, [e1, e3]α) = −cJ ′(e1, e2, e3) = 2b2c2(b2 + c + b2c)e4. On the other hand, we have [J ′(e1, e2, e3), e1]α = −2b2c(b2 + c + b2c)[e4, e1]α = 2(b2c)2(b2 + c + b2c)e4. This proves that (2.14.2) holds whenever b2 6= 1 and b2c(b2 + c + b2c) 6= 0. So (A, [−, −]α1) is not a Maltsev algebra in general. Using Theorem 2.13 it is also possible to twist the Maltsev algebra A into a (Hom-)Lie algebra. For example, consider the following class of algebra morphisms α2 : A → A: α2(e1) = −e1 + a2e2 + a3e3 + a4e4, α2(e2) = be4, α2(e3) = 0 = α2(e4), HOM-MALTSEV, HOM-ALTERNATIVE, AND HOM-JORDAN ALGEBRAS 9 where a2, a3, a4, and b are arbitrary scalars in k. By Theorem 2.13 there is a Hom-Maltsev algebra with multiplication table: Aα2 = (A, [−, −]α2 = α2 ◦ [−, −], α2) [−, −]α2 e1 e2 e3 e4 e2 e1 0 −be4 be4 0 0 0 0 0 e3 0 0 0 0 e4 0 0 0 0 From this multiplication table, it is easy to check that JAα2 algebra. Likewise, one can check that (A, [−, −]α2) is a Lie algebra. = 0, i.e., Aα2 is actually a Hom-Lie (cid:3) Example 2.15. There is a five-dimensional non-Lie Maltsev algebra (A, [−, −]) [26] (Example 3.4) with basis {e1, e2, e3, e4, e5} and multiplication table: [−, −] e1 e2 e3 e4 e5 e1 0 0 0 −e2 0 e2 0 0 0 0 −e3 e3 0 0 0 0 0 e4 e2 0 0 0 0 e5 0 e3 0 0 0 Let us classify all the algebra morphisms α : A → A. From the multiplication table of A, it follows that α is determined by its values at e1, e4, and e5. With a bit of computation, one can show that α : A → A is an algebra morphism if and only if it has the form α(e1) = a1e1 + a2e2 + a3e3 + a4e4 + a5e5, α(e2) = (a1b4 − a4b1)e2 + (a2b5 − a5b2)e3, α(e3) = (a1b4 − a4b1)c5e3, α(e4) = b1e1 + b2e2 + b3e3 + b4e4 + b5e5, α(e5) = c1e1 + c2e2 + c3e3 + c4e4 + c5e5 with ai, bj, ck ∈ k, such that a5(a4b1 − a1b4) = 0 = b5(a4b1 − a1b4), a1c4 = a4c1, a2c5 = a5c2, b1c4 = b4c1, b2c5 = b5c2. For each such algebra morphism α : A → A, by Theorem 2.13 there is a Hom-Maltsev algebra whose multiplication table is: Aα = (A, [−, −]α = α ◦ [−, −], α) [−, −]α e1 e2 e3 e4 e5 e1 0 0 0 −α(e2) e2 0 0 0 0 0 −α(e3) e4 e3 0 α(e2) 0 0 0 0 0 0 0 0 e5 0 α(e3) 0 0 0 10 DONALD YAU The Hom-Maltsev algebra Aα is in general not Hom-Lie, since JAα (e1, e4, e5) = α2(JA(e1, e4, e5)) (by (2.13.1)) = α2(e3) = (a1b4 − a4b1)2c2 5e3, which is not equal to 0 whenever (a1b4 − a4b1)c5 6= 0. In strong contrast with Example 2.14, we claim that (A, [−, −]α) is always a Maltsev algebra, regardless of what algebra morphism α : A → A we choose. Since [−, −]α is anti-symmetric, we only need to see that the Maltsev identity (1.0.1) holds. Indeed, the images of [−, −] and [−, −]α are both contained in span{e2, e3}, and α(e3) lies in span{e3}. It follows that [[x, y]α, z]α ⊆ span{e3}, which implies that for all w, x, y, z ∈ A. Thus, if J ′ denotes the usual Jacobian of (A, [−, −]α) as in (2.14.1), then [[[x, y]α, z]α, w]α = 0 (2.15.1) [J ′(x, y, z), w]α = 0, (2.15.2) which is in particular true when w = x. On the other hand, we have J ′(x, y, [x, z]α) = [[x, y]α, [x, z]α]α + [[[x, z]α, x]α, y]α + [[y, [x, z]α]α, x]α = α2([[x, y], [x, z]]) (by (2.15.1) and multiplicativity of α). Since both [x, y] and [x, z] lie in span{e2, e3}, it follows from the multiplication table of A that [[x, y], [x, z]] = 0. Thus, we have J ′(x, y, [x, z]α) = 0 (2.15.3) for all x, y, z ∈ A. It follows from (2.15.2) and (2.15.3) that the Maltsev identity (1.0.1) holds in (A, [−, −]α). Moreover, certain choices of algebra morphisms β : A → A make Aβ into a Hom-Lie algebra and (A, [−, −]β) into a Lie algebra. For example, consider the algebra morphism β on A given by β(e1) = ae1, β(e2) = abe2, β(e4) = be4, β(e3) = 0 = β(e5), where a, b ∈ k are arbitrary scalars. The only non-zero brackets in Aβ involving the basis elements are [e1, e4]β = abe2 = −[e4, e1]β. This implies that JAβ and J ′ (the usual Jacobian in (A, [−, −]β)) are both equal to 0, so Aβ is a Hom-Lie algebra and (A, [−, −]β) is a Lie algebra. (cid:3) 3. Hom-alternative algebras are Hom-Maltsev-admissible The main purpose of this section is to show that every Hom-alternative algebra [15] gives rise to a Hom-Maltsev algebra via the commutator bracket (Theorem 3.8). This means that Hom-alternative algebras are all Hom-Maltsev-admissible algebras, generalizing the well-known fact that alternative algebras are Maltsev-admissible. More properties of Hom-alternative algebras are considered in sections 5 and 6. At the end of this section, we consider an eight-dimensional, non-Hom-Lie Hom- Maltsev algebra arising from the octonions (Example 3.19). Let us begin with some relevant definitions. HOM-MALTSEV, HOM-ALTERNATIVE, AND HOM-JORDAN ALGEBRAS 11 Definition 3.1. Let V be a k-module and f : V ⊗n → V be an n-linear map for some n ≥ 2. We say that f is alternating if for each permutation π on n letters, where ǫ(π) ∈ {±1} is the signature of π. f = ǫ(π)f ◦ π Since we are working over a field k of characteristic 0, the following characterizations of alternating maps are well-known facts in basic linear algebra and group theory. We, therefore, omit the proof. We will use the following Lemma without further comment. Lemma 3.2. Let V be a k-module and f : V ⊗n → V be an n-linear map. Then the following statements are equivalent: (1) f is alternating. (2) f (x1, . . . , xn) = 0 whenever xi = xj for some i 6= j. (3) f = −f ◦ ι for each transposition ι on n letters. (4) f = (−1)n−1f ◦ ξ and f = −f ◦ η, where ξ is the cyclic permutation (12 · · · n) and η is the adjacent transposition (n − 1, n). Definition 3.3. Let (A, µ, α) be a Hom-algebra (Definition 2.2). Then A is called a: (1) Hom-associative algebra [17] if asA = 0, where asA is the Hom-associator (2.5.1); (2) Hom-alternative algebra [15] if asA is alternating; (3) Hom-flexible algebra [17] if asA(x, y, x) = 0 for all x, y ∈ A. It follows from the above definitions that a Hom-associative algebra is also a Hom-alternative algebra and that a Hom-alternative algebra is also a Hom-flexible algebra. Also, when α = Id in Definition 3.3, we recover the usual notions of associative, alternative, and flexible algebras, respectively. Remark 3.4. In [15] a Hom-alternative algebra was actually defined as a Hom-algebra (A, µ, α) that satisfies both and asA(x, x, y) = 0 (left Hom-alternativity) asA(x, y, y) = 0 (right Hom-alternativity) (3.4.1) (3.4.2) for all x, y ∈ A. It is shown in [15] that this definition is equivalent to the one in Definition 3.3. Indeed, if asA is alternating, then it is clearly also left and right Hom-alternative. Conversely, left (right) Hom-alternativity is equivalent to asA being alternating in the first (last) two variables. Since the transpositions (1 2) and (2 3) generate the symmetric group on three letters, one infers that left and right Hom-alternativity together imply that asA is alternating. To state the main result of this section, we need the following definition. Recall that τ : V ⊗ W ∼= W ⊗ V denotes the twist isomorphism, τ (v ⊗ w) = w ⊗ v. Definition 3.5. Let (A, µ, α) be a Hom-algebra. Define its commutator Hom-algebra as the Hom-algebra A− = (A, [−, −] = µ ◦ (Id − τ ), α). The multiplication [−, −] = µ ◦ (Id − τ ) is called the commutator bracket of µ. We call a Hom- algebra A Hom-Maltsev-admissible (resp. Hom-Lie-admissible [17]) if A− is a Hom-Maltsev (resp. Hom-Lie) algebra (Definition 2.6). 12 DONALD YAU Example 3.6. Since Hom-Lie algebras are all Hom-Maltsev algebras (Example 2.7), every Hom- Lie-admissible algebra is also Hom-Maltsev-admissible. In particular, since every G-Hom-associative algebra (e.g., Hom-associative, Hom-Lie, or Hom-pre-Lie algebra) is Hom-Lie-admissible [17] (Propo- sition 2.7), it is also Hom-Maltsev-admissible. In section 4, we will give examples of Hom-Maltsev- admissible algebras that are not Hom-Lie-admissible. (cid:3) Example 3.7. A Maltsev-admissible algebra is defined as an algebra (A, µ) for which the commutator algebra A− = (A, [−, −] = µ ◦ (Id − τ )) is a Maltsev algebra, i.e., A− satisfies the Maltsev identity (1.0.1) (or equivalently (2.6.3)). Identifying algebras as Hom-algebras with identity twisting maps (Remark 2.3), a Maltsev-admissible algebra is equivalent to a Hom-Maltsev-admissible algebra with α = Id. (cid:3) It is proved in [17] that, given a Hom-associative algebra A, its commutator Hom-algebra A− is a Hom-Lie algebra. Also, the commutator algebra of any alternative algebra is a Maltsev algebra. The following main result of this section generalizes both of these facts. It gives us a large class of Hom-Maltsev-admissible algebras that are in general not Hom-Lie-admissible. Theorem 3.8. Every Hom-alternative algebra is Hom-Maltsev-admissible. In particular, Hom-alternative algebras are all Hom-flexible, Hom-Maltsev-admissible algebras. Examples of Hom-flexible, Hom-Maltsev-admissible algebras that are not Hom-alternative are con- sidered in section 4. The proof of Theorem 3.8 depends on the Hom-type analogues of some identities in alternative algebras, most of which are from [5]. We will first establish some identities about the Hom-associator and the Hom-Jacobian. Then we will go back to the proof of Theorem 3.8. In what follows, we often write µ(a, b) as ab and omit the subscript in the Hom-associator asA (2.5.1) when there is no danger of confusion. The following result is a sort of cocycle condition for the Hom-associator of a Hom-alternative algebra. Lemma 3.9. Let (A, µ, α) be a Hom-alternative algebra. Then the identity as(wx, α(y), α(z)) − as(xy, α(z), α(w)) + as(yz, α(w), α(x)) = α2(w)as(x, y, z) + as(w, x, y)α2(z) (3.9.1) holds for all w, x, y, z ∈ A. Proof. First we claim that for any Hom-algebra (A, µ, α) (Hom-alternative or otherwise), we have as(wx, α(y), α(z)) − as(α(w), xy, α(z)) + as(α(w), α(x), yz) = α2(w)as(x, y, z) + as(w, x, y)α2(z) (3.9.2) for all w, x, y, z ∈ A. Indeed, starting from the left-hand side of (3.9.2), we have: as(wx, α(y), α(z)) − as(α(w), xy, α(z)) + as(α(w), α(x), yz) = ((wx)α(y))α2(z) − α(wx)(α(y)α(z)) − (α(w)(xy))α2(z) + α2(w)((xy)α(z)) + (α(w)α(x))α(yz) − α2(w)(α(x)(yz)) = {(wx)α(y) − α(w)(xy)} α2(z) + α2(w) {(xy)α(z) − α(x)(yz)} − α(wx)α(yz) + α(wx)α(yz) = α2(w)as(x, y, z) + as(w, x, y)α2(z). HOM-MALTSEV, HOM-ALTERNATIVE, AND HOM-JORDAN ALGEBRAS 13 In the second equality above, we used the multiplicativity of α twice. We have established (3.9.2). Now for a Hom-alternative algebra A, its Hom-associator as is alternating, so (3.9.2) implies (3.9.1). (cid:3) Remark 3.10. Note that when α = Id, the condition (3.9.2) says that the (Hom-)associator as ∈ Hom(A⊗3, A) is a Hochschild 3-cocycle. In a Hom-alternative algebra, the Hom-associator is an alternating map on three variables. We now build a map on four variables using the Hom-associator that, as we will prove shortly, is alternating in a Hom-alternative algebra. Definition 3.11. Let (A, µ, α) be a Hom-algebra. Define the Hom-Bruck-Kleinfeld function f : A⊗4 → A as the multi-linear map f (w, x, y, z) = as(wx, α(y), α(z)) − as(x, y, z)α2(w) − α2(x)as(w, y, z) (3.11.1) for w, x, y, z ∈ A. Define another multi-linear map F : A⊗4 → A as where [−, −] = µ ◦ (Id − τ ) is the commutator bracket of µ and ξ is the cyclic permutation F = [−, −] ◦ (cid:0)α2 ⊗ as(cid:1) ◦ (Id − ξ + ξ2 − ξ3), (3.11.2) ξ(w ⊗ x ⊗ y ⊗ z) = z ⊗ w ⊗ x ⊗ y. In terms of elements, the map F is given by F (w, x, y, z) = [α2(w), as(x, y, z)] − [α2(z), as(w, x, y)] + [α2(y), as(z, w, x)] − [α2(x), as(y, z, w)]. (3.11.3) The Hom-Bruck-Kleinfeld function f is the Hom-type analogue of a map studied by Bruck and Kleinfeld ([5] (2.7)). It is closely related to the map F , as we now show. Lemma 3.12. In a Hom-alternative algebra (A, µ, α), we have F = f ◦ (Id − ρ + ρ2), where ρ = ξ3 is the cyclic permutation ρ(w ⊗ x ⊗ y ⊗ z) = x ⊗ y ⊗ z ⊗ w. Proof. By Lemma 3.9 and (3.11.1), we have α2(w)as(x, y, z) + as(w, x, y)α2(z) = as(wx, α(y), α(z)) − as(xy, α(z), α(w)) + as(yz, α(w), α(x)) = f (w, x, y, z) + as(x, y, z)α2(w) + α2(x)as(w, y, z) − f (x, y, z, w) − as(y, z, w)α2(x) − α2(y)as(x, z, w) + f (y, z, w, x) + as(z, w, x)α2(y) + α2(z)as(y, w, x). Since the Hom-associator as is alternating, we have as(y, w, x) = as(w, x, y), as(x, z, w) = as(z, w, x), and as(w, y, z) = as(y, z, w). Therefore, rearranging terms in the above equality, we obtain F = f ◦ (Id − ρ + ρ2) in the explicit form (3.11.3). (cid:3) The following result is the Hom-type analogue of part of [5] (Lemma 2.1). Proposition 3.13. Let (A, µ, α) be a Hom-alternative algebra. Then the Hom-Bruck-Kleinfeld function f is alternating. 14 DONALD YAU Proof. First observe that −F = F ◦ ξ, which follows immediately from the definition (3.11.2) of F . This implies −F = F ◦ ρ, where ρ = ξ3. Note that ρ3 = ξ. Thus, we have: 0 = F ◦ (Id + ρ) = f ◦ (Id − ρ + ρ2) ◦ (Id + ρ) = f ◦ (Id + ρ3) = f ◦ (Id + ξ). (by Lemma 3.12) Equivalently, we have so f changes sign under the cyclic permutation ξ. From the definition (3.11.1) of f and the fact that the Hom-associator as is alternating in a Hom-alternative algebra, we infer also that f = −f ◦ ξ, (3.13.1) f = −f ◦ η, (3.13.2) where η is the adjacent transposition η(w ⊗ x ⊗ y ⊗ z) = w ⊗ x ⊗ z ⊗ y. So f changes sign under the transposition η. Since the cyclic permutation ξ and the adjacent transposition η generate the symmetric group on four letters, we infer from (3.13.1) and (3.13.2) that f is alternating. (cid:3) The following identities are consequences of Proposition 3.13 and are the Hom-type analogues of part of [5] (Lemma 2.2). In what follows, we write µ(x, x) as x2. Corollary 3.14. Let (A, µ, α) be a Hom-alternative algebra. Then: as(x2, α(y), α(z)) = α2(x)as(x, y, z) + as(x, y, z)α2(x), as(α(x), xy, α(z)) = as(x, y, z)α2(x) = as(α(x), α(y), xz), as(α(x), yx, α(z)) = α2(x)as(x, y, z) = as(α(x), α(y), zx) (3.14.1) (3.14.2) (3.14.3) and for all x, y, z ∈ A. Proof. To obtain (3.14.1), set w = x in the definition (3.11.1) of f and use the fact that f is alternating (Proposition 3.13). For (3.14.2) we compute as follows: as(α(x), xy, α(z)) = as(xy, α(z), α(x)) (by alternativity of as) = f (x, y, z, x) + as(y, z, x)α2(x) + α2(y)as(x, z, x) = as(x, y, z)α2(x) (by alternativity of f and as). (by (3.11.1)) This proves half of (3.14.2). For the other half of (3.14.2), we compute similarly as follows: as(α(x), α(y), xz) = as(xz, α(x), α(y)) (by alternativity of as) = f (x, z, x, y) + as(z, x, y)α2(x) + α2(z)as(x, x, y) = as(x, y, z)α2(x) (by alternativity of f and as). (by (3.11.1)) This finishes the proof of (3.14.2). For (3.14.3) we compute as follows: as(α(x), yx, α(z)) = as(yx, α(z), α(x)) (by alternativity of as) = f (y, x, z, x) + as(x, z, x)α2(y) + α2(x)as(y, z, x) = α2(x)as(x, y, z) (by alternativity of f and as). (by (3.11.1)) HOM-MALTSEV, HOM-ALTERNATIVE, AND HOM-JORDAN ALGEBRAS 15 This proves half of (3.14.3). For the other half of (3.14.3), we compute similarly as follows: as(α(x), α(y), zx) = as(zx, α(x), α(y)) (by alternativity of as) = f (z, x, x, y) + as(x, x, y)α2(z) + α2(x)as(z, x, y) = α2(x)as(x, y, z) (by alternativity of f and as). (by (3.11.1)) This finishes the proof. (cid:3) The following result says that every Hom-alternative algebra satisfies a variation of the Hom- Maltsev identity (2.6.2) in which the Hom-Jacobian is replaced by the Hom-associator. Corollary 3.15. Let (A, µ, α) be a Hom-alternative algebra. Then as(α(x), α(y), [x, z]) = [as(x, y, z), α2(x)] for all x, y, z ∈ A, where [−, −] = µ ◦ (Id − τ ) is the commutator bracket. Proof. Indeed, we have as(α(x), α(y), [x, z]) = as(α(x), α(y), xz) − as(α(x), α(y), zx) = as(x, y, z)α2(x) − α2(x)as(x, y, z) = [as(x, y, z), α2(x)], (by (3.14.2) and (3.14.3)) as desired. (cid:3) Next we consider the relationship between the Hom-associator (2.5.1) in a Hom-algebra A and the Hom-Jacobian (2.5.2) in its commutator Hom-algebra A− (Definition 3.5). Lemma 3.16. Let (A, µ, α) be any Hom-algebra. Then JA− = asA ◦ (Id + σ + σ2) ◦ (Id − δ), where σ(x ⊗ y ⊗ z) = z ⊗ x ⊗ y and δ(x ⊗ y ⊗ z) = x ⊗ z ⊗ y. Proof. For x, y, z ∈ A, we have: JA−(x, y, z) = [[x, y], α(z)] + [[z, x], α(y)] + [[y, z], α(x)] = (xy)α(z) − (yx)α(z) − α(z)(xy) + α(z)(yx) + (zx)α(y) − (xz)α(y) − α(y)(zx) + α(y)(xz) + (yz)α(x) − (zy)α(x) − α(x)(yz) + α(x)(zy) = asA(x, y, z) + asA(z, x, y) + asA(y, z, x) − asA(x, z, y) − asA(y, x, z) − asA(z, y, x) = asA ◦ (Id + σ + σ2) ◦ (Id − δ)(x, y, z). This proves the Lemma. Proposition 3.17. Let (A, µ, α) be a Hom-alternative algebra. Then we have JA− = 6asA. Proof. Since the Hom-associator asA is alternating, with the notations in Lemma 3.16 we have The result now follows from Lemma 3.16. asA ◦ σ = asA = −asA ◦ δ. (cid:3) (cid:3) 16 DONALD YAU We are now ready to prove Theorem 3.8. Proof of Theorem 3.8. Let (A, µ, α) be a Hom-alternative algebra and A− = (A, [−, −], α) be its commutator Hom-algebra. The commutator bracket [−, −] = µ ◦ (Id − τ ) is anti-symmetric. Thus, it remains to show that the Hom-Maltsev identity (2.6.2) holds in A−, i.e., To prove this, we compute as follows: JA−(α(x), α(y), [x, z]) = [JA−(x, y, z), α2(x)]. JA−(α(x), α(y), [x, z]) = 6asA(α(x), α(y), [x, z]) (by Proposition 3.17) = [6asA(x, y, z), α2(x)] = [JA−(x, y, z), α2(x)] (by Corollary 3.15) (by Proposition 3.17). We have shown that A− is a Hom-Maltsev algebra, so A is Hom-Maltsev-admissible. (cid:3) Remark 3.18. By Theorem 3.8 the map A 7→ A− defines a functor from the category of Hom- alternative algebras to the category of Hom-Maltsev algebras. Using the combinatorial objects of weighted trees and an argument similar to that in [30], one can show that this functor has a left adjoint M 7→ U (M ). However, we do not know whether there is an analogue of the Poincar´e- Birkhoff-Witt (PBW) Theorem, i.e., whether the canonical map M → U (M ) is injective. In fact, even in the non-Hom case of Maltsev algebras, it is not known whether there is a PBW Theorem with alternative algebras in place of associative algebras. Probably the closest result to a PBW Theorem for Maltsev algebras is in [24]. Example 3.19. In this example, we describe a Hom-alternative algebra (hence Hom-Maltsev- admissible by Theorem 3.8) that is not Hom-Lie-admissible and not alternative. Recall that the octonions is an eight-dimensional alternative (but not associative) algebra O with basis {e0, . . . , e7} and the following multiplication table, where µ denotes the multiplication in O. e6 e6 µ e0 e4 e3 e1 e2 e3 e4 e0 e0 e1 e2 e7 −e2 e1 −e0 e1 e4 e2 −e4 −e0 e2 e5 e3 −e7 −e5 −e0 e3 e4 e4 e5 −e6 e5 e6 e6 e7 e7 e7 e5 e5 e7 e6 −e5 −e3 e7 −e6 e1 e3 −e5 e4 e1 e4 −e3 −e1 −e0 e2 e5 −e4 −e2 −e0 e5 −e7 e3 e1 −e3 e6 e2 −e1 −e6 −e0 e2 −e4 e7 e3 −e2 −e7 −e0 e6 −e1 The reader is referred to [4, 9, 23, 28] for discussion about the roles of the octonions in exceptional Lie groups, projective geometry, physics, and other applications. One can check that there is an algebra automorphism α : O → O given by α(e0) = e0, α(e1) = e5, α(e2) = e6, α(e3) = e7, α(e4) = e1, α(e5) = e2, α(e6) = e3, α(e7) = e4. (3.19.1) There is a more conceptual description of this algebra automorphism on O. Note that e1 and e2 anti-commute, and e3 anti-commutes with e1, e2, and e1e2 = e4. Such a triple (e1, e2, e3) is called a basic triple in [4]. Another basic triple is (e5, e6, e7). Then α is the unique automorphism on O that sends the basic triple (e1, e2, e3) to the basic triple (e5, e6, e7). HOM-MALTSEV, HOM-ALTERNATIVE, AND HOM-JORDAN ALGEBRAS 17 Using [15] (Theorem 2.1), which is the analogue of Theorem 2.13 for Hom-alternative algebras, we obtain a Hom-alternative (hence Hom-Maltsev-admissible) algebra with the following multiplication table. Oα = (O, µα = α ◦ µ, α) µα e0 e1 e2 e3 e4 e5 e6 e7 e6 e3 e4 e3 e2 e1 e0 e1 e7 e6 e5 e0 e4 −e6 e5 −e0 e1 e6 −e1 −e0 e2 e7 −e4 −e2 −e0 e1 e2 −e3 e3 e4 e7 e5 e4 e2 e3 −e2 −e7 e4 −e3 e5 e7 −e2 e1 e5 e1 −e7 −e5 −e0 e6 e2 −e1 −e6 −e0 e5 −e7 e3 e6 −e5 −e3 −e0 e6 −e1 e4 e7 −e6 −e4 −e0 e2 −e4 e7 e3 −e5 Note that Oα is not alternative because µα(µα(e0, e0), e1) = e5 6= e2 = µα(e0, µα(e0, e1)). Since Oα is Hom-alternative, by Theorem 3.8 it is also Hom-Maltsev-admissible, i.e., its commutator Hom-algebra O− α = (O, [−, −]α, α), where [−, −]α = µα ◦ (Id − τ ) = α ◦ µ ◦ (Id − τ ), is a Hom-Maltsev algebra. Observe that Oα is not Hom-Lie-admissible, i.e., O− α is not a Hom-Lie algebra. Indeed, with O− denoting the Maltsev algebra (O, [−, −] = µ ◦ (Id − τ )), we have JO− α = [−, −]α ◦ ([−, −]α ⊗ α) ◦ (Id + σ + σ2) = α2 ◦ [−, −] ◦ ([−, −] ⊗ Id) ◦ (Id + σ + σ2) = α2 ◦ JO− = 6α2 ◦ asO (3.19.2) by Proposition 3.17. Here we are regarding O as the Hom-alternative algebra (O, µ, Id) with identity twisting map. Since asO 6= 0 (because O is not associative) and α is an automorphism, it follows that JO− 6= 0. For example, we have α and asO(e5, e6, e7) = −2e3 6= 0 JO− α (e5, e6, e7) = −12α2(e3) = −12e4 6= 0. Therefore, Oα is a Hom-alternative (and hence Hom-Maltsev-admissible) algebra that is neither alternative nor Hom-Lie-admissible. Also, (O, [−, −]α) is not a Maltsev algebra. Indeed, let J ′ denote the usual Jacobian in (O, [−, −]α) as in (2.14.1). Then we have [J ′(e5, e6, e7), e5]α = [[[e5, e6]α, e7]α, e5]α + [[[e7, e5]α, e6]α, e5]α + [[[e6, e7]α, e5]α, e5]α = 8(e3 − e7) 18 and DONALD YAU J ′(e5, e6, [e5, e7]α) = 2J ′(e5, e6, e1) = −8(e1 + e3 + e7). So (O, [−, −]α) does not satisfy the Maltsev identity (1.0.1). Finally, observe that as long as β : O → O is an algebra automorphism, (3.19.2) implies that O− β is a Hom-Maltsev algebra that is not Hom-Lie. There are plenty of algebra automorphisms on O other than the one in (3.19.1). In fact, the automorphism group of O is the 14-dimensional exceptional Lie group G2 [4, 6]. (cid:3) 4. Hom-Maltsev-admissible algebras In Theorem 3.8 we showed that every Hom-alternative algebra is Hom-Maltsev-admissible, i.e., its commutator Hom-algebra is a Hom-Maltsev algebra. Since Hom-alternative algebras are al- ways Hom-flexible (Definition 3.3), we know that Hom-alternative algebras are Hom-flexible, Hom- Maltsev-admissible algebras. The purpose of this section is to study the (strictly larger) class of Hom-flexible, Hom-Maltsev-admissible algebras. We give several characterizations of Hom-flexible algebras that are Hom-Maltsev-admissible in terms of the cyclic Hom-associator (Proposition 4.3). Then we prove the analogues of the construction results, Theorems 2.11 and 2.13, for Hom-flexible and Hom-Maltsev-admissible algebras (Theorems 4.4 and 4.5). We then consider examples of Hom- flexible, Hom-Maltsev-admissible algebras that are neither Hom-alternative nor Hom-Lie-admissible. To state our characterizations of Hom-flexible algebras that are Hom-Maltsev-admissible, we need the following definition. Definition 4.1. Let (A, µ, α) be a Hom-algebra. Define the cyclic Hom-associator SA : A⊗3 → A as the multi-linear map where asA is the Hom-associator (2.5.1) and σ(x ⊗ y ⊗ z) = z ⊗ x ⊗ y. SA = asA ◦ (Id + σ + σ2), We will use the following preliminary observations about the relationship between the cyclic Hom-associator and the Hom-Jacobian (2.5.2) of the commutator Hom-algebra (Definition 3.5). Lemma 4.2. Let (A, µ, α) be a Hom-flexible algebra. Then we have 2SA = JA− , (4.2.1) where A− = (A, [−, −], α) is the commutator Hom-algebra. Proof. Let (cid:8) denote the cyclic sum (Id + σ + σ2). We have: SA(x, y, z) = (cid:8) asA(x, y, z) = −(cid:8) asA(z, y, x) (by Hom-flexibility) = −(cid:8) asA(x, z, y) = −SA(x, z, y). (4.2.2) Let δ denote the permutation δ(x ⊗ y ⊗ z) = x ⊗ z ⊗ y. Then (4.2.2) is equivalent to SA = −SA ◦ δ. HOM-MALTSEV, HOM-ALTERNATIVE, AND HOM-JORDAN ALGEBRAS 19 This implies that 2SA = SA ◦ (Id − δ) = asA ◦ (Id + σ + σ2) ◦ (Id − δ) = JA−, where the last equality is by Lemma 3.16. (cid:3) The following result gives characterizations of Hom-Maltsev-admissible algebras in terms of the cyclic Hom-associator, assuming Hom-flexibility. The condition (4.3.2) below is the Hom-type ana- logue of [21] (Lemma 1.2(ii)). Proposition 4.3. Let (A, µ, α) be a Hom-flexible algebra and A− = (A, [−, −], α) be its commutator Hom-algebra. Then the following statements are equivalent: (1) A is Hom-Maltsev-admissible (Definition 3.5). (2) The equality JA−(α(x), α(y), [x, z]) = [JA−(x, y, z), α2(x)] holds for all x, y, z ∈ A. (3) The equality SA(α(x), α(y), [x, z]) = [SA(x, y, z), α2(x)] holds for all x, y, z ∈ A. (4) The equality SA(α(w), α(y), [x, z]) + SA(α(x), α(y), [w, z]) = [SA(w, y, z), α2(x)] + [SA(x, y, z), α2(w)] holds for all w, x, y, z ∈ A. (4.3.1) (4.3.2) (4.3.3) Proof. The equivalence of the first two statements is immediate, since the commutator bracket [−, −] = µ ◦ (Id − τ ) is anti-symmetric and (4.3.1) is the Hom-Maltsev identity (2.6.2) for A−. The equivalence of (4.3.1) and (4.3.2) follows from (4.2.1), which uses the Hom-flexibility of A. Finally, that (4.3.2) is equivalent to (4.3.3) follows from linearization. In other words, starting from (4.3.2), one replaces x by w + x to obtain (4.3.3). Conversely, starting from (4.3.3), one sets w = x to obtain (4.3.2). (cid:3) The following construction results for Hom-flexible and Hom-Maltsev-admissible algebras are the analogues of Theorems 2.11 and 2.13. Theorem 4.4. (1) Let (A, µ) be a flexible algebra (i.e., (xy)x = x(yx) for all x, y ∈ A) and α : A → A be an algebra morphism. Then the induced Hom-algebra Aα = (A, µα = α ◦ µ, α) is a Hom-flexible algebra. (2) Let (A, µ, α) be a Hom-flexible algebra. Then the derived Hom-algebra An = (A, µ(n), α2n ) is also a Hom-flexible algebra for each n ≥ 0, where µ(n) = α2n−1 ◦ µ. 20 DONALD YAU Proof. For the first assertion, for any algebra (A, µ), we regard it as the Hom-algebra (A, µ, Id) with identity twisting map. Then we have: asAα = µα ◦ (µα ⊗ α − α ⊗ µα) = α2 ◦ µ ◦ (µ ⊗ Id − Id ⊗ µ) = α2 ◦ asA. (by multiplicativity of α) (4.4.1) Now for a flexible algebra (A, µ), this implies that asAα(x, y, x) = α2(asA(x, y, x)) = 0, so Aα is Hom-flexible. For the second assertion, we have that for any Hom-algebra (A, µ, α): asAn = µ(n) ◦ (µ(n) ⊗ α2n − α2n ⊗ µ(n)) = α2(2n−1) ◦ µ ◦ (µ ⊗ α − α ⊗ µ) = α2(2n−1) ◦ asA. (by multiplicativity of α) (4.4.2) Now for a Hom-flexible algebra (A, µ, α), this implies that asAn(x, y, x) = α2(2n−1)(asA(x, y, x)) = 0, so An is Hom-flexible. (cid:3) Theorem 4.5. (1) Let (A, µ) be a Maltsev-admissible algebra and α : A → A be an algebra morphism. Then the induced Hom-algebra Aα = (A, µα = α ◦ µ, α) is a Hom-Maltsev- admissible algebra. (2) Let (A, µ, α) be a Hom-Maltsev-admissible algebra. Then the derived Hom-algebra An = ) is also a Hom-Maltsev-admissible algebra for each n ≥ 0, where µ(n) = α2n−1 ◦ (A, µ(n), α2n µ. Proof. For the first assertion, the commutator algebra of (A, µ) is A− = (A, [−, −] = µ ◦ (Id − τ )), which is a Maltsev algebra by assumption. In particular, the Maltsev identity holds. The commutator Hom-algebra of the induced Hom-algebra Aα is A− α = (A, [−, −]α, α), where JA−(x, y, [x, z]) = [JA− (x, y, z), x] (4.5.1) [−, −]α = µα ◦ (Id − τ ) = α ◦ [−, −], (4.5.2) which is anti-symmetric. We must show that A− have: α satisfies the Hom-Maltsev identity (2.6.2). We JA− α = [−, −]α ◦ ([−, −]α ⊗ α) ◦ (Id + σ + σ2) = α2 ◦ [−, −] ◦ ([−, −] ⊗ Id) ◦ (Id + σ + σ2) = α2 ◦ JA−. (by (4.5.2)) (4.5.3) Therefore, we have: JA− α (α(x), α(y), [x, z]α) − [JA− α (x, y, z), α2(x)]α = α2{JA−(α(x), α(y), α[x, z])} − α[α2(JA−(x, y, z)), α2(z)] = α3 {JA− (x, y, [x, z]) − [JA− (x, y, z), x]} (by (2.13.2)) (by (4.5.3)) = 0 (by (4.5.1)). This shows that the Hom-Maltsev identity holds in A− α , proving the first assertion. HOM-MALTSEV, HOM-ALTERNATIVE, AND HOM-JORDAN ALGEBRAS 21 For the second assertion, assume that (A, µ, α) is a Hom-Maltsev-admissible algebra. Note that ), the commutator Hom-algebra of the nth derived Hom-algebra An is (An)− = (A, [−, −](n), α2n where [−, −](n) = µ(n) ◦ (Id − τ ) = α2n−1 ◦ [−, −]. Thus, we have (4.5.4) where A− = (A, [−, −], α) is the commutator Hom-algebra of A and (A−)n is its nth derived Hom- algebra (Definition 2.9). Since [−, −](n) is anti-symmetric, we must show that (An)− satisfies the Hom-Maltsev identity (2.6.2). To do that, observe that (An)− = (A−)n, J(An)− = J(A−)n (by (4.5.4)) = α2(2n−1) ◦ JA− (by (2.10.2)). (4.5.5) In the computation below, we write k = 3(2n − 1). Using (2.10.1) and (4.5.5) we compute as follows: J(An)−(α2n (x), α2n = α2(2n−1) nJA−(α2n (y), [x, z](n)) − [J(An)− (x, y, z), (α2n (y), α2n−1[x, z])o (x), α2n )2(x)](n) − α2n−1[α2(2n−1) ◦ JA−(x, y, z), α2n+1 (x)] = αk (cid:8)JA−(α(x), α(y), [x, z]) − [JA−(x, y, z), α2(x)](cid:9) = 0. This last equality follows from the Hom-Maltsev identity (2.6.2) in A−. We have shown that (An)− satisfies the Hom-Maltsev identity, so An is Hom-Maltsev-admissible. (cid:3) Below we consider examples of Hom-flexible, Hom-Maltsev-admissible algebras that are not Hom- alternative, not Hom-Lie-admissible, and not Maltsev-admissible. In particular, these Hom-Maltsev- admissible algebras cannot be obtained from Theorem 3.8. Therefore, the class of Hom-flexible, Hom-Maltsev-admissible algebras is strictly larger than the class of Hom-alternative algebras. Example 4.6. There is a five-dimensional flexible, Maltsev-admissible algebra (A, µ) ([21] Example 1.5, p.29) with basis {e1, . . . , e5} and multiplication table: µ e1 e2 e3 e4 e5 2 e4 e1 0 e5 − 1 0 − 1 2 e1 0 2 e4 e2 e5 + 1 0 0 1 2 e2 0 e4 e3 1 2 e1 0 − 1 2 e2 0 1 2 e3 0 − 1 2 e3 −e5 0 0 e5 0 0 0 0 0 This Maltsev-admissible algebra A is neither alternative nor Lie-admissible. Let λ, ξ ∈ k be arbitrary scalars with λ 6∈ {0, ±1}. There is an algebra morphism α : A → A given by α(e1) = λe1, α(e2) = λ−1e2, α(e3) = ξe3, α(e4) = e4, α(e5) = e5. By Theorems 4.4 and 4.5 the induced Hom-algebra Aα = (A, µα = α ◦ µ, α) is a Hom-flexible, Hom-Maltsev-admissible algebra. Its multiplication table is: 22 DONALD YAU µα e1 e2 e3 e4 e5 2 e4 e1 0 e5 − 1 0 − λ 2 e1 0 2 e4 e2 e5 + 1 0 0 λ−1 2 e2 0 e3 0 0 0 − ξ 2 e3 0 e4 λ 2 e1 − λ−1 2 e2 ξ 2 e3 −e5 0 e5 0 0 0 0 0 Note that Aα is not Hom-alternative because asAα(e1, e2, e2) = λ−2 4 e2 6= 0. Also, Aα is not Hom-Lie-admissible because J(Aα)−(e1, e2, e3) = −b2e3 6= 0. Finally, (A, µα) is not Maltsev-admissible, i.e., (A, [−, −]α) is not a Maltsev algebra, where [−, −]α = µα ◦ (Id − τ ) = α ◦ [−, −]. Indeed, let J ′ denote the usual Jacobian of (A, [−, −]α) as in (2.14.1). Then, on the one hand, we have J ′(e1, e2, [e1, e4]α) = [[e1, e2]α, [e1, e4]α]α + [[[e1, e4]α, e1]α, e2]α + [[e2, [e1, e4]α]α, e1]α = [e4, λe1]α + [[λe1, e1]α, e2]α + [[e2, λe1]α, e1]α = −λ2e1 + 0 + λ2e1 On the other hand, we have = 0. [J ′(e1, e2, e4), e1]α = [(cid:8) [[e1, e2]α, e4]α, e1]α = [(λ−1 − λ)e4, e1]α = (λ2 − 1)e1, which is not equal to 0 because λ 6= ±1. So (A, [−, −]α) does not satisfy the Maltsev identity (2.6.3) and, therefore, is not a Maltsev algebra. (cid:3) Example 4.7. There is a six-dimensional flexible, Maltsev-admissible algebra (A, µ) ([21] Table 5.10, p.301) with basis {e, h, f, u, v, w} and multiplication table: µ e h f − 1 u v w 2 h + λu e 0 e 0 −w 0 h −e 2λu f 0 −v w f u v 1 2 h + λu 0 w 0 v 0 0 0 0 0 0 0 − 1 2 u −f 0 0 0 −v w 0 −w v 0 1 2 u 0 In the table above, λ ∈ k is an arbitrary but fixed scalar. This Maltsev-admissible algebra A is neither alternative nor Lie-admissible. Let γ ∈ k be a scalar with γ 6= 0 and γ8 6= 1. Then there is an algebra morphism α : A → A given by α(e) = γ−2e, α(h) = h, α(f ) = γ2f, α(u) = u, α(v) = γv, α(w) = γ−1w. HOM-MALTSEV, HOM-ALTERNATIVE, AND HOM-JORDAN ALGEBRAS 23 By Theorems 4.4 and 4.5 the induced Hom-algebra Aα = (A, µα = α ◦ µ, α) is a Hom-flexible, Hom-Maltsev-admissible algebra. Its multiplication table is: µα e h f − 1 u v w e 0 0 0 γ−2e 2 h + λu −γ−1w h −γ−2e 2λu γ2f 0 −γv γ−1w f u v 1 2 h + λu 0 γ−1w −γ2f γv 0 0 0 0 0 0 0 0 − 1 2 u 0 0 0 −γv w 0 −γ−1w γv 0 1 2 u 0 Note that Aα is not Hom-alternative because asAα(h, h, f ) = −γ4f 6= 0. Also, Aα is not Hom-Lie-admissible because J(Aα)−(h, f, w) = −12γ2v 6= 0. Finally, (A, µα) is not Maltsev-admissible, i.e., (A, [−, −]α) is not a Maltsev algebra, where [−, −]α = µα ◦ (Id − τ ) = α ◦ [−, −]. Indeed, with J ′ denoting the usual Jacobian of (Aα)− = (A, [−, −]α) as in (2.14.1), we have J ′(e, h, [e, f ]α) − [J ′(e, h, f ), e]α = 4(γ−4 − 1)e − 4(γ4 − 1)e = 4(γ−4 − γ4)e. This is not equal to 0 because γ8 6= 1. So (A, [−, −]α) does not satisfy the Maltsev identity (2.6.3) and, therefore, is not a Maltsev algebra. (cid:3) Example 4.8. There is an eight-dimensional flexible, Maltsev-admissible algebra (A, µ) [21] (The- orems 4.11 and 5.7) with basis {a, e0, e±i : i = 1, 2, 3}, whose multiplication is determined by: e0e±i = −e±ie0 = ±e±i for i = 1, 2, 3, e±ie±j = −e±je±i = ±e∓k for (ijk) = (123), (312), (231), eie−i = 1 2 e0 + γa, e−iei = − 1 2 e2 0 = 2γa, a2 = δa, e0 + γa for i = 1, 2, 3, ax = xa = εx for x ∈ {e0, e±i}i=1,2,3. The unspecified products of the basis elements are 0, and γ, δ, ε are arbitrary but fixed scalars. This Maltsev-admissible algebra A is neither alternative nor Lie-admissible. Let λ, ξ ∈ k be non-zero scalars. Then there is an algebra morphism α : A → A given by α(a) = a, α(e0) = e0, α(e±1) = λ±1e±1, α(e±2) = ξ±1e±2, α(e±3) = (λξ)∓1e±3. By Theorems 4.4 and 4.5 the induced Hom-algebra Aα = (A, µα = α ◦ µ, α) is a Hom-flexible, Hom-Maltsev-admissible algebra. Note that Aα is not Hom-alternative because asAα(e0, e0, a) = 2γ(δ − ε)a, which is not equal to 0 in general. Also, Aα is not Hom-Lie-admissible because J(Aα)−(e0, e1, e2) = 12(λξ)2e−3 6= 0. 24 DONALD YAU Finally, (A, µα) is not Maltsev-admissible, i.e., (A, [−, −]α) is not a Maltsev algebra, where [−, −]α = µα ◦ (Id − τ ) = α ◦ [−, −]. Indeed, with J ′ denoting the usual Jacobian of (A, [−, −]α) as in (2.14.1), we have J ′(e0, e1, [e0, e2]α) − [J ′(e0, e1, e2), e0]α = 8γe−3, where γ = λξ2(λ + ξ − 2λξ − λ2 + λ2ξ), which is not equal to 0 in general. (cid:3) 5. Hom-alternative algebras are Hom-Jordan-admissible In this section, we define Hom-Jordan(-admissible) algebras. Some alternative characterizations of the Hom-Jordan identity (5.2.1) are given in Proposition 5.5. The main result of this section is Theorem 5.6, which says that Hom-alternative algebras are Hom-Jordan-admissible. Then we prove Theorems 5.8 and 5.9, which are construction results for Hom-Jordan and Hom-Jordan-admissible algebras. In Example 5.10 we construct Hom-Jordan algebras from the 27-dimensional exceptional simple Jordan algebra of 3 × 3 Hermitian octonionic matrices. Let us begin with some relevant definitions. Definition 5.1. Let (A, µ, α) be a Hom-algebra. Define its plus Hom-algebra as the Hom-algebra A+ = (A, ∗, α), where ∗ = (µ + µ ◦ τ )/2. With µ(x, y) = xy, the product ∗ is given by 1 2 which is commutative. Also, we have x ∗ y = (µ(x, y) + µ(y, x)) = 1 2 (xy + yx), x ∗ x = µ(x, x) = x2 (5.1.1) for x ∈ A. In other words, µ and ∗ have the same squares. In what follows, we will often abbreviate x ∗ x to x2. Definition 5.2. (1) A Hom-Jordan algebra is a Hom-algebra (A, µ, α) such that µ = µ ◦ τ (commutativity) and the Hom-Jordan identity asA(x2, α(y), α(x)) = 0 (5.2.1) is satisfied for all x, y ∈ A, where asA is the Hom-associator (2.5.1). (2) A Hom-Jordan-admissible algebra is a Hom-algebra (A, µ, α) whose plus Hom-algebra A+ = (A, ∗, α) is a Hom-Jordan algebra. The Hom-Jordan identity (5.2.1) can be rewritten as µ(µ(x2, α(y)), α2(x)) = µ(α(x2), µ(α(y), α(x))). Since the product ∗ is commutative, using (5.1.1) a Hom-algebra A is Hom-Jordan-admissible if and only if A+ satisfies the Hom-Jordan identity asA+ (x2, α(y), α(x)) = 0, (5.2.2) or equivalently for all x, y ∈ A. (x2 ∗ α(y)) ∗ α2(x) = α(x2) ∗ (α(y) ∗ α(x)), HOM-MALTSEV, HOM-ALTERNATIVE, AND HOM-JORDAN ALGEBRAS 25 Example 5.3. A Jordan(-admissible) algebra is a Hom-Jordan(-admissible) algebra with α = Id, since the Hom-Jordan identity (5.2.1) with α = Id is the Jordan identity (1.0.2). The reader is referred to [1, 11, 27] for discussions about structures of Jordan algebras. Other ways of constructing Hom-Jordan(-admissible) algebras are given below. (cid:3) Remark 5.4. In [15] Makhlouf defined a Hom-Jordan algebra as a commutative Hom-algebra satisfying asA(x2, y, α(x)) = 0, which becomes our Hom-Jordan identity (5.2.1) if y is replaced by α(y). This seemingly minor difference is, in fact, very significant with respect to Hom-Jordan- admissibility of Hom-alternative algebras. Using Makhlouf's definition of a Hom-Jordan algebra, Hom-alternative algebras are not Hom-Jordan-admissible, although Hom-associative algebras are still Hom-Jordan-admissible [15] (Theorem 3.3). Let us give some alternative characterizations of the Hom-Jordan identity (5.2.1), including a linearized version of it (5.5.2). The ordinary (non-Hom) version of the following result can be found in, e.g., [27] (Chapter IV). Proposition 5.5. Let (A, µ, α) be a Hom-algebra with µ commutative, i.e., µ = µ ◦ τ . Then the following statements are equivalent: (1) A is a Hom-Jordan algebra, i.e., A satisfies the Hom-Jordan identity (5.2.1). (2) A satisfies 2asA(xz, α(y), α(x)) + asA(x2, α(y), α(z)) = 0 (5.5.1) for all x, y, z ∈ A. (3) A satisfies asA(zx, α(y), α(w)) + asA(wz, α(y), α(x)) + asA(xw, α(y), α(z)) = 0 (5.5.2) for all w, x, y, z ∈ A. Proof. We will show the implications (1) ⇒ (2) ⇒ (3) ⇒ (1). First assume that A is a Hom-Jordan algebra. Replace x with x+ λz for λ ∈ k in the Hom-Jordan identity (5.2.1). Using the commutativity of µ and (5.2.1), the result is 0 = λ(cid:8)2asA(xz, α(y), α(x)) + asA(x2, α(y), α(z))(cid:9) + λ2 (cid:8)2asA(xz, α(y), α(z)) + asA(z2, α(y), α(x))(cid:9) . (5.5.3) Since (5.5.3) holds for both λ = 1 and λ = −1, it follows that the coefficient of λ in (5.5.3) is equal to 0, which is exactly the condition (5.5.1). So statement (1) implies statement (2). Next assume that A satisfies (5.5.1). Replace x with x + γw for γ ∈ k in (5.5.1). By the same reasoning as in the previous paragraph, in the resulting expression the coefficient of γ must be equal to 0. A simple computation shows that this coefficient of γ is twice the left-hand side of (5.5.2). Therefore, statement (2) implies statement (3). Finally, starting from (5.5.2), one sets w = z = x to obtain the Hom-Jordan identity (5.2.1). (cid:3) Note that the linearized Hom-Jordan identity (5.5.2) can be written as (cid:8) x,w,z asA(xw, α(y), α(z)) = 0, where (cid:8) x,w,z is the cyclic sum over (x, w, z). Here is the main result of this section. 26 DONALD YAU Theorem 5.6. Every Hom-alternative algebra is Hom-Jordan-admissible. To prove Theorem 5.6, we will use the following preliminary observation. Lemma 5.7. Let (A, µ, α) be any Hom-algebra and A+ = (A, ∗, α) be its plus Hom-algebra. Then we have 4asA+(x2, α(y), α(x)) = asA(x2, α(y), α(x)) − asA(α(x), α(y), x2) + asA(α(y), x2, α(x)) − asA(α(x), x2, α(y)) + asA(x2, α(x), α(y)) − asA(α(y), α(x), x2) + [α2(y), asA(x, x, x)] (5.7.1) for all x, y ∈ A, where [−, −] = µ ◦ (Id − τ ) is the commutator bracket. Proof. As usual we write µ(a, b) as the juxtaposition ab, and µ(x, x) = x2 = x ∗ x. Starting from the left-hand side of (5.7.1), we have: 4asA+(x2, α(y), α(x)) = 4(x2 ∗ α(y)) ∗ α2(x) − 4α(x2) ∗ (α(y) ∗ α(x)) = (x2α(y))α2(x) + (α(y)x2)α2(x) + α2(x)(x2α(y)) + α2(x)(α(y)x2) − α(x2)(α(y)α(x)) − α(x2)(α(x)α(y)) − (α(y)α(x))α(x2 ) − (α(x)α(y))α(x2) = asA(x2, α(y), α(x)) − asA(α(x), α(y), x2) + (α(y)x2)α2(x) + α2(x)(x2α(y)) − α(x2)(α(x)α(y)) − (α(y)α(x))α(x2). Using the definition of the Hom-associator (2.5.1), the last four terms in (5.7.2) are: (α(y)x2)α2(x) = asA(α(y), x2, α(x)) + α2(y)(x2α(x)), α2(x)(x2α(y)) = −asA(α(x), x2, α(y)) + (α(x)x2)α2(y), −α(x2)(α(x)α(y)) = asA(x2, α(x), α(y)) − (x2α(x))α2(y), −(α(y)α(x))α(x2) = −asA(α(y), α(x), x2) − α2(y)(α(x)x2). Note that [α2(y), asA(x, x, x)] = [α2(y), (x2)α(x) − α(x)x2] = α2(y)(x2α(x)) − α2(y)(α(x)x2) − (x2α(x))α2(y) + (α(x)x2)α2(y). The desired condition (5.7.1) now follows from (5.7.2), (5.7.3), and (5.7.4). (5.7.2) (5.7.3) (5.7.4) (cid:3) Proof of Theorem 5.6. Let (A, µ, α) be a Hom-alternative algebra. To show that it is Hom-Jordan- admissible, it suffices to prove the Hom-Jordan identity for its plus Hom-algebra A+ (5.2.2). To do this, first observe that A itself satisfies the Hom-Jordan identity: asA(x2, α(y), α(x)) = α2(x)asA(x, y, x) + asA(x, y, x)α2(x) (by (3.14.1)) Using again the alternativity of asA, this implies that = 0 (by alternativity of asA). 0 = (asA ◦ θ)(x2, α(y), α(x)) for every permutation θ on three letters. Since asA(x, x, x) = 0 as well, it follows from Lemma 5.7 that 4asA+(x2, α(y), α(x)) = 0, HOM-MALTSEV, HOM-ALTERNATIVE, AND HOM-JORDAN ALGEBRAS from which the desired Hom-Jordan identity for A+ (5.2.2) follows. 27 (cid:3) The following construction results are the analogues of Theorems 2.11 and 2.13 for Hom-Jordan and Hom-Jordan-admissible algebras. Theorem 5.8. (1) Let (A, µ) be a Jordan algebra and α : A → A be an algebra morphism. Then the induced Hom-algebra Aα = (A, µα = α ◦ µ, α) is a Hom-Jordan algebra. (2) Let (A, µ, α) be a Hom-Jordan algebra. Then the derived Hom-algebra An = (A, µ(n) = α2n−1 ◦ µ, α2n ) is also a Hom-Jordan algebra for each n ≥ 0. Proof. For the first assertion, first note that µα = α ◦ µ is commutative. To prove the Hom-Jordan identity (5.2.1) in Aα, regard (A, µ) as the Hom-algebra (A, µ, Id). Then we have: asAα(µα(x, x), α(y), α(x)) = asAα(α(x2), α(y), α(x)) = α2 (cid:0)asA(α(x2), α(y), α(x))(cid:1) = α3 (cid:0)asA(x2, y, x)(cid:1) = 0 (by (1.0.2) in A). (by (4.4.1)) This shows that Aα is a Hom-Jordan algebra. For the second assertion, first note that µ(n) = α2n−1 ◦ µ is commutative. To prove the Hom- Jordan identity (5.2.1) in An, we compute as follows: asAn (µ(n)(x, x), α2n (y), α2n (x)) = α2(2n−1) ◦ asA(α2n−1(x2), α2n = α3(2n−1) ◦ asA(x2, α(y), α(x)) = 0 (by (5.2.1) in A). (y), α2n (x)) (by (4.4.2)) This shows that An is a Hom-Jordan algebra. (cid:3) Theorem 5.9. (1) Let (A, µ) be a Jordan-admissible algebra and α : A → A be an algebra morphism. Then the induced Hom-algebra Aα = (A, µα = α ◦ µ, α) is a Hom-Jordan- admissible algebra. (2) Let (A, µ, α) be a Hom-Jordan-admissible algebra. Then the derived Hom-algebra An = (A, µ(n) = α2n−1 ◦ µ, α2n ) is also a Hom-Jordan-admissible algebra for each n ≥ 0. Proof. For the first assertion, first note that the plus Hom-algebra (Aα)+ = (A, ∗α, α) satisfies ∗α = 1 2 (µα + µα ◦ τ ) = α ◦ 1 2 (µ + µ ◦ τ ) = α ◦ ∗. Therefore, we have (Aα)+ = (A+)α, where A+ is the Jordan-algebra (A, ∗). Since ∗α is commutative, it remains to prove the Hom-Jordan identity in (Aα)+ = (A+)α. We compute as follows: (by (4.4.1) in A+) as(A+)α(µα(x, x), α(y), α(x)) = α2 ◦ asA+(α(x2), α(y), α(x)) = α3 (cid:0)asA+ (x2, y, x)(cid:1) = 0 (by (1.0.2) in A+). This shows that (Aα)+ satisfies the Hom-Jordan identity, so Aα is Hom-Jordan-admissible. For the second assertion, first note that the plus Hom-algebra (An)+ = (A, ∗(n), α2n ) satisfies ∗(n) = 1 2 (µ(n) + µ(n) ◦ τ ) = α2n−1 ◦ (µ + µ ◦ τ ) = α2n−1 ◦ ∗. 1 2 28 DONALD YAU Therefore, we have (An)+ = (A+)n, where A+ is the Hom-Jordan algebra (A, ∗, α) and (A+)n is its nth derived Hom-algebra. Since ∗(n) is commutative, it remains to prove the Hom-Jordan identity in (An)+ = (A+)n. We compute as follows: (y), α2n as(A+)n (µ(n)(x, x), α2n (x)) = α2(2n−1) ◦ asA+ (α2n −1(x2), α2n = α3(2n−1) ◦ asA+ (x2, α(y), α(x)) = 0 (by (5.2.1) in A+). (y), α2n (x)) (by (4.4.2) in A+) This shows that (An)+ is a Hom-Jordan algebra, so An is Hom-Jordan-admissible. (cid:3) Example 5.10. In this example, we discuss how (non-Jordan) Hom-Jordan algebras can be con- structed from the 27-dimensional exceptional simple Jordan algebra M 8 3 . First recall from Example 3.19 the octonions O, which is an eight-dimensional alternative (but not associative) algebra with i=0 biei with basis {e0, . . . , e7}, where e0 is a two-sided multiplicative unit. For an octonion x = P7 each bi ∈ k, its conjugate is defined as the octonion x = b0e0 − P7 i=1 biei. The elements of M 8 3 are 3 × 3 Hermitian octonionic matrices, i.e., matrices of the form X =  a1 x  y x a2 z y z a3   with each ai ∈ k and x, y, z ∈ O. Here we are using the convention ai = aie0 for the diagonal elements. This k-module M 8 3 becomes a Jordan algebra with the multiplication X ∗ Y = 1 2 (XY + Y X), where XY and Y X are the usual matrix multiplication. The reader is referred to [9, 12, 23, 27] for discussions about the Jordan algebra M 8 3 and its relationship with the exceptional Lie algebra F4. Let α : O → O be any unit-preserving and conjugate-preserving algebra morphism, i.e., α(e0) = e0 and α(x) = α(x) for all x ∈ O. Then it extends entrywise to a linear map α : M 8 3 . It is easy to see that this extended map α respects matrix multiplication and hence also the Jordan product ∗, i.e., α is an algebra morphism on (M 8 3 , ∗). By the first part of Theorem 5.8, the induced Hom-algebra (M 8 3 → M 8 3 )α = (M 8 3 , ∗α = α ◦ ∗, α) is a Hom-Jordan algebra. Note that (M 8 3 , ∗α = α ◦ ∗) is in general not a Jordan algebra. For instance, consider the algebra automorphism α : O → O defined in (3.19.1), which is both unit-preserving and conjugate- preserving. We claim that (M 8 3 , ∗α) is not a Jordan algebra, i.e., the Jordan identity ((X ∗α X) ∗α Y ) ∗α X = (X ∗α X) ∗α (Y ∗α X) (5.10.1) is not satisfied for some X, Y ∈ M 8 left-hand side in (5.10.1) is 3 . Indeed, using the multiplicativity of α with respect to ∗, the where X 2 = X ∗ X, and its right-hand side is α(α(α(X 2) ∗ Y ) ∗ X), α2(X 2 ∗ (Y ∗ X)). HOM-MALTSEV, HOM-ALTERNATIVE, AND HOM-JORDAN ALGEBRAS 29 Since α is invertible, to show that (M 8 X, Y ∈ M 8 3 such that 3 , ∗α) is not a Jordan algebra, it suffices to exhibit two elements α(α(X 2) ∗ Y ) ∗ X 6= α(X 2 ∗ (Y ∗ X)). (5.10.2) Now we pick X =   1 0 −e1 0 1 0 e1 0 1   and Y =   1 −e2 −e3 e2 0 0 e3 0 0   in M 8 3 . With a little bit of computation, we obtain α(α(X 2) ∗ Y ) ∗ X =   2 − 1 2 e3 − 3 2 e6 −e1 − e2 − 2e7 1 2 e3 + 3 2 e6 0 3 4 e5 − e7 e1 + e2 + 2e7 4 e5 + e7 − 3   and α(X 2 ∗ (Y ∗ X)) = 3  2e6 0  −2e5 − 2e7 − 7 4 e1 −2e6 2e5 + 2e7 7 4 e1 1 0   . Therefore, we have proved (5.10.2), so (M 8 3 , ∗α) is not a Jordan algebra. (cid:3) 6. Further properties of Hom-alternative algebras In this section, we consider further properties of Hom-alternative algebras, including Hom-type analogues of the Moufang identities [20] (Theorem 6.8) and more identities concerning the Hom- Bruck-Kleinfeld function (3.11.1) (Propositions 6.3 and 6.5). By Theorem 3.8 every Hom-alternative algebra is Hom-Maltsev-admissible. As we saw in Example 3.19, there are Hom-alternative algebras that are not Hom-Lie-admissible. It is, therefore, natural to ask which Hom-alternative algebras are Hom-Lie-admissible. The following result says that the intersection of the classes of Hom-alternative algebras and of Hom-Lie-admissible algebras (within the class of Hom-algebras) is precisely the class of Hom-associative algebras. Proposition 6.1. Let (A, µ, α) be a Hom-algebra. Then A is a Hom-associative algebra if and only if it is both a Hom-alternative algebra and a Hom-Lie-admissible algebra. Proof. If A is Hom-associative, then by definition asA = 0, which is alternating, so A is Hom- alternative [15]. It is observed in [17] that Hom-associative algebras are always Hom-Lie-admissible. Conversely, if A is Hom-alternative and Hom-Lie-admissible, then asA = 1 6 JA− (by Proposition 3.17) = 0 (A− is Hom-Lie). So A is Hom-associative. (cid:3) It is proved in [15] that there is an analogue of Theorem 2.13 for Hom-alternative algebras. It is a variation of [31] (Theorem 2.3), which deals with G-Hom-associative algebras. More precisely, if (A, µ) is an alternative algebra and α : A → A is an algebra morphism, then the induced Hom-algebra Aα = (A, µα = α ◦ µ, α) is a Hom-alternative algebra. 30 DONALD YAU The following result is the analogue of Theorem 2.11 for Hom-alternative algebras. It says that the category of Hom-alternative algebras is closed under taking derived Hom-algebras (Definition 2.9). Proposition 6.2. Let (A, µ, α) be a Hom-alternative algebra. Then the nth derived Hom-algebra An = (A, µ(n) = α2n −1 ◦ µ, α2n ) is also a Hom-alternative algebra for each n ≥ 0. Proof. Indeed, asAn is alternating because asAn = α2(2n−1) ◦ asA by (4.4.2) and asA is alternating. (cid:3) Next we provide further properties of the Hom-Bruck-Kleinfeld function f (3.11.1). The following result gives two characterizations of the Hom-Bruck-Kleinfeld function in a Hom-alternative algebra. It is the Hom-type analogue of part of [5] (Lemma 2.1). Proposition 6.3. Let (A, µ, α) be a Hom-alternative algebra. Then the Hom-Bruck-Kleinfeld func- tion f satisfies f = F = as ◦ ([−, −] ⊗ α⊗2) ◦ (Id + ζ), (6.3.1) 1 3 where F is defined in (3.11.2) and ζ is the permutation ζ(w ⊗ x ⊗ y ⊗ z) = y ⊗ z ⊗ w ⊗ x Proof. First note that f = −f ◦ ρ = f ◦ ρ2 because f is alternating (Proposition 3.13), where ρ = ξ3 is the cyclic permutation ρ(w ⊗ x⊗ y ⊗ z) = x ⊗ y ⊗ z ⊗ w. Therefore, we have F = f ◦ (Id − ρ + ρ2) (by Lemma 3.12) = 3f, which proves the first equality in (6.3.1). It remains to prove that f is equal to the last entry in (6.3.1). Since f is alternating, from its definition (3.11.1) we have 2f (w, x, y, z) = f (w, x, y, z) − f (x, w, y, z) = as(wx, α(y), α(z)) − as(x, y, z)α2(w) − α2(x)as(w, y, z) − as(xw, α(y), α(z)) + as(w, y, z)α2(x) + α2(w)as(x, y, z). Rearranging terms we obtain as([w, x], α(y), α(z)) = [α2(x), as(w, y, z)] − [α2(w), as(x, y, z)] + 2f (w, x, y, z) = [α2(x), as(y, z, w)] − [α2(w), as(x, y, z)] + 2f (w, x, y, z), (6.3.2) in which as(w, y, z) = as(y, z, w) because as is alternating. Interchanging (w, x) with (y, z) in (6.3.2) and using the alternativity of f , we obtain as([y, z], α(w), α(x)) = [α2(z), as(w, x, y)] − [α2(y), as(z, w, x)] + 2f (y, z, w, x) = [α2(z), as(w, x, y)] − [α2(y), as(z, w, x)] + 2f (w, x, y, z). (6.3.3) HOM-MALTSEV, HOM-ALTERNATIVE, AND HOM-JORDAN ALGEBRAS 31 Adding (6.3.2) and (6.3.3) we have as ◦ ([−, −] ⊗ α⊗2) ◦ (Id + ζ)(w ⊗ x ⊗ y ⊗ z) = as([w, x], α(y), α(z)) + as([y, z], α(w), α(x)) = (4f − F )(w, x, y, z) (by (3.11.3)) = f (w, x, y, z), since F = 3f . This proves that f is equal to the last entry in (6.3.1). (cid:3) In Proposition 3.13 we showed that the Hom-Bruck-Kleinfeld function f (3.11.1) in a Hom- alternative algebra is an alternating function on four variables. We now discuss a closely related function on five variables in a Hom-alternative algebra. Definition 6.4. Let (A, µ, α) be a Hom-algebra. Define the multi-linear map g : A⊗5 → A by g(u, v, w, x, y) = f (uv, α(w), α(x), α(y)) − α3(u)f (v, w, x, y) − f (u, w, x, y)α3(v) − α (as(u, x, y)α[v, w]) (6.4.1) − α ((α[u, w])as(v, x, y)) for u, v, w, x, y ∈ A, where f is the Hom-Bruck-Kleinfeld function (3.11.1) and [−, −] = µ ◦ (Id − τ ) is the commutator bracket of µ. The following result says that the map g is almost alternating in a Hom-alternative algebra and is the Hom-type analogue of [5] (Lemma 2.3). Since g is constructed using α, µ, [−, −], as, and f (which is defined using α, µ, and as), the following result is ultimately about identities in a Hom-alternative algebra. Proposition 6.5. In a Hom-alternative algebra (A, µ, α), the map g (6.4.1) is alternating in {u, v, w} and also in {x, y}. That is, g changes sign when two of {u, v, w} (or x and y) are in- terchanged. Proof. The map g is alternating in {x, y} because f and as are both alternating (the former by Proposition 3.13). To show that g is alternating in {u, v, w}, first note that it is enough to show that g is alternating in {u, w} and in {v, w}. The map g is alternating in {u, w} if and only if Since f is alternating and [u, u] = 0, (6.5.1) is equivalent to f (uv, α(u), α(x), α(y)) = α3(u)f (v, u, x, y) + α (as(u, x, y)α[v, u]) . (6.5.2) g(u, v, u, x, y) = 0. (6.5.1) Likewise, g is alternating in {v, w} if and only if g(u, v, v, x, y) = 0, which is equivalent to f (uv, α(v), α(x), α(y)) = f (u, v, x, y)α3(v) + α ((α[u, v])as(v, x, y)) . It remains to prove (6.5.2) and (6.5.3), which we do in the following two Lemmas. (6.5.3) (cid:3) Lemma 6.6. In a Hom-alternative algebra (A, µ, α), (6.5.2) holds. 32 DONALD YAU Proof. To prove (6.5.2), we start with f (vu, α(u), α(x), α(y)) = as([vu, α(u)], α2(x), α2(y)) + as([α(x), α(y)], α(vu), α2(u)), (6.6.1) which follows from Proposition 6.3. We have [vu, α(u)] = (vu)α(u) − α(u)(vu) = (vu)α(u) − (uv)α(u) = [v, u]α(u). Therefore, using the definition (3.11.1) of f , the first summand on the right-hand side of (6.6.1) is: as([vu, α(u)], α2(x), α2(y)) = as([v, u]α(u), α2(x), α2(y)) = f ([v, u], α(u), α(x), α(y)) + as(α(u), α(x), α(y))α2([v, u]) + α2(α(u))as([v, u], α(x), α(y)) (6.6.2) = f (vu, α(u), α(x), α(y)) − f (uv, α(u), α(x), α(y)) + α(as(u, x, y)α[v, u]) + α3(u)as([v, u], α(x), α(y)). In the last equality above, we used the multiplicativity of α. On the other hand, the second summand on the right-hand side of (6.6.1) is: as([α(x), α(y)], α(vu), α2(u)) = as(α[x, y], α(v)α(u), α2(u)) (by multiplicativity of α) = −as(α2(u), α(v)α(u), α[x, y]) = −α2(α(u))as(α(u), α(v), [x, y]) = α3(u)as([x, y], α(v), α(u)) (by alternativity of as) (by (3.14.3)) (6.6.3) (by alternativity of as). Using (6.6.2) and (6.6.3) in (6.6.1), we obtain: f (uv, α(u), α(x), α(y)) = α(as(u, x, y)α[v, u]) + α3(u){as([v, u], α(x), α(y)) + as([x, y], α(v), α(u))} = α(as(u, x, y)α[v, u]) + α3(u)f (v, u, x, y), where the last equality follows from Proposition 6.3 again. This finishes the proof. (cid:3) Lemma 6.7. In a Hom-alternative algebra (A, µ, α), (6.5.3) holds. Proof. To prove (6.5.3), we start with f (vu, α(v), α(x), α(y)) = as([vu, α(v)], α2(x), α2(y)) + as([α(x), α(y)], α(vu), α2(v)), (6.7.1) which follows from Proposition 6.3. We have [vu, α(v)] = (vu)α(v) − α(v)(vu) = α(v)(uv) − α(v)(vu) = α(v)[u, v]. Therefore, using the definition (3.11.1) of f , the first summand on the right-hand side of (6.7.1) is: as([vu, α(v)], α2(x), α2(y)) = as(α(v)[u, v], α2(x), α2(y)) = f (α(v), [u, v], α(x), α(y)) + as([u, v], α(x), α(y))α3(v) + α2([u, v])as(α(v), α(x), α(y)) (6.7.2) = f (vu, α(v), α(x), α(y)) − f (uv, α(v), α(x), α(y)) + as([u, v], α(x), α(y))α3(v) + α ((α[u, v])as(v, x, y)) . HOM-MALTSEV, HOM-ALTERNATIVE, AND HOM-JORDAN ALGEBRAS 33 In the last equality above, we used the alternativity of f (Proposition 3.13) and the multiplicativity of α. On the other hand, the second summand on the right-hand side of (6.7.1) is: as([α(x), α(y)], α(vu), α2(v)) = as(α[x, y], α(v)α(u), α2(v)) (by multiplicativity of α) = −as(α2(v), α(v)α(u), α[x, y]) = −as(α(v), α(u), [x, y])α3(v) = as([x, y], α(u), α(v))α3(v) (by alternativity of as) (by (3.14.2)) (6.7.3) (by alternativity of as). Using (6.7.2) and (6.7.3) in (6.7.1), we obtain: f (uv, α(v), α(x), α(y)) = α ((α[u, v])as(v, x, y)) + {as([u, v], α(x), α(y)) + as([x, y], α(u), α(v))}α3(v) = α ((α[u, v])as(v, x, y)) + f (u, v, x, y)α3(v), where the last equality follows from Proposition 6.3 again. This finishes the proof. (cid:3) In any alternative algebra, the following Moufang identities [20] hold: (xyx)z = x(y(xz)), ((zx)y)x = z(xyx), (xy)(zx) = x(yz)x. (6.7.4) Here xyx = (xy)x = x(yx) is unambiguous in an alternative algebra. Now we prove analogues of the Moufang identities in a Hom-alternative algebra. The proof below is the Hom version of that of [5] (Lemma 2.2). Theorem 6.8 (Hom-Moufang identities). Let (A, µ, α) be a Hom-alternative algebra. Then the following Hom-Moufang identities hold for all x, y, z ∈ A: ((xy)α(x))α2(z) = α2(x)(α(y)(xz)), ((zx)α(y))α2(x) = α2(z)(α(x)(yx)), α((xy)(zx)) = α2(x)((yz)α(x)). (6.8.1a) (6.8.1b) (6.8.1c) Proof. For (6.8.1a) we compute as follows: ((xy)α(x))α2(z) = as(xy, α(x), α(z)) + α(xy)(α(x)α(z)) (by (2.5.1)) = as(xy, α(x), α(z)) + (α(x)α(y))α(xz) = as(xy, α(x), α(z)) + as(α(x), α(y), xz) + α2(x)(α(y)(xz)) = −as(α(x), α(y), xz) + as(α(x), α(y), xz) + α2(x)(α(y)(xz)) = α2(x)(α(y)(xz)). (by multiplicativity of α) (by (2.5.1)) (by (3.14.2)) For (6.8.1b) we compute as follows: ((zx)α(y))α2(x) = as(zx, α(y), α(x)) + α(zx)(α(y)α(x)) (by (2.5.1)) = as(zx, α(y), α(x)) + (α(z)α(x))α(yx) = as(zx, α(y), α(x)) + as(α(z), α(x), yx) + α2(z)(α(x)(yx)) = −as(α(z), α(x), yx)) + as(α(z), α(x), yx) + α2(z)(α(x)(yx)) = α2(z)(α(x)(yx)). (by multiplicativity of α) (by (2.5.1)) (by (3.14.3)) 34 DONALD YAU For (6.8.1c) we compute as follows: α((xy)(zx)) = (α(x)α(y))α(zx) (by multiplicativity of α) = as(α(x), α(y), zx) + α2(x)(α(y)(zx)) = α2(x)as(x, y, z) + α2(x)(α(y)(zx)) = α2(x) {as(y, z, x) + α(y)(zx)} = α2(x)((yz)α(x)). (by (2.5.1)) (by (3.14.3)) (by alternativity of as) This completes the proof. (cid:3) References [1] A.A. Albert, A structure theory for Jordan algebras, Ann. Math. 48 (1947) 546-567. [2] F. Ammar and A. Makhlouf, Hom-Lie superalgebras and Hom-Lie admissible superalgebras, arXiv:0906.1668v1. [3] H. Ataguema, A. Makhlouf, and S. Silvestrov, Generalization of n-ary Nambu algebras and beyond, J. Math. Phys. 50, no. 8 (2009), 083501. [4] J.C. Baez, The octonions, Bull. Amer. Math. Soc. 39 (2002) 145-205. [5] R.H. Bruck and E. Kleinfeld, The structure of alternative division rings, Proc. Amer. Math. Soc. 2 (1951) 878-890. [6] ´E. Cartan, Les groupes r´eels simples finis et continus, Ann. ´Ecole Norm. 31 (1914) 263-355. [7] Y. Fr´egier and A. Gohr, On unitality conditions for hom-associative algebras, arXiv:0904.4874v1. [8] A. Gohr, On hom-algebras with surjective twisting, arXiv:0906.3270v2. [9] F. Gursey and C.-H. Tze, On the role of division, Jordan and related algebras in particle physics, World Scientific, Singapore, 1996. [10] J.T. Hartwig, D. Larsson, and S.D. Silvestrov, Deformations of Lie algebras using σ-derivations, J. Algebra 295 (2006) 314-361. [11] N. Jacobson, Structure and representations of Jordan algebras, Amer. Math. Soc., Providence, RI, 1968. [12] P. Jordan, J. von Neumann, and E. Wigner, On an algebraic generalization of the quantum mechanical formal- ism, Ann. Math. 35 (1934) 29-64. [13] F.S. Kerdman, Analytic Moufang loops in the large, Alg. Logic 18 (1980) 325-347. [14] E.N. Kuz'min, The connection between Mal'cev algebras and analytic Moufang loops, Alg. Logic 10 (1971) 1-14. [15] A. Makhlouf, Hom-alternative algebras and Hom-Jordan algebras, arXiv:0909.0326. [16] A. Makhlouf, Paradigm of nonassociative Hom-algebras and Hom-superalgebras, arXiv:1001.4240v1. [17] A. Makhlouf and S. Silvestrov, Hom-algebra structures, J. Gen. Lie Theory Appl. 2 (2008) 51-64. [18] A. Makhlouf and S. Silvestrov, Hom-algebras and Hom-coalgebras, to appear in J. Algebra Appl., arXiv:0811.0400v2. [19] A.I. Mal'tsev, Analytic loops, Mat. Sb. 36 (1955) 569-576. [20] R. Moufang, Zur struktur von alternativkorpern, Math. Ann. 110 (1935) 416-430. [21] H.C. Myung, Malcev-admissible algebras, Progress in Math. 64, Birkhauser, Boston, MA, 1986. [22] P.T. Nagy, Moufang loops and Malcev algebras, Sem. Sophus Lie 3 (1993) 65-68. [23] S. Okubo, Introduction to octonion and other non-associative algebras in physics, Cambridge Univ. Press, Cambridge, UK, 1995. [24] J.M. P´erez-Izquierdo and I.P. Shestakov, An envelope for Malcev algebras, J. Algebra 272 (2004) 379-393. [25] L.V. Sabinin, Smooth quasigroups and loops, Kluwer Academic, The Netherlands, 1999. [26] A.A. Sagle, Malcev algebras, Trans. Amer. Math. Soc. 101 (1961) 426-458. [27] R.D. Schafer, An introduction to nonassociative algebras, Dover, New York, 1996. [28] T.A. Springer and F.D. Veldkamp, Octonions, Jordan algebras, and exceptional groups, Springer, Berlin, 2000. [29] J. Tits and R.M. Weiss, Moufang polygons, Springer-Verlag, Berlin, 2002. [30] D. Yau, Enveloping algebras of Hom-Lie algebras, J. Gen. Lie Theory Appl. 2 (2008) 95-108. [31] D. Yau, Hom-algebras and homology, J. Lie Theory 19 (2009) 409-421. [32] D. Yau, Hom-bialgebras and comodule algebras, arXiv:0810.4866. [33] D. Yau, Hom-Novikov algebras, arXiv:0909.0726. [34] D. Yau, The Hom-Yang-Baxter equation, Hom-Lie algebras, and quasi-triangular bialgebras, J. Phys. A 42 (2009) 165202 (12pp). HOM-MALTSEV, HOM-ALTERNATIVE, AND HOM-JORDAN ALGEBRAS 35 [35] D. Yau, The Hom-Yang-Baxter equation and Hom-Lie algebras, arXiv:0905.1887. [36] D. Yau, The classical Hom-Yang-Baxter equation and Hom-Lie bialgebras, arXiv:0905.1890. [37] D. Yau, Infinitesimal Hom-bialgebras and Hom-Lie bialgebras, arXiv:1001.5000. [38] D. Yau, Hom-quantum groups I: quasi-triangular Hom-bialgebras, arXiv:0906.4128. [39] D. Yau, Hom-quantum groups II: cobraided Hom-bialgebras and Hom-quantum geometry, arXiv:0907.1880. [40] D. Yau, Hom-quantum groups III: representations and module Hom-algebras, arXiv:0911.5402. Department of Mathematics, The Ohio State University at Newark, 1179 University Drive, Newark, OH 43055, USA E-mail address: [email protected]
1010.4898	1	1010	2010-10-23T18:38:17	Group rings of countable non-abelian locally free groups are primitive	[ "math.RA", "math.GR" ]	We prove that every group ring of a non-abelian locally free group which is the union of an ascending sequence of free groups is primitive. In particular, every group ring of a countable non-abelian locally free group is primitive. In addition, by making use of the result, we give a necessary and sufficient condition for group rings of ascending HNN extensions of free groups to be primitive, which extends the known result for the countable case to the general cardinality case. In order to prove the main theorem, we state some graph-theoretic results and apply them to to Formanek's method.	math.RA	math	GROUP RINGS OF COUNTABLE NON-ABELIAN LOCALLY FREE GROUPS ARE PRIMITIVE Tsunekazu Nishinaka Okayama Shoka University Okayama 700-8601 Japan 0 1 0 2 t c O 3 2 ] . A R h t a m [ 1 v 8 9 8 4 . 0 1 0 1 : v i X r a Abstract We prove that every group ring of a non-abelian locally free group which is the In particular, every union of an ascending sequence of free groups is primitive. group ring of a countable non-abelian locally free group is primitive. In addition, by making use of the result, we give a necessary and sufficient condition for group rings of ascending HNN extensions of free groups to be primitive, which extends the main result in [16] to the general cardinality case. 1 INTRODUCTION A ring is (right) primitive if it has a faithful irreducible (right) module. Our purpose in this paper is to study the primitivity of group rings of locally free groups. A group is called locally free if all of its finitely generated subgroups are free. It is well known that there exist locally free groups which are not free. For example, a properly ascending union of non-abelian free groups of bounded finite rank is infinitely generated and Hopfian (see [14]), and so it is a locally free group which is not free, where a group is Hopfian provided that every surjective endomorphism of that group is an automorphism. An example of uncountable non-free locally free groups can be seen in Higman [10]. It has been seen that a locally free group appears in a subgroup of the fundamental group of a three-dimensional manifold ([8], [1], [11], [15]). The fundamental group of the mapping torus of the standard 2-complex of a free group F with bounding maps the identity and it group ring; primitive; locally free group; ascending HNN-extension. 2000 MSC: 6S34, 20C07, 20E25, 20E06 1 an injective endomorphism ϕ of F , which is called the ascending HNN extension of F corresponding to ϕ, also has a locally free group as a subgroup. Recall that the ascending HNN extension Fϕ of F corresponding to ϕ has the pre- sentation Fϕ = hF, tt−1f t = ϕ(f )i. The ascending HNN extension Fϕ of a free group F is a well-studied class of groups. For example, Fϕ is coherent (Feighn and Handel [6]), where a group is coherent if its finitely generated subgroups are finitely presented. If F is finitely generated then Fϕ is Hopfian (Geoghegan, Mihalik, Sapir and Wise [9]). More- over, Borisov and Sapir [4] have recently shown that it is residually finite. The present author [16] has quite recently shown that the group ring KFϕ is semiprimitive for any field K and it is often primitive provided that F is a non-abelian countable free group. In the proof, it was shown that the group ring of a certain countable locally free group is primitive, and therefore we posed a question: Is it true that the group ring of any locally free group is primitive? In the present paper, we shall give a partial answer to the question: Theorem 1.1 Let G be a non-abelian locally free group which has a free subgroup whose cardinality is the same as that of G itself. (1) Let R be a domain (i.e. a ring with no zero divisors). If R ≤ G then the group ring RG is primitive. (2) If K is a field then KG is primitive. In particular, every group ring of the union of an ascending sequence of non-abelian free groups over a field is primitive, and so every group ring of a countable non-abelian locally free group over a field is primitive (see Corollary 4.6). Theorem 1.1 corresponds to the result obtained by Formanek [7]. He showed that every group ring RG of a free product G of non-trivial groups (except G = Z2 ∗ Z2) over a domain R is primitive provided that the cardinality of R is not larger than that of G. Motivated by the result, the primitivity of some interesting rings and algebras has been studied (for example, the primitivity of free products of algebras by Lichtman [13], the primitivity of group rings of amalgamated free products by Balogun [2] and the primitivity of semigroup algebras of free products by Chaudhry, Crabb and McGregor [5]). In their papers [2] and [13] (see also [12] and [16]), the method established in [7], which is based on the construction of comaximal ideals, has been applied to obtain the primitivity. This Formanek's method is also available for our study. 2 In the present paper, we state some graph-theoretic results and apply them to For- manek's method. For the sake of simplicity of the explanation, we consider here the group ring KG over a field K. For a non-zero element u in KG, let ε(u) be an element in the ideal KGuKG generated by u. Then Formanek's method says that KG is primitive, provided the right ideal ρ generated by the elements ε(u) + 1 for all non-zero u in KG is proper. The main difficulty here is how to choose elements ε(u)'s so as to make ρ be proper. That is, the chosen elements ε(u)'s must satisfy that any finite sum of the form r = P(ε(u) + 1)v is not the identity element in KG, where v's are elements in KG. In general, ε(u)'s and v's are linear combinations of elements of G; say f 's and g's are supports of ε(u) and v respectively. If r is not the identity element, then it has at least one support of the form f g or g. On the contrary, if r is the identity element, then almost all elements of the form αf g are vanished in r, where α is a non-zero coefficient in K. Then, what can we say about supports f 's of ε(u)'s? In order to consider this, regarding the elements of the form f g or g appeared in r as vertices and the equalities of their elements as edges, we use a graph-theoretic method. In section 3, we define an R-graph and an R-cycle, and show that under reasonable conditions, two typical R-graphs have an R-cycle (See Theorem 3.11 and Theorem 3.16). Roughly speaking, if a suitable R-graph for vertices f g's and g's has an R-cycle, then it follows that some supports fi's of ε(u)'s satisfy the equation f1f −1 supports fi's never satisfy such equation. n = 1. We should then choose ε(u)'s so that their · · · fn−1f −1 2 Now, if ϕ is an automorphism, that is ϕ(F ) = F , then the ascending HNN extension Fϕ is a cyclic extension of F . On the other hand, If ϕ(F ) 6= F , then it is a cyclic extension of a locally free group (see Lemma 2.3 (3)). Therefore, by applying Theorem 1.1 to this case, we can establish the primitivity of the group ring KFϕ: Theorem 1.2 Let F be a non-abelian free group. Then the following are equivalent: (1) KFϕ is primitive for a field K. (2) K ≤ F or Fϕ is not virtually the direct product F × Z. (3) K ≤ F or △(G) = 1, where △(G) is the FC center of G. In particular, if Fϕ is a strictly ascending HNN extension, that is, ϕ(F ) 6= F , then KFϕ is primitive for any field K. This extends the main result [16, Theorem 1.1], which was given for the countable case, to the general cardinality case, and follows the semiprimitivity of KFϕ with any 3 cardinality ( see Corollary 4.7 ). 2 PRELIMINARIES Let G be a group and N a subgroup of G. Throughout this paper, we denote by [G : N] the index of N in G. For a group H, G is said to be virtually H if H is isomorphic to N and [G : N] < ∞. If g is an element of G, we let CN (g) denote the centralizer of g in N. Let C(G) be the center of G and △(G) the FC center of G, that is △(G) = {g ∈ G [G : CG(g)] < ∞}. Given a set S, let S denote the cardinality of S. If S ⊆ G and S = {s1, · · · , sm}, hSi = hs1, · · · , smi denotes the subgroup of G generated by the elements of S. The method of Formanek [7] based on the construction of comaximal ideals plays also an important role in our study. We shall give it as follows: Let G be an infinite group, R a ring with identity, and X a set with X = G. Suppose that R ≤ G. Let ψ be a bijection from X to the elements of RG except for the zero element. For x ∈ X, let ε∗(ψ(x)) be a non-zero element in the ideal generated by ψ(x) in RG, and let ρ = Px∈X ε(ψ(x))RG be the right ideal of RG, where ε(ψ(x)) = ε∗(ψ(x)) + 1. Proposition 2.1 (See [7]) If ρ is proper then RG is primitive. We shall use some basic results on free groups in the last section. For the details, we refer the reader to Lyndon and Schupp [14]. The next assertions on locally free groups are almost obvious. For the sake of completeness, we include a proof. Lemma 2.2 Let G be a non-abelian locally free group. (1) The FC center of G is trivial; thus ∆(G) = 1. (2) Let S = {v1, · · · , vs, w1, · · · , wt} be a non-empty finite subset of G such that all elements in S are non-trivial, vi 6= vj and wi 6= wj if i 6= j. Then for each m > 0, there exist elements z1, · · · , zm ∈ G which satisfy (i) vkzlwizl = vhznwjzn if and only if (k, l, i) = (h, n, j), (ii) for p > 0, if Qp q ∈ {1, · · · , p − 1} or (lq, iq) = (nq, jq) for some q ∈ {1, · · · , p}. q=1(zlq wiqzlq )−1(znq wjqznq ) = 1, then either nq = l(q+1) for some Proof. Since G is a non-abelian locally free group, there exists a subset X ⊂ G with X > 1 such that hXi is freely generated by X in G and hXi ⊇ S. If v ∈ S then ChXi(v) 4 is cyclic (see [14, Proposition 2.19]), and so [hXi : ChXi(v)] is not finite, which implies v /∈ ∆(G). Hence we see that ∆(G) = 1 and thus (1) holds. Now, let x1, x2 ∈ X with x1 6= x2, and let nS be the maximum length of the words in S on X, where the length of a word v is defined for the reduced word equivalent to v on X. We set n = 2nS and zl = xn+l that the above z1, · · · , zm satisfy (i) and (ii). ✷ , where l = 1, 2, · · · , m. Then it is easily verified 1 x2xn+l 1 Let F be a non-abelian free group, and Fϕ = hF, tt−1f t = ϕ(f )i the ascending HNN extension of F determined by ϕ. It is easily verified that every element g ∈ Fϕ has a representation of the form g = tkf t−l where k, l ≥ 0 and f ∈ F . Combining this with some elementary observations on free groups, we can show the following properties of Fϕ: Lemma 2.3 (See [16, Lemma 2.1, 2.2]) Let F be a non-abelian free group. (1) ∆(Fϕ) = C(Fϕ). (2) The following are equivalent: (i) C(Fϕ) 6= 1. (ii) There exist n > 0 and f ∈ F such that C(Fϕ) = htnf i. (iii) Fϕ is virtually the direct product F × Z. When this is the case, ϕ is an automorphism of F ; thus ϕ(F ) = F . (3) For a non-negative integer i, let Fi be the subgroup of Fϕ generated by {tif t−i f ∈ F }. Then F1 ⊆ F2 ⊆ · · · ⊆ Fi ⊆ · · · is an ascending chain of free groups, and F∞ = i=1 Fi is a normal subgroup of Fϕ. S∞ The next two lemmas are basic results on group rings. We refer the reader to Passman [18] for a more detailed discussion of these questions. Lemma 2.4 Let K be a field, G a group and N a subgroup of G. (1)(See [20, Theorem 1]) Suppose that N is normal. If △(G) is trivial and △(G/N) = G/N, then KN is primitive implies KG is primitive. (2)(See [19, Theorem 3]) If △(G) is torsion free abelian and [G : N] is finite, then KN is primitive implies KG is primitive. Lemma 2.5 (See [17, Theorem 2]) Let K ′ be a field and G a group. If △(G) is trivial and K ′G is primitive, then for any field extension K of K ′, KG is primitive. 5 Formanek [7] asserts that if G is the direct product of a free group F and the infinite cyclic group hti, then KG is primitive if and only if the cardinality of K is not larger than that of F . Combining this with Lemma 2.3 and 2.4 (2), we have Lemma 2.6 (See [16, Theorem 1.1 (i)]) Let F be a non-abelian free group, and suppose that ϕ(F ) = F , that is, KFϕis the cyclic extension of F by hti. Then the following are equivalent: (1) KFϕ is primitive for a field K. (2) K ≤ F or Fϕ is not virtually the direct product F × Z. (3) K ≤ F or △(G) = 1. 3 GRAPHICAL INVESTIGATION Let KG be the group ring of a group G over a field K, and let a = Pm b = Pn i=1 αifi and If ab = 0 then for each figj, there exists fpgq such that figj = fpgq. Suppose that the following k equations hold; f1g1 = f2g2, i=1 βigi be in KG (αi 6= 0, βi 6= 0). f3g2 = f4g3, · · · , f2k−3gk−1 = f2k−2gk and f2k−1gk = f2kg1. Then we can regard the above equations as forming a kind of cycle, and they imply f −1 2k−1f2k = 1. That is, the above equations give us a information on supports of a. We can use this idea for a more 1 f2 · · · f −1 general case; a1b1 + · · · + anbn ∈ K for ai, bi ∈ KG with ai = P αijfij and bi = P βikgik. To do this, regarding the elements fijgik appeared in aibi as vertices and the equalities of their elements as edges, we use a graph-theoretic method. In this section, we shall define an R-graph and an R-cycle, and show that under reasonable conditions, two typical R-graphs have an R-cycle. One is called an R-colouring R-graph and another is called an R-simple R-graph which is a special case of an R- colouring. These two results, Theorem 3.11 and Theorem 3.16, are used to prove our main theorem in the next section. Throughout this section, G = (V, E) denotes a simple graph; a finite undirected graph which has no multiple edges or loops, where V is the set of vertices and E is the set of edges. For terminology and notations not defined here, we refer to Bondy and Murty [3]. A finite sequence v0e1v1 · · · epvp whose terms are alternately elements eq's in E and vq's in V is called a path of length p in G if vq−1vq = eq ∈ E and vq 6= vq′ for any q, q′ ∈ {0, 1, · · · , p} with q 6= q′; simply denoted by v0v1 · · · vp. Two vertices v and w of G are said to be connected if there exists a path from v to w in G. Connection is 6 an equivalence relation on V , and so there exists a decomposition of V into subsets Ci's (1 ≤ i ≤ m) for some m > 0 such that v, w ∈ V are connected if and only if both v and w belong to the same set Ci; each Ci is called a (connected) component of G. Any graph is a disjoint union of components. Definition 3.1 Let G = (V, E) and G∗ = (V, E∗) be simple graphs with the same vertex set V . For v ∈ V , let U(v) be the set consisting of all neighbours of v in G∗ and v itself: U(v) = {w ∈ V vw ∈ E∗} ∪ {v}. A triple (V, E, E∗) is an R-graph (for a relay-like graph) if it satisfies the following condition (R): (R) If v ∈ V and C is a component of G, then U(v) ∩ C ≤ 1. That is, each U(v) has at most one vertex from each component of G. If G has no isolated vertices, that is, if v ∈ V then vw ∈ E for some w ∈ V , then R-graph (V, E, E∗) is called a proper R-graph. We call U(v) the R-neighbour set of v ∈ V , and set U = {U(v) v ∈ V }. For v, w ∈ V with v 6= w, it may happen that U(v) = U(w), and so U ≤ V generally. If w, w′ ∈ U(v) then the minimum length of paths from w to w′ in G∗ is at most 2. Moreover, U(v) ∩ C > 1 for some v ∈ V and for some component C of G if and only if there exists a path from w to w′ for some w, w′ ∈ U(v) in G. Hence we have Proposition 3.2 In the definition 3.1, the condition (R) is equivalent to each of the following conditions: (R′) If C is a component of G and C ′ is a component of G∗ and if there exist v, w ∈ C ∩ C ′ with v 6= w, then the length of any path from v to w in G∗ is longer than 2. (R′′) If U ∈ U and v, w ∈ U, then there exist no paths from v to w; if v0v1 · · · vm is a path in G, then {v0, vm} 6⊆ U for any U ∈ U. By (R′), in particular, if vw ∈ E∗ then vw 6∈ E. We say G = (V, E) to be the base graph of R = (V, E, E∗). If E = ∅ then R is called the empty graph; denoted by R = ∅. Clearly, if R is non-empty then U > 1. In what follows, let R = (V, E, E∗) be a non-empty R-graph with G = (V, E) and G∗ = (V, E∗). For W ⊆ V , we define EW and E∗ W by EW = {ww′ w, w′ ∈ W and either ww′ ∈ E or wv1 · · · vmw′ is a path in G for some vi ∈ V \ W }, E∗ W = {ww′ w, w′ ∈ W and ww′ ∈ E∗}. 7 E∗ W is simply the edges of the subgraph of G∗ generated by W . EW means that if C is a component of G with C ∩W 6= ∅, then C ∩W also becomes a component of GW = (W, EW ). It is obvious that RW = (W, EW , E∗ W ) is an R-graph. We call RW the R-subgraph of R generated by W , and set UW = {UW (v) = U(v) ∩ W v ∈ W, U(v) ∈ U}. If EW coincides with {ww′ w, w′ ∈ W, ww′ ∈ E}, then RW is simply called the subgraph of R generated by W . The degree dW (v) of v ∈ W in RW is the number of edges of GW incident with v; dV (v) is simply denoted by d(v). For X ⊆ W ⊆ V , IW (X) denotes {x ∈ X dW (x) = 0}; IV (W ) is simply denoted by I(W ). In general, IW (W ) ≥ I(W ). In fact, we have Lemma 3.3 Let W ⊆ V and W c = V \ W . Then 0 ≤ IW (W ) − I(W ) ≤ W c − I(W c). The right side equality holds if and only if for each v ∈ W c \ I(W c), d(v) = 1 and there exists w ∈ W with d(w) = 1 such that vw ∈ E. Proof. Let X = IW (W ) \ I(W ) and Y = W c \ I(W c). Since 0 ≤ X is obvious, it suffices to show that X ≤ Y and the assertion on equality in the statement is true. Note that w ∈ X if and only if dW (w) = 0 and d(w) 6= 0. Therefore, if w ∈ X then there exists v ∈ Y such that vw ∈ E. Then vw′ 6∈ E for any w′ ∈ X with w′ 6= w. In fact, if vw′ ∈ E then ww′ ∈ EW , which implies a contradiction that w, w′ 6∈ IW (W ). Hence X ≤ Y . Since X < ∞ and Y < ∞, if X = Y , then for each w ∈ X ( resp. for each v ∈ Y ) there exists only one v ∈ Y ( resp. w ∈ X ) such that vw ∈ E. From this, if for w ∈ X, d(w) > 1 then ww′ ∈ E for some w′ ∈ W with w 6= w′, but this implies w, w′ 6∈ IW (W ), a contradiction. Hence d(w) = 1 for all w ∈ X. On the other hand, if d(v) > 1 for some v ∈ Y then vv′ ∈ E for some v′ ∈ Y with v 6= v′. For these v, v′ ∈ Y , as mentioned above, there exists w, w′ ∈ X with w 6= w′ such that vw, v′w′ ∈ E, which implies a contradiction w, w′ 6∈ IW (W ), because wvv′w′ is a path in R and so ww′ ∈ EW . We have therefore that d(v) = 1 for all v ∈ Y . The converse assertion on equality is obvious. ✷ In what follows, for π = v0v1 · · · vp a path in G, the origin v0 of π and the terminus vp of π are denoted by o(π) and t(π) respectively. 8 Definition 3.4 Let p > 1 and let πq be a path in G with vq = o(πq) and wq = t(πq) (1 ≤ q ≤ p). Then a sequence (π1, π2, · · · , πp) is an R-path of length p in R if it satisfies the following conditions (i) and (ii): (i) All of vq's and wq's are different from each other, (ii) wqvq+1 ∈ E∗ for 1 ≤ q ≤ p − 1. If, in addition, it satisfies the following condition (iii), then it is an R-cycle of length p in R: (iii) wpv1 ∈ E∗. In particular, if the length of πq is 1, that is, πq = eq ∈ E for all q, then (e1, e2, · · · , ep) is called an R-cycle consisting of edges. It is obvious that R has an R-cycle if the R-subgraph RW has an R-cycle for some ∅ 6= W ⊆ V . A proper R-graph R is called a clique R-graph, provided that the base graph G = (V, E) is a clique graph; thus uv, vw ∈ E implies uw ∈ E. Note that R is a clique R-graph if and only if every component of G is a complete graph. Hence, if a clique R-graph has an R-cycle then it also has an R-cycle consisting of edges. In what follows, N(R) = {U ∈ U U = 1}. We note, if a sequence (π1, · · · , πp) is an R-cycle in R, then neither U(o(πq)) nor U(t(πq)) is in N(R) for all 1 ≤ q ≤ p. We consider the following condition (UC) for U in R: (UC) For each U and U ′ in U, either U ∩ U ′ = ∅ or U ∩ U ′ > 1. If U satisfies (UC) then N(R) = ∅, because U = U ∩ U > 1 for each U ∈ U. Lemma 3.5 Let R be a proper R-graph. If U satisfies (UC) then R has an R-cycle. Proof. Let v1 ∈ V . Since R is proper, there exists w1 ∈ V such that e1 = v1w1 ∈ E. Since U(w1) > 1, there exists v2 ∈ V such that w1v2 ∈ E∗. Then v2v1 6∈ E∗ by (R). Since R is proper again, there exists w2 ∈ V such that e2 = v2w2 ∈ E, where w2 6= w1 and w2 6= v1 because of (R). We see then that (e1, e2) satisfies both of (i) and (ii) in Definition 3.4; thus it is an R-path in R. If U(w2) 6⊆ {v1, w1}, then we can proceed with this procedure. Since R is a finite graph, this procedure terminates in a finite number of steps, say, exactly p steps; that is, there exist p > 1 and a sequence σ = (e1, · · · , ep−1) of edges with eq = vqwq such that σ is an R-path in R, and in addition, there exists ep = vpwp ∈ E such that wp−1vp ∈ E∗ and U(wp) ⊆ {vq, wq 1 ≤ q ≤ p − 1}. 9 If there exists q ∈ {1 ≤ q ≤ p − 1} such that vq ∈ U(wp), then the sequence σ contains an R-cycle. In fact, if wp = vq for some q ∈ {1, · · · , p − 1} then q < p − 1 by (R), and then (πq, eq+1, · · · , ep−1) is an R-cycle, where πq = vpvqwq. If vq ∈ U(wp) \ {wp} for some q ∈ {1, · · · , p − 1}; thus wpvq ∈ E∗, then (eq, · · · , ep) is an R-cycle. Therefore, we may assume that for each q ∈ {1 ≤ q ≤ p − 1}, vq 6∈ U(wp), that is wpvq 6∈ E∗ (1 ≤ q ≤ p − 1). Then wq ∈ U(wp) for some q ∈ {1 ≤ q ≤ p − 1}. If wp = wq, (eq+1, · · · , ep) is an R-cycle because wpvq+1 = wqvq+1 ∈ E∗. We may assume therefore that wpwq ∈ E∗ and q is minimal with this property; thus q = min{1 ≤ q′ ≤ p − 1 wpwq′ ∈ E∗}. Note that q < p − 1 by (R). Since U(wp) ∩ U(vq+1) ⊇ {wq} 6= ∅, we have that U(wp) ∩ U(vq+1) > 1 by the hypothesis (UC). Since wpvq+1 6∈ E∗, there exists q′ ∈ {1 ≤ q ≤ p − 2} with q′ 6= q such that wpwq′ ∈ E∗ and vq+1wq′ ∈ E∗. By the minimality of q, q < q′, in fact, q + 1 < q′ by (R). Since wq′vq+1 ∈ E∗, we see then that (eq+1, · · · , eq′) is an R-cycle. ✷ In the above lemma, we cannot replace the condition (UC) by N(R) = ∅. For instance, we have the following example: Example 3.6 Let G = (V, E) be the base graph with V = {v1, · · · , v10} and E = {v1v3, v2v5, v4v7, v6v9, v8v10}. Let E∗ = {v1v2, v3v4, v4v5, v6v7, v7v8, v9v10}. Then we have that U(v1) = U(v2) = {v1, v2}, U(v3) = {v3, v4}, U(v5) = {v4, v5}, U(v4) = {v3, v4, v5}, U(v6) = {v6, v7}, U(v8) = {v7, v8}, U(v7) = {v6, v7, v8} and U(v9) = U(v10) = {v9, v10}. In this case, R = (V, E, E∗) is a non-empty R-graph and N(R) = ∅ but it has no R-cycles. Let Cn(V ) = {V1, · · · , Vn} be the set of components of G∗ = (V, E∗). For 1 ≤ i ≤ n and for v, v′ ∈ Vi, define v ≃ v′ by U o(v) = U o(v′), where U o(v) = U(v) \ {v}; thus U o(v) is the set of neighbours of v in G∗. Clearly, ≃ is an equivalence relation on Vi. If c(Vi) = {Vi1, · · · , Vili} is the set of equivalence class of Vi, then Vi is the disjoint union of non-empty Vij's. We can easily see that if v ∈ Vi, there exists S ⊆ c(Vi) \ {Vij} such that U o(v) = SW ∈S W for all v ∈ Vij. (1) In particular, for each v, w ∈ Vij, w 6∈ U o(v). If for each i ∈ {1, · · · , n} and for each v, v′ ∈ Vi, U(v) ∩ U(v′) 6= ∅, then Cn(V ) is called a colouring of R. Note that Cn(V ) is a colouring if and only if for each i and for each v, v′ ∈ Vi, there exists U ∈ U such that 10 v, v′ ∈ U, and so if Cn(V ) is a colouring then for each i and for each v, v′ ∈ Vi, vv′ 6∈ E by (R′′); thus, in this case, Cn(V ) is a colouring of the base graph G. We here consider the following condition (RC) for c(Vi) which is stronger than (1) above: (RC) For each v ∈ Vij, U o(v) = Vi \ Vij. That is, c(Vi) satisfies (RC) if and only if G∗(Vi, E∗ Vi) is a complete k-partite graph Kl1,··· ,lk, where k = c(Vi) and lj = Vij. Let consider the graph Gi = (c(Vi), E) with the vertex set c(Vi) = {Vi1, · · · , Vili} and the edge set E = {VijVik j 6= k, for v ∈ Vij, U o(v) ⊇ Vik}. If U o(v) ⊇ Vik for v ∈ Vij then U o(v′) ⊇ Vij for v′ ∈ Vik. Combining this with (1), we see that the above definition of E is well defined. Then, the neighbour set of Vij in Gi is c(Vi) \ {Vij} if and only if for each v ∈ Vij, U o(v) = Vi \ Vij in R, and therefore, c(Vi) satisfies (RC) if and only if Gi is isomorphic to the complete graph Kli on li vertices. Now, the definition of Cn(V ) implies that Gi is a connected graph, and the definition of c(Vi) implies that for each distinct vertices Vij, Vik in c(Vi), the neighbour set of Vij does not coincide with that of Vik. It easily follows from these above that Gi is isomorphic to the complete graph Kli, provided li ≤ 3. Hence we have Remark 3.7 If c(Vi) ≤ 3, then G∗ Vi = (Vi, E∗ Vi) is a complete k-partite graph; thus c(Vi) satisfies (RC). In Example 3.6, we can set that V11 = {v1}, V12 = {v2}, V21 = {v3, v5}, V22 = {v4}, V4 ≃ V31 = {v6, v8}, V32 = {v7}, V41 = {v9}, V42 = {v10} and Vi = Sj Vij. Then G∗ V1 ≃ G∗ K1,1 and G∗ V2 ≃ G∗ V3 ≃ K1,2. Let li = c(Vi). If li > 3, Gi need not be isomorphic to the complete graph Kli and thus c(Vi) need not satisfy (RC) (See Example 3.9 below). Our purpose is to make use of results on R-graphs for proving our main theorem in the next section. From this point of view, it suffices to consider the case when li = 3. However, in the following consideration, we only need the assumption (RC); we need not to assume the condition li ≤ 3. We therefore define Cn(V ) to be an R-colouring of R if for each i, c(Vi) satisfies (RC), and investigate when R-colouring R-graphs have an R-cycle. As a result, we can use R-graph theory to analyze more general case than the one of our main theorem (See Corollary 4.5). 11 Definition 3.8 Let R be an R-graph with Cn(V ) = {V1, · · · , Vn}. If for each 1 ≤ i ≤ n, G∗ Vi = (Vi, E∗ Vi) is a complete k-partite graph; thus c(Vi) satisfies (RC), then Cn(V ) is an R-colouring of R or R is an R-colouring R-graph with Cn(V ). If Cn(V ) is an R-colouring of R, then it is a colouring of R. As has been mentioned in Remark 3.7, in case of li = c(Vi) ≤ 3, an R-graph is always an R-colouring. However, in case of li > 3, it is not true. In fact, a colouring need not be an R-colouring. If Cn(V ) is a colouring of R, then each Vij, Vik ∈ c(Vi), there exists a path π in Gi = (c(Vi), E) whose length is 1 or 2 such that the origin o(π) = Vij and the terminus t(π) = Vik. Hence, for example, if li = 4, Gi is isomorphic to either the complete graph K4 or the graph described in the following example: Example 3.9 Let Vi = {v1, v2, v3, v4}, U(v1) = {v1, v2}, U(v2) = Vi, U(v3) = U(v4) = {v2, v3, v4}. In this case, we have that Vij = {vj} for 1 ≤ j ≤ 4; thus c(Vi) = 4. Then, v2 ∈ U(vj) for all 1 ≤ j ≤ 4; thus U(vp) ∩ U(vq) 6= ∅ for each p, q ∈ {1, 2, 3, 4}, but for v1 ∈ Vi1, U o(v1) = {v2} = Vi2 6= Vi \ Vi1. Hence, c(Vi) satisfies the colouring condition but it fails to satisfy (RC), and certainly, in the graph Gi = (c(Vi), E), we see that E coincides with {Vi1Vi2, Vi2Vi3, Vi3Vi4, Vi4Vi2}; thus Gi is not isomorphic to the complete graph K4. Let R be an R-colouring R-graph with Cn(V ), where Cn(V ) = {V1, · · · , Vn} and c(Vi) = {Vi1, · · · , Vili}. For W ⊆ Vi, We denote by m(W ) the maximum number in {W ∩ Vij 1 ≤ j ≤ li} and by JW the set {j W ∩ Vij 6= ∅}. In general, m(Vi) ≥ m(W ), and clearly, if m(W ) = 1 then G∗ W ) is a complete graph, that is, UW (v) = W for all v ∈ W , and also if JW > 2 then UW (v) ∩ UW (w) > 1 for all v, w ∈ W . Suppose W = (W, E∗ W > m(W ) + 1. Then li > 1 and JW > 1. In this case, if JW = 2, say JW = {1, 2}, and m(W ) = W ∩ Vi1, then m(W ) > 1 and W ∩ Vi2 = W − m(W ) > 1, which implies that UW (v) ∩ UW (w) > 1 for all v, w ∈ W . Hence we have Remark 3.10 Let W ⊆ Vi. (i) If m(W ) = 1 then UW (v) = W for all v ∈ W . (ii) If JW > 2 then UW (v) ∩ UW (w) > 1 for all v, w ∈ W . (iii) If W > m(W ) + 1 then UW (v) ∩ UW (w) > 1 for all v, w ∈ W . Recall that for W ⊆ V , I(W ) denotes {w ∈ W dV (w) = 0}. 12 Theorem 3.11 Let n > 1, and let R = (V, E, E∗) be an R-colouring R-graph with Cn(V ) = {V1, · · · , Vn}. Suppose that Vi ≥ 2m(Vi) + 1 for each i ∈ {1, · · · , n}. If I(V ) ≤ n then R has an R-cycle. Proof. Let W = V \ I(V ), Wi = Vi \ I(Vi) and mi = m(Vi) for i = 1, · · · , n. We prove the statement above by induction on n. First, let n = 2. By the assumption, Wi = Vi − I(Vi) ≥ 2mi + 1 − 2 ≥ mi > 0 (i = 1, 2), and so the R-subgraph RW ( in this case, it is simply the subgraph generated by W ) is non-empty; thus JWi > 0 for i = 1, 2. Moreover, W = V − I(V ) ≥ V − 2 ≥ 2 Xi=1 (2mi + 1) − 2 = 2(m1 + m2). (2) If JWi = 1 for i = 1, 2, then W ≤ m1 + m2, which contradicts (2) above, and so JWi > 1 for i = 1 or i = 2. If JW1 = 1 then JW2 > 1 and W1 < W2 because W1 ≤ m1 < m1 + 2m2 ≤ W2 by (2). Since RW is proper and W1 < W2, there exist v ∈ W1 and v1, v2 ∈ W2 with v1 6= v2 such that vv1, vv2 ∈ E. However, since R is an R-colouring R-graph and JW2 > 1, there exists U ∈ UW2 such that v1, v2 ∈ U; this contradicts (R). We see therefore that JWi > 1 for both i = 1 and i = 2, and also that d(v) = 1 for all v ∈ W and W1 = W2. Again by (2), we have that Wi ≥ m1 + m2. In particular, Wi ≥ 2 (i = 1, 2). If m1 = 1 or m2 = 1, say m1 = 1, then m(W1) = 1 and so UW (v) = W1 for all v ∈ W1 by Remark 3.10 (i). For v1 ∈ W1, there exists w1 ∈ W2 such that e1 = v1w1 ∈ E. Since JW2 > 1, there exists v2 ∈ W2 such that w1v2 ∈ E∗, and for this v2, there exists w2 ∈ W1 with w2 6= v1 such that e2 = v2w2 ∈ E. Certainly, (e1, e2) is an R-cycle because w2v1 ∈ E∗. In case of m1 > 1 and m2 > 1, since Wi ≥ m1 + m2 > m(Wi) + 1 for i = 1, 2, by virtue of Remark 3.10 (iii), UW satisfies the condition (UC). Hence, by Lemma 3.5, RW has an R-cycle and so does R. Suppose next that n > 2 and the statement holds for all numbers between 2 and n − 1. If UW satisfies the condition (UC), then it has an R-cycle by Lemma 3.5. We may assume therefore that UW fails to satisfy the condition (UC); thus there exists i such that UWi(v) ∩ UWi(w) ≤ 1 for some v, w ∈ Wi. By Remark 3.10, for such i, m(Wi) = 1 if and only if Wi = 1, and it holds that either JWi ≤ 1 and Wi = m(Wi) JWi = 2 and Wi = m(Wi) + 1. or (3) 13 In case that there exists i ∈ {1, · · · , n} such that Wi = m(Wi), say i = n, we consider the R-subgraph RV ′ with V ′ = V \Vn = V1∪· · ·∪Vn−1. By Lemma 3.3 and the assumption of the statement, IV ′(V ′) − I(V ′) ≤ Wn = m(Wn), I(V ′) = I(V ) − I(Vn) and I(Vn) = Vn − Wn ≥ (2mn + 1) − m(Wn), and so we have that IV ′(V ′) ≤ I(V ′) + m(Wn) = I(V ) − I(Vn) + m(Wn) ≤ n − (2mn + 1) + 2m(Wn) ≤ n − 1. By our induction hypothesis, RV ′ has an R-cycle. We may assume therefore that Wi > m(Wi) for all i ∈ {1, · · · , n}. When this is the case, m(Wi) > 0 for all i. Then, by (3), there exists i ∈ {1, · · · , n} such that Wi = m(Wi) + 1 and JWi = 2; say JWi = {1, 2}. In addition, in this case, as mentioned at (3) above, m(Wi) > 1 because of Wi > 1. We may here assume i = n. Let Wn = Wn1 ∪ Wn2 with Wn1 = {v1} and Wn2 = {v2, · · · , vq+1}, where q = m(Wn). Since Vn ≥ 2mn + 1 > Vn1 + Vn2, there exists k ∈ {1, · · · , ln} such that k 6= 1, 2 and I(Vn) ⊇ Vnk 6= ∅, where ln = c(Vn). We may assume k = 3. Let v0 ∈ Vn3 and set Xn = {v0, v1, v2} and V (1) = V1 ∪ · · · ∪ Vn−1 ∪ Xn. We should note V (1) < V because Wn2 = m(Wn) > 1. We consider here the R-subgraph RV (1). It is obvious that UXn satisfies (UC). Let Xn1 = {v1}, Xn2 = {v2} and Xn3 = {v0}. Then c(Xn) = {Xn1, Xn2, Xn3} satisfies the R-colouring condition (RC) because Xnj ⊆ Vnj and {Vn1, Vn2, Vn3} satisfies (RC). Hence Cn(V (1)) = {V1, · · · , Vn−1, Xn} is an R-colouring of RV (1). We set Yn = Vn \ Xn; thus Yn = V \ V (1). By Lemma 3.3, IV (1)(V (1)) ≤ I(V (1)) + Yn − I(Yn). Since I(V (1)) = I(V ) − (I(Vn) − 1) and Yn − I(Yn) = Wn − 2 = m(Wn) − 1, we have that IV (1)(V (1)) ≤ I(V ) − (I(Vn) − 1) + (m(Wn) − 1) ≤ n − I(Vn) + m(Wn). Moreover, because of Vn ≥ 2mn + 1, we see that I(Vn) = Vn − Wn ≥ (2mn + 1) − (m(Wn) + 1) ≥ mn, and hence, IV (1)(V (1)) ≤ n − mn + m(Wn) ≤ n. In addition, Xn ≥ 2m(Xn) + 1, in fact, m(Xn) = 1 and Xn = 3 = 2m(Xn) + 1. That is, RV (1) satisfy all of the conditions supposed for R in the statement. Let W (1) = W (1) i = n , where W (1) 1 ∪ · · · ∪ W (1) 14 i > m(W (1) n = {v1, v2}. If U Vi \ IV (1)(Vi) (i = 1, · · · , n − 1) and W (1) W (1) fails to the condition (UC) and W (1) ) for all i ∈ {1, · · · , n}, then we can proceed with this procedure, and get R-subgraphs RV (1), RV (2), · · · . On the other hand, V > V (1) > V (2) > · · · , and therefore, there exists p > 0 such that RV (p) satisfies either (UC) or W (p) = m(W (p) ) for some i ∈ {1, · · · , n}. In either case, we have already seen that RV (p) has an R-cycle. i i i ✷ In the above theorem, the assumption that Vi ≥ 2m(Vi) + 1 for each i ∈ {1, · · · , n} cannot be dropped. Let R = (V, E, E∗) be the R-graph which is described in Example 3.6, and let R′ = (V ′, E′, E∗′) be the R-graph with V ′ = V ∪ {w1, w2, w3, w4}, E′ = E and E∗′ = E∗ ∪ {viw1, vjw2, vkw3, vlw4 i = 1, 2, j = 3, 4, 5, k = 6, 7, 8, l = 9, 10}. j=1 V ′ = (V ′ 22 = {v4}, V ′ 11 = {v1}, V ′ 12 = {v2}, V ′ 13 = {w1}, V ′ 41 = {v9}, V ′ 32 = {v7}, V ′ 33 = {w3}, V ′ i = S3 21 = {v3, v5}, V ′ ij. That is, R′ is an R-colouring R-graph with C4(V ′) = {V ′ i , E∗′ V ′ i Then, V ′ 23 = {w2}, V ′ 31 = {v6, v8}, V ′ 43 = {w4} and V ′ 3, V ′ 4} and G∗ ) is a complete 3-partite graph. In addition, I(V ′) = {w1, w2, w3, w4} V ′ i and so I(V ′) = 4. Since m(V1) = m(V4) = 1 and m(V2) = m(V3) = 2, we see that Vi = 2m(Vi) + 1 for i = 1, 4 but Vi = 2m(Vi) for i = 2, 3. As has been pointed out, R is an R-colouring R-graph with C4(V ) and it has no R-cycles. Hence R′ has also no R-cycles, because V = V ′ \ I(V ′) in R′ and R′ 42 = {v10}, V ′ V is isomorphic to R. 1, V ′ 2, V ′ Now, let R be an R-colouring R-graph with Cn(V ) = {V1, · · · , Vn}. By Remark 3.10 (i), if m(Vi) = 1 for i ∈ {1, · · · , n}, U(v) = Vi for all v ∈ Vi; thus G∗ Vi) is a complete graph. Therefore, if m(Vi) = 1 for every 1 ≤ i ≤ n, then G∗ = (V, E∗) is a disjoint union of complete graphs and U coincides with the set of components of G∗; thus Vi = (Vi, E∗ U = Cn(V ). In such case, we define Cn(V ) to be a simple R-colouring of R. It is obvious that Cn(V ) is a simple R-colouring of R if and only if R is an R-graph satisfying the condition (SC): (SC) U, U ′ ∈ U =⇒ either U ∩ U ′ = ∅ or U = U ′. Definition 3.12 Let R = (V, E, E∗) be an R-graph. R is R-simple if the following conditions are satisfied: (i) R has a simple R-colouring; thus R satisfies (SC), (ii) there exist no R-cycles of length 2 consisting of edges; if there exists vw, v′w′ ∈ E such that vv′ ∈ E∗, then ww′ 6∈ E∗. 15 Let R be an R-simple R-graph with the set U of R-neighbour sets. By (i), we can define the graph whose vertex set V = U and whose edge set E = {UU ′ U, U ′ ∈ U, there exist v ∈ U and v′ ∈ U ′ such that vv′ ∈ E}; the graph (V, E) is denoted by R/U. Then V = U, and for U ∈ V, dV(U) = Pv∈U dV (v). Moreover, R/U has no loops by (R), and has no multiple edges by (ii); that is, R/U is a simple graph. We call R/U the induced simple graph of R. If R has an R-cycle then it induces the cycle in R/U. Conversely, if R/U has a cycle then the origin of it in R is either an R-cycle or a cycle in the base graph G = (V, E). Hence we have Lemma 3.13 Let R be an R-simple R-graph with the base graph G = (V, E) and the set U of R-neighbour sets. Suppose that G has no cycles. Then R has an R-cycle if and only if R/U has a cycle. Definition 3.14 Let R be an R-simple R-graph with the set U of R-neighbour sets. Then U and U ′ in U are said to be R-connected if there exists a finite sequence U0e1U1 · · · epUp whose terms are alternately R-neighbour sets Uq's and edges eq's in E with eq = vqwq such that U0 = U, Up = U ′, vq ∈ Uq−1, wq ∈ Uq. In an R-simple R-graph, 'R-connected' means simply 'connected' in the induced simple graph of it. Since R-connection is an equivalence relation on U, there exists a decomposi- tion of U into subsets Ui's (1 ≤ i ≤ m) for some m > 0 such that U, U ′ ∈ U are R-connected if and only if both U and U ′ belong to the same set Ui. The subgraphs RW1, · · · , RWm U for of R generated by Wi's are called the R-components of R, provided Wi = SU ∈Ui each i ∈ {1, · · · , m} and V = Sm If R has exactly one R-component then R is i=1 Wi. R-connected. Recall that N(R) = {U ∈ U U = 1}. We set L(R) = {U ∈ U U > 2} and M(R) = {U ∈ U U = 2}. Lemma 3.15 Let R = (V, E, E∗) be an R-simple R-graph with the set U of R-neighbour sets. Suppose that R is R-connected. Then R has an R-cycle if and only if V − U − ω + 1 6= 0, where ω is the number of components of G. In particular, if R is proper and L(R) ≥ N(R) then R has an R-cycle. Proof. Let Gi = (Vi, Ei) (i = 1, · · · , ω) be the components of the base graph G = (V, E). Since Gi is connected, there exists a spanning tree G′ i=1 E′ i, i = (Vi, E′ i) of Gi. Let E′ = Sω 16 G′ = (V, E′) and R′ = (V, E′, E∗). It is obvious that R′ is R-simple and R-connected. Moreover, R has an R-cycle if and only if so does R′. Since G′ has no cycles, by virtue of Lemma 3.13, R′ has an R-cycle if and only if the induced simple graph R′/U = (V′, E′) has a cycle. On the other hand, since R′ is R-connected, R′/U is connected, and so R′/U is a tree if and only if E′ = V′ − 1. (4) We set D = PU ∈V′ dV′(U) and di = Pv∈Vi dV (v), where dV (v) means the degree of v not in G but in G′. Recall that di = 2Vi − 2 because G′ i is a tree. Hence we have D = di = ω Xi=1 ω Xi=1 (2Vi − 2) = 2V − 2ω. Since V′ = U and generally D = 2E′, the condition (4) can be replaced by V − U − ω + 1 = 0. That is, R′/U has a cycle if and only if V − U − ω + 1 6= 0. Now, if R is proper, 2V − 2ω = D ≥ 3L(R) + 2M(R) + N(R) and U = L(R) + M(R) + N(R). Hence, in particular, if L(R) ≥ N(R), then V − U − ω + 1 ≥ 1 2 (L(R) − N(R)) + 1 > 0, and so R has an R-cycle. ✷ Theorem 3.16 Let R = (V, E, E∗) be an R-simple R-graph with the set U of R-neighbour sets. Then R has an R-cycle if and only if there exists an R-component RW = (W, EW , E∗ W ) of R with the set UW of R-neighbour sets such that W − UW − ω + 1 6= 0, where ω is the number of components of GW = (W, EW ). In particular, if R is proper and L(R) ≥ N(R) then R has an R-cycle. Proof. If there exists an R-cycle in R, then it exists in an R-component. Hence R has an R-cycle if and only if there exists an R-component RW of R which has an R-cycle. Let RW = (W, EW , E∗ W ) be an R-component of R with the set UW of R-neighbour sets, and let ω be the number of components of GW = (W, EW ). By Lemma 3.15, RW has an R-cycle if and only if W − UW − ω + 1 6= 0. Now, if L(R) ≥ N(R) then there exists an R-component RW such that L(RW ) ≥ N(RW ), and in addition, if R is proper then so is RW . Hence, in this case, RW has an R-cycle by Lemma 3.15. This completes the proof. ✷ 17 4 PROOF OF THEOREMS In what follows, let G be a non-abelian locally free group in which there exists a free subgroup H with basis X such that H = G. Since G is non-abelian, we may assume that H is non-abelian. If the rank of H is infinite then X = H. On the other hand, if the rank of H is finite then there exists a free subgroup of H whose rank is countable; for instance, the derived subgroup [H, H] of H is a free group of countable rank. Therefore, we may assume here that X = G. Let R (∋ 1) be a ring with no zero divisors. We suppose that R ≤ G. Since X = G ≥ ℵ0, we can divide X into three subsets X1, X2 and X3 each of whose cardinality is X. Let σi be a bijection from X to Xi (i = 1, 2, 3). For x ∈ X, x(i) denote the image of x by σi. Let RG be the group ring of G over R. Since RG = X, there exists a bijection ψ from X to RG \ {0}. Let x ∈ X, and let ψ(x) = mx Xi=1 αxifxi, where αxi ∈ R, fxi ∈ G, mx > 0, (5) each of them depends on x, and fxi 6= fxj if i 6= j. Since G is locally free, for the subset {x(1), x(2), x(3), fxi 1 ≤ i ≤ mx} of G, there exist elements zx1, zx2, zx3 in G which satisfy the assertions of Lemma 2.2 (2). We define ε(x) by ε(x) = = 3 3 Xk=1 Xk=1 3 3 Xl=1 Xl=1 x(k)zxlψ(x)zxl + 1 αxix(k)zxlfxizxl + 1. mx Xi=1 (6) Let ξx(k, l, i) = x(k)zxlfxizxl. Then the assertions of Lemma 2.2 (2) mean the the following: Remark 4.1 (i) ξx(k, l, i) = ξx(h, n, j) if and only if (k, l, i) = (h, n, j). (ii) Let p > 0, 1 ≤ lq, nq ≤ 3 and 1 ≤ iq, jq ≤ mx, where 1 ≤ q ≤ p. If q=1 ξx(1, lq, iq)−1ξx(1, nq, jq) = 1, then either nq = lq+1 for some q ∈ {1, · · · , p − 1} Qp or (lq, iq) = (nq, jq) for some q ∈ {1, · · · , p}. Let ρ be the right ideal of RG generated by all ε(x)'s, that is, ε(x)RG. ρ = Xx∈X (7) Let r = Pm there exist nt > 0, βtj ∈ R with βtj 6= 0 and gtj ∈ G such that rt = ε(xt)Pnt t=1 rt be a non-zero element of ρ, where 0 6= rt ∈ ε(xt)RG with xt ∈ X. Then j=1 βtjgtj with 18 gtj 6= gti (j 6= i). In what follows, we simply write mt, αti, xtk, ztl and fti instead of mxt, αxti, x(k) , zxtl and fxti, respectively. By the expression of ε(xt) as in (6), we have that t rt = 3 Xk,l=1 mt Xi=1 nt Xj=1 αtiβtjxtkztlftiztlgtj + nt Xj=1 βtjgtj, where mt, nt > 0. (8) To prove Theorem 1.1, by virtue of Proposition 2.1, we shall show that ρ is a proper right ideal of RG. By making use of graph-theoretic results obtained in the previous section, we shall prove r = Pm prepare the following notations. t=1 rt 6= 1. To connect the problem with our graphical method, we For 1 ≤ t ≤ m, let nt and mt be as described in (8). we set Pt = {(t, k, l, i, j) 1 ≤ k, l ≤ 3, 1 ≤ i ≤ mt, 1 ≤ j ≤ nt} and Qt = {(t, j) 1 ≤ j ≤ nt}. Moreover, for v = (t, k, l, i, j) ∈ Pt and w = (t, j) ∈ Qt, we set η(v) = xtkztlftiztlgtj and η(w) = gtj. Then we can replace the expression (8) of rt by the following expression: (9) (10) γwη(w), where γv = αtiβtj and γw = βtj. (11) γvη(v) + Xw∈Qt rt = Xv∈Pt t=1 Pt and Q = Sm Let P = Sm t=1 Qt. We regard W = P ∪ Q as the set of vertices and E = {vw v, w ∈ W, v 6= w and η(v) = η(w)} as the set of edges, and consider the R-graph R = (W, E, E∗), where vv′ ∈ E∗ if and only if v, v′ ∈ Pt such that v = (t, k, l, i, j) and v′ = (t, k′, l, i, j) with k 6= k′; thus U(v) = {(t, k′, l, i, j) 1 ≤ k′ ≤ 3} for v = (t, k, l, i, j) ∈ Pt, U(w) = {w} for w ∈ Qt and U = {U(v) v ∈ W } (in the proof of Theorem 1.1, in fact, the above vertices set W = P ∪ Q is replaced by V = P ∗ ∪ Q∗; see below for detail). We shall then show that there exist some isolated vertices in R which make r = 1 false. To do this, by making use of Theorem 3.11, we shall first show that there exists a suitable number of isolated vertices in the subgraph of G = (W, E) generated by Pt1 in Lemma 4.4 after preparing two remarks, where Pt1 = {v v = (t, 1, l, i, j) ∈ Pt}. Let Mt = {(l, i, j) 1 ≤ l ≤ 3, 1 ≤ i ≤ mt, 1 ≤ j ≤ nt}. For v = (t, k, l, i, j) ∈ Pt and µ = (l, i, j) ∈ Mt, we write v = (t, k, µ). Let µ = (l, i, j) and µ′ = (l′, i′, j′) be in Mt. If j = j′ then η(t, k, µ) = η(t, k′, µ′) if and only if (k, l, i) = (k′, l′, i′) by Remark 4.1 (i). If (l, i) = (l′, i′), then η(t, 1, µ) = η(t, 1, µ′) implies gtj = gtj ′, and so j = j′; thus µ = µ′. Therefore, if µ 6= µ′, then η(t, 1, µ) = η(t, 1, µ′) implies j 6= j′ and (l, i) 6= (l′, i′). Moreover, it is obvious that η(t, 1, µ) = η(t, 1, µ′) holds if and only if η(t, k, µ) = η(t, k, µ′) holds for all k ∈ {1, 2, 3}. Hence we have 19 Remark 4.2 Let t ∈ {1, · · · , m}, and let v, v′ ∈ Pt with v = (t, k, µ) and v′ = (t, k′, µ′), where µ = (l, i, j) and µ′ = (l′, i′, j′). (i) Suppose that η(v) = η(v′). If j = j′, then v = v′. (ii) Suppose that η(v) = η(v′). If k = k′ and either j = j′ or (l, i) = (l′, i′), then µ = µ′. (iii) η(t, k, µ) = η(t, k, µ′) holds for some k ∈ {1, 2, 3} if and only if it holds for any k ∈ {1, 2, 3}. For µ and µ′ in Mt, define the relation µ ∼ µ′ by η(t, 1, µ) = η(t, 1, µ′). It is obvious that ∼ is a equivalence relation on Mt. Let Ct(µ) be the equivalence class of µ ∈ Mt and let Nt = {µ ∈ Mt Ct(µ) = {µ}}. Since Nt ⊆ Mt and ν 6∼ ν′ for ν, ν′ ∈ Nt with ν 6= ν′, as a complete set of representatives for Mt/ ∼, we can choose a set Tt which satisfies Nt ⊆ Tt. For µ = (l, i, j) ∈ Tt, let γ(t,1,µ) = αtiβtj, where αtiβtj is as described in (11). We set γ∗ (t,1,µ) = Xµ′∈Ct(µ) γ(t,1,µ′) and M ∗ t = {µ ∈ Tt γ∗ (t,1,µ) 6= 0}. By the definition of M ∗ t and Remark 4.2 (i) (iii), we have Remark 4.3 Let t ∈ {1, · · · m} and k, k′ ∈ {1, 2, 3}. For µ, µ′ ∈ M ∗ η(t, k′, µ′). Then k = k′ if and only if µ = µ′. t , suppose η(t, k, µ) = Now in (11), replacing Pt by P ∗ t = {(t, k, µ) µ ∈ M ∗ t , 1 ≤ k ≤ 3}, we can use the following expression of rt: rt = Xv∈P ∗ t γ∗ vη(v) + Xw∈Qt γwη(w). (12) Lemma 4.4 Let nt = Qt be as described in (9), and let M ∗ Nt > nt for all t ∈ {1, · · · , m}. In particular, M ∗ t > nt. t and Nt as above. Then Proof. Since M ∗ t ⊇ Nt, it suffices to show that Nt > nt. Suppose, to the contrary, that Nt ≤ nt for some t ∈ {1, · · · , m}. If nt = 1 then Nt = Mt by Remark 4.2 (i), which implies Nt = 3mt ≥ 3 > 1 = nt, a contradiction. Hence we have nt > 1. Let G = (V, E) be the graph with the vertex set V = Mt and the edge set E defined by {vw v, w ∈ V, v 6= w, η(t, 1, v) = η(t, 1, w)}. 20 It is obvious that GW = (W, E) is a clique graph, where W = V \ Nt. We set Vjl = l=1 Vjl; thus V = Snt {(l, i, j) ∈ V 1 ≤ i ≤ mt} and Vj = S3 let G∗ = (V, E∗) = Snt Vj ) (j = 1, · · · , nt) be the complete 3-partite graph with the partite set {Vj1, Vj2, Vj3} and Vj . Since U(v) = Vj \ Vjl ∪ {v} for v ∈ Vjl and GW is a clique graph, by Remark 4.2 (i), R = (V, E, E∗) satisfies (R), and so R is an R-graph; in fact, it is a non-empty clique R-graph. In addition, since G∗ Vj is the complete 3-partite graph, R is an R-colouring R-graph with Cnt(V ) = {V1, · · · , Vnt}. Since mt > 0, we see then j=1 Vj. Let G∗ Vj = (Vj, E∗ j=1 G∗ that Vj = 3mt ≥ 2mt + 1 = 2m(Vj) + 1 for each j ∈ {1, · · · , nt}. Moreover, according to our hypothesis, Nt ≤ nt, that is, I(V ) ≤ nt. Hence, by virtue of Theorem 3.11, a clique R-graph R has an R-cycle consisting of edges. That is, there exist p > 1 and edges e1, · · · , ep ∈ E with eq = vqwq (1 ≤ q ≤ p) such that all of vq's and wq's are different from each other, wqvq+1 ∈ E∗ (1 ≤ q ≤ p − 1) and wpv1 ∈ E∗. Let vq = (lq, iq, jq) and wq = (l′ q), where 1 ≤ q ≤ p. Let ξt(1, lq, iq) = xt1ztlq ftiq ztlq . Then eq = vqwq ∈ E q, j′ q, i′ implies ξt(1, lq, iq)gtjq = η(t, 1, vq) = η(t, 1, wq) = ξt(1, l′ q, i′ q)gtj ′ q. Moreover, wqvq+1 ∈ E∗ and wpv1 ∈ E∗ mean that j′ q = jq+1, j′ p = j1, and l′ q 6= lq+1. Hence we have p Yq=1 ξt(1, lq, iq)−1ξt(1, l′ q, i′ q) = 1 with l′ q 6= lq+1 (1 ≤ q ≤ p − 1). Since vq 6= wq and η(t, 1, vq) = η(t, 1, wq) for all 1 ≤ q ≤ p, it follows from Remark 4.2 (ii) that (lq, iq) 6= (l′ q) for all 1 ≤ q ≤ p. However, this contradicts the assertion of Remark q, i′ 4.1 (ii). ✷ We are now in a position to prove Theorem 1.1. Proof. [Proof of Theorem 1.1] (1): Let ρ be as described in (7); a non-trivial right ideal of RG. By virtue of Proposition 2.1, it suffices to show that ρ is proper. Let r = Pm t=1 rt be as described in (11); a non-zero element of ρ. For t ∈ {1, · · · , m}, recall t=1P ∗ t t=1Qt) ∪ {w0}, where w0 = (0, 0). We define γw0 = −1 (∈ R) and η(w0) = 1 t , 1 ≤ k ≤ 3} and Qt = {(t, j) 1 ≤ j ≤ nt}. We set P ∗ = ∪m P ∗ and Q∗ = (∪m t = {(t, k, µ) µ ∈ M ∗ (∈ G) respectively. Then r = Pv∈P ∗ γ∗ vη(v) +Pw∈Q∗ γwη(w) + 1 by (12). In order to prove that ρ is proper, it suffices to show that r 6= 1. Suppose, to the contrary, that r − 1 = 0, that is, 21 Xv∈P ∗ γ∗ v η(v) + Xw∈Q∗ γwη(w) = 0. (13) Now, we set V = P ∗ ∪ Q∗ and let G = (V, E) be the graph whose vertices are the elements of V and whose edge set E is defined as E = {vw v, w ∈ V, v 6= w, η(v) = η(w) in G}. By (13), G is proper and it is a non-empty clique graph. Let E∗ = {vw v = (t, k, µ), w = (t, k′, µ) ∈ P ∗, k 6= k′}. By Remark 4.3, R = (V, E, E∗) satisfies (R′), and so R is an R-graph, and in fact, it is a non-empty clique R-graph. We shall show that R has an R-cycle. If v ∈ V , U(v) = ( {(t, k′, l, i, j) 1 ≤ k′ ≤ 3} if v = (t, k, l, i, j) ∈ P ∗ {v} if v ∈ Q∗ , and so U = {U(v) v ∈ V } satisfies the condition (SC). Hence, either R is R-simple or it has an R-cycle of length 2 consisting of edges. We may assume, therefore, that R is R-simple. If v ∈ P ∗ then U(v) = 3, and if v ∈ Q∗ then U(v) = 1. This means that L(R) = {U(v) v ∈ P ∗} and N(R) = {U(v) v ∈ Q∗}. By Lemma 4.4, M ∗ t > nt, which implies On the other hand, L(R) = (1/3)P ∗ = Pm t=1 M ∗ t=1 nt. t > Pm N(R) = Q∗ = Pm t=1 Qt + 1 = Pm t=1 nt + 1, and so L(R) ≥ N(R). Hence, by Theorem 3.16, a clique R-graph R has an R-cycle consisting of edges, as desired. By the definition of an R-cycle, there exist p > 1 and vq = (tq, kq, µq), wq = (tq, hq, µq) ∈ P ∗ with µq = (lq, iq, jq) ∈ M ∗ tq (1 ≤ q ≤ p) such that all of vq's and wq's are different from each other, wqvq ∈ E∗, eq = vqwq+1 ∈ E (1 ≤ q ≤ p − 1) and vpw1 ∈ E. Hence, we have η(vq) = η(wq+1) (1 ≤ q ≤ p − 1), η(vp) = η(w1) and kq 6= hq (1 ≤ q ≤ p). 22 (14) (15) For the sake of simplicity of notation, the subscript p + 1 means the subscript 1; we set wp+1 = w1, tp+1 = t1, jp+1 = j1 · · · . Since vq 6= wq+1, by Remark 4.2 (i), η(vq) = η(wq+1) implies either tq 6= tq+1 or jq 6= jq+1. (16) By (10), the definition of η(vq), η(vq) = xtq kqζq and η(wq+1) = xtq+1hq+1ζq+1, where ζq = ztq lqftq iq ztq lqgtq jq, and so (14) implies that p (xtq kq)−1xtq+1hq+1 = 1. Yq=1 (17) Recall that xtk's are elements in X which is a basis of a free group and that xtk = xt′k′ if and only if (t, k) = (t′, k′). Now, in (17), if tq 6= tq+1, it is obvious that xtq kq 6= xtq+1hq+1. If tq = tq+1 then jq 6= jq+1 because of (16), and so µq 6= µq+1. Since η(vq) = η(wq+1) by (14), Remark 4.3 implies kq 6= hq+1, and hence, xtq kq 6= xtq+1hq+1 again. Moreover, 6= xtq+1kq+1 by (15). Therefore it follows a contradiction that we have that xtq+1hq+1 Qp q=1(xtq kq)−1xtq+1hq+1 6= 1. This complete the proof of (1). (2): If K ′ is the prime field of K then K ′ ≤ G, and therefore K ′G is primitive by (1). Since ∆(G) = 1 by Lemma 2.2 (1), the conclusion follows from Lemma 2.5. ✷ Now, a ring R is called a (right) strongly prime ring if for each 0 6= α ∈ R, there exists a finite subset S(α) of R such that αS(α)β 6= 0 for all non-zero β ∈ R. S(α) is called a (right) insulator of α. For instance, domains and simple rings are strongly prime. Formanek's result [7, Theorem] on primitivity of RG for a domain R was generalized to one for a strongly prime ring R by Lawrence [12]. The same situation holds for the case of our theorem. Corollary 4.5 The assertion of Theorem 1.1 (1) holds also for a strongly prime ring R. Proof. Let ϕ(x) = Pmx i=1 αxifxi (x ∈ X) be as described in (5) and S(αx1) = {δxq 1 ≤ q ≤ dx} a right insulator of αx1. Going back to the beginning of this section, for {x(1), x(2), x(3), fxi 1 ≤ i ≤ mx}, there exist elements zxlq (1 ≤ l ≤ 3, 1 ≤ q ≤ dx) 23 which satisfy assertion of Lemma 2.2 (2). We replace (6) by ε(x) = = 3 3 Xk=1 Xk=1 3 3 Xl=1 Xl=1 dx dx Xq=1 Xq=1 x(k)zxlqψ(x)δxqzxlq + 1 αxiδxqx(k)zxlqfxizxlq + 1. mx Xi=1 Then (8) is replaced by rt = 3 dt Xk,l=1 Xq=1 mt Xi=1 nt Xj=1 αtiδtqβtjxtkztlqftiztlqgtj + nt Xj=1 βtjgtj, where mt, nt, dt > 0. (18) Let Atj = {q 1 ≤ q ≤ dx, αtiδtqβtj 6= 0 for some i}. Since S(αt1) is a right insulator of αt1, for each j ∈ {1, · · · , nt}, there exists q ∈ {1, · · · , dt} such that αt1δtqβtj 6= 0, and so Atj 6= ∅. For q ∈ Atj, let Btj(q) = {i 1 ≤ i ≤ mt, αtiδtqβtj 6= 0}. It is obvious that Btj(q) 6= ∅. The non-zero parts of (18) is here replaced by the following expression: 3 nt rt = αtiδtqβtjxtkztlqftiztlqgtj (19) nt Xk,l=1 Xj=1 Xq∈Atj Xi∈Btj (q) Xj=1 + βtjgtj, where nt > 0, Atj > 0, Btj(q) > 0. After this, we renumber the elements in Atj and Btj(q), and we can then follow the same proof as in Theorem 1.1 (1). We can summarize the procedure as follows: Let Atj = {q1, · · · , qatj }; atj = Atj, and tj = {1, · · · , 3atj} tjl = {1, · · · , mtjs}, where for 1 ≤ h ≤ 3, l = 3(s − 1) + h. We here replace Pt in for qs ∈ Atj, let Btj(qs) = {p1, · · · , pmtjs}; mtjs = Btj(qs). We set A∗ and B∗ (9) by Pt = {(t, k, l, i, j) 1 ≤ k ≤ 3, l ∈ A∗ tj, i ∈ B∗ tjl, 1 ≤ j ≤ nt}. Then η(t, k, l, i, j) = xtkzthqsftpizthqsgtj, where l = 3(s − 1) + h with 1 ≤ h ≤ 3 and pi ∈ Btj(qs). We also replace Mt by Mt = {(l, i, j) l ∈ A∗ tjl, 1 ≤ j ≤ nt}. Let R = (V, E, E∗) be as described in the proof of Lemma 4.4, where V = Mt as above. Then tj, i ∈ B∗ R is an R-colouring R-graph with Cnt = {V1, · · · , Vnt}, and the difference between this R l=1 Vjl and Vjl = mt there 3atj l=1 Vjl and Vjl = mtjs here. Since atj > 0 and mtjs > 0, we can easily see that Theorem 3.11 is also valid in this case and that the same assertion as Lemma 4.4 and the one in the proof of Lemma 4.4 is simply that Vj = S3 whereas Vj = S holds. The remains of the proof are the same as the proof of Theorem 1.1 (1). ✷ 24 i=1 Fi contains a free subgroup F with F = F∞. In fact, if either Fi ≤ ℵ0 for all i or Fi is a maximal cardinality for If F1 ⊆ F2 ⊆ · · · are free groups, then F∞ = S∞ some i, then the assertion is obvious. Hence, it suffices to consider the case that their cardinalities are not bounded above. We may then assume that Fi < Fi+1 for all i. Since each element of Fi is a product of finitely many basis elements of Fi+1, each Fi+1 can be written as a free product Gi+1 ∗ Hi+1, where Gi+1 and Hi+1 are free subgroups of Fi+1 with Fi ⊆ Gi+1 and Fi+1 = Hi+1. Then H2 ∗ H3 ∗ · · · is a free subgroup of F∞ with the same cardinality as F∞. Now, it is well known that a countable locally free group is the union of an ascending sequence of free subgroups. Hence, by Theorem 1.1 (2), we have Corollary 4.6 Let F1 ⊆ F2 ⊆ · · · ⊆ Fn ⊆ · · · be an ascending chain of non-abelian free groups, and F∞ = ∪∞ i=1Fi. Then the group ring KF∞ is primitive for any field K. In particular, every group ring of a countable non-abelian locally free group over a field is primitive. We are now in a position to prove easily Theorem 1.2: Proof. [Proof of Theorem 1.2] By virtue of Lemma 2.6, we may assume that ϕ(F ) 6= F . Then ∆(Fϕ) = 1 by lemma 2.3 (2). Let Fi be the subgroup of Fϕ generated by {tif t−i f ∈ F }, and F∞ = S∞ i=1 Fi. By lemma 2.3 (3), F∞ is a normal subgroup of Fϕ, and it is also a locally free group which is of type as described in Corollary 4.6. Hence, KF∞ is primitive by Corollary 4.6. It is obvious that Fϕ/F∞ is isomorphic to hti, and thereby, it follows from Lemma 2.4 (1) that KFϕ is primitive. ✷ Finally, we state the semiprimitivity of group rings of ascending HNN extensions of free groups, which extends [16, Corollary 3.7] to the general cardinality case: Corollary 4.7 Let F be a non-abelian free group, and Fϕ the ascending HNN extension of F determined by ϕ. If K is any field then the group ring KFϕ is semiprimitive. Proof. Let K ′ be the prime field of K. Since K ′ ≤ F , by virtue of Theorem 1.2, K ′Fϕ is primitive and so semiprimitive. As is well known, semiprimitive group rings are separable algebras, thus semiprimitivity of group rings close under extensions of coefficient fields, and therefore KFϕ is semiprimitive. ✷ 25 References [1] J. W. Anderson, Finite volume hyperbolic 3-manifolds whose fundamental group con- tains a subgroup that is locally free but not free Sci. Ser. A Math. Sci.(N.S), 8(1)(2002), 13-20 [2] B. O. Balogun, On the primitivity of group rings of amalgamated free products Proc. Amer. Math. Soc., 106(1)(1989), 43-47 [3] J. A. Boundy and U. S. R. Murty, Graph Theory with Application Macmillan, London, Elsevier, New York, 1979. [4] A. Borisov and M. Sapir, Polynomial maps over finite fields and residual finiteness of mapping tori of group endomorphisms Invent. Math., 160(2)(2005), 341-356 [5] M. A. Chaudhry, M. J. Crabb and M. McGregor, The primitivity of semigroup algebras of free products Semigroup Forum, 54(2)(1997), 221-229 [6] M. Feighn and M. Handel, Mapping tori of free group automorphisms are coherent Ann. Math., 149(1999), 1061-1077. [7] E. Formanek, Group rings of free products are primitive J. Algebra, 26(1973), 508-511 [8] B. Freedman and M. H. Freedman, Kneser-Haken finiteness for bounded 3-manifolds locally free groups, and cyclic covers, Topology, 37(1998), 133-147 [9] R. Geoghegan, M. L. Mihalik, M. Sapir and T. Wise, Ascending HNN extensions of finitely generated free groups are Hopfian Bull. London Math. Soc., 33(3)(2001), 292-298 [10] G. Higman, Almost free groups Proc. London Math.Soc., 1(3)(1951), 284-290 [11] R. P. Kent IV, Bundles, handcuffs, and local freedom Geom. Ded., 106(1)(2004), 145-159 [12] J. Lawrence, The coefficient ring of primitive group ring Canad. J. Math., 27(3)(1975), 489-494 [13] A. I. Lichtman, The primitivity of free products of associative algebras J. Algebra, 54(1)(1978), 153-158 26 [14] R. C. Lyndon and P. E. Schupp, Combinatorial Group Theory Ergeb. der Math. und ihrer Grenzgeb., 89, Springer, Berlin 1977 [15] B. Maskit, A locally free Kleinian group Duke Math. J., 50(1)(1983), 227-232 [16] T. Nishinaka, Group rings of proper ascending HNN extensions of countably infinite free groups are primitive J. Algebra, 317(2007), 581-592 [17] D. S. Passman, Primitive group rings Pac. J. Math., 47(1973), 499-506. [18] D. S. Passman, The algebraic structure of group rings Wiley-Interscience, New York, 1977. 2nd ed., Robert E. Krieger Publishing, Melbourne, FL, 1985. [19] A. Rosenberg, On the primitivity of the group algebra Can. J. Math., 23(1971), 536-540. [20] A. E. Zalesskii, The group algebras of solvable groups Izv. Akad. Nauk BSSR, ser Fiz. Mat., (1970), 13-21. 27
1611.04306	2	1611	2017-08-12T10:08:13	A new family of Poisson algebras and their deformations	[ "math.RA", "math.AG" ]	Let $\Bbbk$ be a field of characteristic zero. For any positive integer $n$ and any scalar $a\in\Bbbk$, we construct a family of Artin-Schelter regular algebras $R(n,a)$, which are quantisations of Poisson structures on $\Bbbk[x_0,\dots,x_n]$. This generalises an example given by Pym when $n=3$. For a particular choice of the parameter $a$ we obtain new examples of Calabi-Yau algebras when $n\geq 4$. We also study the ring theoretic properties of the algebras $R(n,a)$. We show that the point modules of $R(n,a)$ are parameterised by a bouquet of rational normal curves in $\mathbb{P}^{n}$, and that the prime spectrum of $R(n,a)$ is homeomorphic to the Poisson spectrum of its semiclassical limit. Moreover, we explicitly describe ${\rm Spec}\ R(n,a)$ as a union of commutative strata.	math.RA	math	A NEW FAMILY OF POISSON ALGEBRAS AND THEIR DEFORMATIONS C. LECOUTRE AND S. J. SIERRA Abstract. Let k be a field of characteristic zero. For any positive integer n and any scalar a ∈ k, we construct a family of Artin-Schelter regular algebras R(n, a), which are quantisations of Poisson structures on k[x0, . . . , xn]. This generalises an example given by Pym when n = 3. For a particular choice of the parameter a we obtain new examples of Calabi-Yau algebras when n ≥ 4. We also study the ring theoretic properties of the algebras R(n, a). We show that the point modules of R(n, a) are parameterised by a bouquet of rational normal curves in Pn, and that the prime spectrum of R(n, a) is homeomorphic to the Poisson spectrum of its semiclassical limit. Moreover, we explicitly describe Spec R(n, a) as a union of commutative strata. 7 1 0 2 g u A 2 1 ] . A R h t a m [ 2 v 6 0 3 4 0 . 1 1 6 1 : v i X r a 1. Introduction One of the fundamental notions of noncommutative algebraic geometry is that the noncommutative analogues of polynomial rings (or, alternatively, the coordinate rings of "noncommutative projective spaces") are algebras which are Artin-Schelter regular [AS87]. Constructing and, if possible, classifying, Artin-Schelter regular algebras of various global dimensions is thus one of the core problems in the subject. Within the class of quadratic Artin-Schelter regular algebras one also seeks to construct those which are Calabi-Yau in the sense of Ginzburg [Gin06]. In general, the problem of classifying Artin-Schelter regular algebras (or Calabi-Yau algebras) is unsolved, even for global dimension 4. In 2014, Brent Pym classified 4-dimensional Calabi-Yau algebras arising as deformation-quantisations C. He showed that there are six families of Calabi-Yau algebras of torus-invariant Poisson structures on A4 which arise this way; the most interesting example has generators x0, x1, x2, x3 and relations (1.1) [x0, x1] = 5x2 0 45 [x0, x2] = − 2 195 [x0, x3] = x2 0 + 5x0x1 45 2 x0x1 + 5x0x2 x2 0 − x0x1 + 3x0x2 + x2 1 2 3 [x1, x2] = − 2 [x1, x3] = 5x0x1 − 3x0x2 + 7x0x3 − 21 [x2, x3] = − 2 x0x2 − x0x3 + 77 2 77 2 x2 1 + x1x2 5 2 x1x2 + 7x1x3 − 3x2 2. In this paper we generalize Pym's example to arbitrary dimensions, and study the properties of the resulting algebras. In particular, we obtain new examples of Calabi-Yau algebras in all global dimensions ≥ 5. Pym's example comes from an action of the two-dimensional solvable Lie algebra on C[x0, x1, x2, x3], inducing a Poisson bracket. He shows, using a deformation formula of Coll, Gerstenhaber, and Giaquinto, that this Poisson bracket quantizes to give the algebra (1.1). We generalize Pym's methods to construct Date: August 15, 2017. 2010 Mathematics Subject Classification. Primary: 16S38; Secondary 14A22, 17B63, 16W70, 16S80. Key words and phrases. Artin-Schelter regular, Calabi-Yau algebra, Poisson algebra, semiclassical limit. The first author was partially supported by an LMS mobility grant. The second author was partially supported by EPSRC grant EP/M008460/1. 1 a family of algebras R(n, a), depending on a scalar a and an integer n ≥ 1. The R(n, a) are graded, quadratic, noetherian, AS-regular domains of global dimension n + 1 (Theorem 3.8, Proposition 3.15). When n = 3 and a = − 5 4 we obtain Pym's algebra (1.1). For another example, R(1, a) is isomorphic to the Jordan plane Chx, yi/(xy − yx − x2) (if a 6= 0) or to the polynomial ring C[x, y] (if a = 0). For each n, there is one algebra R(n, a) that is Calabi-Yau (Definition 4.14). We have: Theorem 1.2. (Corollary 4.17) The algebra R(n, a) is Calabi-Yau if and only if a = − (n + 2)(n − 1) 2(n + 1) . Example (1.1) is the algebra arising from Theorem 1.2 when n = 3. The examples for n ≥ 4 have not to our knowledge been studied before. Each R(n, a) induces a Poisson bracket {−,−}a on C[x0, . . . , xn] in the semiclassical limit. Let A(n, a) be the Poisson algebra (C[x0, . . . , xn],{−,−}a). In such a deformation-quantisation context it is often expected that the algebra R(n, a) and the Poisson algebra A(n, a) share similarities. For instance we make a detailed study of the prime spectrum Spec R(n, a) of R(n, a) and the Poisson prime spectrum PSpec A(n, a) of A(n, a) and show that they are homeomorphic (Theorem 7.1). We further investigate the structure of this space and prove: Theorem 1.3. (Corollary 7.6) Let n ≥ 3. The space Spec R(n, a) ∼= PSpec A(n, a) is a union of quasiprojective strata, and has dimension (n − 2) (n − 1) if a 6∈ Q if a ∈ Q.   Moreover we show that R(n, a) satisfies the Dixmier-Moeglin equivalence (Theorem 8.3) and, using a transfer result, we prove that A(n, a) satisfies the Poisson Dixmier-Moeglin equivalence (Theorem 8.6). Taking advantage of the fact that Spec R(n, a) ∼= PSpec A(n, a) we compute many examples of prime spectra of the R(n, a). We investigate various ring theoretic properties of the rings R(n, a) and A(n, a). For n ≥ 2 we show that R(n, a) ∼= R(n, b) if and only if A(n, a) is Poisson isomorphic to A(n, b) if and only if a = b (Theorem 5.1). We compute the skewfield of fractions of R(n, a) and show that it is isomorphic to a Weyl skewfield if and only if a ∈ Q (Theorem 8.11). We also show that the graded automorphism group of R(n, a) is isomorphic to the graded Poisson automorphism group of A(n, a). Finally, we compute the point modules of the R(n, a). We show: Theorem 1.4. (Theorem 6.1) Let n ≥ 1. For any a, the point schemes of R(n, a) are isomorphic. The point modules of R = R(n, a) are parameterised by the union C1∪···∪ Cn, where Ck is a rational normal curve of degree k in P(R∗ 1) ∼= Pn. As with Pym's original example, these curves correspond to various nice (equivariant with respect to the appropriate group action) ways to embed P1 in Symn(P1) ∼= Pn. The organisation of the paper is as follows. In Section 2 we define the R(n, a) and A(n, a), and in Section 3 we prove that A(n, a) is the associated graded ring of R(n, a) and that R(n, a) is an Artin- Schelter regular noetherian domain. We prove Theorem 1.2 in Section 4, and also calculate the graded automorphism group of R(n, a) and the graded Poisson automorphism group of A(n, a). We study when two R(n, a) are isomorphic in Section 5 and prove Theorem 1.3 in Section 6. In the final three sections we describe the (Poisson) prime and primitive ideals of R(n, a) and A(n, a). We prove Theorem 1.4 in Section 7. In Section 8 we show that R(n, a) satisfies the Dixmier-Moeglin equivalence describing primitive ideals, and that A(n, a) satisfies the related Poisson Dixmier-Moeglin equivalence. We also compute the fraction skewfield of R(n, a). Finally, in Section 9 we give many explicit examples of prime spectra. 2 Acknowledgements. We thank Michel Van den Bergh, Fran¸cois Dumas, Gene Freudenburg, and Diane Maclagan for helpful comments. We particularly thank Brent Pym for extremely useful discussion on a visit of the second author to Oxford in early 2016 (including suggesting that the algebras R(n, a) are all twists of each other). This project began when the first author was supported by an LMS mobility grant to visit the University of Edinburgh, and we thank the LMS and the University of Edinburgh for their support. We thank the anonymous referee for their helpful comments. 2. Notation and definitions Throughout we work over a field k of characteristic zero (in the introduction for simplicity we worked over C). In this section we define the algebras R(n, a) and the Poisson algebras A(n, a) that are the subject of the paper, and describe how they arise from k×-invariant actions of the two-dimensional solvable Lie algebra on polynomial rings. We begin by discussing such actions. Fix an integer n ∈ Z>0, and let ∆ be the "downward derivation" (where ∂i = ∂/∂Xi). For a0, . . . , an ∈ k, let Γ = Γ(a0, . . . , an) be the weighted Euler operator ∆ = X0∂1 + ··· + Xn−1∂n We are interested in when ∆ and Γ generate a copy of the two-dimensional solvable Lie algebra inside Derk(k[X0, . . . , Xn]). Γ =X aiXi∂i. Lemma 2.1. We have ∆Γ − Γ∆ = ∆ if and only if aj = a0 + j for j ∈ {0, . . . , n}. Proof. This follows from the computation (∆Γ − Γ∆)Xi+1 = (ai+1 − ai)Xi, so ∆Γ − Γ∆ = ∆ if and only if ai+1 = ai + 1 for all i. (cid:3) The importance of Lemma 2.1 is the following result of [Pym15], based on the universal deformation formula of [CGG89]. Proposition 2.2. ([Pym15, Lemma 3.3.]) Let A be a commutative k-algebra, and let ∆, Γ : A → A be k-derivations so that ∆ is locally nilpotent and [∆, Γ] = ∆. For k ∈ N, define (cid:0)Γ k! Γ · (Γ − 1)··· (Γ − (k − 1)). Then k(cid:1) = 1 ∞ f ∗ g = Xk=0 k∆k(f )(cid:18)Γ k(cid:19)(g) defines an associative product on A[] whose semiclassical limit as → 0 is the Poisson bracket {f, g} = ∆(f )Γ(g) − Γ(f )∆(g). Further, evaluating at a particular ∈ k gives an associative product The rings (A,∗) are isomorphic for any 6= 0. 1 ∗ : A ⊗k A → A. 1We note that our sign convention in Proposition 2.2 differs from that of Pym's original result. 3 Thus let A = k[X0, . . . , Xn]. For a ∈ k, define Xi=0 Γa = n Let ∗a : A ⊗k A → A be defined by (a + i)Xi∂i. ∞ (2.3) Xk=0 By Lemma 2.1 and Proposition 2.2, this defines an associative multiplication on A. Let {−,−}a : A ⊗k A → A be the associated Poisson bracket (2.4) f ∗a g = ∆k(f )(cid:18)Γa k (cid:19)(g). {f, g}a = ∆(f )Γa(g) − Γa(f )∆(g). Going forward, we define R(n, a) to be the associative algebra (A,∗a) and A(n, a) to be the Poisson algebra (A,{−,−}a). The goal of this paper is to study these two algebras. To end the section, we describe how the construction above relates to the canonical action of the standard Borel subgroup of PGL2(k) on P1. Let G =( a b c! ac 6= 0) / k× 0 be the standard Borel subgroup of PGL2(k). Note that G ∼= k ⋊ k× and is the subgroup of PGL2(k) that fixes the point ∞ = [1 : 0] ∈ P1. The group G acts on Symn(P1) ∼= Pn and thus on the homogeneous coordinate ring B(Pn,O(1)) ∼= k[X0, . . . , Xn]. Thus the two-dimensional solvable Lie algebra Lie G acts by derivations on k[X0, . . . , Xn] for all n. All of these actions are induced by taking symmetric powers of the standard action on V = k · {X, Y }. To see how this works explicitly, fix a ∈ k and let Γa = aX∂X + (a + 1)Y ∂Y and ∆ = X∂Y as above. Fix also n ≥ 1. If we set Xj = X n−jY j/j! then X0, . . . , Xn form a basis for Symn(V) with ∆(Xj) = Xj−1 and Γa(Xj) = ((n − j)a + j(a + 1))Xj = (na + j)Xj. We see that n'th symmetric power of the bracket {−,−}a on k[X, Y ] induces {−,−}na on k[X0, . . . , Xn]. Consider the surjection φ : k[X0, . . . , Xn] ։ k[X0, X1](n) induced from the Veronese embedding of P1 as a rational normal curve in Pn. The Veronese embedding is G-equivariant by construction, and from the discussion above we expect φ to be a Poisson homomorphism from A(n, a) → A(1, a/n)(n). We will see in Section 6 that this does happen, and furthermore that there is also a surjection R(n, a) → R(1, a/n)(n). 3. First properties Fix n ≥ 1 and a ∈ k, and let A = A(n, a) and R = R(n, a). In this section, we give an explicit presentation of R. We show that R is a noetherian Artin-Schelter regular domain of global dimension n + 1. Finally we study the localisations of A and R at the (Poisson) normal element X0. In the sequel, we will suppress the subscript a where it is clear from context, so will use ∗ to denote multiplication in R and {−,−} to denote the Poisson bracket on A. We use concatenation to denote the (commutative) multiplication in A. Let zj = a + j and set X−1 := 0. Since (cid:0)Γa ℓ (cid:1)Xj we deduce from (2.3) that, for all 0 ≤ i, j ≤ n, we have: Xi ∗ Xj = ℓ (cid:1)(Xj) = (cid:0)a+j ℓ(cid:19)Xi−ℓXj Xℓ=0(cid:18)zj (3.1) i and (3.2) {Xi, Xj} = zjXi−1Xj − ziXj−1Xi. 4 We will use two gradings on A: the standard degree grading d defined by d(Xi) = 1 for all i and the weight grading ǫ defined by ǫ(Xi) = i. For k ∈ N, let Ak be the d-homogeneous component of d-degree k, and let Ak be the ǫ-homogeneous component of ǫ-degree k. Let Aj k = Aj ∩ Ak. Note that we have (3.3) ∆(Aj k) ⊆ Aj−1 k , whereas Γa preserves both d-and ǫ-degree. It follows immediately that R is also d-graded. Thus the Hilbert series of R (with respect to d) is (3.4) hilb R = hilb A = 1 (1 − t)n+1 . Observe also that the Poisson bracket on A is d-homogeneous: {Ak, Aℓ} ⊆ Ak+ℓ. On the other hand, ǫ defines a filtration on R: Lemma 3.5. Let A = A(n, a) with Poisson bracket {−,−} and let R = R(n, a), with multiplication ∗. Let R≤k := Lℓ≤k Aℓ, considered as a subspace of R. Then R≤k ∗ R≤ℓ ⊆ R≤k+ℓ, so R is filtered by the R≤k. The associated graded algebra of R is naturally isomorphic as a graded algebra to A, and under this identification we have {gr f, gr g} = gr(f ∗ g − g ∗ f ). i=0 fi, g = Pℓ Proof. As graded vector spaces, since R = A certainly gr R = Lk R≤k/R≤k−1 ∼= Lk Ak = A. Let f = Pk i=0 gi ∈ A, where fi, gi ∈ Ai and fk, gℓ 6= 0. From (2.3) and (3.3) we have that the ǫ-degree k + ℓ components of both f ∗ g and g ∗ f are equal to fkgℓ = gℓfk so the multiplications on gr R and A agree. Finally, gr(f ∗ g − g ∗ f ) lies in ǫ-degree k + ℓ − 1 and is thus equal to ∆(fk)Γa(gℓ) − ∆(gℓ)Γa(fk) = {gr f, gr g}, as needed. (cid:3) Recall that a k-algebra R is strongly noetherian if R⊗k C is noetherian for any commutative noetherian k-algebra C. (For example, polynomial rings are strongly noetherian.) We have: Corollary 3.6. For any n ∈ Z>0 and a ∈ k, the ring R(n, a) is a strongly noetherian domain. Proof. This follows by standard arguments (see [MR01, Proposition 1.6.6, Theorem 1.6.9]) from the corresponding properties for A(n, a) = gr R(n, a). (cid:3) The next result is useful because it allows us to use inductive arguments to establish properties of the R(n, a) or A(n, a). Proposition 3.7. Suppose that n ≥ 1. Let A = A(n, a) and let R = R(n, a). Then (1) X0 is normal in R, and R/hX0i ∼= R(n − 1, a + 1). (2) X0 is Poisson normal in A, and A/hX0i is Poisson isomorphic to A(n − 1, a + 1). Proof. (1) That X0 ∗ R = R ∗ X0 = X0A is immediate from (3.1), and so X0 is normal. Let π : R → R/hX0i be the canonical map, and let Yi := π(Xi+1) for all 0 ≤ i < n. Clearly R/hX0i = R/X0 ∗ R may be identified with A/X0A = A/hX0i as a graded vector space, and this and R(n − 1, a + 1) are isomorphic as graded vector spaces. From (3.1), the multiplication ∗ on R/hX0i satisfies Xℓ=0(cid:18)a + j + 1 ℓ which is precisely the multiplication on R(n − 1, a + 1). Yi ∗ Yj = (cid:19)Yi−ℓYj, i Note that the isomorphism R/hX0i ∼= R(n − 1, a + 1) respects both the d-grading and the ǫ-filtration on R. 5 For (2), it is clear that X0 is Poisson normal. Since A/hX0i is the associated graded of the ǫ-filtration on R/hX0i ∼= R(n − 1, a + 1), the remaining statement follows immediately from Lemma 3.5. (cid:3) We next prove that the algebras R(n, a) have good homological properties. Recall that an N-graded k-algebra R with R0 = k is Artin-Schelter regular or AS-regular if: (1) gldim R < ∞; (2) R has finite Gelfand-Kirillov-dimension; if i 6= gldim R if i = gldim R. (3) Exti Rk[ℓ] 0 . R(kR, RR) ∼=  (Here k[ℓ] means that the module is degree-shifted by some amount ℓ ∈ Z.) The Artin-Schelter regular condition is a noncommutative analogue of the good properties of commutative polynomial rings [AS87]. Condition (3) above is called the AS-Gorenstein condition. Theorem 3.8. Fix n ∈ Z>0 and let a ∈ k. The algebra R = R(n, a) is Artin-Schelter regular of global dimension n + 1. Proof. We prove by induction on n that R is AS-regular, Auslander-Gorenstein, and Cohen-Macaulay. We do not give the definitions of Auslander-Gorenstein or Cohen-Macaulay; the unfamiliar reader may treat them as technical terms internal to this proof. It is well-known that the Jordan plane R(1, a) and the polynomial ring R(1, 0) satisfy all of the above properties, so that the base case n = 1 is clear. Thus we may assume that n > 1. By Lemma 3.5 A(n, a) is the associated graded ring of R under the ǫ-filtration. Thus by [MR01, Corollary 7.6.18], gldim R ≤ gldim A(n, a) = n + 1. By Proposition 3.7 and [MR01, Theorem 7.3.5], gldim R ≥ gldim R(n − 1, a + 1) + 1. This last is n by induction. Thus gldim R = n + 1. By induction, R(n−1, a+1) is Auslander-Gorenstein and Cohen-Macaulay. By [Lev92, Theorem 5.10], the same holds for R. By [Lev92, Theorem 6.3] and (3.4), R is AS-Gorenstein and thus AS-regular. (cid:3) We now compute the relations for R. We will use the Vandermonde identity: for all a, b ∈ k and k Xu=0(cid:18)a u(cid:19)(cid:18) b k − u(cid:19) =(cid:18)a + b k (cid:19). k ∈ N, we have (3.9) We will also use the following lemma: Lemma 3.10. Let V = A1 = spank(X0, . . . , Xn). For b ∈ k, define a linear map φb : V → V by φb(Xi) = Then φaφb = φa+b. Proof. By (3.9) we have: i Xℓ=0(cid:18)b ℓ(cid:19)Xi−ℓ. φaφb(Xj ) = φa j Xi=0(cid:18)b i(cid:19)Xj−i! = = j Xi=0(cid:18)b Xk=0 k i(cid:19) j−i Xℓ=0(cid:18)a Xu=0(cid:18)b u(cid:19)(cid:18) a ℓ(cid:19)Xj−i−ℓ! k − u(cid:19)! Xj−k = j j Xk=0(cid:18)a + b k (cid:19)Xj−k = φa+b(Xj). (cid:3) The following lemma gives us quadratic relations that are satisfied in R. We also give an equation that reverses the deformation formula (3.1) and thus allows us to obtain the commutative product from the noncommutative product ∗. Recall that we set zj = a + j. 6 i j i i (3.12) and (3.13) XiXj = Xi ∗ Xj + ℓ (cid:19)Xi−ℓ ∗ Xj = Xℓ=1(cid:18)−zj Xk=1(cid:18)−zi Proof. Fix j, and recall the definition of the linear map φb from Lemma 3.10. Since for any F ∈ R1 we have F ∗ Xj = φzj (F )Xj, clearly XiXj = φ−1 zj (Xi) ∗ Xj. By Lemma 3.10, we immediately obtain (3.12). Xℓ=0(cid:18)−zj Xℓ=1(cid:18)−zj k (cid:19)Xj−k ∗ Xi − ℓ (cid:19)Xi−ℓ ∗ Xj. ℓ (cid:19)Xi−ℓ ∗ Xj Xi ∗ Xj − Xj ∗ Xi = Applying (3.12) to the equation XiXj = XjXi, we obtain (3.13). Lemma 3.11. For all 0 ≤ i, j ≤ n we have (cid:3) Note that relations (3.13) can be rewritten as: (3.14) We now have: j Xk=0(cid:18)−zi k (cid:19)Xj−k ∗ Xi = i Xℓ=0(cid:18)−zj ℓ (cid:19)Xi−ℓ ∗ Xj. Proposition 3.15. The relations in R = R(n, a) are exactly the (cid:0)n+1 2 (cid:1) relations given by (3.13) for 0 ≤ i < j ≤ n. Proof. By comparing dim R1 ⊗k R1 and dim R2 it is clear that R satisfies (cid:0)n+1 2 (cid:1) quadratic relations; since the relations in (3.13) are linearly independent for 0 ≤ i < j ≤ n they are precisely the quadratic relations of R. By Theorem 3.8 R is AS-regular, and by (3.4) it is what is referred to as a quantum Pn in [ST01]. Thus by [ST01, Theorem 2.2], R is Koszul and in particular is given by quadratic relations. Thus the relations in (3.13) are precisely the relations of R. (cid:3) Example 3.16. When n = 1 we obtain: {X0, X1} = −aX 2 0 and X0 ∗ X1 − X1 ∗ X0 = −aX0 ∗ X0. For a 6= 0 the algebra R(1, a) is isomorphic to the well-known Jordan plane, and the Poisson algebra A(1, a) is isomorphic to the Poisson-Jordan plane: the polynomial algebra k[X0, X1] endowed with Poisson bracket {X0, X1} = X 2 0 . Example 3.17. When n = 2 the algebra R(2, a) is given by generators X0, X1, X2 and relations: (3.18) X0 ∗ X1 − X1 ∗ X0 = −aX0 ∗ X0, X0 ∗ X2 − X2 ∗ X0 = −aX0 ∗ X1 −(cid:18)a X1 ∗ X2 − X2 ∗ X1 = (a + 2)X0 ∗ X2 − (a + 1)X1 ∗ X1 +(cid:18)a + 2 2(cid:19)X0 ∗ X0, 2 (cid:19)X0 ∗ X1. The Poisson bracket on A(2, a) is given by: (3.19) {X0, X1} = −aX 2 0 , {X0, X2} = −aX0X1, {X1, X2} = (a + 2)X0X2 − (a + 1)X 2 1 . Example 3.20. When n = 3 and a = −5/4 we obtain an algebra isomorphic to Pym's example (1.1). The isomorphism sends Xi 7→ xi/4i for i = 0, . . . , 3. We now prove a technical result which gives equivalent conditions for an element to be normal (or Poisson normal), under mild conditions on a and n. In particular we show that, up to multiplication by nonzero scalars, X0 is the only homogeneous element of d-degree 1 that is (Poisson) normal. 7 Proposition 3.21. Assume either that n ≥ 2 or that n = 1 and a 6= 0. Then: (1a) N is Poisson normal in A = A(n, a) ⇐⇒ (1b) N is normal in R = R(n, a) ⇐⇒ (1c) we have ∆(N ) = 0 and Γa(N ) = uN for some u ∈ k. (2a) N is Poisson central in A ⇐⇒ (2b) N is central in R ⇐⇒ (2c) we have ∆(N ) = Γa(N ) = 0. Proof. Without loss of generality, n > 0. The implications (1c) ⇒ (1a), (1b) and (2c) ⇒ (2a), (2b) are clear from (2.3). To prove the other implications, we first prove: Claim: There is an irreducible G ∈ A with ∆(G) = 0 and Γ(G) = λG for some λ 6= 0, and further such that hGi = G ∗ R = R ∗ G is a completely prime ideal of R. isomorphism R/hX0i ∼= R(n − 1, a + 1). To prove the claim, if a 6= 0 then we may take G = X0, so λ = a. The last statement follows from the 2 X 2 If a = 0 and n ≥ 2 then let G = X0X2− 1 1 and λ = 2. We have ∆(G) = 0 and Γa(G) = 2(a+ 1)G = λG. Note that G is normal in R and ǫ-homogeneous and so the ǫ-filtration on R descends to R/hGi. It is clear that gr(R/hGi) is isomorphic to A/hGi which is a domain, so G ∗ R is completely prime. We now prove (1a) ⇒ (1c). Let G, λ be as in the Claim. Suppose that N is Poisson normal in A. Without loss of generality we can assume that N /∈ GA and N 6∈ X0A since G and X0 are Poisson normal. Because {G, N} = −λG∆(N ) ∈ N A, there exists G′ ∈ A such that G∆(N ) = G′N . Since N /∈ GA and G is irreducible there exists v ∈ A such that G′ = vG and we obtain ∆(N ) = vN after dividing by G. Since ∆ is d-homogeneous we must have v ∈ A0 = k. But ∆ is locally nilpotent so we must have v = 0 and ∆(N ) = 0. There therefore exists X ′ 1 ∈ A such that: X ′ 1N = {X1, N} = X0Γa(N ). Again since N /∈ X0A we can write X ′ Γa being d-homogeneous, we conclude as before that u ∈ A0 = k is a scalar. 1 = X0u for some u ∈ A, and we have Γa(N ) = uN . The derivation We now prove (1b) ⇒ (1c). Suppose that N is normal in R. Without loss of generality we can assume that N /∈ X0 ∗ R and N 6∈ G ∗ R since X0 and G are normal. By normality of N there exists G′ such that G∗ N = N ∗ G′. Since N /∈ G∗ R we must have G′ ∈ G∗ R as G∗ R is completely prime. Moreover, using the equation G ∗ N = N ∗ G′ we deduce that G′ is d-homogeneous and d(G′) = d(G). So G′ = wG for a scalar w, and we conclude that N must satisfy the equation G ∗ N = wN ∗ G. Using the definition of the product ∗ we obtain the following equation after dividing by G: ∆2(N ) + ··· +(cid:18)λ (3.22) w(cid:18)N + λ∆(N ) + for some k ≥ 0. We decompose N = Pℓ we deduce from equation (3.22) that Nℓ = wNℓ, i.e. w = 1 and ∆(Nℓ) = 0. Finally we obtain by (a decreasing) induction that ∆(Ni) = 0 for all i, so that N ∈ ker ∆. By normality of N there exists F ∈ R such that X1 ∗ N = N ∗ F . Since ∆(N ) = 0 we have λ(λ − 1) i=0 Ni ∈ L Ai into ǫ-homogeneous pieces. Since ∆(Ai) ⊆ Ai−1 k(cid:19)∆k(N )(cid:19) = N 2 N F = N ∗ F = X1 ∗ N = X1N + X0Γa(N ), so N divides X0Γa(N ) ∈ A. As N 6∈ X0 ∗ R = X0A we have that N divides Γa(N ), so there exists u ∈ A with Γa(N ) = uN . Comparing d-degrees we have u ∈ A0 = k. For the remaining equivalences note that if N is either Poisson central in A or central in R then certainly N is (Poisson) normal. We have seen that ∆(N ) = 0 and Γa(N ) = uN for some u ∈ k. We then have either: 0 = {X1, N} = uX0N (if N is Poisson central) or 0 = X1 ∗ N − N ∗ X1 = uX0N (if N is central). In either case u = 0. Thus (2a) ⇒ (2c) and (2b) ⇒ (2c). Remark 3.23. We note that Proposition 3.21 does not make it easy to find the normal (or central) elements of R. In particular, it is a famously difficult problem in symbolic dynamics to calculate A∆ = {f ∈ A ∆(f ) = 0}; see [Fre13]. In fact, A∆ is not known explicitly for n ≥ 9. (cid:3) 8 0 ] and A◦ := A[X −1 The rings R(n, a) are complicated to study but they become much simpler after localisation at the normal element X0. To end this section we study the localisations of R = R(n, a) and A = A(n, a) at the (Poisson) normal element X0. Let R◦ := R[X −1 0 ]. We note that ∆ and Γa extend to A◦, and that the Poisson bracket {−,−}a extends to A◦ and is still defined by (2.4). Likewise, the multiplication ∗a on R◦ is still defined by (2.3). The d-grading extends to A◦ and to R◦. The ǫ-grading extends to A◦, and ǫ defines a filtration on R◦ with associated graded A◦, as with the non-localised rings. We further define a new grading A◦ = Le∈k A◦(e) by defining A◦(e) to be the e-eigenspace of Γa. d, then F ∈ A◦(e) for Notice that A◦(e) 6= 0 if and only if e ∈ aZ + N. Further note that if F ∈ (A◦)ǫ e = da + ǫ. It follows from (3.3) that A◦ is also the associated graded ring of R◦ with respect to the e-filtration, and that the ǫ- and e-filtrations on R◦ both induce the Poisson bracket {−,−}a on A◦. In contrast to A∆, which is unknown in general, B is easy to compute and is Let B = (A◦)∆. isomorphic to a polynomial ring in n variables. Explicitly, for j ≥ 1 define ∞ Yj = Xp=0 (−X1)p p!X p 0 ∆p(Xj). This is well-defined since ∆ is locally nilpotent. Note that Yj ∈ (A◦)j with e(Yj) = e(Xj) = zj. 1; in particular, Yj is e-homogeneous, Lemma 3.24. We have that A◦ is freely generated by X1, Y2, . . . , Yn and B is freely generated by Y2, . . . , Yn as commutative algebras over k[X ±1 0 ]. Proof. That B is generated by X ±1 1. If i ≥ 2, then (3.25) Xi − Yi ∈ k[X ±1 0 , X2, . . . , Xi−1]. 0 , Y2, . . . , Yn follows from [vdE93, Proposition 2.1], since ∆(X1/X0) = This shows that the elements X0, X1, Y2, . . . , Yn are algebraically independent, proving that B is freely generated as claimed. By (3.25) and induction, A◦ is generated over B by X1, and thus over k[X ±1 Kdim A◦ = n + 1, thus A◦ is freely generated as well. 0 ] by X1, Y2, . . . , Yn. Since (cid:3) Lemma 3.26. Let δ = X0Γa, so δ ∈ Derk(B). Define a map β : B[z] → A◦ by β(P bizi) = P biX i Then β is a ring isomorphism B[z; δ] → R◦ and a Poisson isomorphism (B[z],{−.−}δ) → A◦. Proof. Since B is an e-graded subring of A◦, therefore δ(B) ⊆ B and so δ ∈ Derk(B). Let ⋆ be the multiplication in B[z; δ]. Note that if we put e(z) = a + 1 then β is clearly e-graded as an isomorphism of vector spaces and e-filtered as a map from B[z; δ] → R◦. Since A◦ is the associated graded of R◦ it suffices to prove that β : B[z; δ] → R◦ is a ring homomorphism. It is enough to check that β(z ⋆ g) = β(z) ∗a g for g ∈ B(e). But we have 1. β(z) ∗a g = X1 ∗a g = gX1 + egX0 = β(gz + δ(g)) = β(z ⋆ g) as needed. (cid:3) 4. Graded automorphisms and the Nakayama automorphism In this section we compute the graded automorphism group of R(n, a) and determine its Nakayama automorphism. In particular we prove Theorem 1.2. We begin with the graded automorphism group of R(n, a). In order to prove that the maps φb defined in 3.10 are well-defined automorphisms of R(n, a) for any a, b ∈ k we use the theory of Zhang twists. More specifically we prove that R(n, a) and R(n, b) are Zhang twists of each other for any a, b ∈ k. We now recall the definition of Zhang twist from [Zha96]. If S is an N-graded ring with multiplication ∗, and φ ∈ Aut (S) is a graded automorphism of S, then the Zhang twist of S by φ is written Sφ. As a 9 i = Si and s ∈ Sφ graded vector space, Sφ is isomorphic to S. The multiplication ◦ on Sφ is defined by r ◦ s = r ∗ φi(s) for all r ∈ Sφ j = Sj. This is associative by [Zha96, Proposition 2.3]. In the terminology of [Zha96] the family of maps {φi i ∈ N} is a twisting system of S. We note that the definition in [Zha96] is slightly more general than the one we give here, but we do not need this greater generality. Also recall that we may define a left-hand Zhang twist of S by φ, which we write φS. The multiplication ◦′ on φS is defined by r ◦′ s = φj (r) ∗ s for all r ∈ φSi = Si and s ∈ φSj = Sj. This is associative by [Zha96, Proposition 4.2]. We have the following easy lemma. Lemma 4.1. Let S be an N-graded ring with multiplication ∗, and let φ ∈ Aut (S) be a graded automor- phism of S. Then Sφ ∼= φ−1 Proof. Define Φ : Sφ → φ−1 S. S by Φ(t) = φ−k(t) for all t ∈ Sφ k . We denote the multiplication on Sφ by ◦ and the multiplication on φ−1 S by ◦′. Then for r ∈ Si, s ∈ Sj, we have since φ is an algebra automorphism of S. Thus Φ(r ◦ s) = Φ(r ∗ φi(s)) = φ−(i+j)(r ∗ φi(s)) = φ−i−j(r) ∗ φ−j(s) as needed. (cid:3) Φ(r) ◦′ Φ(s) = φ−j (φ−i(r)) ∗ φ−j(s) = Φ(r ◦ s), The first goal of this section is to prove the following theorem Theorem 4.2. Fix n ∈ Z≥0. For any a, b ∈ k, the rings R(n, a) and R(n, b) are right and left Zhang twists of each other. We will prove this by constructing some explicit graded automorphisms of R(n, a). Lemma 4.3. For any a ∈ k, the map φa(Xj) = j Xi=0(cid:18)a i(cid:19)Xj−i induces a d-graded automorphism of R(n, a), which we also denote φa. Proof. Note that Thus φa is simply the automorphism of R(n, a) induced by conjugating by the normal element X0. (cid:3) Xj ∗ X0 = φa(Xj)X0 = X0φa(Xj) = X0 ∗ φa(Xj). Proposition 4.4. For any a ∈ k, let φa be the automorphism of R(n, a) defined in Lemma 4.3. Then R(n, 0) ∼= R(n, a)φa, under the map induced by sending Xi 7→ Xi. Proof. By Lemma 4.1, and Lemma 3.10, it suffices to prove that R(n, 0) ∼= φ−aR(n, a). Let ◦ denote the multiplication on φ−aR(n, a). Then for any i and j, we have Xi ◦ Xj = φ−a(Xi) ∗ Xj = = = = i i j−ℓ Xℓ=0(cid:18)−a ℓ (cid:19)Xi−ℓ ∗ Xj Xk=0(cid:18)−a ℓ (cid:19)(cid:18)a + j Xℓ=0 Xv=0(cid:18)−a v (cid:19)(cid:18)a + j Xu=0 Xu=0(cid:18)j u(cid:19)Xi−uXj, u i i 10 k (cid:19)Xi−ℓ−kXj u − v(cid:19)Xi−uXj where we have used the Chu-Vandermonde identity at the end. This agrees with the formula for multi- plication in R(n, 0). (cid:3) As a corollary of Proposition 4.4 we obtain that φb induces an automorphism of R(n, a) for any a, b. For we have Lemma 4.5. Let S be an N-graded ring and let φ, ψ be graded automorphisms of S such that φψ = ψφ. Then ψ is also an automorphism of Sφ and φS. Proof. We prove the lemma for Sφ; note the statement makes sense because as a graded vector space S = Sφ. Let ∗ denote the multiplication on S and let ◦ denote the multiplication on Sφ. We check that for any r ∈ Si and s ∈ Sj, we have ψ(r ◦ s) = ψ(r ∗ φi(s)) = ψ(r) ∗ ψφi(s) = ψ(r) ∗ φiψ(s) = ψ(r) ◦ ψ(s), as needed. (cid:3) Corollary 4.6. Fix n ∈ Z>0. For any a, b ∈ k, the action of φb on R(n, a) induces an automorphism of R(n, a). Proof. That φb induces an automorphism of R(n, 0) for any b follows from Lemma 4.5, with φ = ψ = φb. We then apply Lemma 4.5 again, using the facts that φaφb = φa+b = φbφa and that R(n, a) ∼= φaR(n, 0) proved in Proposition 4.4. Corollary 4.7. For any a, b ∈ k, we have φb R(n, a) ∼= R(n, a + b), under the map Xi 7→ Xi. Proof. The proof is very similar to the proof of Proposition 4.4 and is left to the reader. (cid:3) (cid:3) Proof of Theorem 4.2. The result follows immediately from Corollary 4.7, using Lemma 4.1 to move from left to right twists. (cid:3) We now show that up to composition with nonzero scalar multiplication, the only non trivial graded automorphisms of R(n, a) are the maps φb. We will need the following lemma. Lemma 4.8. Let n ≥ 2. Suppose that ψ ∈ Aut R(n, a) is such that ψ(X0) = X0 and ψ(Xj) = Xj +αjX0 with αj ∈ k for 1 ≤ j ≤ n. Then ψ is the identity. Proof. By applying ψ to the second equation of (3.18) we obtain [X0, X2] = [X0, X2 + α2X0] = ψ([X0, X2]) = [X0, X2] − aα1X0 ∗ X0. Thus we have α1 = 0 when a 6= 0. If a = 0 we apply ψ to the third equation of (3.18) to get (X1 + α1X0) ∗ (X2 + α2X0) − (X2 + α2X0) ∗ (X1 + α1X0) = [X1, X2] = [X1, X2] − 2α1X0 ∗ X1 + (2α2 − α2 1 + α1)X0 ∗ X0. Again we must have α1 = 0. Now suppose that αj = 0 for 1 ≤ j < k. We apply the automorphism ψ to the relation (3.13) with i = 1 and j = k. After rearranging we get (4.9) X1 ∗ (Xk + αkX0) − (a + k)X0 ∗ (Xk + αkX0) = (Xk + αkX0) ∗ X1 + k Xj=1(cid:18)−a − 1 j (cid:19)Xk−j ∗ X1. Thanks to equation (3.1) we compare the coefficients of the X 2 and we obtain αk(a − (a + k)) = 0. So −kαk = 0 and αk = 0 since k 6= 0. 0 term in both sides of the equality (4.9) (cid:3) We can now determine the graded automorphism group Aut grR(n, a) of R(n, a). For λ ∈ k×, let ξλ be the automorphism that scales all Xi by λ. Theorem 4.10. Assume either that n ≥ 2 or that n = 1 and a 6= 0. The d-graded automorphisms of R(n, a) are of the form ξλφc, for some λ ∈ k× and c ∈ k. In particular we have Aut grR(n, a) ∼= k× × k. 11 Proof. It follows from Corollary 4.6 that the maps ξλφc are indeed automorphisms of R(n, a). Recip- rocally, we proceed by induction on n. We first assume that a 6= −(n − 1). For n = 1 (and a 6= 0), the result follows from [Shi05, Theorem 3.1]. We assume that the result is true for R(n − 1, b) with b 6= −(n − 2). Let ψ be a d-graded automorphism of R(n, a). Thanks to Proposition 3.21 the only normal elements of R(n, a) with d-degree 1 are the nonzero scalar multiples of X0. Thus, up to com- position with some ξλ, we have that ψ(X0) = X0. In particular ψ induces a d-graded automorphism of i=0 λ(cid:0)c R(n, a)/hX0i ∼= R(n − 1, a + 1). By the induction hypothesis we have ψ(Xj) = Pj−1 i(cid:1)Xj−i + αjX0 for any j ≥ 1, where λ ∈ k× and c, αj ∈ k. We obtain that λ = 1 by applying φ to the first equation of (3.18) when a 6= 0, or to the third equation of (3.18) when a = 0. In particular we have ψ = φc modulo X0, and by applying Lemma 4.8 to ψ ◦ φ−c we conclude that ψ = φc. We finally deal with the case a = −(n − 1). We prove by induction on n ≥ 2 that the d-graded auto- morphisms of R(n,−(n− 1)) are of the form ξλφc. We only prove the base case of the induction since the induction step of the previous induction will also apply to that case. Let ψ be a d-graded automorphism of R(2,−1). Then ψ(X0) = λX0 for some nonzero λ, and by rescaling we may assume that λ = 1. Further, ψ induces a d-graded automorphism of the commutative polynomial ring R(2,−1)/hX0i ∼= R(1, 0) ∼= k[X1, X2]. Therefore there exist c, d, α, β, γ, δ such that ψ(X1) = αX2 + βX1 + cX0 and ψ(X2) = γX2 + δX1 + dX0, where αδ − βγ 6= 0. Applying ψ to the relations (3.18) we obtain that α = 0, β = γ = 1, δ = c and d = (cid:0)c 2(cid:1). This shows One can prove a similar version of Proposition 4.10 for the Poisson algebra A(n, a). For any c ∈ k the maps ϕc := exp(c∆) are well-defined automorphisms of the commutative polynomial ring A(n, a) since ∆ is a locally nilpotent derivation. Moreover we observe that that ψ = φc and concludes the proof. (cid:3) ∆({Xi, Xj}) = {∆(Xi), Xj} + {Xi, ∆(Xj)} for all 0 ≤ i, j ≤ n. This shows that ∆ is a Poisson derivation of A(n, a), and thus the maps ϕc are Poisson automorphisms of A(n, a). We state the following result without proof. Proposition 4.11. Assume either that n ≥ 2 or that a 6= 0. The d-graded Poisson automorphisms of A(n, a) are of the form ξλϕc, for some λ ∈ k× and c ∈ k. In particular the graded Poisson automorphism group PAut grA(n, a) of A(n, a) is isomorphic to k× × k. Corollary 4.12. Assume either that n ≥ 2 or that a 6= 0. Then PAut grA(n, a) ∼= k× × k ∼= Aut grR(n, a). When n = 1 and a = 0 we have PAut grA(1, 0) = Aut grR(1, 0) = Aut grk[X, Y ] ∼= GL 2(k). Remark 4.13. We note that for certain values of a, there exist non-graded (Poisson) automorphisms. For instance it is well known that for any polynomial P ∈ k[X0] the map sending X0 7→ X0 and X1 7→ X1 +P defines an automorphism of the Jordan plane R(1, a) and the Poisson-Jordan plane A(1, a). When n = 2 and a = 1/q for some q ∈ Z>0, the map f defined by f (X1) = X1 + X q+1 f (X2) = X2 + X q f (X0) = X0, 0 X1 + X 2q+1 0 , 0 is a non-graded Poisson automorphism of A(2, 1/q). To conclude this section we calculate the Nakayama automorphism of R(n, a) and prove, in particular, that R(n,− 1 2 (n+2)(n−1)/(n+1)) is (n+1)-Calabi-Yau for every n > 0. We first recall some definitions. Let R be an a k-algebra, and let Re = R ⊗k Rop be the enveloping algebra of R. An (R, R)-bimodule M can be considered as a left Re-module by defining r ⊗ s · m = rms. 12 Definition 4.14. We say that R is skew Calabi-Yau (or skew CY) if (i) R is homologically smooth: R has a finite projective resolution as a left Re-module such that each term is finitely generated; (ii) There are an algebra automorphism µ of R and an integer d such that Exti Re (R, Re) ∼=  0 1Rµ if i 6= 0 if i = d. (Here 1Rµ is the R-bimodule which is isomorphic to R as a k-vector space and such that r· s· t = rsµ(t).) If R is skew CY, the automorphism µ is called the Nakayama automorphism of R. If µ is inner, then R is Calabi-Yau or CY. By [RRZ14, Lemma 1.2], any AS-regular connected graded algebra is skew CY. In particular, the algebras R(n, a) are skew CY. We will need the following lemma before calculating the Nakayama automorphism of R(n, a). Lemma 4.15. Let µ be the Nakayama automorphism of R = R(n, a) for any n ≥ 1. Then for all 0 ≤ j ≤ n, we have µ(Xj) − Xj ∈ span(X0, . . . , Xj−1). In particular µ(X0) = X0. Proof. The Nakayama automorphism of the commutative polynomial ring A = k[X0, . . . , Xn] is well- known to be trivial. Since A is the associated graded of R with respect to the ǫ-filtration and clearly Ae ∼= k[X0, . . . , X2n+1] is the associated graded of Re, by [Bjo89, Proposition 3.1], we have that the associated graded of Extn Ae(A, Ae) = 1A1. This shows that the ǫ-leading term of µ(Xj ) must be Xj. (cid:3) Re (R, Re) = 1Rµ is a subquotient of Extn Theorem 4.16. For any n ≥ 1, the Nakayama automorphism of R(n, a) is φc, where c = (n + 1)a +(cid:18)n + 1 2 (cid:19) − 1. Proof. We begin by calculating the Nakayama automorphism of R(n, 0). For n = 1 we have R(1, 0) ∼= k[X0, X1], so the Nakayama automorphism is the identity φ0; we have 0 =(cid:0)2 2(cid:1)−1. Suppose now that n > 1 and we wish to calculate the Nakayama automorphism µ of R(n, 0). Let b = n+(cid:0)n 2 (cid:1)−1. Since by Lemma 4.15 we have µ(X0) = X0, thus µ induces an automorphism of R(n, 0)/hX0i ∼= R(n − 1, 1). By [RRZ14, Lemma 1.5], using the fact that X0 is central, this induced automorphism is equal to the Nakayama automorphism of R(n − 1, 1), which by induction on n is φb. Thus modulo X0, we have that µ = φb. By applying Lemma 4.8 to ψ = µ ◦ φ−b we see that µ = φb. 2(cid:1)−1 =(cid:0)n+1 Let µ be the Nakayama automorphism of R(n, 0) and ν be the Nakayama automorphism of R(n, a) = ξλ for some λ ∈ k×. By Lemma 4.15, we must φaR(n, 0). By [RRZ14, Theorem 0.3], we have ν = µφn+1 have λ = 1. Thus a ν = µφn+1 a = φ(n+1 2 )−1+(n+1)a, as claimed. (cid:3) Corollary 4.17. For any n ≥ 1, the algebra R(n, a) is Calabi-Yau if and only if a = n + 1 = − 1 −(cid:0)n+1 2 (cid:1) 2(n+1) = − 5 (n + 2)(n − 1) . 2(n + 1) 4 , so Pym's example 1.1 is Calabi-Yau, using Example 4.18. When n = 3, we have − (n+2)(n−1) Example 3.20. 13 5. Isomorphisms We have seen in Example 3.16 that we have R(1, 0) ∼= R(1, a) ⇐⇒ a = 0 and that R(1, a) ∼= R(1, b) for any nonzero a, b ∈ k. A similar statement holds for the Poisson algebras A(1, a). In this section we analyse the isomorphism question for R(n, a) and A(n, a) where n ≥ 2. Moreover we show that each R(n, a) is isomorphic to its opposite ring. The main theorem of this section is the next result. Theorem 5.1. Let n ≥ 2 and let a, a′ ∈ k. The following are equivalent: (1) a = a′; (2) R(n, a) ∼= R(n, a′); (3) A(n, a) is Poisson isomorphic to A(n, a′). We remark that Theorem 5.1 is not particularly surprising given Corollary 4.17, since for each n there is a unique a with R(n, a) Calabi-Yau. Proof of Theorem 5.1. (1) ⇒ (2), (3) is trivial. (3) ⇒ (1). Let A = A(n, a) and let A′ = A(n, a′). To avoid any confusion, we will denote the 0, . . . , X ′ n. generators of A by X0, . . . , Xn as usual and denote the corresponding generators of A′ by X ′ We will let ∆ also denote the downward derivation on A′, so ∆(X ′ i) = X ′ i−1. Suppose that there is a Poisson isomorphism α : A → A′. For 0 ≤ i ≤ n set α(Xi) = Ti = Li + Pi, where Li ∈ A′ k. Note that since α can be seen as an algebra automorphism of the polynomial ring A, the linear parts Li must be nonzero for all 0 ≤ i ≤ n (using the fact that the Jacobian of α must be nonzero at the origin). 1 and Pi ∈ k ⊕Lk≥2 A′ If a = 0 then since X0 is Poisson central T0 must be as well. As the Poisson bracket respects the d- grading on A′, the degree 1 part L0 must be Poisson central and by Proposition 3.21 we have L0 ∈ ker ∆. Thus L0 = λ0X ′ 0 is Poisson central and so a′ = 0. This shows that a = 0 ⇐⇒ a′ = 0. Suppose now that aa′ 6= 0. Now T0 is Poisson normal and by Proposition 3.21 we have T0 ∈ ker ∆ and 0 for some 0 6= λ0 ∈ k. Since Γa′(L0) = a′L0 T0 is a Γa′ -eigenvector. Thus L0 ∈ ker ∆ and so L0 = λ0X ′ we have Γa′(T0) = a′T0. Let c = a/a′ 6= 0. Applying α to the equation a∆(g)X0 = {g, X0}a we obtain aα∆(g)T0 = {α(g), T0}a′ = a′∆α(g)T0 and so (5.2) as maps from A → A′. Since ∆ respects the d-degree we have ∆(Li) = cLi−1 for 1 ≤ i ≤ n. Thus there are λ1, λ2 ∈ k so that ∆α = cα∆ L0 = λ0X ′ 0, 1 + λ1X ′ L1 = cλ0X ′ 0, 2 + cλ1X ′ L2 = c2λ0X ′ 1 + λ2X ′ 0. From (5.2) we also conclude that T0, T1 ∈ Im ∆ and so have no constant term. Considering the equality (5.3) (a + 2)T0T2 − (a + 1)T 2 1 = α({X1, X2}a) = {T1, T2}a′ then gives that T2 has no constant term. Therefore the X ′ (a + 2)c2λ2 c3λ2 1X ′ 0(a′ + 2)X ′ (2) ⇒ (1). Let R = R(n, a) and let R′ = R(n, a′). As above, we denote the generators of R′ by n. If R ∼= R′ then by [BZ17, Theorem 0.1] there exists a graded isomorphism α : R → R′. Since 0, . . . , X ′ 2 term of the left-hand side of (5.3) is 2) = 2. On the other hand, the X ′ 2. Thus a+2 2 term of the right-hand side is ∆(cλ0X ′ 0X ′ and a′ = a. 1)Γa′ (c2λ0X ′ 0X ′ 0X ′ 0X ′ a = a′+2 a′ X ′ 14 X0 is normal in R1 its image α(X0) is normal in R′ We prove the result by induction on n. 1 and α(X0) ∈ span(X ′ 0) thanks to Proposition 3.21. We first consider the case n = 2. Since X0 is central if and only if a = 0 we deduce that a = 1 there exist 2. We fix λ0 ∈ k× such that 0 ⇐⇒ a′ = 0. Therefore we may assume that a, a′ 6= 0 in the following. Since α(X1) ∈ R′ scalars µ0, µ1 and µ2 not all zero such that α(X1) = µ0X ′ α(X0) = λ0X ′ 0 + µ1X ′ 1 + µ2X ′ 0. By applying α to the first equality in (3.18) we obtain 2 ∗ X ′ 0) + λ0µ2(X ′ 1 − X ′ 2 − X ′ 0 ∗ X ′ 0 ∗ X ′ 1 ∗ X ′ λ0µ1(X ′ 0) = −aλ2 0X ′ After simplification using the relations (3.18) we compare the coefficients of the X ′ 1 term. We get −a′λ0µ2 = 0 which implies that µ2 = 0 since a′ 6= 0. In particular we have α(X1) ∈ span(X ′ 1) and α respects the ǫ-filtrations on R and R′. By Lemma 3.5 α induces a Poisson isomorphism between A(2, a) and A(2, a′), so a = a′. 0, X ′ 0 ∗ X ′ 0. 0 ∗ X ′ Suppose now n ≥ 3 and the result true for n − 1. Since α(X0) ∈ span(X ′ R(n − 1, a + 1) ∼= R/hX0i → R′/hX ′ 0i ∼= R(n − 1, a′ + 1) and we have a + 1 = a′ + 1. 0) there is an isomorphism (cid:3) To end the section we prove that R(n, a) and R(n, a)op are isomorphic. This will be used in the next section. Theorem 5.4. For any u ∈ k, the map ωu from R(n, a) to R(n, a) defined by ℓ (cid:19)Xi−ℓ = (−1)iφi−u(Xi) ωu(Xi) = (−1)i i Xℓ=0(cid:18)i − u is an anti-isomorphism. In particular R(n, a) and R(n, a)op are isomorphic for any a ∈ k and any n > 0. We will need the following lemmas. Lemma 5.5. For any v ∈ N and a, m ∈ k we have (5.6) v (−1)k(cid:18)a Xk=0 k(cid:19)(cid:18)m − k v − k(cid:19) =(cid:18)m − a v (cid:19). Proof. The proof is a combination of the Chu-Vandermonde identity (3.9) and the identity for any ℓ ∈ N and u ∈ k. Lemma 5.7. For a, u ∈ k we have (5.8) Proof. We have ℓ(cid:19) = (−1)ℓ(cid:18)ℓ − u − 1 (cid:19) (cid:18)u ℓ ωuφa = ωu+a. ωuφa(Xi) = i i i ℓ i−k (cid:19)Xi−k−ℓ k(cid:19) i−k Xk=0(cid:18)a (−1)i−k(cid:18)i − k − u Xℓ=0 (−1)−k(cid:18)a Xℓ=0 Xk=0 = (−1)i (−1)−k(cid:18)a Xk=0 Xv=0 Xv=0(cid:18)i − (u + a) k(cid:19)(cid:18)i − k − u (cid:19)Xi−k−ℓ k(cid:19)(cid:18)i − k − u v − k (cid:19)Xi−v (cid:19)Xi−v = wa+u(Xi), = (−1)i = (−1)i v ℓ v i where we have set v = k + ℓ and then use Lemma 5.5 with m = i − u. 15 (cid:3) (cid:3) Proof of Theorem 5.4. Note that the relations (3.14) in R(n, a) can be rewritten as (5.9) φ−zi (Xj) ∗ Xi = φ−zj (Xi) ∗ Xj. for any 0 ≤ i, j ≤ n. To see that ωu is a well-defined anti-automorphism of R(n, a) it is enough to apply ωu to both side of (5.9) and check that we obtain the same result. Using (5.8) we have ωu(φ−zi (Xj) ∗ Xi) = ωu(Xi) ∗ ωuφ−zi(Xj) = ωu(Xi) ∗ ωu−zi(Xj) = (−1)i+jφi−u(Xi) ∗ φj−u+zi (Xj) and similarly ωu(φ−zj (Xi) ∗ Xj) = (−1)i+jφj−u(Xj) ∗ φi−u+zj (Xj). For any c ∈ k the map φc is an automorphism of R(n, a), and by applying φc to (5.9) we obtain that φc−zi(Xj) ∗ φc(Xi) = φc−zj (Xi) ∗ φc(Xj). Letting c = a + i + j − u = i − u + zj = j − u + zi we deduce that φj−u(Xj) ∗ φi−u+zj (Xi) = φi−u(Xi) ∗ φj−u+zi (Xj). This proves that ωu(φ−zi (Xj) ∗ Xi) = ωu(φ−zj (Xi) ∗ Xj) and finishes the proof. (cid:3) The opposite algebra A(n, a)op of the Poisson algebra A(n, a) is the same associative k-algebra endowed with the opposite Poisson bracket {−,−}op := −{−,−}. The analogue of Theorem 5.4 for A(n, a) follows easily from Lemma 3.5. It is straightforward to check that the map β from A(n, a)op to A(n, a) sending Xi to (−1)iXi for all i is a Poisson algebra isomorphism. 6. Point modules Let R be an N-graded ring. A (left or right) point module over R is a cyclic graded module M with hilb(M ) = 1/(1 − t). In this section we study the point modules of the R(n, a) and prove: Theorem 6.1. Let R = R(n, a), where n ≥ 1. If M = R/J is a right or left point module, then J is generated by J1 as a right (respectively, left) ideal of R. There is a projective scheme X that parameterises both left and right point modules, and X red is isomorphic to n copies of P1, where the k'th copy is embedded in V (X0, . . . , Xn−k−1) ⊆ P(R∗ 1) as a rational normal curve of degree k. Recall that there are normal elements Y2, . . . , Yn in R◦ = R[X −1 0 ]. Let T be the subalgebra of R◦ generated by X0, X1, Y2, . . . , Yn. We will compute the point modules for R by relating points of R to those of T . Let C = k[X0, Y2, . . . , Yn]. We know from Lemma 3.26 that R◦ = C[X −1 0 ][X1; δ] where δ is the derivation X0Γa of C[X −1 0 ]. The proof of that result also gives that T = C[X1; δ]. We then have: Proposition 6.2. 2 (cid:1) quadratic relations. They are: X1X0 − X0X1 − aX 2 (1) T has (cid:0)n+1 A : A(k) : X1Yk − YkX1 − (a + k)X0Yk B(k) : C(j, k) : X0Yk − YkX0 YkYj − YjYk 0 for 2 ≤ k ≤ n for 2 ≤ k ≤ n for 2 ≤ j < k ≤ n. (2) T is Artin-Schelter regular. 16 (cid:3) Proof. (1) is a straightforward computation. For (2), note that the weight grading ǫ can be used to define a filtration on T whose associated graded ring is the commutative polynomial ring C[X1]. Thus T has finite global dimension. Clearly T has finite Gelfand-Kirillov dimension. Note that Y2, . . . , Yn are normal in T , and T /(Y2, . . . , Yn) is isomorphic to the Jordan plane and is AS-regular. Thus T is AS-Gorenstein by [Lev92, Corollary 5.10, Theorem 6.3]. (cid:3) By Corollary 3.6 R is strongly noetherian. Thus by [AZ01, Theorem E4.3] there is a projective scheme X that parameterises right R-point modules up to isomorphism. For x ∈ X, let M (x) be the corresponding point module. By [KRS05, Proposition 10.2], there is σ ∈ Aut (X) so that M (x)[1]≥0 ∼= M (σ(x)) for all x ∈ X. We also have: Lemma 6.3. If M is an R-point module or a T -point module, then either M X0 = 0 or M is X0- torsionfree. Proof. This is a consequence of the facts that R and T are AS-regular with Hilbert series 1/(1 − t)n+1 and X0 is a normal element. See [LBS93, p. 728] for a summary of the argument. (cid:3) We now relate X0-torsionfree point modules over R to point modules over T . Proposition 6.4. Let M be an X0-torsionfree point module over R. Then there is a unique T -action on M that extends the action of T ∩ R and makes M a T -point module. Similarly, if L is an X0-torsionfree point module over T then there is a unique R-action on L that extends the action of T ∩ R and makes L also a point module over R. Proof. Let x ∈ X and recall the isomorphism M (x)[1]≥0 ∼= M (σ(x)). Equivalently, there are inclusions M (x) ⊆ M (σ−1(x))[1] for all x ∈ X. Note also that because σ ∈ Aut (X), the point σ−1(x) is the unique y ∈ X so that M (x) ⊆ M (y)[1]. Let The module N (x) is Z-graded with dim N (x)k = 1 for all k ∈ Z. (In fact, N (x) is the injective hull of M (x) in the category of graded R-modules.) N (x) := lim←− M (σ−n(x))[n]. Suppose now that M (x) is X0-torsionfree. By Lemma 6.3 so is each M (σ−n(x)), as M (x) ⊆ M (σ−n(x))[n]. Thus N (x) is X0-torsionfree. We may thus choose a basis {nk k ∈ Z} of N (x) with nk ∈ N (x)k and nkX0 = nk+1. If we define nkX −1 0 = nk−1 for all k ∈ Z, we obtain an action of R[X −1 0 ] on N (x), and thus an action of T on N (x)≥0 = M (x). The action is clearly unique. The proof that an X0-torsionfree point module over T has an induced R-action is similar. If K is an X0-torsionfree point module over T (or over R) then K is cyclic as a module over k[X0]. (cid:3) Thus K is also cyclic under the induced R-action (or T -action). We now compute the point modules of T , at least up to radical. Let V := T /hY2, . . . , Yni. We have 0i. If a 6= 0, this ring is the Jordan plane; in particular, the V are V ∼= khX0, X1i/hX1X0 − X0X1 − aX 2 isomorphic for any a 6= 0. If a = 0, then V is a commutative polynomial ring. Proposition 6.5. The reduced point scheme of T is isomorphic to V (X0) ∪ V (Y2, . . . , Yn) ⊆ P(T ∗ particular, an X0-torsionfree point module over T must be annihilated by Y2, . . . , Yn. 1 ). In The point modules over T that are annihilated by Y2, . . . , Yn are parameterised by P1. Proof. The final statement comes directly from the isomorphism of V with the Jordan plane (if a 6= 0) or with k[X0, X1] (if a = 0). We multilinearise the relations of T as in [ATVdB90], to compute the scheme X(2) ⊆ P(T ∗ 1 )×2 parameterising truncated point modules over T with Hilbert series 1 + t + t2. Let the coordinates on 17 1 )×2 be X0, X1, Y2, . . . , Yn, X ′ P(T ∗ entries in T1 so that X(2) is defined by the equations 2 , . . . , Y ′ 0, X ′ 1, Y ′ 2 (cid:1) × (n + 1) matrix A with The rows of A are given by: A : n. We thus obtain an (cid:0)n+1   A = 0. X ′ 0 X ′ 1 Y ′ 2 ... Y ′ n   hX1 − aX0 −X0 0 . . .i . . . X1 − (a + k)X0 (k + 1) . . . # 0 . . . X0 . . . (k + 1) # Yk . . . (j + 1) −Yj (k + 1) . . . # . A(k) : B(k) : C(j, k) : "0 −Yk "−Yk "0 . . . (Here we use (ℓ) to indicate the column of an entry.) Let X ′ be the projection of X(2) onto the first coordinate. It is standard that X ′ is defined by the locus where rank(A) < n + 1. Consider the minor Ak of A given by rows A, A(k), and B(2) − B(n). In columns j ≥ 2, j 6= k the only nonzero entry of Ak is the (j + 1, j + 1) entry X0. Thus det Ak = ±X n−2 0 X1 − aX0 −X0 0 −Yk X1 − (a + k)X0 0 X0 = ±kX n 0 Yk. (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) 0 −Yk (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Since k 6= 0 we have that X ′ is a closed subscheme of V (X n 1 Y2, . . . , X n 1 Yn). Let Y be the reduced point scheme of T . Since T is strongly noetherian, for some N we have Y ⊆ P(T ∗ 1 )×N . Let X be the projection of Y to the first factor; we have X ⊆ X ′. To prove that this projection induces an isomorphism Y ∼= V (X0) ∪ V (Y2, . . . Yn) it suffices to prove that each point of V (X0) ∪ V (Y2, . . . Yn) corresponds to a point module. That is, if W is a codimension 1 subspace of T1 with either X0 ∈ W or Y2, . . . Yn ∈ W we must show that T /W T is a point module. If X0 ∈ W then T /W T is isomorphic to the right module over T /X0T defined by factoring out the image of W . Since T /X0T is commutative any codimension 1 subspace of (T /X0T )1 defines a point module, so T /W T is a point module over T . Now suppose that Y2, . . . , Yn ∈ W . Then T /W T is isomorphic to the right module over V given by factoring out the image of W . As V is isomorphic either to the Jordan plane or to k[X0, X1], we see likewise that any codimension 1 subspace of T1 that contains Y2, . . . , Yn defines a point module. (cid:3) The natural map T → V induces a graded homomorphism ζ : T [X −1 0 ] → D, where D = Qgr(V ) = V [v−1 : v ∈ V is nonzero and homogeneous]. Let u = X1X −1 and let t = X0. It is well-known that D ∼= k(u)[t, t−1; σ], where σ(u) = u − a. To check this we verify that we have X1X0 = ut2 = t(u + a)t = X0(X1 + aX0). 0 Our next result computes ζ(R) ⊆ k(u)[t; σ]. To prove Theorem 6.1, the case a = 0 is the only one needed, but we give the general result because it is of independent interest. Proposition 6.6. We have ζ(R) ∼= V (n), the n-th Veronese of V . 18 Proof. Recall that R = R(n, a), where n ≥ 1. Note also that the subring of R generated by X0, . . . , Xk is isomorphic to R(k, a) for all 1 ≤ k ≤ n. We will abuse notation and write R(1, a) ⊆ R(2, a) ⊆ ··· ⊆ R(n, a), and will use this to prove the theorem by induction on n. Likewise, we write A(1, a) ⊆ A(n, a). We will show, for all k, that ζ(Xk) ∈ k[u] · t and, more specifically, that k! uk. ζ(Xk)t−1 ∈ k[u] and has leading term 1 (6.7) We need a subsidiary lemma. Recall that the underlying space of R is the commutative ring A, with 1, we multiplication indicated by ∗ in R and juxtaposition in A. When we write an expression like X j mean the commutative power; we will write X ∗j 1 to mean the noncommutative power. Lemma 6.8. For 0 ≤ k ≤ n, the following hold. (1) For any f ∈ R(1, a), we have (2) Xk ∗ f − Xkf ∈ k−1 Xi=0 Xi ∗ R(1, a) = XiA(1, a). k−1 Xi=0 k Xi=0 Xi ∗ R(1, a) = XiA(1, a). k Xi=0 Proof. Again, we induct on k; the result is trivial for k = 0. Assume we know the result for k. (1). We may assume that f is e-homogeneous. We have Xk+1 ∗ f − Xk+1f = k+1 Xi=1 Xk+1−i(cid:18)Γ i(cid:19)(f ). i(cid:1)(f ) is a scalar multiple of Xk+1−if . Thus the right hand side is in Pk Each Xk+1−i(cid:0)Γ Pk i=0 Xi ∗ R(1, a) by induction. (2) follows directly from (1). Since ζ(X0), ζ(X1) ∈ k[u][t; σ] we obtain that ζ(R(1, a)) ⊆ k[u][t; σ]. Since ζ is graded, we have ζ(X j 1 ) = (ut)j = u(u − a)(u − 2a)··· (u − (j − 1)a)tj, which has leading term ujtj. We immediately obtain from Lemma 6.8(1) (and an elementary induction) that 1 ) = fj(u)tj. We have ζ(X ∗j i=0 XiA(1, a) = (cid:3) (6.9) fj = uj + (lower order terms) ∈ k[u]. Assume now that we have shown that (6.7) holds for 0, . . . , k − 1, and consider the image of X k 0 Xk = X ∗(k) 0 ∗ Xk under ζ. Recall that Yk = Xk + k Xp=1 (−1)pXk−pX p 1 p!X p 0 . ζ(X k 0 Xk) = ζ X k 0 Yk + 0 Xk−pX p (−1)p+1X k−p p! 1 ! . k Xp=1 Thus (6.10) We know that ζ(X k 0 Yk) = 0 since ζ(Yk) = 0. Applying (6.7), Lemma 6.8(1), and (6.9), we obtain that ζ(X k−p 0 Xk−pX p 1 ) = Thus (6.10) reduces to ζ(X k 0 Xk) = (cid:16) k Xp=1 1 (k − p)!(cid:0)uk + lower terms(cid:1)tk+1. p!(k − p)!(cid:17)uk + lower terms! tk+1. (−1)p+1 19 Since commuting with t does not effect the leading term of a polynomial in u, it follows that ζ(Xk) = t−kζ(X k 0 Xk) = (cid:16) k Xp=1 p!(k − p)!(cid:17)uk + lower terms! t, (−1)p+1 and that ζ(Xk)t−1 ∈ k[u]. Finally, Xp=1 k (−1)p+1 p!(k − p)! = k Xp=0 (−1)p+1 p!(k − p)! + 1 k! = 1 k! , as needed. We have thus established (6.7) for Xk. It follows from (6.7) that (6.11) for all k. ζ(cid:0)R(k, a)(cid:1)1 = k · {t, ut, . . . , ukt} To complete the proof of the proposition, we must identify the ring S = ζ(R) = kht, ut, . . . , unti ⊆ k(u)[t; σa]. Denoting the point [1 : 0] ∈ P1 by ∞, we may identify S1 with global sections of the sheaf OP1(n∞), multiplied by t. It is then clear that S is isomorphic to the nth Veronese of the twisted homogeneous coordinate ring (see [AVdB90]) B(P1,O(1), σa/n). We have that B(P1,O(1), σa/n) is isomorphic to the Jordan plane if a 6= 0 and to k[X0, X1] if a = 0 by [Rog14, Example 4.13]. Thus, no matter the value of a we have S ∼= V (n). (cid:3) Note that we have constructed above the homomorphism R(n, a) → R(1, a/n)(n) that was predicted at the end of Section 2. Taking associated graded rings, we obtain also the predicted Poisson homomorphism A(n, a) → A(1, a/n)(n). Further, if we let 1 ≤ k ≤ n, then factoring out X0, . . . , Xn−k and applying this construction to the factor we obtain a surjection R(n, a) → R(1, (a + n− k)/k)(k) and thus a P1 of point modules corresponding to point modules of R(1, (a + n − k)/k). These modules are all predicted by the discussion in Section 2. In a sense the striking content of Theorem 6.1 is that there are no other point modules for R(n, a). We now prove Theorem 6.1. Proof of Theorem 6.1. By Theorem 5.4, it suffices to prove the result for right point modules. We prove the result by induction on n, first noting the result is trivial for n = 1, since point modules over R = R(1, a) are in bijection with codimension-1 subspaces of R1 for any a. Thus we may suppose that n > 1. Since R(n, a) is a Zhang twist of R(n, 0) (Theorem 4.2), by [Zha96] it suffices to prove the result for R = R(n, 0). Let M = R/J be a point module, and recall the definition of the homomorphism ζ from just before Proposition 6.6. We claim that J = J1R. If X0 6∈ J then by Lemma 6.3 M is X0-torsionfree and by Proposition 6.4, M is also an X0-torsionfree T -point module. Thus M Yk = 0 for k ∈ {2, . . . , n} by Proposition 6.2, so M is annihilated by ker ζ. By Proposition 6.6, R/ ker ζ is isomorphic to k[X, Y ](n). The claim follows from the fact that if J is a right ideal of k[X, Y ](n) such that (k[X, Y ](n))/J is a point module, then J is generated in degree one. We may thus view the point scheme X of R as contained in P(R∗ 1), where M = R/J corresponds to the codimension 1 subspace J1 ⊆ R1. By slight abuse of notation, we regard a point module M as a closed point of the point scheme X ⊆ P(R∗ 1). Let Z(1) be the Zariski closure of {M = R/J X0 6∈ J}. Then ker ζ ⊆ J for all M = R/J in Z(1). Thus Z(1) is isomorphic to the point scheme of R/ ker ζ ∼= k[X, Y ](n), which is the image of the degree n Veronese map from P1 → Pn; in other words, a degree n rational normal curve in P(k[X, Y ]∗ n) = P(R∗ 1). It remains to consider the case of point modules annihilated by X0. By induction (using Proposition 3.7 (cid:3) as usual), these form a bouquet of n − 1 rational normal curves in V (X0), and the result follows. 20 Remark 6.12. Let a, b ∈ k, and let φb be the automorphism of R(n, a) defined in Lemma 4.3. It is easy to see that φb extends to an automorphism of R(n, a)[X −1 0 ] =: R◦(n, a). By Corollary 4.7, we have φb R◦(n, a) ∼= R◦(n, a + b). Let ∗a denote multiplication in R(n, a). Since [X0, Yk] = 0 in R◦(n, a + b), we have φb(Yk) ∗a X0 = φb(X0) ∗a Yk = X0 ∗a Yk = Yk ∗a X0, so φb(Yk) = Yk. It follows that ker ζ is φb-invariant. One can use this to give an alternate proof of the X0-torsionfree case of Theorem 6.1. 7. Primes of R and Poisson primes of A The results of the previous section construct surjections from R(n, a) to R(1, a′)(n−k) for all 0 ≤ k < n (where a′ depends on a, n, and k). The kernels of these maps are of course prime ideals, and are in some sense independent of a: for example, the kernel of the map R(n, a) → R(1, a/n)(n) is generated by Y2, . . . , Yn no matter the value of a. (This is a slight abuse of notation, since Y2, . . . , Yn are in the localisation R(n, a)[X −1 0 ].) We will see that, similar to the above, the d-graded prime spectrum of R(n, a) is largely independent of a. On the other hand, the ungraded primes of R(n, a) depends very sensitively on a. First, though, we explore the connection between primes of R(n, a) and Poisson primes of A(n, a). It is well-known that there is often a close relationship between prime ideals of a noncommutative ring R and Poisson primes of its semiclassical limit. Fix n and a and let R = R(n, a) with multiplication ∗a. Let A = A(n, a) with Poisson bracket {−,−}a. We will show that Spec R = PSpec A in the strongest possible sense. That is, we prove: Theorem 7.1. Let P ⊆ A = R. Then: (1) P ∈ Spec R if and only if P ∈ PSpec A, and further every prime ideal of R is completely prime; (2) P is a primitive ideal of R if and only if P is a Poisson primitive ideal of A. In the statement of Theorem 7.1, recall that an ideal I of a ring R is called left (resp. right) primitive if it is the annihilator of a simple left (resp. right) R-module. Thanks to Theorem 5.4 the two notions coincide for the ring R(n, a) and we don't specify left or right for a primitive ideal in this article. An ideal P of a Poisson ideal A is called Poisson primitive if it is the largest Poisson ideal contained in a maximal ideal of the commutative ring A. We further give a stratification of Spec R = PSpec A, and show the strata are homeomorphic to commutative (projective, projective-over-affine, or affine) varieties. We compute also the d-homogeneous primes of R, and show that (if a 6∈ Z) they do not depend on the precise value of a. David A. Jordan's work [Jor14] on Poisson algebras and Ore extensions is crucial to this section. The following is a slight strengthening of [Jor14, Theorem 3.6]. Proposition 7.2. Let B be a noetherian k-algebra that is a domain and let δ be a nonzero derivation of (7.3) for b ∈ B. B. Let R = B[z; δ], which we write as the left B-module R =Ln≥0 Bzn, with multiplication ∗ such that Let A be the Poisson-Ore extension (B[z],{−,−}δ) (see [Jor14, Lemma 3.1]); as a ring A = B[z], with Poisson bracket defined by z ∗ b = b ∗ z + δ(b) (7.4) {azm, bzn}δ = (maδ(b) − nbδ(a))zm+n−1. Let P ⊆ R = A. Then: P ∈ Spec R if and only if P ∈ PSpec A. Further, if P ∈ Spec R = PSpec A, then either P ⊇ δ(B) or P is generated by the δ-invariant prime ideal P ∩ B of B. For the proof of Proposition 7.2, recall that a δ-ideal I of B is an ideal I with δ(I) ⊂ I. The δ-ideal I is δ-prime if for all δ-ideals J, K of B, we have that JK ⊆ I implies that J ⊆ I or K ⊆ I. Since char k = 0, a δ-prime ideal of B is prime by [Goo06, Lemma 1.1]. 21 Proof of Proposition 7.2. This proof is largely a recapitulation of the proof of [Jor14, Theorem 3.6], pointing out that the homeomorphism constructed there is in fact the identity map. Let J = δ(B)B. For any δ-ideal Q of B, we have Q ∗ R = R ∗ Q = QA = Q[z] (using the identification of R with A as left B-modules). In particular, this holds for Q = J. Let P ⊆ A = R and suppose that either P ∈ Spec R or P ∈ PSpec A. If J ⊆ P then P ⊇ JA = J ∗ R = R∗ J. From (7.3), the two multiplications on the graded vector space A/JA = R/J ∗ R are equal, and the induced Poisson bracket on A/JA is trivial. Let C = R/J ∗ R = A/JA. Then J ⊆ P ∈ Spec R holds if and only if P/J ∗ R = P/JA ∈ Spec C = PSpec C, if and only if J ⊆ P ∈ PSpec A. Now suppose that J 6⊆ P and let Q = P ∩ B; note Q is a δ-prime ideal of B. By [Jor14, Lemma 3.2], if P ∈ PSpec A then P = QA = Q ∗ R, and thus P ∈ Spec R. But by [Jor14, Lemma 3.3], if P ∈ Spec R then P = Q ∗ R = QA and P ∈ PSpec A. Finally, we note that the results in [Jor14] are stated for k = C, but are valid over any field of characteristic 0. (cid:3) We now prove Theorem 7.1. Recall from Section 3 that R◦ = R[X −1 0 ] and A◦ = A[X −1 0 ]. Proof of Theorem 7.1. (1). If n = 1 the result is well-known. So let n > 1 and let P ⊆ A = R. First suppose that X0 6∈ P . Let P ◦ := P [X −1 0 P + ··· ⊆ A◦ = R◦. If P ∈ Spec R or P ∈ PSpec A then P = P ◦ ∩ A. So it suffices to prove that P ◦ ∈ Spec R◦ ⇐⇒ P ◦ ∈ PSpec A◦. This is an immediate application of Lemma 3.26 and Proposition 7.2. That prime ideals of R◦ are completely prime is shown in [Sig84] (see also [Jor14, Remark 3.7]) and the rest of (1) follows immediately for P . Now suppose that X0 ∈ P . If P ∈ Spec R or P ∈ PSpec A then X0 ∗ R = X0A ⊆ P , and (1) follows 0 ] = P + X −1 by induction, considering the image of P in PSpec A/hX0i = Spec R/hX0i. For (2) the case X0 6∈ P follows from [Jor14, Corollary 4.4], applied to P ◦ ⊆ R◦ = A◦, and the case X0 ∈ P follows by induction, as above. (cid:3) We turn now to describing the topological space Spec R = PSpec A. We note that this space has a natural stratification: for 0 ≤ j ≤ n + 1, let Specj(R) = {P ∈ Spec(R) Xi ∈ P if and only if 0 ≤ i < j}. It is immediate that Spec R is the disjoint union of the Speci(R). By Proposition 3.7, we have Specj R ∼= Specj−1 R(n − 1, a + 1). Thus to describe the primes of R explicitly, it suffices by induction to describe the open stratum Spec0 R. We have Spec0(R) ∼= Spec R◦ = PSpec A◦, using (the proof of) Theorem 7.1. Before describing PSpec A◦ = Spec R◦, we establish some notation. Let K be a commutative ring. By PK(2, . . . , n) we denote the weighted projective space P(2, . . . n) with base Spec K: explic- itly, PK(2, . . . , n) = Proj K[Y2, . . . , Yn] (see [Har77, page 76]), where K is assumed to be concentrated in degree 0, and deg Yi = i. If C is a graded ring, let Specgr(C) be the set of prime graded ideals of C, under the Zariski topology; if C has multiple gradings, say d and e, we will write Specd−gr(C) or Spece−gr(C) to indicate which grading is being used. Likewise, let PSpecd−gr(A◦) = { d-graded Poisson primes of A◦ }. For n ≥ 2 let C(n) = k[Z2, . . . , Zn], graded with deg Zi = i. Let X(n) = Specgr C(n). Note that X(n) = Pk(2, . . . , n) ⊔ {C(n)+ = hZ2, . . . , Zni} and is (n − 2)-dimensional. The structure of PSpec A◦ = Spec R◦ depends sensitively on the value of a, as shown in the next result. Theorem 7.5. Assume that n ≥ 2. (1) If a 6∈ Q, then PSpec A◦ is homeomorphic to X(n). Further, all primes of R◦ and Poisson primes of A◦ are d-graded. 22 (2) If a ∈ Q×, then PSpec A◦ is homeomorphic to the rational affine variety Spec Z, where Z is the Poisson centre of A◦, and has dimension (n − 1). Further, PSpecd−gr(A◦) is homeomorphic to X(n). (3) If a = 0, then PSpec A◦ is the disjoint union of a stratum homeomorphic to Pk[X ±1 0 ](2, . . . , n) and PSpec A◦/(Y2, . . . , Yn) ∼= A2 r V (X0) and has dimension max{n − 1, 2}. We have that PSpecd−gr(A◦) is the disjoint union of a stratum homeomorphic to P(2, . . . , n) and a stratum homeomorphic to A1. (4) Further, as long as a 6= 0, then PSpecd−gr(A◦) does not depend on a: that is, an ideal P of k[X0, . . . , Xn] is a Poisson prime of some A(n, a)◦ (with a 6= 0) if and only if P is a Poisson prime of all A(n, a)◦. We immediately obtain: Corollary 7.6. Let n ≥ 1 and let a ∈ k. Then Spec R(n, a) ∼= PSpec A(n, a) is a union of quasiprojective rational varieties and has dimension: • max{n − 2, 1} if a 6∈ Q; • max{n − 1, 1} if a ∈ Q and (n, a) 6∈ {(2,−1), (1, 0)}; • 2 if (n, a) ∈ {(2,−1), (1, 0)}. Proof. Combine Theorem 7.5, Proposition 3.7, and Example 3.16. (cid:3) Example 7.7. In the case a ∈ Q×, it is not necessarily true that Poisson primes of A◦ are centrally generated. For example, let n = 2 and a = −7/4. Recall that Y2 = X2 − X 2 . Then ∆(Y2) = 0 and Γ(Y2) = (1/4)Y2. By Proposition 3.21, Y2 is Poisson normal in A◦. 1 2X0 We will see that the Poisson centre of A◦ is Z = k[X0Y 7 2 ], so the Poisson ideal hY2i of A◦ is not centrally generated. However, Y2 is the unique Poisson prime of A◦ with hY2i ∩ Z = hX0Y 7 2 i. The situation in the example is typical. Theorem 7.5(2) will follow from: Proposition 7.8. Assume that a ∈ Q×. The map φ : PSpec A◦ → Spec Z P 7→ P ∩ Z is a homeomorphism. The inverse map is defined by Further, Spec Z is a rational variety of dimension n − 1. ψ : Spec Z → PSpec A◦ Q0 7→pQ0A◦. As an immediate consequence, we have: Corollary 7.9. If a ∈ Q×, the map P 7→ P [X −1 V (X0) ∼= PSpec(A◦) and Spec(Z). 0 ] ∩ Z induces a homeomorphism between PSpec(A) r (cid:3) Proposition 7.8 will follow from a general lemma on gradings of localised polynomial rings. Lemma 7.10. Let B = k[X ±1 0 , X1, . . . , Xm], where deg Xi = ai ∈ Z. Assume that a0 6= 0. Write the Z-grading on B as B =Ln∈Z Bn, and let Z = B0. Let Q0 ∈ Spec Z and let N = √Q0B. Then: (1) N is prime; (2) Kdim(B/N ) = 1 + Kdim(Z/Q0). 23 (3) The map Q0 7→ √Q0B gives a homeomorphism η : Spec Z → Specgr(B). The inverse map is given by Q 7→ Q ∩ Z. Further, Spec Z is a rational variety of dimension n − 1: that is, the fraction field Frac(Z) of Z is isomorphic to k(t0, . . . , tn−2). Proof. (1). Let B′ =Lk∈Z Bka0 ∼= Z[X ±1 0 ]. Suppose that N is not prime; then there are x, y 6∈ N with xy ∈ N , so (xy)m ∈ Q0B and (xy)mp ∈ Q0B′ for some m, p ∈ Z≥1. Thus either xmp or ymp is in Q0B, since Q0B′ ⊆ Q0B is a prime ideal of B′. This is a contradiction. (2). Since B is module-finite over B′, therefore B/Q0B is module-finite over B′/Q0B′ ∼= (Z/Q0)[X ±1 0 ]. Thus Kdim B/N = Kdim B/Q0B = Kdim B′/Q0B′ = 1 + Kdim Z/Q0. Let d = min{k ∈ Z≥0 Tk 6= 0} and let 0 6= x ∈ Td. (3). Let C = Z r Q0 and let T = BC−1/NC−1. Since N is graded and prime, T is a graded domain. We claim that T ∼= Frac(Z/Q0)[x±1]. Certainly T0 = ZC−1/Q0C−1 ∼= Frac(Z/Q0). Now, X0 and X −1 0 map to nonzero elements of T under the natural map B → T . By abuse of notation, let X0 also denote the image of X0 in T . A straightforward combinatorial argument shows that T =Lk∈Z Tkd. In particular, a0 = dℓ for some ℓ and so xℓ(X0)−1 ∈ T0 is invertible. Thus x is invertible in T . It follows that Tdk = T0xk for all k, completing the proof of the claim. Since T ∼= Frac(Z/Q0)[x±1] is a Laurent polynomial ring over a field, it has no non trivial graded ideals. Now, Specgr T is in bijection with {Q ∈ Specgr B Q ⊇ N, Q ∩ C = ∅} = {Q ∈ Specgr B Q ∩ Z = Q0}, where we have used part (1) of the lemma. Thus there is only one such Q, namely N . Define θ : Specgr B → Spec Z by θ(Q) = Q ∩ Z. The argument above shows that if Q ∈ Specgr B, then Q = ηθ(Q). Since θη is easily seen to be the identity on Spec Z, therefore θ = η−1. As η and θ clearly preserve inclusions, they are continuous and thus homeomorphisms. (cid:3) an e-graded subalgebra of A◦. Since Z is a normal semigroup algebra, Spec Z is rational. We next give an explicit characterisation of Poisson primitive ideals of A◦ if a ∈ Q. By Theorem 7.1 these are the same as the primitive ideals of R◦. Recall from the end of Section 3 that we define a grading A◦ =Le∈k A◦(e) by setting A◦(e) to be the e-eigenspace of Γa. Moreover notice that (A◦)∆ is Corollary 7.11. If a ∈ Q× then PSpec A◦ is homeomorphic to Spec Z. Furthermore, a Poisson prime P of PSpec A◦ is Poisson primitive if and only if P ∩ Z is a maximal ideal of Z. Proof. Let B = k[X ±1 0 , Y2, . . . , Yn] = (A◦)∆. Let Z be the Poisson centre of A◦. By Proposition 3.21, Z = BΓa = B(0). Since a 6= 0, X0 ∈ δ(B) so no nontrivial ideal of A◦ can contain δ(B). By Proposition 7.2, there is an inclusion-preserving bijection ρ between PSpec A◦ and the set of δ-prime ideals of B, defined by ρ(P ) = P ∩ B. Note that a δ-prime of B is the same as an e-graded prime of B. Thus by Lemma 7.10 the map φ = θρ : Spec A◦ → Spec Z is a homeomorphism. The inverse to φ is ψ = ρ−1η : Q0 7→ A◦(pQ0B) =pQ0A◦. By [Jor14, Corollary 4.4], P ∈ PSpec A◦ is Poisson primitive if and only if P ∩ B is a δ-primitive ideal of B, i.e. P ∩ B is the largest δ-stable ideal of B contained in some maximal ideal M . Since an ideal of B is δ-stable ⇐⇒ it is Γa-stable, a δ-primitive ideal of B is a maximal e-graded ideal of B, and by Lemma 7.10 these are precisely ideals of the form ψ(M0) for M0 = P ∩ B ∩ Z = P ∩ Z ∈ maxspec Z. (cid:3) We next assume that a 6= 0, and consider d-graded Poisson primes of A◦, which by Theorem 7.1 may and C = k[Z2, . . . , Zn]. The be identified with d-graded primes of R◦. For 2 ≤ i ≤ n, let Zi = YiX −1 e-grading on A◦ restricts to C, with e(Zi) = i. Proposition 7.12. If a 6= 0, then PSpecd−gr(A◦) is homeomorphic to Spece−gr(C) ∼= X(n). Further, as long as a 6= 0, then PSpecd−gr(A◦) does not depend on a in the sense of Theorem 7.5(4). 0 24 Proof. Note that C is precisely B0 = {b ∈ B b is d-homogeneous of degree 0 }. As in the proof of Corollary 7.11, there is an inclusion-preserving bijection between PSpec(A◦) and Spece−gr(B), and it is clear that this bijection takes d-graded primes to d-graded primes: in other words, PSpecd−gr(A◦) is homeomorphic to the set of (d, e)-bigraded primes of B, via the map P 7→ P ∩ B. Since B = C[X ±1 0 ] is strongly d-graded, we have Specd−gr(B) ∼= Spec C. Thus PSpecd−gr(A◦) is homeomorphic to Spece−gr(C) via P 7→ P ∩ C. But Spece−gr(C) is homeomorphic by definition to X(n). For the final statement, note that the definition of C, the e-grading on C, and the restriction homeo- morphism PSpecd−gr(A◦) → Spece−gr(C) do not depend on the value of a. The result above is particularly strong if a 6∈ Q, since we then have: (cid:3) Proposition 7.13. If a 6∈ Q, then all Poisson primes of A◦ are d-graded. Proof. We have seen that PSpec(A◦) ∼= Spece−gr(B), where B = k[X ±1 0 , Y2, . . . , Yn]. Let S = {e(b) b ∈ B}, which is clearly equal to Za + N. Since a 6∈ Q, we have S ∼= Z ⊕ N as a semigroup, and it follows that if b ∈ B is e-homogeneous then b is d-homogeneous. (cid:3) We now combine the previous results to prove Theorem 7.5. Proof of Theorem 7.5. (1) follows from Propositions 7.12 and 7.13, and (2) is Corollary 7.11 and Propo- sition 7.12 again. For (3), let B+ =P BYi; we have B+ = δ(B)B. Note that by Proposition 7.2, PSpec A◦ is equal to the disjoint union of: (7.14) and (7.15) {P = QA◦ Q 6⊇ B+ is a δ-prime ideal of B} {P P is a prime ideal of A◦ with P ⊇ B+}. As we have repeatedly seen above, δ-prime ideals of B are the same as e-graded ideals: that is, (7.14) is homeomorphic to {Q ∈ Spece−gr B Q 6⊇ B+} ∼= Proj B = Pk[X ±1 0 ](2, . . . , n). On the other hand, (7.15) is clearly homeomorphic to the spectrum of A◦/(Y2, . . . , Yn) ∼= k[X ±1 (with trivial Poisson bracket). 0 , X1] Finally, (4) follows from Proposition 7.12. (cid:3) Corollary 7.16. If a 6∈ {−n + 1, . . . ,−1, 0} then PSpecd−gr A(n, a) = Specd−gr R(n, a) does not depend on a. In particular Spec R(n, a) ∼= Spec R(n, b) for any a, b /∈ Q. Proof. Combine Theorem 7.5(4), Proposition 3.7, and induction. The second assertion follows from Proposition 7.13. (cid:3) 8. Dixmier-Moeglin equivalences and skewfields In this section we show that the algebra R(n, a) satisfies the Dixmier-Moeglin equivalence (DME). Recall that primitive ideals are annihilators of simple modules. In particular they are not easily distin- guished among the primes. The DME characterises them with algebraic and topological properties. A prime ideal P in a noetherian ring R is said rational provided that the field Z(Frac R/P ) is algebraic over the ground field, and is said locally closed if the point {P} is locally closed in Spec R (with respect to the Zariski topology). We say that the Dixmier-Moeglin equivalence holds for a given noetherian algebra if the sets of primitive ideals, locally closed ideals and rational ideals coincide. This idea originated in the work of Dixmier and Moeglin who showed that for any finite dimensional complex Lie algebras, theses sets are equal. 25 Thanks to our result on the DME we prove a transfer result which says that the Poisson algebra A(n, a) satisfies a similar equivalence, the so-called Poisson Dixmier-Moeglin equivalence (PDME), which we will recall later. To prove our results on the DME we relate R(n, a) to the enveloping algebra of a solvable Lie algebra sitting inside the localisation R(n, a)◦. With the Gelfand-Kirillov conjecture [GK66, Section 5] in mind, this motivated us to investigate the skewfield of fractions of R(n, a) at the end of this section. 0 ]. We denote by T the subalgebra of R◦ generated by X0, X1, Y2, . . . , Yn. Setting C := k[X0, Y2, . . . , Yn] we have that T = C[X1; δ], where δ = X0Γa is a derivation of the commutative ring C. Moreover we set Y0 := X0 and we denote by ga the n-dimensional solvable Lie algebra with basis elements Y0, X, Y2, . . . , Yn and Lie brackets Let R = R(n, a) and recall that R◦ = R[X −1 (8.1) [X, Yi] = (a + i)Yi [Yi, Yj] = 0 for all i, j. In particular sending Yi to Yi and X to Y −1 0 X1 we obtain the isomorphism (8.2) U (ga)[Y −1 0 ] ∼= T [Y −1 0 ] ∼= R◦. We now state the main result of this section. Theorem 8.3. The algebra R(n, a) satisfies the DME for any a ∈ k and n ≥ 1. The proof of Theorem 8.3 relies on the following lemmas. Lemma 8.4. Let P ∈ Spec R and suppose that X0 ∈ P . Then P is locally closed in Spec R if and only if P/hX0i is locally closed in Spec R/hX0i. Proof. This follows from the isomorphism R/hX0i (cid:3) P/hX0i ∼= R P . For P ∈ Spec R with X0 /∈ P , we set P ◦ := P [X −1 0 ] ∈ Spec R◦. Lemma 8.5. Let P ∈ Spec R and suppose that X0 /∈ P . Then P is locally closed in Spec R if and only if P ◦ is locally closed in Spec R◦. Proof. It is enough to prove the lemma for P = h0i. Recall that h0i is locally closed if and only if h0i 6=T{Q ∈ Spec R Q 6= h0i}. We set T. I1 = \h0i6=Q∈Spec R Q and I2 = \h0i6=T ∈Spec R◦ Suppose that I1 6= h0i and let h0i 6= U ∈ I1. Then U ∈ I2 and I2 6= 0. Reciprocally if I2 6= h0i, then there exists U nonzero inside Th0i6=Q∈Spec R and X0 /∈Q Q (recall that prime ideals are completely prime in R, see assertion (1) of Theorem 7.1). Then U X0 belongs to any nonzero Q ∈ Spec R and I1 6= h0i. (cid:3) We can now prove Theorem 8.3. Proof of Theorem 8.3. We proceed by induction on n. It is well-known that the Jordan plane R(1, a) (a 6= 0) and the commutative polynomial ring R(1, 0) satisfy the DME, so that the base case n = 1 is true. Suppose that R(n− 1, b) satisfies the DME for any b ∈ k and n > 1. By [BG02, II.7.17] the algebra R satisfies the (noncommutative) Nullstellensatz. Then by [BG02, II.7.15] we have the implications locally closed ⇒ primitive ⇒ rational. It remains to prove that rational implies locally closed. Let P ∈ Spec R be rational. Suppose first that X0 ∈ P . Then P/hX0i(cid:19) ∼= Z(cid:18)Frac R/hX0i Z(cid:18)Frac R P(cid:19) and P/hX0i is rational in R/hX0i. Since R/hX0i ∼= R(n − 1, a + 1) satisfies the DME by the induction hypothesis, the prime P/hX0i is locally closed. We conclude that P is locally closed by Lemma 8.4. Suppose now that X0 /∈ P . Then Z(cid:18)Frac R◦ P ◦(cid:19) ∼= Z(cid:18)Frac 26 R P(cid:19) and P ◦ is rational in R◦. Since the algebra U (ga) satisfies the DME over k by [IS80], the localisation 0 ] ∼= R◦ satisfies the DME. Then P ◦ is locally closed in Spec R◦ and we conclude that P is U (ga)[X −1 locally closed in Spec R by Lemma 8.5. (cid:3) The second main theorem of the section is that A(n, a) satisfies the Poisson Dixmier-Moeglin equiv- alence, which we define here. Recall from Section 7 that a Poisson primitive ideal is by definition the largest Poisson ideal contained inside a maximal ideal. Let A be a Poisson k-algebra and P ∈ PSpec(A). The ideal P is said locally closed if the point {P} is a locally closed point of PSpec(A) and is said Poisson rational provided the field ZP(cid:0)Frac (A/P )(cid:1) is algebraic over the ground field k. We say that the Poisson Dixmier-Moeglin equivalence holds for the Poisson algebra A if the sets of Poisson primitive ideals, of locally closed Poisson ideals and of Poisson rational ideals coincide. Our proof proceeds via a transfer result for the PDME. Theorem 8.6. The algebra A(n, a) satisfies the PDME for any a ∈ k and n ≥ 1. Recall that we set A◦ := A[X −1 0 ]. We first prove two lemmas. An algebra is catenary if for every pair of distinct prime ideals P ⊂ Q all saturated chains of prime ideals from P to Q have the same length. Lemma 8.7. Let S be a catenary and noetherian k-algebra with finite GK dimension and let P ∈ Spec S. Then {Q ∈ Spec S ht(Q) = ht(P ) + 1} is finite ⇒ P is locally closed, where ht(P ) is the height of P , i.e. the supremum of the length of chains of prime ideals descending from P . Proof. Since S is catenary we can assume that P = h0i. Suppose that S has only finitely many height one prime ideals, namely, P1, . . . , Pℓ. Since S has finite GK dimension, it satisfies the DCC on prime ideals. In particular any nonzero P ∈ Spec S contains one of the Pi and therefore {h0i} = Spec S \ V (∩iPi) is open in its closure. (cid:3) Lemma 8.8. The ring R = R(n, a) is catenary for any a ∈ k and n ∈ Z>0. Proof. Thanks to [BG02, II.9.5], Lemma 3.6 and the proof of Theorem 3.8, we only need to prove that Spec R has normal separation, that is, for every distinct pair P ⊂ Q of comparable primes in R the ideal Q/P of R/P contains a nonzero normal element of R/P . We proceed by induction on n. Let P ⊂ Q be a pair of comparable primes in R. The proof split into First assume that X0 ∈ P , so that X0 ∈ Q. Then P/hX0i ⊂ Q/hX0i inside R/hX0i ∼= R(n− 1, a + 1), Next assume that X0 /∈ Q, so that X0 /∈ P . Then P ◦ ⊂ Q◦ inside R◦ ∼= U (ga)[X −1 0 ]. Since ga is solvable Spec U (ga) has normal separation by [GW89, Theorem 12.19]. It is then easy to see that 0 ] ∼= R◦ has normal separation. Hence there exists a nonzero normal element U in Spec U (ga)[X −1 Q◦/P ◦ ∼= Q/P [X −1 0 ] (where we denote again by X0 its image in the quotient Q/P ). In particular there is an integer ℓ ≥ 0 such that U X ℓ 0 is a nonzero normal element in Q/P (recall that X0 is normal in R). Finally suppose that X0 /∈ P and X0 ∈ Q. Then X0 ∈ Q \ P and is normal modulo P as it is already three cases. and we are done by induction. normal in R. (cid:3) We now prove Theorem 8.6. Proof of Theorem 8.6. By [Oh99, Propositions 1.7, 1.10] we have the implications Poisson locally closed ⇒ Poisson primitive ⇒ Poisson rational. Let P be a Poisson rational ideal of A. Then by [BLLM, Theorem 8.3] the set {Q ∈ PSpec A ht(Q) = ht(P ) + 1} is finite. Since PSpec A = Spec R, the set {Q ∈ Spec R ht(Q) = ht(P ) + 1} = {Q ∈ PSpec A ht(Q) = ht(P ) + 1} 27 is finite. By Lemma 8.7 we conclude that P is locally closed in Spec R, hence P is locally closed in PSpec A since PSpec A = Spec R. (cid:3) To end the section we investigate the structure of the skewfield of fractions Frac R of the noetherian domain R. Recall the definition of the solvable Lie algebra ga from (8.1). By equation (8.2), R◦ is isomorphic to a localisation of the enveloping algebra U (ga). Then by [BGR73, Jos77, McC74], the skewfield Frac R is isomorphic to a Weyl skewfield, when k is algebraically closed and when the Lie algebra ga is algebraic. From [GK66, Section 8] we note that this algebra is algebraic if and only if a ∈ Q. In that case we provide an explicit description of this Weyl skewfield, and we show that Frac R is not isomorphic to a Weyl skewfield when a /∈ Q. Moreover we prove these results over a field of characteristic zero that is not necessarily algebraically closed. Recall that R◦ = B[X1; X0Γa], where B = k[X ±1 0 , Y2, . . . , Yn]. For any field K and ε, a ∈ K, let hε,a(K) be the 3-dimensional solvable Lie algebra over K with basis {x, y, z} and Lie bracket [x, y] = εy, [x, z] = az and [y, z] = 0. Proposition 8.9. We have Frac R ∼= Frac U (ga) ∼= Frac U (hǫ,a(K)) where K =(cid:0)Qgr(B)(cid:1)0 is a field of transcendence degree n − 2 over k, and where ε = 2 if n = 2 and ε = 1 if n > 2. Proof. We define a Z2-grading f on B as follows. Set f (X ±1 1 ) = (±1, 0) and f (Yi) = (1, i) for all i = 2, . . . , n. This is a combination of the d-grading and the ǫ-grading. Note that if u ∈ B has degree (0, 0), then u ∈ ker Γa. We now form the graded quotient ring E := Qgr(B) of B by inverting all its homogeneous elements Note that E is also Z2-graded by a grading denoted f again. It is a standard fact that E ∼= K[s±1, t±1], where K = E(0,0) is a field such that E = B[h−1 h f -homogeneous]. trdeg kK = trdeg kB − 2 = n − 2, and where 0 6= s ∈ E(1,0) and 0 6= t ∈ E(0,ε). By definition ε := min{α ≥ 1 E(0,α) 6= 0}. For instance we can choose s = X0 and t = Y2X −1 0 when n = 2, or t = Y3Y −1 2 when n > 2. Thus we have Frac R = Frac E[X1; X0Γa] = Frac K[s±1, t±1][X1; X0Γa]. For u ∈ K we have X1u − uX1 = Γa(u) = 0 since f (u) = (0, 0), and u commutes also with s and t since E is commutative. Moreover we get X1s − sX1 = X0Γa(X0) = aX 2 X1t − tX1 = X0Γa(t) = εX0t. 0 We conclude by setting X ′ we have st = ts and the result follows by setting x := X ′ 1s − sX ′ 1 := X1X −1 that X ′ 0 1 = aX0 = as and X ′ 1, y := t and z := s. 1t − tX ′ 1 = εt. Since s, t ∈ B (cid:3) Remark 8.10. Generators of the field K can be obtained by solving the system of equations   u0 + u2 + ··· + un = 0, 2u2 + 3u3 + ··· + nun = 0. Y n−i−1 n−1 Setting Y0 := X0 and Zi := YiY −(n−i) Note that K is purely transcendental over k of transcendence degree n − 2. Theorem 8.11. When a ∈ Q the skewfield Frac R is isomorphic to the first Weyl skewfield D1(F ) over a field F of transcendence degree n − 1 over k. When a /∈ Q, we have Z(Frac R) = K, and Frac R is not isomorphic to any Weyl skewfield. for i ∈ {0, 2, . . . , n− 2} we have K = k (Z0, Z2, . . . , Zn−2). n 28 Proof. The skewfield of the enveloping algebra of the Lie algebra hε,a(k) is either isomorphic to the first Weyl skewfield over a field of transcendence degree 1 over k when a is rational, or has a trivial centre when a is irrational. This is a classical fact that can be found in [GK66, Section 8] and [Ric02, Proposition 1.2.4.1 and Remark after]. The result follows from Proposition 8.9 and the isomorphism U (hε,a(K)) ∼= U (hε,a(k)) ⊗k K. (cid:3) Remark 8.12. (1) Let a ∈ Q and suppose that a = p/q ∈ Q with gcd(p, q) = 1 (when a = 0 we have p = 0 and we set q = 1). Setting Z := X q 3 we obtain 0 Y p 2 Y −p F = k (Z, Z0, Z1, . . . , Zn−2) where the Zi's are defined in Remark 8.10. (2) For a, b ∈ Q it is clear that Frac R(n, a) ∼= Frac R(n, b) since both are isomorphic to a Weyl skewfield over a field of transcendence degree n − 2. When a /∈ Q and b = ±a + n for some n ∈ Z, it is easy to verify that the skewfields Frac R(n, a) and Frac R(n, b) are isomorphic. However it remains unclear whether or not this condition is also necessary for an isomorphism Frac R(n, a) ∼= Frac R(n, b) when a, b /∈ Q. Using similar methods we can prove the following result about the Poisson structure of the field Frac (A). For a Lie algebra g we denote by S(g) its symmetric algebra that we endow with the so- called Kirillov-Kostant-Souriau Poisson bracket, that is the Poisson bracket obtained by extending by bi-derivation and bi-linearity the Lie bracket of g inside S(g) {X, Y } := [X, Y ]g for any X, Y ∈ g. Theorem 8.13. As Poisson algebras, we have Frac A ∼= Frac S(ga) ∼= Frac S(hε,a(K)). Moreover, if a ∈ Q then Frac A is isomorphic to the field of fractions of the Poisson algebra P = F [X, Y ], where {X, Y } = 1 and where F is the field described in Remark 8.12. When a /∈ Q, the Poisson centre of Frac A is K and Frac A is not isomorphic to FracP. 9. Examples of spectra In this final section we study PSpec A(n, a) and Spec R(n, a) for small values of n. Since these two spectra are equal it is enough to describe PSpec A(n, a). Because we will give explicit description of these spectra we assume that k is algebraically closed in this section. Example 9.1. Let n = 1 and consider A = A(1, a). If a = 0 then A(1, a) = k[X0, X1] with trivial Poisson bracket and thus PSpec A = Spec k[X0, X1]. Suppose that a 6= 0. Then A is isomorphic to the Poisson-Jordan plane (see Example 3.16). well-known that its Poisson spectrum is It is {h0i,hX0i,hX0, X1 − λi λ ∈ k} and that the d-graded primes are hX0i,hX0, X1i. Moreover only hX0i is not Poisson primitive. Assume that n ≥ 2 and that a ∈ Q×. We denote by PSpec1 A(n, a) the set of Poisson prime ideals of A that contain X0 and by PSpec0 A(n, a) the set of Poisson prime ideals of A that do not contain X0. Since A(n, a)/hX0i ∼= A(n − 1, a + 1) by Proposition 3.7 there is a homeomorphism: PSpec1 A(n, a) ∼= PSpec A(n − 1, a + 1). On the other hand by Theorem 7.5 there is a homeomorphism PSpec0 A(n, a) ∼= Spec(Z), 29 where Z is the Poisson centre of A◦ = A(n, a)[X −1 0 ]. Thus to describe PSpec A(n, a) completely we must study the ring Z. We know that Kdim Z = n − 1, so we first construct n − 1 algebraically independent elements of Z. Write a = p/q with gcd(p, q) = 1 and −p > 0. For i = 2, . . . , n let di = gcd(p, i) > 0, and set ui = p+iq di and vi = −p di (9.2) . Note that gcd(ui, vi) = 1. Finally we set i = X ui Y ′ By construction e(Y ′ i ) = 0, and we have Z ′ := k[Y ′ We can be more precise Lemma 9.3. Z is a free Z ′-module with basis . i 0 Y vi 2 , . . . , Y ′ n] ⊆ Z. S =(X s (9.4) 0 Y s2 2 ··· Y sn In particular its rank is S ≤Qn n i=2 vi. 0 ≤ si < vi, i = 2, . . . , n and ps + uidisi = 0) . n Xi=2 Proof. By the extension of Proposition 3.21 to A◦ we have Z = (A◦)Γa,∆ = B0 and so Z = Span(X s = Span(X s ··· Y sn 0Y s2 2 0 Y s2 2 0 Y s2 2 ··· Y sn n s ∈ Z, s2, . . . , sn ∈ N, and as + ··· Y sn n s ∈ Z, s2, . . . , sn ∈ N, and ps + n n (a + i)si = 0) Xi=2 uidisi = 0) . Xi=2 i ≥ 0 and 0 ≤ εi < vi. Then Let M = X s we rewrite n ∈ Z and for i = 2, . . . , n set si = vis′ n (cid:1)(cid:0)X s−Ω 0 2 n 2 M =(cid:0)Y ′s′ ··· Y ′s′ n ∈ Z. In particular this shows that a generating set of Z as a module over Z ′ is given by the set S. The result follows since the elements of S are linearly independent over Z ′. (cid:3) ··· Y ′s′ ··· Y εn n (cid:1), Y ε2 n ∈ Z ′ we have X s−Ω 2 where Ω = Pn iui. Since M, Y ′s′ ··· Y εn i=2 s′ Y ε2 2 0 2 n 2 i + εi, where s′ Note that it is possible that S = {1}. For example, let n = 2 and set M = X s 2 ∈ S. Then ps + u2d2s2 = 0, i.e. v2s = u2s2 and v2 must divide s2 since gcd(u2, v2) = 1. This implies that s2 = 0 since 0 ≤ s2 < v2. Therefore Z = Z ′ = k[Y ′ We next work out PSpec A(2, a) explicitly. By Example 9.1, for a, b ∈ Qr{−1} the sets PSpec1 A(2, a) 2 ] when n = 2. 0Y s2 and PSpec1 A(2, b) are homeomorphic. More precisely we have PSpec1 A(2, a) ∼= PSpec1 A(2, b) ∼= {hX0i,hX0, X1i,hX0, X1, X2 − µi µ ∈ k}. The following result explicitly describes the stratum PSpec0 A(2, a). Proposition 9.5. For a ∈ Q× we have PSpec0 A(2, a) = {h0i,hX0Y2i, Pλ λ ∈ k×}, where Pλ =( hX u2 0 Y v2 2 − λi h(X0Y2)v2 − λX v2−u2 0 −1 ≤ a < 0, a < −1 or a > 0. i The ideal Pλ is Poisson maximal if and only if −1 < a < 0. Proof. Let A = A(2, a) and let A◦ = A[X −1 discussion after Lemma 9.3 that Z = k[Y ′ 2 ] where Y ′ Spec Z and PSpec0 A sending Q ∈ Spec Z to ϕ(Q) = √QA◦ ∩ A. It is clear that ϕ(Y ′ If λ 6= 0, the ideal (Y ′ X u2−v2 2 − λ)A◦ of A◦ is prime and so ϕ((Y ′ (X0Y2)v2 , we have 0 Y v2 2 = X u2 0 ]. Let Z be the Poisson centre of A◦. Recall from the 2 . We denote by ϕ the bijection between 2 Z) = hX0Y2i. 2 − λ)A◦(cid:1) ∩ A. Since Y ′ 2 = u2 − v2 ≥ 0 else. 2 − λ)Z) = (cid:0)(Y ′ 2 − λi h(X0Y2)v2 − λX v2−u2 i 0 30 0 (9.6) Pλ := ϕ((Y ′ 2 − λ)Z) =( hY ′ But u2 − v2 = 2 d2 (p + q) and we have: Thus PSpec A is as described. (u2 − v2) ≥ 0 ⇐⇒ −1 ≤ a < 0. If −1 < a < 0 then u2 − v2 > 0 and Pλ = hX u2−v2 0 all primes in PSpec1(A). If a = −1 then Pλ = hX0X2 − X 2 hX0, X 2 1 /2 − λi. If a < −1 or a > 0 then v2 − u2 > 0 and Pλ is clearly contained in hX0, X1i. (X0Y2)v2 − λi is comaximal with X0 and thus with 1 /2 − λi is contained in the Poisson ideal (cid:3) We deduce the following result. Theorem 9.7. Let a ∈ Q× r {−1}. Then we have PSpec A(2, a) = {h0i,hX0i,hX0, X1i,hX0, X1, X2 − µi, Pλ,hX0Y2i µ ∈ k, λ ∈ k×} where the ideals Pλ are described in Proposition 9.5. Moreover only h0i and hX0, X1i are not Poisson primitive. Further, PSpec A(2, a) is homeomorphic to PSpec A(2, b) if and only if (a2 + a)/(b2 + b) > 0. Proof. The description of PSpec A(2, a) is immediate from Proposition 9.5. Pictorially, PSpec A(2, a) is given by the following diagrams, where lines represent inclusion of ideals. hX0, X1, X2 − µi hX0, X1i hX0, X1, X2 − µi hX0, X1i hX0i hX0Y2i Pλ hX0i hX0Y2i Pλ h0i h0i −1 < a < 0 a < −1 or a > 0 . Note that despite the fact that the algebras A(2, a) are not isomorphic for different values of a (see Theorem 5.1), their spectra fall generically into only two non homeomorphic families. Since the topology on PSpec A(2, a) is governed by inclusions, it is clear that PSpec A(2, a) and PSpec A(2, b) are homeomorphic if and only if a, b fall into the same case of (9.6), which is if and only if (a2 + a)/(b2 + b) > 0. (cid:3) Example 9.8. Let a = 1/2. Then p = −1, q = −2, d2 = 1, u2 = −5 and v2 = 1. Thus Pλ = hX0(X2 − λX 5 0 ) − 1/2X 2 1i ⊆ hX0, X1i. On the other hand when a = −1/2 we have p = −1, q = 2, d2 = 1, u2 = 3 and v2 = 1. Thus Pλ = hX 2 0 (X0X2 − 1/2X 2 1 ) − λi, which is Poisson maximal. Therefore the two spectra are not homeomorphic. Remark 9.9. From Theorem 7.5, when a 6∈ Q then PSpec A(2, a) = {h0i,hX0i,hX0Y2i,hX0, X1i,hX0, X1, X2 − µi µ ∈ k}. Note that these are the prime ideals that are Poisson for all values of a. For all a 6= −1, 0 there are five d-graded Poisson primes of A(2, a): the ideals {h0i,hX0i,hX0Y2i,hX0, X1i,hX0, X1, X2i}. 31 For the remainder of this section, we will study PSpec A(3, a) for various values of a. Again, we are most interested in PSpec0 A(3, a) which is homeomorphic to Spec Z as above. From Lemma 9.3 we know that Z is a free module over Z ′ = k[Y ′ 2 , Y ′ 3] with basis S = {X s 0Y s2 2 Y s3 3 0 ≤ si < vi, i = 2, 3 and ps + u2d2s2 + u3d3s3 = 0}. Proposition 9.10. The cardinality of S is equal to α := −p d2d3 . Proof. Since p = −d2v2 = −d3v3 the equation ps + u2d2s2 + u3d3s3 = 0 implies that d2 divides s3 (since gcd(d2, u3d3) = 1) and d3 divides s2 (since gcd(d3, u2d2) = 1). Set s2 := k2d3 and s3 := k3d2. Then the equation ps + u2d2s2 + u3d3s3 = 0 becomes (9.11) − αs + u2k2 + u2k2 = 0. To solve this Diophantine equation we first set t = −αs + u2k2 and we solve t + u3k3 = 0. We get t = −u3k, k3 = k,   for k ∈ Z. We now solve the Diophantine equation −αs + u2k2 = t = −u3k. Since gcd(−α, u2) = 1 there exist m, n ∈ Z such that −αm + u2n = 1. The solution of the equation −αs + u2k2 = −u3k is then for k, ℓ ∈ Z, and the solution of (9.11) is     For i = 2, 3 we have 0 ≤ ki < α = −p for k, ℓ ∈ Z. (9.12) s = −mu3k + u2ℓ, k2 = −nu3k + αℓ, s = −mu3k + u2ℓ, k2 = −nu3k + αℓ, k3 = k, since 0 ≤ si < vi. Fix k3 ∈ {0, . . . , α − 1}. Since k2 = −nu3k3 + αℓ for some ℓ ∈ Z, there exists a unique ℓ ∈ Z such that k2 ∈ {0, . . . , α − 1}. The integer s is uniquely determined by k2 and k3, so we conclude that to each k3 ∈ {0, . . . , α − 1} corresponds a unique monomial X s d2d3 (cid:3) 0Y s2 2 Y s3 3 ∈ S. Since α = 1 ⇐⇒ S = {1} we deduce the following corollary. Corollary 9.13. We have Z = Z ′ if and only if α = 1 if and only if p ∈ {−1,−2,−3,−6}. (cid:3) From the proof of Proposition 9.10 we observe that S = {X s 0Y d3k2 2 Y d2k3 3 s ∈ Z, 0 ≤ ki < α, and − αs + u2k2 + u3k3 = 0}, and that a recipe for finding these basis elements consists of computing m, m′ such that −αm+u2m′ = 1, plugging these values into (9.12) and, for each 0 ≤ k3 < α, finding the unique l ∈ Z such that k2 = −m′u3k3 + αl ∈ {0, . . . , α − 1}. We conclude this section with a couple of examples with α 6= 1 which illustrate possibilities for the ring structure of Z. But first we describe the d-graded Poisson prime that are common to A(3, a) for any generic a. By Theorem 7.5 we have PSpecgr A◦ ∼= Pk(2, 3) ⊔ {(Y2, Y3)}. In particular, for any [α : β] ∈ P1, the element 0 )3 + β(Y3X −1 F[α:β] = α(Y2X −1 32 0 )2 is Poisson central. Multiplying by X 6 0 , we obtain a pencil of Poisson normal sextic elements of A(4, a) X 6 0 F[α:β] = α(X 3 0 X 3 2 − 3 2 X 2 0 X 2 1 X 2 2 + 3 4 X0X 4 X 6 1 ) 1 X2 − 1 X 2 2 + 1 8 1 9 + β(X 4 0 X 2 3 + X 2 0 X 2 X 6 1 − 2X 3 0 X1X2X3 + 2 3 X 2 0 X 3 1 X3 − 2 3 X0X 4 1 X2). Therefore the d-graded Poisson prime ideals that are common to A(3, a) for any a 6= −2,−1, 0 are {h0i,hX0i,hX0Y2i,hX 2 0 Y3i, hX 6 0 F[α:β]i,hX0, X1i,hX0Y2, X 2 0 Y3i, hX0, X1, X2i,hX0, X1, X2, X3i [α : β] ∈ P1}. Among them the Poisson primitive ideals are hX0, X1i,hX0Y2, X 2 Example 9.14. For a = −5/4 (the algebra in Example 1.1) we have α = 5, Y ′ Moreover 0 Y3i and hX0, X1, X2i,hX0, X1, X2, X3i. 0 Y 5 3 . 2 and Y ′ 2 = X 3 3 = X 7 0 Y 5 S = {(X 2 0 Y2Y3)i i = 0, . . . , 4}, 2 Y ′ with (X 2 0 Y2Y3)5 = Y ′ 3 . Thus Z ∼= k[A, B, C]/hC5 = ABi. Example 9.15. Choose a = −24/5. We have α = 4, Y ′ S = {1, C, D, E}, where C := X −4 and B := Y ′ 3 , D := X −5 0 Y 6 0 Y 3 2 Y 4 2 Y 6 3 . We have 2 = X −7 0 Y 12 2 3 and E := X −6 0 Y 9 and Y ′ 2 Y 2 3 = X −3 2 . We have 3 . We further set A := Y ′ 2 0 Y 8 Z ∼= hC2 = BD, D2 = AB = CE, CD = BE, E2 = AD, DE = ACi k[A, B, C, D, E] . References [AS87] M. Artin and W. F. Schelter, Graded algebras of global dimension 3, Adv. Math. 66 (1987), no. 2, 171–216. MR 917738 [ATVdB90] M. Artin, J. Tate, and M. Van den Bergh, Some algebras associated to automorphisms of elliptic curves, The Grothendieck Festschrift, Vol. I, Progr. Math., vol. 86, Birkhauser Boston, Boston, MA, 1990, pp. 33–85. MR 1086882 [AVdB90] M. Artin and M. Van den Bergh, Twisted homogeneous coordinate rings, J. Algebra 133 (1990), no. 2, 249–271. MR 1067406 [AZ01] M. Artin and J. J. Zhang, Abstract Hilbert schemes, Algebr. Represent. Theory 4 (2001), no. 4, 305–394. MR 1863391 [BG02] K. A. Brown and K. R. Goodearl, Lectures on algebraic quantum groups, Advanced Courses in Mathematics. CRM Barcelona, Birkhauser Verlag, Basel, 2002. MR 1898492 [BGR73] W. Borho, P. Gabriel, and R. Rentschler, Primideale in Einhullenden auflosbarer Lie-Algebren (Beschrei- bung durch Bahnenraume), Lecture Notes in Mathematics, Vol. 357, Springer-Verlag, Berlin-New York, 1973. MR 0376790 [Bjo89] J. Bjork, The Auslander condition on Noetherian rings, S´eminaire d'Alg`ebre Paul Dubreil et Marie-Paul Malliavin, 39`eme Ann´ee (Paris, 1987/1988), Lecture Notes in Math., vol. 1404, Springer, Berlin, 1989, pp. 137– 173. MR 1035224 [BLLM] J. Bell, S. Launois, O. Leon Sanchez, and R. Moosa, Poisson algebras via model theory and differential-algebraic [BZ17] [CGG89] [Fre13] [Gin06] [GK66] [Goo06] geometry, to appear in Journal of the European Mathematical Society. J. Bell and J. J. Zhang, An isomorphism lemma for graded rings, Proc. Amer. Math. Soc. 145, 2017, no. 3, 989–994. MR 3589298 V. Coll, M. Gerstenhaber, and A. Giaquinto, An explicit deformation formula with noncommuting deriva- tions, Ring theory 1989 (Ramat Gan and Jerusalem, 1988/1989), Israel Math. Conf. Proc., vol. 1, Weizmann, Jerusalem, 1989, pp. 396–403. MR 1029329 G. Freudenburg, Foundations of invariant theory for the down operator, J. Symbolic Comput. 57 (2013), 19–47. MR 3066449 V. Ginzburg, Calabi-Yau algebras, arxiv:math/0612139, 2006. I. M. Gelfand and A. A. Kirillov, Sur les corps li´es aux alg`ebres enveloppantes des alg`ebres de Lie, Inst. Hautes ´Etudes Sci. Publ. Math. (1966), no. 31, 5–19. MR 0207918 K. R. Goodearl, A Dixmier-Moeglin equivalence for Poisson algebras with torus actions, Algebra and its applications, Contemp. Math., vol. 419, Amer. Math. Soc., Providence, RI, 2006, pp. 131–154. MR 2279114 33 [GW89] K. R. Goodearl and R. B. Warfield, Jr., An introduction to noncommutative Noetherian rings, London Math- [Har77] [IS80] ematical Society Student Texts, vol. 16, Cambridge University Press, Cambridge, 1989. MR 1020298 R. Hartshorne, Algebraic geometry, Springer-Verlag, New York-Heidelberg, 1977, Graduate Texts in Mathe- matics, No. 52. MR 0463157 R. S. Irving and L. W. Small, On the characterization of primitive ideals in enveloping algebras, Math. Z. 173 (1980), no. 3, 217–221. MR 592369 [Jor14] [Jos77] D. A. Jordan, Ore extensions and Poisson algebras, Glasgow Math. J. 56 (2014), no. 2, 355–368. MR 3187902 A. Joseph, A generalization of the Gelfand-Kirillov conjecture, Amer. J. Math. 99 (1977), no. 6, 1151–1165. MR 0460397 [KRS05] D. S. Keeler, D. Rogalski, and J. T. Stafford, Naıve noncommutative blowing up, Duke Math. J. 126 (2005), no. 3, 491–546. MR 2120116 [LBS93] L. Le Bruyn and S. P. Smith, Homogenized sl(2), Proc. Amer. Math. Soc. 118 (1993), no. 3, 725–730. [Lev92] [McC74] [MR01] MR 1136235 T. Levasseur, Some properties of noncommutative regular graded rings, Glasgow Math. J. 34 (1992), no. 3, 277–300. MR 1181768 J. C. McConnell, Representations of solvable Lie algebras and the Gelfand-Kirillov conjecture, Proc. London Math. Soc. (3) 29 (1974), 453–484. MR 0357529 J. C. McConnell and J. C. Robson, Noncommutative Noetherian rings, revised ed., Graduate Studies in Mathematics, vol. 30, American Mathematical Society, Providence, RI, 2001, With the cooperation of L. W. Small. MR 1811901 [Oh99] S. Q. Oh, Symplectic ideals of Poisson algebras and the Poisson structure associated to quantum matrices, Comm. Algebra 27 (1999), no. 5, 2163–2180. MR 1683857 [Pym15] [Ric02] B. Pym, Quantum deformations of projective three-space, Adv. Math. 281 (2015), 1216–1241. MR 3366864 L. Richard, Equivalence rationnelle et homologie de Hochschild pour certaines alg`ebres polynomiales classiques et quantiques, Ph.D. thesis, Universit´e Blaise Pascal, 2002. [Rog14] [RRZ14] M. Reyes, D. Rogalski, and J. J. Zhang, Skew Calabi-Yau algebras and homological identities, Adv. Math. D. Rogalski, An introduction to noncommutative projective geometry, arxiv: 1403.3065, 2014. 264 (2014), 308–354. MR 3250287 [Shi05] [Sig84] [ST01] [vdE93] E. N. Shirikov, Two-generated graded algebras, Algebra Discrete Math. (2005), no. 3, 60–84. MR 2237896 G. Sigurdsson, Differential operator rings whose prime factors have bounded Goldie dimension, Arch. Math. (Basel) 42 (1984), no. 4, 348–353. MR 753356 B. Shelton and C. Tingey, On Koszul algebras and a new construction of Artin-Schelter regular algebras, J. Algebra 241 (2001), no. 2, 789–798. MR 1843325 A. van den Essen, An algorithm to compute the invariant ring of a Ga-action on an affine variety, J. Symbolic Comput. 16 (1993), no. 6, 551–555. MR 1279532 [Zha96] J. J. Zhang, Twisted graded algebras and equivalences of graded categories, Proc. London Math. Soc. (3) 72 (1996), no. 2, 281–311. MR 1367080 (Lecoutre) School of Mathematics, Statistics and Actuarial Science, University of Kent, Canterbury CT2 7NF, UK. E-mail address: [email protected] (Sierra) School of Mathematics, Peter Guthrie Tait Road, King's Buildings, University of Edinburgh, Edinburgh EH9 3FD, UK. E-mail address: [email protected] 34
1109.3290	1	1109	2011-09-15T08:47:04	Some problems in operad theory	[ "math.RA", "math.QA" ]	This is a list of some problems and conjectures related to various types of algebras, that is to algebraic operads. Some comments and hints are included.	math.RA	math	October 24, 2018 20:15 WSPC - Proceedings Trim Size: 9in x 6in PbsOperadicWS6 1 Some problems in operad theory Jean-Louis Loday Institut de Recherche Math´ematique Avanc´ee CNRS et Universit´e de Strasbourg 7 rue R. Descartes 67084 Strasbourg Cedex, France e-mail: [email protected] This is a list of some problems and conjectures related to various types of algebras, that is to algebraic operads. Some comments and hints are included. Keywords: Operad, dendriform, algebra up to homotopy, Hopf algebra, octo- nion, Manin product. Introduction Since 1991 I got involved in the operad theory, namely after an enlightening lecture by Misha Kapranov in Strasbourg. During these two decades I came across many questions and problems. The following is an excerpt of this long list which might be helpful to have in mind while working in this theme. Of course this is a very personal choice. Notation and terminology are those of [14] and [17]. Various types of algebras, i.e. algebraic operads, can be found in [23]. 1. On the notion of group up to homotopy The notion of associative algebra up to homotopy is well-known: it is called A∞-algebra and was devised by Jim Stasheff in [21]. It has the following important property: starting with a differential graded associative algebra (A, d), if (V, d) is a deformation retract of (A, d), then (V, d) is not a dg associative algebra in general, but it is an A∞-algebra. This is Kadeishvili's theorem [9], see [17] for a generalization and variations of it. It is called the Homotopy Transfer Theorem. Let us now start with a group G. What is the notion of a "group up to homotopy" ? To make this question more precise we move, as in quantum group theory, to the group algebra K[G] October 24, 2018 20:15 WSPC - Proceedings Trim Size: 9in x 6in PbsOperadicWS6 2 of the group G over a field K. It is well-known that this is not only a unital associative algebra, but it is a cocommutative Hopf algebra. So, it has a cocommutative coproduct ∆ (induced by the diagonal on G), and the existence of an inverse in G translates to the existence of an antipode on the group algebra. So we can now reformulate the question as follows: "What is the notion of cocommutative Hopf algebra up to homotopy ?" One of the criterions for the answer to be useful would be the existence of a Homotopy Transfer Theorem for cocommutative Hopf algebras. One has to be careful enough to take into account that the existence of a unit and a counit is part of the structure of a Hopf algebra. In the associative case the operad A∞ does not take the unit into account. See [14] for the Hopf relation of a nonunital bialgebra. The fact that the tensor product of two associative algebras is still an associative algebra plays a prominent role in the definition of a bialgebra (a fortiori a Hopf algebra). So it is clear that a first step in analyzing this problem is to check whether one can put an A∞-structure on the tensor product of two A∞-algebras and to unravel the properties of such a con- struction. A first answer has been given by Saneblidze and Umble in [20]. But this tensor product is not associative. This problem has been addressed in [15]. 2. Subgroup of free group It is well-known that a subgroup of a free group is free. The proof is topo- logical in the sense that it consists in letting the free group act on a tree. Could one find a proof by looking at the properties of the associated group algebra (which is a Hopf algebra) ? I am thinking about something sim- ilar to the theorems which claim that some algebra is free under certain condition (PBW type theorems, see [14]). 3. The octonions as an algebra over a Koszul operad The octonions form a normed division algebra O of dimension 8, see for instance [1]. The product is known not to be associative contrarily to the other normed division algebras R, C and H. However it does satisfy some algebraic relation: it is an alternative algebra. Let us recall that an alterna- tive algebra is a vector space equipped with a binary operation x · y, such that the associator (x, y, z) := (x · y) · z − x · (y · z) is antisymmetric: (x, y, z) = −(y, x, z) = −(x, z, y). October 24, 2018 20:15 WSPC - Proceedings Trim Size: 9in x 6in PbsOperadicWS6 3 It turns out that the operad of alternative algebras is not too good, because it is not a Koszul operad, cf. [7]. Whence the question: Find a (small) binary operad such that the octonions form an algebra over this operad and such that this operad is Koszul. Of course the ambiguity of the question is in the adjective "small". Because such an operad exists: it suffices to take the magmatic operad on one binary operation. But there may exists a smaller operad (i.e. a quotient of M ag), which is best. This operad need not be quadratic, that is, we may look for relations involving 4 variables, like in Jordan algebras. 4. Commutative algebras up to homotopy in positive characteristic In positive characteristic p it is best to work with divided power algebras rather than commutative algebras. The notion of commutative algebra up to homotopy is well-known: it is the C∞-algebras (also denoted Com∞, see [17]). What is, explicitly, the notion of divided power algebra up to homotopy in characteristic p ? Theoretically the problem can be solved as follows. One can perform the theory of Koszul duality for operads with divided powers, cf. [8]. Any such operad with divided powers ΓP, which is Koszul, gives rise to a dg operad with divided powers ΓP∞. A divided power algebra up to homotopy is an algebra over ΓP∞. The point is to make all the steps of the theory explicit in the case P = Com. 5. Manin black product for operads What is the operad Com • Ass ? One is asking for a small presentation by generators and relations. 6. L-dendriform algebras and operadic black product By definition an L-dendriform algebra is a vector space equipped with two operations x ≺ y and x ≻ y satisfying (x ≺ y) ≺ z + y ≻ (x ≺ z) = x ≺ (y ∗ z) + (y ≻ x) ≺ z, (x ∗ y) ≻ z + y ≻ (x ≻ z) = x ≻ (y ≻ z) + (y ≺ x) ≻ z, where x ∗ y = x ≺ y + x ≻ y (cf. [3]). This is one of the numerous ways of splitting the associativity of the operation ∗. Conjecture: the operad encoding L-dendriform algebras is a Manin black October 24, 2018 20:15 WSPC - Proceedings Trim Size: 9in x 6in PbsOperadicWS6 4 product: preLie • preLie = L-Dend. In favor of this conjecture we have the following facts (cf. [22]): preLie • Com = Zinb, preLie • Ass = Dend, preLie • Lie = preLie, and also preLie • Dend = Quad. Similarly it seems that preLie • Zinb = ComQuad and that the operads Octo, L-Quadri, L-Octo introduced in [10] are also black products: preLie • Quadri = Octo, preLie • L-Dend = L-Quadri, preLie • L-Quadri = L-Octo. Note. Some of these questions have been recently settled in [2]. 7. Resolutions of associative algebras Koszul duality theory for associative algebras gives a tool to construct free resolutions for some associative algebras, and even the minimal resolutions in certain cases. When the algebra is a group algebra, then there are tools to construct (at least the beginning of) a resolution by taking the free module on the set of generators, then of relations, then of relations between the relations, and so forth (syzygies), see for instance [12]. It would be very interesting to compare these various methods. 8. Hidden structure for EZ-AW maps Given a deformation retract h (A, dA) p i (V, dV ) , pi = idV , IdA − ip = dAh + hdA, the HTT says that an algebraic structure on (A, dA) can be transferred to some other algebraic structure (the hidden one) on (V, dV ). This principle is not special to chain complexes and can be applied to other situations as shown for crossed modules in [17]. Apply this principle to the Eilenberg- Zilber and Alexander-Whitney quasi-isomorphisms. Let us recall that, for X and Y being simplicial modules, these isomorphisms relate the chain complex C•(X × Y ) to the tensor product C•(X) ⊗ C•(Y ). % % / / o o October 24, 2018 20:15 WSPC - Proceedings Trim Size: 9in x 6in PbsOperadicWS6 5 9. Interpolating between Dend and C om The so-called En-operads are operads which interpolate between the homo- topy class of the operad Ass, which contains A∞, and the homotopy class of the operad Com, which contains C∞. So, it solves the question: what is an associative algebra which is more or less commutative. The answer is: an En-algebra ; the larger n is, the more commutative it is. Question: what is a dendriform algebra which is more or less commuta- tive ? In other words we are looking for an interpolation Dn between the op- erads Dend∞ and Zinb∞, where Dend is the operad of dendriform algebras (two generating operations ≺ and ≻ and three relations) and Zinb is the operad of Zinbiel algebras (dendriform algebras such that x ≻ y = y ≺ x). One of the motivation for finding D2 is the following. It is known that the Grothendieck-Teichmuller group is related to the operad E2. Knowing D2 could lead to a dendriform version of the Grothendieck-Teichmuller group. 10. Good triples of binary quadratic operads Let P be a binary quadratic operad which is Koszul (cf. for instance [17,18]). It gives a notion of P-algebra and also a notion of P-coalgebra. We conjec- ture that there is a compatibility relation which defines a notion of P c-P- bialgebra such that (P, P, Vect) is a good triple of operads in the sense of [14]. Comments. There are many examples known: P = Com, As, Dend, M ag, 2-as. When it holds, it gives a criterion for proving that a given P-algebra is free. 11. On the coalgebra structure of Connes-Kreimer Hopf algebra The Connes-Kreimer Hopf algebra is an algebra of polynomials endowed with an ad hoc coproduct, cf. [5]. It is known that the indecomposable part is not only coLie, but in fact co-pre-Lie, cf. [4]. If we linearly dualize (as graded modules), the Hopf algebra is the Grossman-Larson Hopf algebra, which is cocommutative and its primitive part is pre-Lie. I conjecture that there is some type of algebras, that is some operad X , and some type of Comc-X -bialgebras, which fit into a good triple of operads (Com, X , preLie). October 24, 2018 20:15 WSPC - Proceedings Trim Size: 9in x 6in PbsOperadicWS6 6 If so, then the Grossman-Larson algebra would be the free X -algebra on one generator. The solution of this problem in the noncommutative framework is given by the operad Dend, cf. [16]. 12. Generalized bialgebras in positive characteristic The Poincar´e-Birkhoff-Witt theorem and the Cartier-Milnor-Moore theo- rem are structure theorems for cocommutative bialgebras in characteristic zero. They can be summarized by saying the triple of operads (Com, As, Lie) is a good triple. Several other good triples have been described in [14], some of them being valid in any characteristic, like the triple (As, Dup, M ag) for instance. For the classical case, it is known that, in order for the CMM theorem to be true in characteristic p, one has to replace the notion of Lie algebra by the notion of restricted Lie algebras. Operadically, restricted Lie algebras, divided power algebras and the like are obtained by replacing the "coinvariants" in the definition of an operad by the "invariants", cf. [8], ΓP(V ) := X (P(n) ⊗ V ⊗n)Sn . n So the PBW-CMM theorem in characteristic p can be phrased by saying that (Com, As, ΓLie) is a good triple. Note that As = ΓAs, so equivalently (Com, ΓAs, ΓLie) is a good triple. It would be very interesting to generalize the results on generalized bialgebras to positive characteristic along these lines, that is, to show that, when (C, A, P) is a good triple, then so is (C, ΓA, ΓP). Similarly, (ΓCom, As, Lie) is a good triple. One should be able to show that, when (C, A, P) is a good triple, then so is (ΓC, A, P). 13. Higher Dynkin diagrams and operads Show that there exists some types of algebras (i.e. some operads) for which the finite dimensional simple algebras are classified by the diagrams de- scribed by Ocneanu in [19]. The toy-model is the operad Lie and the Dynkin diagrams. October 24, 2018 20:15 WSPC - Proceedings Trim Size: 9in x 6in PbsOperadicWS6 7 14. Coquecigrues It is well-known that a Lie group admits a tangent space at the unit element which is a Lie algebra. But there is also another relationship between groups and Lie algebras, more specifically between discrete groups and Lie algebras (over Z). It is given by the descending central series. For G a discrete group, G(1) = [G, G] is its commutator subgroup, and, more generally, G(n) = [G, G(n−1)] is the nth term of the descending central series. It is well-known that the graded abelian group Ln G(n)/G(n+1) is a Lie algebra whose bracket is induced by the commutator in G. The Jacobi identity is a consequence of a nice (and not so well-known) relation, valid in any group G, called the Philip Hall relation (see for instance [12] for some drawing of it related to the Borromean rings). A natural question is the following. Let P be a variation of the operad Lie (we have in mind preLie and Leib). Is there some structure playing the role of groups in this realm ? For Leibniz algebras the question arised natu- rally in my research on the periodicity properties of algebraic K-theory, cf. [11,13]. I called this conjectural object a coquecigrue. In fact I was more in- terested in the cohomology theory which should come with this new notion, to apply it further to groups. Recent progress using the notion of racks was achieved by Simon Covez in [6]. 15. Homotopy groups of spheres Let p be a prime number. Let (πS • (X))p be the stable homotopy groups of the pointed connected topological space X, localized at p. Find a type of algebras such that (πS • (X))p is an algebra of this type and such that (πS • (∗))p is the free algebra of this type over one generator (in degree 2p−3). Comments. The Toda brackets are likely to play a role in this problem. References 1. J.C. Baez, The octonions. Bull. Amer. Math. Soc. (N.S.) 39 (2002), no. 2, 145 -- 205. 2. C. Bai, O. Bellier, L. Guo, X. Ni, Splitting of operations, Manin products and Rota-Baxter operators. Preprint arXiv:1106.6080 3. C. Bai, L. Liu and X. Ni, Some results on L-dendriform algebras, J. Geom. Phys. 60 (2010), 940 -- 950. 4. F. Chapoton, and M. Livernet, Pre-Lie algebras and the rooted trees operad. Internat. Math. Res. Notices 2001, no. 8, 395 -- 408. 5. A. Connes, D. Kreimer, Hopf algebras, renormalization and noncommutative geometry. Comm. Math. Phys. 199 (1998), no. 1, 203 -- 242. October 24, 2018 20:15 WSPC - Proceedings Trim Size: 9in x 6in PbsOperadicWS6 8 6. S. Covez, The local integration of Leibniz algebras. Preprint: arXiv:1011.4112. 7. A. Dzhumadil'daev and P. Zusmanovich, The alternative operad is not Koszul, Experimental Mathematics 20 (2011), 138-144. 8. B. Fresse, On the homotopy of simplicial algebras over an operad. Trans. Amer. Math. Soc. 352 (2000), no. 9, 4113 -- 4141. 9. T. V. Kadeishvili, The algebraic structure in the homology of an A(∞)- algebra, Soobshch. Akad. Nauk Gruzin. SSR 108 (1982), no. 2, 249 -- 252 (1983). 10. L. Liu, X. Ni, C. Bai, L-quadri-algebras (in Chinese), Sci. Sin. Math. 42 (2011) 105 -- 124. 11. J.-L. Loday, Comparaison des homologies du groupe lin´eaire et de son alg`ebre de Lie. Ann. Inst. Fourier (Grenoble) 37 (1987), no. 4, 167 -- 190. 12. J.-L. Loday, Homotopical syzygies, in "Une d´egustation topologique: Homo- topy theory in the Swiss Alps", Contemporary Mathematics no 265 (AMS) (2000), 99 -- 127. 13. J.-L. Loday, Algebraic K-theory and the conjectural Leibniz K-theory, K- theory (2003), 105 -- 127. 14. J.-L. Loday, Generalized bialgebras and triples of operads. Ast´erisque No. 320 (2008), x+116 pp. 15. J.-L. Loday, Geometric diagonals for the Stasheff associahedron and products of A-infinity algebras, preprint (2011). 16. J.-L. Loday and M.O. Ronco, Combinatorial Hopf algebras, in Quanta of Maths, Clay Mathematics Proceedings, vol. 11, Amer. Math. Soc., Provi- dence, RI, 2011, pp. 347 -- 383. 17. J.-L. Loday and B. Vallette, Algebraic Operads, (2010), book, submitted. 18. M. Markl, S. Shnider, J. Stasheff, Operads in algebra, topology and physics. Mathematical Surveys and Monographs, 96. American Mathematical Society, Providence, RI, 2002. x+349 pp. 19. A. Ocneanu, The classification of subgroups of quantum SU(N), Quantum symmetries in theoretical physics and mathematics (Bariloche, 2000), Con- temp. Math., vol. 294, Amer. Math. Soc., Providence, RI, 2002, pp. 133 -- 159. 20. S. Saneblidze, R. Umble, Diagonals on the permutahedra, multiplihedra and associahedra. Homology Homotopy Appl. 6 (2004), no. 1, 363 -- 411. 21. J. D. Stasheff, Homotopy associativity of H-spaces. I, II, Trans. Amer. Math. Soc. 108 (1963), 275-292; ibid. 108 (1963), 293 -- 312. 22. B. Vallette, Manin products, Koszul duality, Loday algebras and Deligne conjecture. J. Reine Angew. Math. 620 (2008), 105 -- 164. 23. G.W. Zinbiel, Encyclopedia of types of algebras 2010, this volume and arXiv:1101.0267.
1303.0920	1	1303	2013-03-05T03:54:11	Free associative algebras, noncommutative Grobner bases, and universal associative envelopes for nonassociative structures	[ "math.RA", "math.RT" ]	These are the lecture notes from my short course of the same title at the CIMPA Research School on Associative and Nonassociative Algebras and Dialgebras: Theory and Algorithms - In Honour of Jean-Louis Loday (1946-2012), held at CIMAT, Guanajuato, Mexico, February 17 to March 2, 2013. The underlying motivation is to apply the theory of noncommutative Grobner bases in free associative algebras to the construction of universal associative envelopes for nonassociative structures defined by multilinear operations. Trilinear operations were classified by the author and Peresi in 2007. In her Ph.D. thesis of 2012, Elgendy studied the universal associative envelopes of nonassociative triple systems obtained by applying these trilinear operations to the 2-dimensional simple associative triple system. In these notes I use computer algebra to extend some aspects of her work to the 4-dimensional and 6-dimensional simple associative triple systems.	math.RA	math	FREE ASSOCIATIVE ALGEBRAS, NONCOMMUTATIVE GR OBNER BASES, AND UNIVERSAL ASSOCIATIVE ENVELOPES FOR NONASSOCIATIVE STRUCTURES MURRAY R. BREMNER Abstract. These are the lecture notes from my short course of the same title at the CIMPA Research School on Associative and Nonassociative Algebras and Dialgebras: Theory and Algorithms - In Honour of Jean-Louis Loday (1946 -- 2012), held at CIMAT, Guanajuato, Mexico, February 17 to March 2, 2013. The underlying motivation is to apply the theory of noncommuta- tive Grobner bases in free associative algebras to the construction of universal associative envelopes for nonassociative structures defined by multilinear oper- ations. Trilinear operations were classified by the author and Peresi in 2007. In her Ph.D. thesis of 2012, Elgendy studied the universal associative envelopes of nonassociative triple systems obtained by applying these trilinear opera- tions to the 2-dimensional simple associative triple system. In these notes I use computer algebra to extend some aspects of her work to the 4-dimensional and 6-dimensional simple associative triple systems. 1. Introduction The primary goal of these lecture notes is to apply the theory of noncommuta- tive Grobner bases in free associative algebras to the construction of universal as- sociative envelopes for nonassociative structures defined by multilinear operations. Throughout I will take an algorithmic approach, developing just enough theory to motivate the computational methods. Some of the easier proofs and examples are left as exercises for the reader. Along the way, I will mention a number of open research problems. I begin by recalling the basic definitions of the most familiar examples of nonassociative structures: finite dimensional Lie and Jordan algebras and their universal associative enveloping algebras. Unless otherwise indicated, I will work over an arbitrary field F . 1.1. Lie algebras. Lie algebras are defined by the polynomial identities of degree ≤ 3 satisfied by the Lie bracket [x, y] = xy − yx in every associative algebra, namely anticommutativity and the Jacobi identity: [x, x] ≡ 0, [[x, y], z] + [[y, z], x] + [[z, x], y] ≡ 0. Every polynomial identity satisfied by the Lie bracket in every associative algebra is a consequence of these two identities; see Corollary 7.2. Definition 1.1. Let A be an associative algebra with product denoted xy. We write A− for the Lie algebra which has the same underlying vector space as A, but the original associative operation is replaced by the Lie bracket [x, y] = xy − yx. Let L be a Lie algebra over F . If L is isomorphic to a subalgebra of A− then we call A an associative envelope for L. 1 2 MURRAY R. BREMNER Example 1.2. Let L = sln(F ) be the special linear Lie algebra of all n×n matrices of trace 0 over F . Then clearly L is a subalgebra of A− where A = Mn(F ) is the associative algebra of all n × n matrices. Definition 1.3. The universal associative envelope U (L) of the Lie algebra L is the unital associative algebra satisfying the following universal property, which implies that U (L) is unique up to isomorphism: • There is a morphism of Lie algebras α : L → U (L)− such that for any unital associative algebra A and any morphism of Lie algebras β : L → A−, there is a unique morphism of associative algebras γ : U → A satisfying β = γ ◦ α. In the terminology of category theory, this says that the functor sending a Lie algebra L to its universal associative envelope U (L) is the left adjoint of the functor sending an associative algebra A to the Lie algebra A−. Lemma 1.4. The subset α(L) generates U (L). If A is an associative envelope for L, and A is generated by the subset L, then A is isomorphic to a quotient of U (L); that is, A ≈ U (L)/I for some ideal I. Proof. Exercise. (cid:3) We will see later that U (L) is always infinite dimensional, and that the map α is always injective, so that L is isomorphic to a subalgebra of U (L)−. These are corollaries of the PBW theorem (Theorem 7.1) that we will prove using the theory of noncommutative Grobner bases. Example 1.5. Let L be the n-dimensional Lie algebra with basis {x1, . . . , xn} and trivial commutation relations [xi, xj ] = 0 for all i, j. Then U (L) ≈ F [x1, . . . , xn], the algebra of commutative associative polynomials in n variables over F . 1.2. Jordan algebras. Assume that char F 6= 2. Jordan algebras are defined by the polynomial identities of degree ≤ 4 satisfied by the Jordan product x ◦ y = 1 2 (xy + yx) in every associative algebra, commutativity and the Jordan identity: x ◦ y ≡ y ◦ x, ((x ◦ x) ◦ y) ◦ x ≡ (x ◦ x) ◦ (y ◦ x). In contrast to Lie algebras, there exist further identities satisfied by the Jordan product in every associative algebra which are not consequences of these two iden- tities. The simplest such identities were discovered almost 50 years ago; they have degree 8 and are called the Glennie identities [56]. Definition 1.6. Let A be an associative algebra with product denoted xy. We write A+ for the Jordan algebra which has the same underlying vector space as A, but the original associative operation is replaced by the Jordan product x ◦ y = 1 2 (xy + yx). Let J be a Jordan algebra over F . If J is isomorphic to a subalgebra of A+ then we call A an associative envelope for J. Example 1.7. Let Sn(F ) be the Jordan algebra of symmetric n × n matrices with entries in F , and let A = Mn(F ) be the associative algebra of all n × n matrices. Exercise 1.8. Modify Definition 1.3 to define universal associative envelopes for Jordan algebras. State and prove the analogue of Lemma 1.4 for Jordan algebras. If J is finite dimensional, then so is its universal associative envelope U (J). On the other hand, the natural map from J to U (J) may not be injective; hence, strictly speaking, the universal associative envelope U (J) may not be an associative envelope in the sense of Definition 1.6. GR OBNER BASES AND UNIVERSAL ENVELOPES 3 Example 1.9. Let J be the n-dimensional Jordan algebra with basis {x1, . . . , xn} and trivial products xi ◦ xj = 0 for all i, j. Then U (J) ≈ Λ(x1, . . . , xn), the exterior (Grassmann) algebra on n generators over F , and so dim U (J) = 2n. We have the following definition, which has no analogue for Lie algebras. Definition 1.10. If a Jordan algebra J has an associative envelope then we call J a special Jordan algebra. Otherwise, we call J an exceptional Jordan algebra. Example 1.11. The vector space H3(O) of 3 × 3 Hermitian matrices over the 8- dimensional division algebra O of real octonions is closed under the Jordan product and is a 27-dimensional exceptional Jordan algebra. 2. Free Associative Algebras These lecture notes on the theory of noncommutative Grobner bases follow closely the exposition by de Graaf [43, §§6.1-6.2]. The most famous paper on this topic is by Bergman [9], but similar results were published a little earlier by Bokut [11]. Bokut's approach was based on Shirshov's work on Lie algebras [99]. (Shirshov's papers have appeared recently in English translation [100].) For further references, including current research directions, see Section 11. Definition 2.1. Let X = {x1, x2, . . . , xn, . . . } be an alphabet: a set of indeter- minates (sometimes called letters), finite or countably infinite. We impose a total order on X by setting xi ≺ xj if and only if i < j. We write X ∗ for the set of words (also called monomials) w = xi1 xi2 · · · xik where xi1 , xi2 , . . . , xik ∈ X and k ≥ 0. (If k = 0 then we have the empty word denoted w = 1.) The degree of a word w = xi1 xi2 · · · xik is the number of letters it contains, counting repetitions: deg(w) = k. We define concatenation on X ∗ by (u, v) 7→ uv for any u, v ∈ X ∗; this associative operation makes X ∗ into the free monoid generated by X. Example 2.2. If X = {a} has only one element, then X ∗ = { ak k ≥ 0 } is the set of all non-negative powers of a. The multiplication on X ∗ is given by aiaj = ai+j, so X ∗ is commutative. If X has two or more elements, then X ∗ is noncommutative. For example, if X = {a, b} then there are 2k distinct words of degree k for all k ≥ 0: k = 0 : k = 1 : k = 2 : k = 3 : k = 4 : 1 a, b a2, ab, ba, b2 a3, a2b, aba, ab2, ba2, bab, b2a, b3 a4, a3b, a2ba, a2b2, aba2, abab, ab2a, ab3, ba3, ba2b, baba, bab2, b2a2, b2ab, b3a, b4 Definition 2.3. A nonempty word u ∈ X ∗ is a subword (also called a factor or a divisor) of w ∈ X ∗ if w = v1uv2 for some v1, v2 ∈ X ∗. If v1 = 1 then u is a left subword of w; if v2 = 1 then u is a right subword of w. We say that u is a proper subword of w if u 6= w. Definition 2.4. The total order on X extends to a total order on X ∗, called the deglex (degree lexicographical) order, as follows: If u, w ∈ X ∗ then u ≺ w (we say u precedes w) if and only if either (i) deg(u) < deg(w), or (ii) deg(u) = deg(w) where u = vxiu′ and w = vxj w′ for some v, u′, w′ ∈ X ∗ and xi, xj ∈ X with xi < xj. 4 MURRAY R. BREMNER In condition (ii) we find the common left subword v of highest degree, and then compare the next letters xi and xj using the total order on X. We write u (cid:22) v when u ≺ v or u = v. We often write v ≻ u to mean u ≺ v. Example 2.5. Let X = {a, b} with a ≺ b. We list the words in X ∗ of degree ≤ 3 in deglex order; this is the same order as in Example 2.2: 1 ≺ a ≺ b ≺ a2 ≺ ab ≺ ba ≺ b2 ≺ a3 ≺ a2b ≺ aba ≺ ab2 ≺ ba2 ≺ bab ≺ b2a ≺ b3. Exercise 2.6. Let X = {a, b, c} with a ≺ b ≺ c. List the words in X ∗ of degree ≤ 3 in deglex order. Do the same with c ≺ b ≺ a. Definition 2.7. A total order on X ∗ is multiplicative if for all u, v, w ∈ X ∗ with u ≺ v we have uw ≺ vw and wu ≺ wv. (More concisely, we could require the single condition that w1uw2 ≺ w1vw2 for all u, v, w1, w2 ∈ X ∗.) Definition 2.8. A total order on X ∗ satisfies the descending chain condition (DCC) if whenever w1, w2, . . . , wn, · · · ∈ X ∗ with w1 (cid:23) w2 (cid:23) · · · (cid:23) wn (cid:23) · · · then for some n we have wn = wn+1 = · · · ; that is, there do not exist infinite strictly decreasing sequences. Equivalently, for any w ∈ X ∗ the set {v ∈ X ∗ v ≺ w} is finite. The DCC allows us to use induction on X ∗ with respect to the total order. Lemma 2.9. The total order ≺ on X ∗ from Definition 2.4 is multiplicative and satisfies the descending chain condition. Proof. Exercise. (cid:3) Definition 2.10. We write F hXi for the vector space with basis X ∗ over F . Concatenation in X ∗ extends bilinearly to F hXi: (cid:16)Xi aiui(cid:17)(cid:16)Xj bjvj(cid:17) = Xi,j aibjuivj (ai, bj ∈ F ; ui, vj ∈ X ∗). This multiplication makes F hXi into the free associative algebra generated by X over F . This is a unital algebra, since the empty word acts as the unit element. Elements of F hXi are linear combinations of monomials in X ∗, and we refer to them as noncommutative polynomials in the variables X with coefficients in F . (Here noncommutative means not necessarily commutative.) Example 2.11. If X = {a} has only one element, then F hXi is the same as F [a], the familiar algebra of commutative associative polynomials in one variable. If X has two or more elements, then F hXi and F [X] do not coincide: F [X] is commutative but F hXi is noncommutative. Definition 2.12. Consider a nonzero element f ∈ F hXi. We write f = Xi∈I aiui (ai ∈ F ; ui ∈ X ∗), where I is a nonempty finite index set and ai 6= 0 for all i ∈ I. The support of f is the set of all monomials occurring in f : support(f ) = { ui i ∈ I }. (If f = 0 then by convention its support is the empty set ∅.) For nonzero f ∈ F hXi, the support is a nonempty finite subset of X ∗; the greatest element of support(f ) with respect to the total order ≺ on X ∗ is the leading monomial of f , denoted GR OBNER BASES AND UNIVERSAL ENVELOPES 5 LM (f ). The coefficient of LM (f ) is the leading coefficient of f , denoted lc(f ). We say that f is monic if lc(f ) = 1. For any subset S ⊆ F hXi, we write LM (S) = { LM (f ) f ∈ S }. Example 2.13. For X = {a, b, c} and cab − bca + da − cb + a2 ∈ F hXi we have support(f ) = { a2, cb, da, bca, cab}, LM (f ) = cab, lc(f ) = 1. Definition 2.14. The standard form of a nonzero element f ∈ F hXi consists of f divided by lc(f ) with the monomials in reverse deglex order. Thus the standard form is monic and the leading monomial occurs in the first (leftmost) position. The polynomial f in the previous example is in standard form. 3. Universal Associative Envelopes of Lie and Jordan Algebras We use the concepts of the previous section to construct the universal associative envelopes of Lie and Jordan algebras. Definition 3.1. Every associative algebra A is isomorphic to a quotient F hXi/I for some set X and some ideal I ⊆ F hXi. If I is generated by the subset G ⊂ I then the pair (X, G) is a presentation of A by generators and relations. 3.1. Lie algebras. Let L be a Lie algebra of finite dimension d over F with basis X = { x1, . . . , xd }. The structure constants ck ij ∈ F are given by the equations [xi, xj] = d Xk=1 ck ij xk (1 ≤ i, j ≤ d). Let F hXi be the free associative algebra generated by X. (By a slight abuse of notation, we regard the basis elements of L as formal variables, but this should not cause confusion.) Let I be the ideal in F hXi generated by the d(d−1)/2 elements xixj − xj xi − d Xk=1 ck ijxk (1 ≤ j < i ≤ d). The quotient algebra U (L) = F hXi/I is the universal associative envelope of L. Example 3.2. We consider the Lie algebra sl2(F ) of 2 × 2 matrices of trace 0 over a field F of characteristic 0. We use the following notation for basis elements: h = E11 − E22 = (cid:20) 1 0 −1 (cid:21) , 0 e = E12 = (cid:20) 0 0 1 0 (cid:21) , f = E21 = (cid:20) 0 1 0 0 (cid:21) . The structure constants are given by these equations: [h, e] = 2e, [h, f ] = −2f, [e, f ] = h. From these equations we obtain the following set of generators G for the ideal I: he − eh − 2e, hf − f h + 2f, ef − f e − h. The universal associative envelope of sl2(F ) is the quotient U (sl2(F )) = F hh, e, f i/I. 6 MURRAY R. BREMNER 3.2. Jordan algebras. If J is a Jordan algebra with structure constants xi ◦ xj = d Xk=1 ck ij xk (1 ≤ i, j ≤ d), then we consider the ideal I generated by the d(d+1)/2 elements 1 2 (xixj + xjxi) − d Xk=1 ck ijxk (1 ≤ j ≤ i ≤ d), and U (J) = F hXi/I is the universal associative envelope of J. Example 3.3. We consider the Jordan algebra S2(F ) of symmetric 2 × 2 matrices over a field F of characteristic 0. We use the following notation for basis elements: a = E11 = (cid:20)1 0 0 0(cid:21) , b = E22 = (cid:20)0 0 0 1(cid:21) , c = E12 + E21 = (cid:20)0 1 1 0(cid:21) . The structure constants are given by these equations: a ◦ a = 2a, a ◦ b = 0, a ◦ c = c, b ◦ b = 2b, b ◦ c = c, c ◦ c = 2a + 2b. From these equations we obtain the following set G of generators for the ideal I: a2 − a, ba + ab, ca + ac − c, b2 − b, cb + bc − c, c2 − b − a. The universal associative envelope of S2(F ) is the quotient U (S2(F )) = F ha, b, ci/I. 4. Normal Forms of Noncommutative Polynomials To understand the structure of the quotient algebra F hXi/I, we need to find a basis for F hXi/I and express the product of any two basis elements as a linear combination of basis elements. This can be achieved easily if we can construct a Grobner basis for the ideal I: a set of generators (not a linear basis) for I with special properties which will be explained in detail in this section and the next. 4.1. Normal forms modulo an ideal. A basis for F hXi/I is a subset B of F hXi consisting of coset representatives: the elements b + I for b ∈ B are linearly inde- pendent in F hXi/I and span F hXi/I. Equivalently, B is a basis for a complement C(I) to I in F hXi, meaning that F hXi = I ⊕ C(I), the direct sum of subspaces. Lemma 4.1. Assume that I is an ideal in F hXi, and that B is a subset of F hXi. Then the set { b + I b ∈ B } is a basis of the quotient F hXi/I if and only if B is a basis for a complement of I in F hXi; that is, the elements of B are linearly independent in F hXi and F hXi = I ⊕ span(B). Proof. Exercise. (cid:3) Definition 4.2. Let I be an ideal in F hXi. The set N (I) of normal words modulo I is the subset of X ∗ consisting of all monomials which are not leading monomials of elements of I: N (I) = { w ∈ X ∗ w /∈ LM (I) }. The complement to I in F hXi is the subspace C(I) ⊆ F hXi with basis N (I). Proposition 4.3. We have F hXi = I ⊕ C(I). GR OBNER BASES AND UNIVERSAL ENVELOPES 7 Proof. We follow de Graaf [43, Proposition 6.1.1] but fill in some details. The proof consists for the most part of writing out the details in the division algorithm for noncommutative polynomials. First, we prove that I ∩ C(I) = {0}. Assume that f ∈ I and f ∈ C(I). If f 6= 0 then since f ∈ I, its leading monomial LM (f ) belongs to LM (I); but since f ∈ C(I), its leading monomial belongs to N (I), and hence does not belong to LM (I). This contradiction implies that f = 0. Second, we prove that any f ∈ F hXi can be written as f = g + h where g ∈ I and h ∈ C(I). This is clear for f = 0 (take g = h = 0), so we assume that f 6= 0. We use induction on leading monomials with respect to the total order ≺ on X ∗. For the basis of the induction, assume that LM (f ) = 1 (the empty word). Then f = α ∈ F \ {0}. If I = F hXi then N (I) = ∅ and C(I) = {0}; we have f = α + 0 where α ∈ I and 0 ∈ C(I). If I 6= F hXi then 1 /∈ LM (I) so 1 ∈ N (I); we have f = 0 + α where 0 ∈ I and α ∈ C(I). Since X ∗ satisfies the DCC, we may now assume the claim for all f0 ∈ F hXi with LM (f0) ≺ LM (f ). This is the inductive hypothesis, which depends on the fact that only finitely many elements of X ∗ precede LM (f ). We have f = αLM (f ) + f0 where α = lc(f ) ∈ F , and either f0 = 0 or LM (f0) ≺ LM (f ). If f0 = 0 then f = αLM (f ); if LM (f ) ∈ I then f = αLM (f ) + 0 ∈ I + C(I), and if LM (f ) /∈ I then LM (f ) ∈ N (I) and f = 0 + αLM (f ) ∈ I + C(I). If f0 6= 0 then LM (f0) ≺ LM (f ), and by induction we have f0 = g0 + h0 where g0 ∈ I and h0 ∈ C(I). We now have two cases: LM (f ) ∈ N (I) and LM (f ) /∈ N (I). If LM (f ) ∈ N (I) then f = αLM (f ) + (g0 + h0) = g0 +(cid:0) αLM (f ) + h0(cid:1) ∈ I + C(I). If LM (f ) /∈ N (I) then by definition of N (I) we have LM (f ) = LM (k) for some k ∈ I \ {0}. (We cannot assume that LM (f ) ∈ I. This raises an important issue: we are non-constructively choosing an element k ∈ I which has the same leading monomial as the element f . Finding an algorithm to construct such an element k is one of the main goals of the theory of noncommutative Grobner bases.) Write k = βLM (k) + k0 where β = lc(k) ∈ F \ {0}, and either k0 = 0 or LM (k0) ≺ LM (k) = LM (f ). Then f − α β k = (cid:16)αLM (f ) + (g0 + h0)(cid:17) − α β(cid:16)βLM (k) + k0(cid:17) = αLM (f ) + g0 + h0 − αLM (k) − k0 α β If k0 = 0 then = g0 + h0 − α β k0 since LM (f ) = LM (k). f = (cid:16) α β k + g0(cid:17) + h0 ∈ I + C(I). If k0 6= 0 then by induction k0 = ℓ0 + m0 where ℓ0 ∈ I and m0 ∈ C(I). We have k + g0 + h0 − k0 α β α f = = α β α β k + g0 + h0 − = (cid:16) α β k + g0 − α β β(cid:0)ℓ0 + m0(cid:1) ℓ0(cid:17) +(cid:16)h0 − α β m0(cid:17). 8 MURRAY R. BREMNER The first three terms belong to I, and the last two terms belong to C(I). (cid:3) Corollary 4.4. Let I be an ideal in F hXi. Then every element f ∈ F hXi has a unique decomposition f = g + h where g ∈ I and h ∈ C(I). Proof. This follows immediately from the definition of direct sum. (cid:3) Definition 4.5. For any element f ∈ F hXi and any ideal I ⊆ F hXi, the element h ∈ C(I) which is uniquely determined by Corollary 4.4 is called the normal form of f modulo I, and is denoted NFI (f ) or NF (f ) if I is understood. Lemma 4.6. Let I ⊆ F hXi be an ideal. Define a product f · g on C(I) as follows: For any f, g ∈ C(I) set f · g = NFI (f g). Then the algebra consisting of the vector space C(I) with the product f · g is isomorphic to the quotient algebra F hXi/I. Proof. Exercise. (cid:3) Lemma 4.6 shows how to find a basis and structure constants for F hXi/I. But this depends on being able to determine the basis N (I) of the complement C(I), and to calculate the normal form NFI (f ) for every element f ∈ F hXi. 4.2. Computing normal forms. Our next task is to find an algorithm for which the input is an element f ∈ F hXi and an ideal I ⊆ F hXi given by a set G of generators, and the output is the normal form NFI (f ). We present an algorithm for computing the normal form NF (f, G) of f with respect to the set G. Unfortunately, the output of this algorithm depends on the set G; that is, if G1 and G2 are two generating sets for the same ideal I, then we may have NF (f, G1) 6= NF (f, G2). Furthermore, even for one set G, the output may depend on the choice of reductions performed at each step of the algorithm; see Example 4.9 below. Therefore in general the output is not the normal form of f modulo I. The important property of a Grobner basis is that if G is a Grobner basis for I then NF (f, G) = NFI (f ). Definition 4.7. Let f be an element of F hXi and let G be a finite subset of F hXi. We say that f is in normal form with respect to G if the following condition holds: • For every generator g ∈ G and every monomial w ∈ support(f ), the leading monomial LM (g) is not a subword of w. We first give an informal description of the algorithm for computing the normal form of f with respect to G. This algorithm is similar to the calculation in the proof of Proposition 4.3; it is a division algorithm for noncommutative polynomials. We may assume without loss of generality that the elements of G are monic. Consider the set LM (G) of leading monomials of the elements of G. For each v ∈ LM (G) and w ∈ support(f ) we can easily determine if v is a subword of w. If this never occurs, then f is in normal form with respect to G, and the algorithm terminates. Otherwise, w = u1vu2 for some u1, u2 ∈ X ∗, and f contains the term αw for some α ∈ F \ {0}. There exists g ∈ G with LM (g) = v; we replace f by f2 = f − αu1gu2. This reduction step eliminates from f the term αw. Repeating this procedure, we obtain a sequence f1 = f, f2, f3, . . . , fn, . . . of elements of F hXi; this sequence converges since X ∗ satisfies the DCC. This algorithm is given in pseudocode in Figure 1. GR OBNER BASES AND UNIVERSAL ENVELOPES 9 NormalForm(f, G) Input: An element f ∈ F hXi and a finite monic subset G ⊂ F hXi. Output: The normal form of f with respect to G. (1) Set n ← 0, f0 ← 0, f1 ← f . (2) While fn 6= fn+1 do: (a) Set n ← n + 1. (b) If w = u1vu2 for some v ∈ LM (G) and w ∈ support(fn) then set fn+1 ← fn − αu1gu2 where v = LM (g) else set fn+1 ← fn. (3) Return fn. Figure 1. Algorithm for a normal form of f with respect to G Lemma 4.8. For the algorithm of Figure 1, we have LM (f1) (cid:23) LM (f2) (cid:23) LM (f3) (cid:23) · · · (cid:23) LM (fn) (cid:23) · · · , and so LM (fn) = LM (fn+1) = · · · for some n ≥ 1. Hence the algorithm termi- nates, and its output fn is a normal form of f with respect to G. Furthermore, fn + I = f + I in F hXi/I; that is, fn is congruent to f modulo the ideal I generated by G. Proof. Exercise. (cid:3) A normal form of f with respect to G is not uniquely determined by the algorithm of Figure 1: the output depends on the choices made of v and w in step (2)(b). In particular, it follows that the output of the algorithm does not necessarily equal NFI (f ), which is uniquely determined by Corollary 4.4. Example 4.9. Let X = {a, b, c} and let I ⊂ F hXi be the ideal generated by G = { a2 − a, ba + ab, b2 − b, ca + ac − c, cb + bc − c, c2 − b − a }. (We have seen this set before in Example 3.3.) For convenience, we write each generator in standard form, and the generators are sorted in deglex order of their leading monomials. We compute the normal form of f1 = c2b with respect to G in two different ways, and obtain two different answers. We will see in Example 6.9 that NFI (c2b) = b, so neither of these two calculations produces the desired result. (1) Starting with g6 = c2 − b − a we obtain f2 = f1 − g6b = c2b − (c2b − b2 − ab) = b2 + ab. Next using g3 = b2 − b we obtain f3 = f2 − g3 = b2 + ab − (b2 − b) = ab + b. No further reductions are possible; the algorithm terminates with output ab + b. (2) Starting with g5 = cb + bc − c we obtain f2 = f1 − cg5 = c2b − (c2b + cbc − c2) = −cbc + c2. Next using g5 again we obtain f3 = f2 + g5c = −cbc + c2 + (cbc + bc2 − c2) = bc2. 10 MURRAY R. BREMNER Using g6 = c2 − b − a gives f4 = f3 − bg6 = bc2 − (bc2 − b2 − ba) = b2 + ba. Using g3 = b2 − b gives f5 = f4 − g3 = b2 + ba − (b2 − b) = ba + b. Finally, using g2 = ba + ab we obtain f6 = f5 − g2 = ba + b − (ba + ab) = −ab + b. No further reductions are possible; the algorithm terminates with output −ab + b. 5. Grobner Bases for Ideals in F hXi If the set G of generators of the ideal I has a certain special property, stated in the next definition, then the output of the algorithm of Figure 1 is uniquely determined, and equals the normal form of f modulo I. Definition 5.1. Let X be a finite set and let G be a set of generators for the ideal I in the free associative algebra F hXi. We say that G is a Grobner basis for I if the following condition holds: • For every nonzero element f ∈ I there is a generator g ∈ G such that LM (g) is a subword of LM (f ). In other words, the leading monomial of every nonzero element of the ideal contains a subword equal to the leading monomial of some generator of the ideal. Remark 5.2. A Grobner basis is not a basis in the sense of linear algebra: it is not a basis for I as a vector space over F , but rather a set of generators for I. In this context, basis means set of generators. Unfortunately, this misleading terminology is so well-established that we have no choice but to accept it. The next theorem shows why Grobner bases are so important. Recall that the set N (I) of all normal words modulo I is the complement of LM (I) in X ∗: the set of all words which are not leading monomials of elements of I. If we have a Grobner basis for I, then we can easily compute N (I) using part (a) of the next theorem, and we can easily compute NFI (f ) for all f ∈ F hXi using part (b): Theorem 5.3. If G is a Grobner basis for the ideal I ⊆ F hXi then: (a) N (I) = { w ∈ X ∗ for all g ∈ G, LM (g) is not a subword of w }. (b) For all f ∈ F hXi we have NFI (f ) = NF (f, G): the normal form of f modulo I equals the normal form of f with respect to G. Proof. Part (a) follows immediately from Definitions 4.2 and 5.1. For part (b), consider f ∈ F hXi and let h = NF (f, G) be the normal form of f with respect to G computed by the algorithm of Figure 1. For any w ∈ support(h), since h ∈ I and G is a Grobner basis for I, we know by Definition 4.7 that for all g ∈ G, LM (g) is not a subword of w. Part (a) of the theorem now shows that w ∈ N (I); since this holds for all w ∈ support(h), we have h ∈ C(I). By the last statement of Lemma 4.8 we know that f − h ∈ I. Clearly f = (f − h) + h ∈ I ⊕ C(I), and hence the uniqueness of the decomposition in Corollary 4.4 implies that h = NFI (f ). (cid:3) Theorem 5.3 is a beautiful result, but we still have the following problem: • Find an algorithm for which the input is a set G of generators for the ideal I ⊆ F hXi, and for which the output is a Grobner basis of I. GR OBNER BASES AND UNIVERSAL ENVELOPES 11 This requires defining overlaps and compositions for two generators g1, g2 ∈ G (Definition 5.10), and proving the Composition (Diamond) Lemma (Lemma 6.2). Definition 5.4. Let X be a finite set and let G be a finite subset of F hXi. We say that G is self-reduced if the following two conditions hold: (1) Every g ∈ G is in normal form with respect to G \ {g}. (2) Every g ∈ G is in standard form; in particular, lc(g) = 1. Remark 5.5. Condition (1) in Definition 5.4 is stronger than the condition given by de Graaf [43, Definition 6.1.5], which requires only that for all g ∈ G and for all h ∈ G \ {g}, LM (h) is not a subword of LM (g). The definition of de Graaf is analogous to the row-echelon form of a matrix, whereas our definition is analogous to the reduced row-echelon form (and is therefore somewhat more canonical). Exercise 5.6. Referring to Remark 5.5, explain the analogy between row-echelon forms of matrices and self-reduced sets of noncommutative polynomials in F hXi. (Consider finite sets of homogeneous polynomials of degree 1.) By calling the algorithm of Figure 1 repeatedly, we can create an algorithm for which the input is a finite subset G ⊂ F hXi generating an ideal I ⊆ F hXi and the output is a self-reduced set which generates the same ideal. A naive approach would compute the set { NF ( g, G \ {g} ) g ∈ G }. However, this set may not generate the same ideal, and it may not be self-reduced; so we have to be careful. Example 5.7. Let X = {a, b, c} with a ≺ b ≺ c, and let G = {c − a, c − b}. Then G is not self-reduced; computing the normal form of each element with respect to the other gives c − a − (c − b) = b − a and c − b − (c − a) = −b + a (with standard form b − a). Clearly the set {b − a} does not generate the same ideal as G. Example 5.8. Let X = {a, b, c, d} with a ≺ b ≺ c ≺ d, and consider the set G = { d − a, d − b, d − c }, which is not self-reduced. One way to compute the normal form of each element with respect to the others is as follows, replacing each result by its standard form: d − a − (d − b) = b − a, d − b − (d − c) = c − b, d − c − (d − a) = a − c → c − a. Clearly the set { b − a, c − b, c − a } is not self-reduced. Exercise 5.9. Using the algorithm of Figure 1, compose an algorithm whose input is a finite subset G ⊂ F hXi generating an ideal I ⊆ F hXi and whose output is a self-reduced set generating the same ideal. Hint: Avoid the problems illustrated by the last two examples by sorting G using deglex order of leading monomials. Definition 5.10. Consider two nonzero elements g1, g2 ∈ F hXi in standard form; we allow g1 = g2. Set w1 = LM (g1) and w2 = LM (g2). Assume that (1) w1 is not a proper subword of w2, and w2 is not a proper subword of w1 (we say "proper" because we allow g1 = g2). (Condition (1) is satisfied if g1, g2 belong to a self-reduced set.) Assume also that (2) for some words u1, u2, v ∈ X ∗ with v 6= 1 we have w1 = u1v and w2 = vu2 (condition (1) implies that u1 6= 1 and u2 6= 1). 12 MURRAY R. BREMNER In this case, we call v an overlap between w1 and w2, and we have w1u2 = u1w2, where u1 is a proper right subword of w1, and u2 is a proper left subword of w2: w1u2 = u1vu2 = u1w2. The element g1u2 − u1g2 is called a composition of g1 and g2; the common term, a scalar multiple of u1vu2, cancels, since both g1 and g2 are monic. (In the theory of commutative Grobner bases, compositions are often called S-polynomials.) Example 5.11. Consider the following two words in X ∗ where X = {a, b, c}: w1 = a2bcba, w2 = bacba2. These words have the following overlaps: • w1 has a self-overlap: w1 = u1v = vu2 for u1 = a2bcb, v = a, u2 = abcba. • w1 and w2 overlap: w1 = u1v, w2 = vu2 for u1 = a2bc, v = ba, u2 = cba2. • w2 and w1 have overlaps of length 1 and length 2: (cid:5) w2 = u2v, w1 = vu1 for u2 = bacba, v = a, u1 = abcba. (cid:5) w2 = u2v, w1 = vu1 for u2 = bacb, v = a2, u1 = bcba. Example 5.12. Consider the last two generators from Example 3.3: g5 = cb + bc − c, g6 = c2 − b − a. There is a composition of g6 and g5 corresponding to w6 = c2, w5 = cb, u6 = c, u5 = b, v = c. We obtain g6u5 − u6g5 = (c2 − b − a)b − c(cb + bc − c) = c2b − b2 − ab − c2b − cbc + c2 = −b2 − ab − cbc + c2 sf−−→ cbc − c2 + b2 + ab, where the arrow denotes replacing the polynomial by its standard form. Remark 5.13. The motivation for considering compositions is as follows. Suppose that s = g1u2−u1g2 is a composition of g1 and g2, and that the normal form of s with respect to G is nonzero. Then NF (s, G) is an element of the ideal I whose leading monomial is not divisible by any element of G. If we replace G by G ∪ {NF (s, G)}, then we are one step closer to having a Grobner basis for I. 6. The Composition (Diamond) Lemma This lemma is fundamental to the theory of Grobner bases, and leads to an algorithm for constructing a Grobner basis for an ideal from a given set of generators for the ideal; the basic idea underlying this algorithm was given in Remark 5.13. The origin of the name Diamond Lemma is roughly as follows; see also [9, 89]. We have an element f ∈ F hXi, and we want to compute its normal form with respect to a finite subset G ⊂ F hXi. At every step in the computation, there may be many different choices of reduction: many leading monomials of elements of G may occur as subwords of many monomials in f . We want to be sure that whatever sequence of reductions we perform, the final result will be the same. This condition GR OBNER BASES AND UNIVERSAL ENVELOPES 13 is called the "resolution of ambiguities", and is illustrated by this "diamond": gi ւ ց g0 = f = h0 gm = hn ց ւ hj Definition 6.1. Let G = {g1, . . . , gn} be a set of generators for the ideal I ⊆ F hXi. For any word w ∈ X ∗ we define I(G, w) to be the subspace of I spanned by the elements of the form ugv where g ∈ G, u, v ∈ X ∗, and LM (ugv) ≺ w: I(G, w) = n n Xi=1 αiuigivi(cid:12)(cid:12)(cid:12) αi ∈ F ; ui, vi ∈ X ∗; LM (uigivi) ≺ wo. Thus I(G, w) is the subspace of I, relative to the set G of generators, consisting of the elements all of whose monomials precede w in the total order on X ∗. Lemma 6.2. Composition (Diamond) Lemma. Let G be a monic self-reduced set generating the ideal I ⊆ F hXi. Then these conditions are equivalent: (1) G is a Grobner basis for I. (2) For every pair of generators g, h ∈ G, if LM (g)u = vLM (h) for some u, v ∈ X ∗, then gu − vh ∈ I(G, t) where t = LM (g)u = vLM (h). Remark 6.3. Condition (2) implies that every composition gu−vh of the elements of G is a linear combination of elements of the form uigivi where gi ∈ G and ui, vi ∈ X ∗ with uiLM (gi)vi ≺ LM (g)u = vLM (h). The crucial point here is that we are only allowed to use elements of the form uigivi. Proof. (of Lemma 6.2) We follow closely the proof by de Graaf [43, Theorem 6.1.6]. (1) =⇒ (2): Assume that G is a Grobner basis. For g, h ∈ G let f = gu−vh where LM (g)u = vLM (h) for some u, v ∈ X ∗. Clearly f ∈ I and so NFI (f ) = 0. For t = LM (g)u = vLM (h) we have LM (f ) ≺ t since the leading terms of LM (g)u and vLM (h) cancel. When we apply the algorithm of Figure 1 to compute NF (f, G), we repeatedly subtract terms of the form αu1ku2 (α ∈ F ; k ∈ G; u1, u2 ∈ X ∗; LM (u1ku2) ≺ t). Clearly all these terms belong to I and hence to I(G, t). Since G is a Grobner basis, we have NF (f, G) = NFI (f ) = 0. It follows that f is a sum of terms in I(G, t), and hence f ∈ I(G, t). (2) =⇒ (1): We assume condition (2) and prove that G is a Grobner basis for I. Let f ∈ I be arbitrary; we have (1) f = n Xi=1 αiuigivi ( αi ∈ F ; ui, vi ∈ X ∗; gi ∈ G ). We need to show that LM (g) is a subword of LM (f ) for some g ∈ G. We write si = LM (uigivi). Renumbering the generators in G if necessary, we may assume that (2) s1 = · · · = sℓ ≻ sℓ+1 (cid:23) · · · (cid:23) sn. 14 MURRAY R. BREMNER Thus ℓ is the number of equal highest monomials in deglex order; the remaining monomials strictly precede these highest monomials; and we sort the remaining monomials in weak reverse deglex order. If ℓ = 1 then s1 ≻ s2 and so LM (f ) = u1s1v1 = u1LM (g1)v1 as required. We now assume ℓ ≥ 2. In this case we can rewrite equation (1) as follows: (3) f = α1(u1g1v1 − u2g2v2) + (α1 + α2)u2g2v2 + Since ℓ ≥ 2, we have (4) u1LM (g1)v1 = u2LM (g2)v2. n Xi=3 αiuigivi. If u1 = u2 then LM (g1)v1 = LM (g2)v2. Hence either LM (g1) is a left subword of LM (g2), or LM (g2) is a left subword of LM (g1). But this contradicts the assumption that G is self-reduced. Hence u1 6= u2, and so either u1 is a proper left subword of u2, or u2 is a proper left subword of u1. Assume that u1 is a proper left subword of u2; a similar argument applies when u2 is a proper left subword of u1. We have u2 = u1u′ u1LM (g1)v1 = u1u′ 2LM (g2)v2 2 where u′ and so LM (g1)v1 = u′ 2 6= 1. Then 2LM (g2)v2. If v1 is a right subword of v2 then LM (g2) is a subword of LM (g1), again contra- dicting the assumption that G is self-reduced. Hence v2 is a right subword of v1, giving v1 = v′ 1v2 where v′ 1 6= 1. Then LM (g1)v′ 1v2 = u′ 2LM (g2)v2 and so LM (g1)v′ 1 = u′ 2LM (g2). By the assumption that condition (2) holds, it follows that g1v′ 1 − u′ 2g2 ∈ I(G, s) where s = LM (g1)v′ 1 = u′ 2LM (g2). Therefore u1(g1v′ 1 − u′ 2g2)v2 = u1g1v′ 1v2 − u1u′ 2g2v2 = u1g1v1 − u2g2v2. But u1LM (g1)v1 = u2LM (g2)v2 (since ℓ ≥ 2) and so cancellation gives u1g1v1 − u2g2v2 ∈ I(G, t), where t = u1LM (g1)v1. It follows that we can rewrite equation (1) to obtain an expression of the same form where either i) the new value of LM (u1g1v1) is lower in deglex order (this happens when ℓ = 2 and α1 + α2 = 0), or ii) the number ℓ, defined by the order relations (2), has decreased. Since the total order on X ∗ satisfies the descending chain condition, after a finite number of steps we obtain an expression for f of the form (1) where ℓ = 1, and then again LM (f ) = u1s1v1 = u1LM (g1)v1 as required. (cid:3) Lemma 6.4. Consider two elements g, h ∈ G in standard form, and let s ∈ X ∗ be an arbitrary monomial. Set u = sLM (h), v = LM (g)s and t = LM (g)sLM (h), so that LM (g)u = vLM (h) = t. Then we have gu − vh ∈ I(G, t). Proof. Separate the leading monomials of g and h: g = LM (g) + g0, h = LM (h) + h0, GR OBNER BASES AND UNIVERSAL ENVELOPES 15 where either g0 = 0 or LM (g0) ≺ LM (g), and either h0 = 0 or LM (h0) ≺ LM (h). We calculate as follows: gu − vh = (cid:0)LM (g) + g0(cid:1)sLM (h) − LM (g)s(cid:0)LM (h) + h0(cid:1) = g0sLM (h) − LM (g)sh0 = g0s(h − h0) − (g − g0)sh0 = g0sh − gsh0. Then clearly gu − vh = (g0s)h − g(sh0) ∈ I(G, t) where t = LM (g)sLM (h). (cid:3) Theorem 6.5. Main Theorem. Suppose that G is a monic self-reduced set of generators for the ideal I ⊆ F hXi. Then these two conditions are equivalent: (1) G is a Grobner basis for I. (2) For every composition f of the generators in G, the normal form of f with respect to G is zero: NF (f, G) = 0. Proof. (1) =⇒ (2): Let G be a Grobner basis for I, and let f = g1u2 − u1g2 be a composition of g1, g2 ∈ G where u1, u2 ∈ X ∗. Clearly f ∈ I, and hence by the definition of Grobner basis, for some g ∈ G the leading monomial LM (g) is a subword of LM (f ); say LM (f ) = v1LM (g)v2. If we define f1 = f − αv1gv2 where α = lc(f ), where the subtracted element belongs to I, then either f1 = 0 or LM (f1) ≺ LM (f ). Repeating this argument, and using the DCC on X ∗, we obtain NF (f, G) = 0 after a finite number of steps. (2) =⇒ (1): Suppose that f = g1u2 − u1g2 is a composition of g1, g2 ∈ G where u1, u2 ∈ X ∗, and set t = LM (g1)u2 = u1LM (g2). Assume that NF (f, G) = 0. Definition 5.10 implies that u2 6= LM (g2) and u1 6= LM (g1). If u2 is longer than LM (g2) then also u1 is longer than LM (g1), and hence by Lemma 6.4 we have f ∈ I(G, t). If u2 is shorter than LM (g2) then u1 is shorter than LM (g1). Since NF (f, G) = 0 by assumption, the algorithm of Figure 1 outputs zero after a finite number of steps. But during each iteration of the loop in step (2) of that algorithm, we set fn+1 ← fn − αu1gu2, where LM (u1gu2) = LM (fn) (cid:22) LM (f ) ≺ t. Thus f is a linear combination of terms u1gu2 which strictly precede t in deglex order, showing that f ∈ I(G, t). In both cases we have f ∈ I(G, t), and now Lemma 6.2 completes the proof. (cid:3) Remark 6.6. Theorem 6.5 suggests the Grobner basis algorithm in Figure 2 for which the input is a set G generating the ideal I ⊆ F hXi and for which the output (assuming that the algorithm terminates) is a Grobner basis for I. Exercise 6.7. (a) Write a complete formal proof by induction (with basis and inductive hypothesis) of the statement "repeating this argument, and using the DCC on X ∗, we obtain NF (f, G) = 0 after a finite number of steps" from part (1) =⇒ (2) of the proof of Theorem 6.5. (b) Write a complete formal proof by induction (with basis and inductive hy- pothesis) of the statement "f is a linear combination of terms u1gu2 which strictly precede t in deglex order" from part (2) =⇒ (1) of the proof of Theorem 6.5. 16 MURRAY R. BREMNER GrobnerBasis(G) Input: A finite subset G ⊂ F hXi generating an ideal I ⊆ F hXi. Output: If step (2) terminates, the output is a Grobner basis of I. (1) Set newcompositions ← true. (2) While newcompositions do: (a) Convert the elements of G to standard form. (b) Sort G by deglex order of leading monomials: G = {g1, . . . , gn}. (c) Convert G to a self-reduced set: • Set selfreduced ← false. • While not selfreduced do: (i) Set selfreduced ← true. (ii) Set H ← { } (empty set). (iii) For i = 1, . . . , n do: -- Set H ← H ∪ { NF ( gi, {g1, . . . , gi−1} ) }. (iv) Convert the elements of H to standard form. (v) Sort H by deglex order of leading monomials. (vi) If G 6= H then set selfreduced ← false. (vii) Set G ← H. (d) Set compositions ← { } (empty set). (e) Set newcompositions ← false. (f) For g ∈ G do for h ∈ G do: • If LM (g) and LM (h) have an overlap w then: (i) Define u, v by LM (g) = vw and LM (h) = wu. (ii) Set s ← gu − vh (the composition of g and h). (iii) Replace s by its standard form. (iv) Set t ← NF (s, G). (v) Replace t by its standard form. (vi) If t 6= 0 and t /∈ compositions then ∗ Set newcompositions ← true. ∗ Set compositions ← compositions ∪ {t}. (3) Return G. Figure 2. Computing a Grobner basis of the ideal I generated by G Remark 6.8. A different approach to the Composition (Diamond) Lemma, em- phasizing Shirshov's point of view which was developed by the Novosibirsk school of algebra, can be found in the works of Bokut and his co-authors. See in particular, Bokut [11], Bokut and Kukin [21, Chapter 1], Bokut and Shum [22], Bokut and Chen [13]. See also Mikhalev and Zolotykh [85]. Example 6.9. We compute a Grobner basis for the ideal appearing in the construc- tion of the universal associative envelope of the Jordan algebra S2(F ) of symmetric 2 × 2 matrices. Let X = {a, b, c} and let I be the ideal in F hXi generated by the self-reduced set G from Example 3.3: (5) n g1 = a2 − a, g4 = ca + ac − c, g2 = ba + ab, g5 = cb + bc − c, g3 = b2 − b, g6 = c2 − b − a. GR OBNER BASES AND UNIVERSAL ENVELOPES 17 The first iteration of the algorithm produces 10 compositions (including 3 self- compositions); after putting them in standard form, denoted p sf−−→ q, we obtain g1a − ag1 g3a − bg2 g4a − cg1 g5b − cg3 g6b − cg5 sf−−→ 0, sf−−→ s2 = bab + ba, sf−−→ s3 = aca, sf−−→ s5 = bcb, sf−−→ s7 = cbc − c2 + b2 + ab, g2a − bg1 g3b − bg3 g5a − cg2 g6a − cg4 g6c − cg6 sf−−→ s1 = aba + ba, sf−−→ 0, sf−−→ s4 = cab − bca + ca, sf−−→ s6 = cac − c2 + ba + a2, sf−−→ s8 = cb + ca − bc − ac. Computing normal forms of these compositions with respect to the set G using the algorithm of Figure 1, we obtain only two distinct nonzero results: s1 − ag2 + g1b − g2 = −2ab s2 − g2b + ag3 − g2 = −2ab sf−−→ 0, s3 − ag4 + g1c = 0 sf−−→ ab, sf−−→ ab, s4 − g4b + bg4 − g2c + ag5 − g5 − g4 = −2bc − 2ac + 2c sf−−→ bc + ac − c, s5 − bg5 + g3c = 0 sf−−→ 0, s6 − g4c + ag6 − g2 = −2ab sf−−→ ab, s7 − g5c + bg6 + g2 = 2ab sf−−→ ab, s8 − g5 − g4 = −2bc − 2ac + 2c sf−−→ bc + ac − c. So we define (6) t1 = ab, t2 = bc + ac − c. We include the new generators (6) in the original set (5), obtaining a new set H of generators for the ideal I: g1 = a2 − a, g3 = b2 − b, g5 = cb + bc − c, t1 = ab, t2 = bc + ac − c, g6 = c2 − b − a. g2 = ba + ab, g4 = ca + ac − c, For each element h ∈ H, we compute its normal form with respect to the elements which precede it in the total order on H (deglex order of leading monomials). In this simple example, all we need to do is to replace g2 by g2 − t1 = ba:     g1 = a2 − a, g3 = b2 − b, g5 = cb + bc − c, t1 = ab, t2 = bc + ac − c, g6 = c2 − b − a. g′ 2 = ba, g4 = ca + ac − c, We now verify that this set is a Grobner basis: all compositions of these generators have normal form 0 with respect to this set. Remark 6.10. Using the Grobner basis of Example 6.9, it is easy to compute the normal form of any element of F hXi using Theorem 5.3(b). In particular, we can compute the normal form of the element f = c2b from Example 4.9: f1 − g6b − g3 − t1 = c2b − (c2 − b − a)b − (b2 − b) − ab = b. 18 MURRAY R. BREMNER In this way, using Lemma 4.6, we can calculate the structure constants of the universal associative envelope U (S2(F )). Exercise 6.11. Let J = S2(F ) be the Jordan algebra of symmetric 2 × 2 matrices. A basis for U (J) = F ha, b, ci/I consists of the cosets of the monomials which do not any leading monomial from the Grobner basis (Example 6.9). (a) Write down the (finite) set of basis monomials for U (J). (b) Using Lemma 4.6, compute the structure constants for U (J): express prod- ucts of basis monomials as linear combinations of basis monomials. (c) Determine explicitly the structure of the associative algebra U (J). (Use the algorithms in my survey paper [24] on the Wedderburn decomposition.) Example 6.12. Here is an example, from de Graaf [43, page 226], of a generating set G for which the algorithm of Figure 2 never terminates. This also shows why we must consider self-compositions of generators. Let X = {a, b} and define G0 = {g1 = aba − ba}. The first iteration of the algorithm produces one composition of g1 with itself: g1ba − abg1 = (aba − ba)ba − ab(aba − ba) = −baba + ab2a sf−−→ baba − ab2a. Computing the normal form of this composition with respect to G0 gives (baba − ab2a) − b(aba − ba) = −ab2a + b2a sf−−→ ab2a − b2a. Including this with g1 gives a new generating set, which is already self-reduced: G1 = { g1 = aba − ba, g2 = ab2a − b2a }. The second iteration produces three compositions: g1b2a − abg2 = (aba − ba)b2a − ab(ab2a − b2a) = −bab2a + ab3a sf−−→ bab2a − ab3a, g2ba − ab2g1 = (ab2a − b2a)ba − ab2(aba − ba) = −b2aba + ab3a sf−−→ b2aba − ab3a, g2b2a − ab2g2 = (ab2a − b2a)b2a − ab2(ab2a − b2a) = −b2ab2a + ab4a sf−−→ b2ab2a − ab4a. Computing the normal forms of these compositions with respect to G1 gives (bab2a − ab3a) − b(ab2a − b2a) = −ab3a + b3a sf−−→ ab3a − b3a, (b2aba − ab3a) − b2(aba − ba) = −ab3a + b3a sf−−→ ab3a − b3a, (b2ab2a − ab4a) − b2(ab2a − b2a) = −ab4a + b4a sf−−→ ab4a − b4a. Including these with g1, g2 gives a new generating set, which is already self-reduced: G2 = { g1 = aba − ba, g2 = ab2a − b2a g3 = ab3a − b3a, g4 = ab4a − b4a }. It is now easy to verify that the algorithm never terminates; see Exercise 6.13. Exercise 6.13. (a) Work out in detail the next iteration for Example 6.12. (b) State and prove a conjecture for the elements of the set Gn obtained at the end of the n-th iteration of the Grobner basis algorithm in Example 6.12. GR OBNER BASES AND UNIVERSAL ENVELOPES 19 Example 6.14. Here is another (much more complicated) example in which self- compositions play an essential role. We set X = {a, b} and consider the following two elements of F ha, bi which clearly form a self-reduced set: g1 = aba − a2b − a, g2 = bab − ab2 − b. (1) The first iteration of the Grobner basis algorithm produces three compositions: s1 = g1ba − abg1 = (aba − a2b − a)ba − ab(aba − a2b − a) = ababa − a2b2a − aba − ababa + aba2b + aba = aba2b − a2b2a, s2 = g2a − bg1 = (bab − ab2 − b)a − b(aba − a2b − a) = baba − ab2a − ba − baba + ba2b + ba = ba2b − ab2a, s3 = g2ab − bag2 = (bab − ab2 − b)ab − ba(bab − ab2 − b) = babab − ab2ab − bab − babab + ba2b2 + bab = ba2b2 − ab2ab. We compute the normal form of each composition with respect to {g1, g2}: s1 − g1ab − a2g2 = aba2b − a2b2a − (aba − a2b − a)ab − a2(bab − ab2 − b) = aba2b − a2b2a − aba2b + a2bab + a2b − a2bab + a3b2 + a2b = −a2b2a + a3b2 + 2a2b sf−−→ a2b2a − a3b2 − 2a2b = h1, s2 = h2, s3 + abg2 + g1b2 = ba2b2 − ab2ab + ab(bab − ab2 − b) + (aba − a2b − a)b2 = ba2b2 − ab2ab + ab2ab − abab2 − ab2 + abab2 − a2b3 − ab2 = ba2b2 − a2b3 − 2ab2 = h3. We combine these compositions with the original generators and sort them: g1 = aba − a2b − a, h1 = a2b2a − a3b2 − 2a2b, h3 = ba2b2 − a2b3 − 2ab2. g2 = bab − ab2 − b, h2 = ba2b − ab2a, Self-reducing this set eliminates h3 since h3 − h2b − abg2 − g1b2 = 0. (2) The second iteration produces five compositions with these normal forms: h4 = ba3b − ab2a2 + ba2, h7 = ba4b2 − ab2a3b + 2ba3b, h5 = ba3b2 − a2b3a, h6 = a3b3a − a4b3 − 3a3b2, h8 = a4b4a − a5b4 + 2a3b3a − 6a4b3 − 6a3b2. Combining these compositions with g1, g2, h2, h1 and self-reducing the resulting set eliminates h5 and replaces h7 and h8 with these elements: 7 = ba4b2 − a2b3a2 + 2ab2a2 − 2ba2, h′ 8 = a4b4a − a5b4 − 4a4b3. h′ (3) The third iteration of the algorithm produces 18 compositions: ba4b − ab2a3 + 2ba3, ba5b2 − a2b3a3 + 2ba4b + 2ab2a3 − 2ba3, ba5b2 − ab2a4b + 3ba4b, 20 MURRAY R. BREMNER ba6b2 − ab2a5b + 4ba5b, ba5b3 − a3b4a2 + 3a2b3a2 − 6ab2a2 + 6ba2, ba6b3 − a2b3a4b + 4ba5b2 + 2ab2a4b, ba6b3 − ab2a5b2 + 4ba5b2, ba5b4 − a4b5a, a5b5a − a6b5 − 5a5b4, ba7b3 − a2b3a5b + 6ba6b2 + 2ab2a5b + 4ba5b, ba7b3 − ab2a6b2 + 5ba6b2, a5b5a2 − a6b5a − 5a6b4 − 20a5b3, a6b6a − a7b6 + 6a5b5a − 12a6b5 − 30a5b4, a6b6a − a7b6 + 8a5b5a − 14a6b5 − 40a5b4, ba8b4 − a2b3a6b2 + 8ba7b3 + 2ab2a6b2 + 10ba6b2, a6b6a2 − a7b6a + 4a5b5a2 − 10a6b5a − 20a6b4 − 80a5b3, a7b7a − a8b7 + 12a6b6a − 19a7b6 + 36a5b5a − 108a6b5 − 180a5b4, a8b8a − a9b8 + 12a7b7a − 20a8b7 + 36a6b6a − 120a7b6 + 24a5b5a − 240a6b5 − 120a5b4. At this point it seems clear that the algorithm will never terminate! Exercise 6.15. Referring to Example 6.14: (a) Verify the statements about the second and third iterations. (b) Prove that the algorithm does not terminate. (c) Determine a closed form for the generators at the end of the n-th iteration. Remark 6.16. A rich source of examples of the behavior of the Grobner basis algorithm comes from the construction of universal associative envelopes for nonas- sociative triple systems obtained from the trilinear operations classified by the au- thor and Peresi [28]. A detailed study of the simplest non-trivial examples of this construction appears in the Ph.D. thesis of Elgendy [45]; see also her forthcoming paper [46]. Similar examples are discussed in §10 of these lecture notes. 7. Application: The PBW Theorem We now present the beautiful combinatorial proof of the Poincar´e-Birkhoff-Witt (PBW) Theorem discovered by Bokut [11] and independently by Bergman [9]. We follow the exposition given by de Graaf [43, Theorem 6.2.1]. The assumption that the Lie algebra is finite dimensional is not essential. Theorem 7.1. PBW Theorem. If L is a finite dimensional Lie algebra over a field F with ordered basis X = {x1, . . . , xn}, then a basis of its universal associative envelope U (L) consists of the monomials xe1 1 · · · xen n (e1, . . . , en ≥ 0). It follows immediately that: (i) U (L) is infinite dimensional. GR OBNER BASES AND UNIVERSAL ENVELOPES 21 (ii) The natural map L → U (L) is injective. (iii) L is isomorphic to a subalgebra of U (L)−. Proof. The structure constants of L have the form [xi, xj] = n Xk=1 ck ijxk (ck ij ∈ F ), where ck ii = 0. The universal associative envelope U (L) is the quo- tient of the free associative algebra F hXi by the ideal I generated by the elements ij and ck ji = −ck gij = xixj − xjxi − [xi, xj] = xixj − xjxi − n Xk=1 ck ij xk. By anticommutativity of the Lie bracket, we may assume that i > j, and hence xixj is the leading monomial of gij. (If i = j then gii = 0.) So we set G = { gij 1 ≤ j < i ≤ n }. We will show that G is a Grobner basis for the ideal I. Consider the leading monomials of two distinct generators, LM (gij) = xixj (i > j), LM (gℓk) = xℓxk (ℓ > k). The only possible compositions of these generators occur when either j = ℓ or k = i. It suffices to assume j = ℓ, so we consider gij and gjk where i > j > k. We have which produces the composition LM (gij) xk = xixjxk = xi LM (gjk), gijxk − xigjk = (cid:0)xixj − xj xi − [xi, xj ](cid:1)xk − xi(cid:0)xjxk − xkxj − [xj , xk](cid:1) = xixj xk − xjxixk − [xi, xj]xk − xixj xk + xixkxj + xi[xj, xk] = −xjxixk − [xi, xj ]xk + xixkxj + xi[xj, xk] = xixkxj − xjxixk − [xi, xj]xk + xi[xj , xk], which is in standard form. (It is convenient to avoid explicit structure constants in this calculation; recall that [xi, xj ] is a homogeneous polynomial of degree 1.) To compute the normal form with respect to G, we first subtract gikxj and add xjgik: xixkxj − xjxixk − [xi, xj ]xk + xi[xj, xk] −(cid:0)xixk − xkxi − [xi, xk](cid:1)xj + xj(cid:0)xixk − xkxi − [xi, xk](cid:1) = xixkxj − xjxixk − [xi, xj]xk + xi[xj , xk] − xixkxj + xkxixj + [xi, xk]xj + xjxixk − xj xkxi − xj[xi, xk] = −[xi, xj]xk + xi[xj , xk] + xkxixj + [xi, xk]xj − xj xkxi − xj [xi, xk] = −xjxkxi + xkxixj − [xi, xj]xk + xi[xj, xk] + [xi, xk]xj − xj [xi, xk]. We next add gjkxi and subtract xkgij: − xjxkxi + xkxixj − [xi, xj ]xk + xi[xj, xk] + [xi, xk]xj − xj [xi, xk] +(cid:0)xjxk − xkxj − [xj , xk](cid:1)xi − xk(cid:0)xixj − xjxi − [xi, xj](cid:1) = −xjxkxi + xkxixj − [xi, xj]xk + xi[xj , xk] + [xi, xk]xj − xj[xi, xk] + xjxkxi − xkxj xi − [xj , xk]xi − xkxixj + xkxj xi + xk[xi, xj] = −[xi, xj]xk + xi[xj, xk] + [xi, xk]xj − xj [xi, xk] − [xj, xk]xi + xk[xi, xj ] 22 MURRAY R. BREMNER = xi[xj, xk] − [xj , xk]xi + xj [xk, xi] − [xk, xi]xj + xk[xi, xj] − [xi, xj]xk. We now observe that this last expression is equal to [xi, [xj , xk]] + [xj, [xk, xi]] + [xk, [xi, xj ]], which is zero by the Jacobi identity. Thus every composition of the generators has normal form zero, and so G is a Grobner basis. The leading monomials of the elements of this Grobner basis have the form xixj where i > j. A basis for U (L) consists of all monomials w which do not have any of these leading monomials as a subword. That is, if w contains a subword xixj then i ≤ j. It follows that the monomials in the statement of this Theorem form a basis for U (L). In particular, the monomials x1, . . . , xn of degree 1 are linearly independent in U (L), and hence the natural map from L to U (L) is injective. (cid:3) Corollary 7.2. Every polynomial identity satisfied by the Lie bracket in every associative algebra is a consequence of anticommutativity and the Jacobi identity. Proof. Suppose that p(a1, . . . , an) ≡ 0 is a polynomial identity which is not a con- sequence of anticommutativity and the Jacobi identity. Then the Lie polynomial p(a1, . . . , an) is a nonzero element of the free Lie algebra L generated by the vari- ables {a1, . . . , an}. Let A be any associative algebra, and let ǫ : L → A− be any morphism of Lie algebras. By definition of polynomial identity, we have ǫ(p) = 0. Take A = U (L) and let ǫ be the injective map L → U (L)− obtained from the PBW theorem. Since p 6= 0 we have ǫ(p) 6= 0, giving a contradiction. (cid:3) Remark 7.3. Lie algebras arose originally as tangent algebras of Lie groups. Weak- ening the requirement of associativity in the definition of Lie groups gives rise to var- ious classes of nonassociative smooth loops, such as Moufang loops, Bol loops, and monoassociative loops. The corresponding tangent algebras are known respectively as Malcev algebras, Bol algebras, and BTQ algebras. Universal nonassociative en- velopes for Malcev and Bol algebras have been constructed by P´erez-Izquierdo and Shestakov [91, 93]. This problem is still open for BTQ algebras, but see my recent paper with Madariaga [27]. All of these tangent algebras are special cases of Akivis and Sabinin algebras; for the universal nonassociative envelopes of these structures, see Shestakov and Umirbaev [97] and P´erez-Izquierdo [92]. The PBW Theorem shows that for every Lie algebra L, the original set of genera- tors obtained from the structure constants is already a Grobner basis. The original generators in I can be interpreted as rewriting rules in U (L) as follows: xixj − xjxi − n Xk=1 ck ijxk ∈ I ⇐⇒ xixj = xj xi + n Xk=1 ck ijxk ∈ U (L). Repeated application of these rewriting rules allows us to work out explicit multi- plication formulas for monomials in U (L). Exercise 7.4. Let L be the 2-dimensional solvable Lie algebra with basis {a, b} where [a, b] = b; the other structure constants follow from anticommutativity. The basis of U (L) obtained from the PBW theorem consists of the monomials aibj for i, j ≥ 0. The ideal I is generated by ab − ba − b, and so in U (L) we have the relation ba = ab − b. Use this and induction on the exponents to work out a formula for the product (aibj)(akbℓ) as a linear combination of basis monomials. GR OBNER BASES AND UNIVERSAL ENVELOPES 23 Exercise 7.5. Let L be the 3-dimensional nilpotent Lie algebra with basis {a, b, c} where [a, b] = c, [a, c] = [b, c] = 0. The PBW basis of U (L) consists of the monomi- als aibjck for i, j, k ≥ 0. In U (L) we have ba = ab − c, ac = ca, bc = cb. State and prove a formula for (aibjck)(aℓbmcn) as a linear combination of basis monomials. Exercise 7.6. Let L be the 3-dimensional simple Lie algebra sl2(F ) with basis {e, f, h} where [h, e] = 2e, [h, f ] = −2f , [e, f ] = h. The PBW basis of U (L) consists of the monomials f ihjek for i, j, k ≥ 0. In U (L) we have eh = he − 2e, hf = f h − 2f, ef = f e + h. State and prove a formula for (f ihjek)(f ℓhmen) as a linear combination of basis monomials. (This exercise is harder than the previous two. Note that {h, e} and {h, f } span 2-dimensional solvable subalgebras. See also Example 3.2.) 8. Jordan Structures on 2 × 2 Matrices In this section we study some examples of nonassociative structures whose uni- versal associative envelopes are finite dimensional. The underlying vector space in all three examples is M2(F ), the 2 × 2 matrices over a field F of characteristic 6= 2. We will use the following notation for the basis of matrix units: a = E11 = (cid:20)1 0 0 0(cid:21) , b = E12 = (cid:20)0 0 1 0(cid:21) , c = E21 = (cid:20)0 0 1 0(cid:21) , d = E22 = (cid:20)0 0 0 1(cid:21) . 8.1. The Jordan algebra of 2 × 2 matrices. We first make M2(F ) into a Jordan algebra J using the Jordan product x ◦ y = xy + yx. (For convenience we omit the scalar 1 2 .) The universal associative envelope U (J) is isomorphic to F ha, b, c, di/I where the ideal I is generated by the following set of 10 elements, obtained from the structure constants of J; this set is already self-reduced: g1 = a2 − a, g5 = cb + bc − d − a, g9 = dc + cd − c, g2 = ba + ab − b, g6 = c2, g10 = d2 − d. g3 = b2, g7 = da + ad, g4 = ca + ac − c, g8 = db + bd − b, We obtain three distinct nonzero compositions from the pairs (g5, g2), (g5, g3), (g6, g5); computing their normal forms with respect to the set of generators gives: s1 = ad, s2 = bd − ab, s3 = cd − ac. Combining these three compositions with the original ten generators gives a new set of 13 generators; self-reduction makes only minor changes ad, ba + ab − b, b2, bd − ab, cb + bc − d − a, a2 − a, c2, cd − ac, da, db + ab − b, dc + ac − c, ca + ac − c, d2 − d. Every composition of these 13 generators has normal form zero, and so this set is a Grobner basis. There are only 9 monomials in F ha, b, c, di which do not have the leading monomial of one of the Grobner basis elements as a subword: u1 = 1, u2 = a, u3 = b, u4 = c, u5 = d, u6 = ab, u7 = ac, u8 = bc, u9 = abc. The cosets of these monomials modulo I form a basis for U (J). The multiplication table of U (J) is displayed in Table 1, where ui is denoted by i and dot indicates 0. 24 MURRAY R. BREMNER 1 2 3 2 2 1 2 3 3−6 4 4−7 2+5−8 5 6 7 8 9 · · · 9 9 3−6 2−9 3 6 · · 3 6 4 4 7 8 · 4−7 9 · · · 5 5 · 6 7 5 6 7 8−9 · 6 6 6 · 5−8+9 · · · 6 6 7 7 7 8−9 · · · · · · 8 8 9 · 4 8−9 · 7 8 9 9 9 9 · 4−7 · · · 9 9 1 2 3 4 5 6 7 8 9 Table 1. Structure constants for U (J) where J = M2(F )+ Exercise 8.1. (a) Verify the multiplication table for U (J) by computing the normal form of each product of basis elements with respect to the Grobner basis. (b) Use the algorithms in my survey paper [24] to compute the structure of U (J). Prove or disprove that U (J) ≈ F ⊕ M2(F ) ⊕ M2(F ). Compare your results with the known representation theory of Jordan algebras; see Jacobson [69]. 8.2. The 2 × 2 matrices as a Jordan triple system. This subsection and the next introduce the topic of multilinear operations, which will be discussed system- atically in Section 9. We consider the vector space T = M2(F ) with a trilinear operation, the Jordan triple product hx, y, zi = xyz + zyx. Working out the struc- ture constants for this operation, we find that the ideal I appearing in the definition of U (T ) is generated by the following self-reduced set of 40 elements: a3 − a, aba, aca, ada, ba2 + a2b − b, bab, b2a + ab2, b3, bca + acb − a, bcb − b, bda + adb, bdb, ca2 + a2c − c, cab + bac − d, cac, cba + abc − a, cb2 + b2c, cbc − c, c2a + ac2, c2b + bc2, c3, cda + adc, cdb + bdc − a, cdc, da2 + a2d, dab + bad, dac + cad, dad, dba + abd − b, db2 + b2d, dbc + cbd − d, dbd, dca + acd − c, dcb + bcd − d, dc2 + c2d, dcd, d2a + ad2, d2b + bd2 − b, d2c + cd2 − c, d3 − d. These elements produce 36 distinct nonzero compositions: ad, b2, bd − ab, c2, cd − ac, da, db − ba, dc − ca, d2 − cb − bc + a2, a2d, ab2, abd − a2b, acb + abc − a, ac2, acd − a2c, adb, adc, ad2, bad, b2c, b2d, bc2, bcd − bac, bdc + acb − a, bdc − abc, bd2 − b2c − a2b, bd2 − abd, bd2 − a2b, cad, cbd + bcd − a2d − d, cbd + bcd − d, cbd + bac − d, c2d, cd2 − acd, cd2 + bc2 − a2c, cd2 − a2c. Taking the union of these two sets gives 76 generators, and self-reducing this set produces a set with only 22 elements: ad, b2, bd − ab, c2, cd − ac, da, db − ba, dc − ca, d2 − cb − bc + a2, a3 − a, aba, aca, acb + abc − a, ba2 + a2b − b, bab, bca − abc, bcb − b, GR OBNER BASES AND UNIVERSAL ENVELOPES 25 ca2 + a2c − c, cab + bac − d, cac, cba + abc − a, cbc − c. All compositions of this new set have normal form zero, so we have a Grobner basis. There are only 17 monomials in F ha, b, c, di which do not have the leading monomial of one of these 22 generators as a subword, and the cosets of these monomials modulo I form a basis for the universal associative envelope U (T ): 1, a, b, c, d, a2, ab, ac, ba, bc, ca, cb, a2b, a2c, abc, bac, a2bc. The multiplication table for U (T ) is an array of size 17 × 17. Exercise 8.2. Use a computer algebra system to calculate the multiplication table for U (T ). Compute the Wedderburn decomposition of U (T ). Prove or disprove that U (T ) ≈ F ⊕ M2(F ) ⊕ M2(F ) ⊕ M2(F ) ⊕ M2(F ). For the structure theory of Jordan triple systems, see Loos [80] and Meyberg [84]. 8.3. The 2 × 2 matrices with the Jordan tetrad. We consider the vector space Q = M2(F ) with a quadrilinear operation, the Jordan tetrad {w, x, y, z} = wxyz + zyxw. Working out the structure constants for this operation, we find that the ideal I is generated by the self-reduced set of 136 elements displayed in Table 2. Remarkably, there are 2769 distinct nontrivial compositions of these 136 generators. The most complicated normal form of these compositions is bcbcdcd + bcbc2d2 + dcd2 + c2bd + cbdc + cbcd + bdc2 − adcd − c. Combining the original 136 generators with the 2769 compositions produces a new generating set of 2905 elements. After two iterations of self-reduction, this large set of generators collapses to the set 25 elements in Table 3 which form a Grobner basis. There are only 25 monomials in F ha, b, c, di which do not have the leading monomial of one of these 25 generators as a subword; the cosets of these monomials form a basis of the universal associative envelope U (Q): 1, a, abc, b, ba2, c, d, a2, ab, bac, ca2, cab, ac, a3b, ba, a3c, bc, ca, a2bc, cb, ba2c, a3, a3bc. a2b, a2c, The multiplication table of U (Q) is an array of size 25 × 25. Exercise 8.3. Use a computer algebra system to calculate the multiplication table of U (Q). Compute the Wedderburn decomposition of U (Q). Prove or disprove that U (T ) ≈ F ⊕ M2(F ) ⊕ M2(F ) ⊕ M2(F ) ⊕ M2(F ) ⊕ M2(F ) ⊕ M2(F ). Remark 8.4. At present there is no general theory of the structures obtained from regarding the Jordan tetrad as a quadrilinear operation on an associative algebra. For the role played by tetrads in Jordan theory, see McCrimmon [83]. Exercise 8.5. Prove that if J is a finite dimensional Jordan algebra then its uni- versal associative envelope is also finite dimensional. Exercise 8.6. Prove that if J is an n-dimensional Jordan algebra with zero prod- uct, then U (J) is the exterior algebra of an n-dimensional vector space. Exercise 8.7. Prove that if J is the Jordan algebra of a symmetric bilinear form, then U (J) is the corresponding Clifford algebra. 26 MURRAY R. BREMNER aba2 + a2ba, ab2a, adba + abda, baca + acab, aca2 + a2ca, adca + acda, bada + adab, acba + abca − a, ad2a, b2a2 + a2b2, ba3 + a3b − b, b2ab + bab2, ac2a, ba2b, b2ca + acb2, bcba + abcb − b, b2da + adb2, bcb2 + b2cb, bca2 + a2cb − a, bc2a + ac2b, bc2b, bda2 + a2db, bdab + badb, bd2a + ad2b, bdba + abdb, bd2b, bdb2 + b2db, ca3 + a3c − c, bdcb + bcdb − b, ca2c, caba + abac, cab2 + b2ac, caca + acac, cada + adac, cb2a + ab2c, cadb + bdac, cb3 + b3c, cba2 + a2bc − a, cb2c, cbdb + bdbc, cbca + acbc − c, c2a2 + a2c2, cbab + babc, c2ab + bac2, cbda + adbc, c2ba + abc2, c2b2 + b2c2, c2bc + cbc2, c3a + ac3, c3b + bc3, c2db + bdc2, cda2 + a2dc, cdab + badc, cdac + cadc, cdb2 + b2dc, cdbc + cbdc − c, cdca + acdc, cdcb + bcdc, cd2a + ad2c, da2d, dac2 + c2ad, cd2b + bd2c − a, cd2c, daba + abad, dab2 + b2ad, da3 + a3d, dabc + cbad, da2b + ba2d, daca + acad, dada + adad, dadb + bdad, dadc + cdad, c2da + adc2, a4 − a, ada2 + a2da, baba + abab, b3a + ab3, b4, bcab + bacb − b, bcda + adcb, bdca + acdb − a, ca2b + ba2c − d, cacb + bcac, cbac + cabc − c, cbcb + bcbc − d − a, c2ac + cac2, c4, cdba + abdc − a, cdc2 + c2dc, da2c + ca2d, dacb + bcad, dba2 + a2bd − b, db3 + b3d, dbda + adbd, dcac + cacd, dcbd + dbcd − d, dcdb + bdcd, d2ad + dad2, d2ca + acd2 − c, d3b + bd3 − b, dbab + babd, db2d, db2c + cb2d, dbac + cabd − d, dbca + acbd, dbad + dabd, dbcb + bcbd − b, dbdb + bdbd, dcad + dacd, dbdc + cdbd, dcba + abcd, dc2a + ac2d, dc2b + bc2d, dcdc + cdcd, d2ba + abd2 − b, d2a2 + a2d2, d2b2 + b2d2, dca2 + a2cd − c, dcb2 + b2cd, dc3 + c3d, d2ab + bad2, d2cb + bcd2 − d, d2c2 + c2d2, d3c + cd3 − c, d4 − d. db2a + ab2d, dbc2 + c2bd, dcab + bacd − d, dcbc + cbcd − c, dc2d, d2ac + cad2, dcda + adcd, d2bc + cbd2 − d, d2cd + dcd2, d2bd + dbd2, d3a + ad3, Table 2. The 136 generators of the ideal I for the Jordan tetrad b2, ad, d2 − cb − bc + a2, bcb − ba2 − a2b, ba3 + a3b − b, cabc + a3c − c. bd − ab, c2, cd − ac, aba, aca, acb + abc − a3, cac, ba2b, cba + abc − a3, ca3 + a3c − c, da, db − ba, bab, cbc − ca2 − a2c, dc − ca, bca − abc, a4 − a, ca2b + ba2c − d, ca2c, Table 3. The Grobner basis of the ideal I for the Jordan tetrad 9. Multilinear Operations We now consider generalizations of the two basic nonassociative bilinear opera- tions, the Lie bracket and the Jordan product, to n-linear operations for any integer n ≥ 2. This discussion is based on my papers with Peresi [28, 29]. 9.1. Multilinear operations. An n-linear operation ω(a1, . . . , an) over a field F is a linear combination of permutations of the monomial a1 · · · an. We regard ω as a multilinear element of degree n in the free associative algebra on n generators: ω(a1, . . . , an) = Xσ∈Sn xσaσ(1) · · · aσ(n) (xσ ∈ F ), GR OBNER BASES AND UNIVERSAL ENVELOPES 27 where the sum is over all permutations in the symmetric group Sn acting on {1, . . . , n}. We may also identify ω(a1, . . . , an) with an element of F Sn, the group algebra of the symmetric group Sn: ω(a1, . . . , an) = Xσ∈Sn xσσ (xσ ∈ F ). The group Sn acts on F Sn by permuting the subscripts of the generators: σ · aτ (1) · · · aτ (n) = aστ (1) · · · aστ (n). Two n-linear operations are said to be equivalent if each is a linear combination of permutations of the other; that is, they generate the same left ideal in F Sn. When discussing n-linear operations, we assume that the characteristic of F is either 0 or a prime p > n; this is a necessary and sufficient condition for the group algebra F Sn to be semisimple. In this case, F Sn is the direct sum of simple two- sided ideals, each isomorphic to a matrix algebra Md(F ), and the projections of Sn to these matrix algebras define the irreducible representations of Sn. 9.2. The case n = 2. Every bilinear operation is equivalent to one of the following: the Lie bracket [x, y] = xy − yx, the Jordan product x ◦ y = 1 2 (xy + yx), the original associative operation xy, and the zero operation. In other words, the only left ideals in the group algebra F S2 ≈ F ⊕ F are {0} ⊕ F , F ⊕ {0}, F ⊕ F , and {0} ⊕ {0}. The first summand F corresponds to the unit representation of S2, and a basis for this summand is the idempotent 1 2 (xy + yx). The second summand corresponds to the sign representation, and a basis for this summand is the idempotent 1 2 (xy − yx). These two idempotents are orthogonal in the sense that their product is zero. 9.3. The case n = 3. Faulkner [49] classified the trilinear polynomial identities satisfied by a large class of nearly simple triple systems. Twenty years later, trilinear operations were classified up to equivalence in my work with Peresi [28]; we also determined the polynomial identities of degree 5 satisfied by these operations. The structure of the group algebra in this case is F S3 ≈ F ⊕ M2(F ) ⊕ F. The first and last summands correspond to the unit and sign representations re- spectively; bases for these summands are the following idempotents: S = 1 A = 1 6 (abc + acb + bac + bca + cab + cba), 6 (abc − acb − bac + bca + cab − cba). The middle summand M2(F ) corresponds to the irreducible 2-dimensional repre- sentation of S3. To find a basis for M2(F ) corresponding to the matrix units Eij (i, j = 1, 2) we use the representation theory of the symmetric group developed by Young [103] and simplified by Rutherford [96] and Clifton [40]. It follows that any trilinear operation can be represented as a triple of matrices: (cid:20) a, (cid:20)b11 b21 b12 b22(cid:21), c(cid:21). As representatives of the equivalence classes we may take the triples in which each matrix is in row canonical form. Using computer algebra [28], it can be shown that there are exactly 19 trilinear operations satisfying polynomial identities in degree 5 which do not follow from their 28 MURRAY R. BREMNER identities in degree 3. Simplified forms of these operations were later discovered by the author [25] and Elgendy [46]. Together with these 19 operations, it is conventional to include the symmetric, alternating and cyclic sums, even though for these operations, every identity in degree 5 follows from those of degree 3; see my paper with Hentzel [26]. These 22 trilinear operations are given in Table 4. The first column gives the name of the operation; the second column gives the row canonical forms of the representation matrices of the corresponding element of the group algebra; the third column gives the the simplest representative of the equivalence class as a linear combination of permutations. (The parameter q represents the (1, 2) entry of the 2 × 2 matrix.) 9.4. Associative n-ary algebras. Simple associative triple systems were classi- fied by Hestenes [64], Lister [75] and Loos [81]; their work was extended to simple associative n-ary systems by Carlsson [35]. The classification by Carlsson can be reformulated as follows. Let (d1, . . . , dn−1) be a sequence of n−1 positive integers; two such sequences are regarded as equivalent if they differ only by a cyclic per- mutation. For each i = 1, . . . , n−1, let Vi be a vector space of dimension di over F , and consider the direct sum V = V1 ⊕ · · · ⊕ Vn−1. Let A be the subspace of EndF (V ) consisting of the linear operators T : V → V which satisfy the conditions T (V1) ⊆ V2, T (V2) ⊆ V3, . . . , T (Vn−2) ⊆ Vn−1, T (Vn−1) ⊆ V1. Then A is a simple associative n-ary system, and every such system has this form. If we choose bases of the subspaces V1, . . . , Vn−1 then we can represent the elements of A as D ×D block matrices where D = d1 +· · ·+dn−1. The block in position (i, j) where 1 ≤ i, j ≤ n−1 has size di × dj; nonzero entries may appear only in blocks (2, 1), . . . , (n−1, n), (n, 1). To illustrate, for n = 3, 4, 5 we obtain the matrices of the following forms, where Tij is an arbitrary block of size di × dj : (cid:20) 0 T21 T12 0 (cid:21) , 0 T21 0   0 0 T32 T13 0 0 ,   0 T21 0 0   0 0 T32 0 0 0 0 T43 T14 0 0 0 .   9.5. Special nonassociative n-ary systems. If A is an associative n-ary system and ω(a1, . . . , an) is an n-linear operation, then we obtain a nonassociative n-ary system Aω by interpreting each monomial in ω as the corresponding product in A. Such a nonassociative n-ary system is called special (by analogy with special Jordan algebras) since it comes from a multilinear operation on an associative system. In order to understand these nonassociative n-ary systems, we construct their universal associative envelopes using the theory of noncommutative Grobner bases. The ultimate goal is to classify all the irreducible finite dimensional representa- tions of these systems. This generalizes the familiar construction of the universal enveloping algebras of Lie and Jordan algebras, where a dichotomy arises: a fi- nite dimensional simple Lie algebra has an infinite dimensional universal envelope and infinitely many isomorphism classes of irreducible finite dimensional represen- tations, but a finite dimensional simple Jordan algebra has a finite dimensional universal envelope and only finitely many irreducible representations. 9.6. Universal associative envelopes. This subsection gives the precise defini- tion of the universal associative envelope of a nonassociative n-ary system relative to an n-linear operation; we consider only the case of a special nonassociative n-ary GR OBNER BASES AND UNIVERSAL ENVELOPES 29 operation F ⊕ M2(F ) ⊕ F F S3 symmetric sum alternating sum cyclic sum Lie q = ∞ Lie q = 1 2 Jordan q = ∞ Jordan q = 0 Jordan q = 1 Jordan q = 1 2 anti-Jordan q = ∞ anti-Jordan q = −1 anti-Jordan q = 1 2 anti-Jordan q = 2 fourth family q = ∞ fourth family q = 0 fourth family q = 1 fourth family q = −1 fourth family q = 2 fourth family q = 1 2 cyclic commutator weakly commutative weakly anticommutative 0 1 1 2 1 1 2 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 (cid:20) 1, (cid:20)0 (cid:20) 0, (cid:20)0 (cid:20) 1, (cid:20)0 (cid:20) 0, (cid:20)0 (cid:20) 0, (cid:20)1 (cid:20) 1, (cid:20)0 (cid:20) 1, (cid:20)1 (cid:20) 1, (cid:20)1 (cid:20) 1, (cid:20)1 (cid:20) 0, (cid:20)0 (cid:20) 0, (cid:20)1 −1 (cid:20) 0, (cid:20)1 (cid:20) 0, (cid:20)1 (cid:20) 1, (cid:20)0 (cid:20) 1, (cid:20)1 (cid:20) 1, (cid:20)1 (cid:20) 1, (cid:20)1 −1 (cid:20) 1, (cid:20)1 (cid:20) 1, (cid:20)1 (cid:20) 0, (cid:20)1 (cid:20) 1, (cid:20)1 (cid:20) 0, (cid:20)1 0(cid:21), 0(cid:21) 0(cid:21), 1(cid:21) 0(cid:21), 1(cid:21) 0(cid:21), 0(cid:21) 0(cid:21), 0(cid:21) 0(cid:21), 0(cid:21) 0(cid:21), 0(cid:21) 0(cid:21), 0(cid:21) 0(cid:21), 0(cid:21) 0(cid:21), 1(cid:21) 0 (cid:21), 1(cid:21) 0(cid:21), 1(cid:21) 0(cid:21), 1(cid:21) 0(cid:21), 1(cid:21) 0(cid:21), 1(cid:21) 0(cid:21), 1(cid:21) 0 (cid:21), 1(cid:21) 0(cid:21), 1(cid:21) 0(cid:21), 1(cid:21) 1(cid:21), 0(cid:21) 1(cid:21), 0(cid:21) 1(cid:21), 1(cid:21) 0 0 1 0 0 0 0 0 0 0 0 0 0 2 1 2 1 2 0 0 1 2 0 0 abc + acb + bac + bca + cab + cba abc − acb − bac + bca + cab − cba abc + bca + cab abc − acb − bca + cba abc + acb − bca − cba abc + cba abc + bac abc + acb abc + 2acb + 2cab + cba abc − 2acb + 2cab − cba abc − acb abc − cba abc − bac abc − acb − bac abc − acb + bca abc − bac + cab abc + bac + cab abc + acb + bca abc + acb + bac abc − bca abc + acb + bac − cba abc + acb − bca − cab Table 4. The twenty-two trilinear operations 30 MURRAY R. BREMNER system. The earliest discussion of this construction appears to be that of Birkhoff and Whitman [10, §2]; the presentation here follows my survey paper [25, §7.2]. Suppose that B is a subspace, of an associative n-ary system A over the field F , which is closed under the n-linear operation ω(a1, . . . , an) = Xσ∈Sn xσaσ(1) · · · aσ(n) (xσ ∈ F ). Set d = dim B and let {b1, . . . , bd} be a basis of B over F ; then we have the structure constants for the resulting nonassociative n-ary system Bω: ω(bi1, . . . , bin ) = d Xj=1 cj i1 ···in bj (1 ≤ i1, . . . , in ≤ d). Let F hXi be the free associative algebra generated by the symbols X = {b1, . . . , bd} and consider the ideal I ⊆ F hXi generated by the following dn elements: xσbiσ(1) · · · biσ(n) − Xσ∈Sn d Xj=1 cj i1 ···in bj (1 ≤ i1, . . . , in ≤ d). The quotient algebra U (Bω) = F hXi/I is the universal associative enveloping algebra of the nonassociative n-ary system Bω. Since the n-ary structure on Bω is special (that is, defined in terms of the associative structure on A), the natural map Bω → U (Bω) will necessarily be injective. From this set of generators for the ideal I, we use the algorithm of Figure 2 to compute a Grobner basis for I. We then use this Grobner basis to obtain a monomial basis for the universal associative envelope U (Bω). The multiplication table for U (Bω) is then obtained by computing normal forms of products of basis monomials. The next section is devoted to examples of this procedure. 10. Special Nonassociative Triple Systems In her Ph.D. thesis [45] and her forthcoming paper [46], Elgendy undertook a detailed study using noncommutative Grobner bases of the universal associative envelopes of the nonassociative triple systems obtained by applying the trilinear operations of Table 4 to the 2-dimensional associative triple system A1 of the form (cid:20)0 ∗ ∗ 0(cid:21) , where ∗ represents an arbitrary scalar. She distinguished two classes of operations: those of Lie type, for which the universal envelopes are infinite dimensional; and those of Jordan type, for which the universal envelopes are finite dimensional. For the operations of Lie type, she discovered that the universal envelopes are closely related to the down-up algebras introduced by Benkart and Roby [8]. For the operations of Jordan type, she determined explicit Wedderburn decompositions of the universal envelopes and classified the irreducible representations; for these cases, she used the algorithms described in my survey paper [24]. GR OBNER BASES AND UNIVERSAL ENVELOPES 31 In this section, I consider the same problem for the 4- and 6-dimensional asso- ciative triple systems A2 and a3 consisting of all matrices of the forms 0 ∗ ∗   ∗ ∗ 0 0 0 0   , 0 ∗ ∗ ∗   ∗ ∗ 0 0 0 0 0 0 . ∗ 0 0 0   The resulting universal envelopes provide many examples of associative algebras, both finite dimensional and infinite dimensional, that deserve further study. It seems reasonable to expect that this will lead to generalizations of down-up algebras, and to nonassociative triple systems with many finite dimensional representations. The computations are described in detail for A2 and the results for A1, A2 and A3 are summarized in Table 5. All calculations were done using Maple worksheets written by the author. 10.1. Symmetric sum. The original set of generators obtained from the structure constants consists of these 20 elements which form a Grobner basis for the ideal: a3, ba2 + aba + a2b, b2a + bab + ab2, b3, ca2 + aca + a2c − a, cba + cab + bca + bac + acb + abc − b, c2b + cbc + bc2, c3, db2 + bdb + b2d − b, da2 + ada + a2d, cb2 + bcb + b2c, c2a + cac + ac2 − c, dba + dab + bda + bad + adb + abd − a, dca + dac + cda + cad + adc + acd − d, dcb + dbc + cdb + cbd + bdc + bcd − c, d2a + dad + ad2, d2b + dbd + bd2 − d, dc2 + cdc + c2d, d2c + dcd + cd2, d3. There are infinitely many monomials in F ha, b, c, di which do not contain the leading monomial of one of these generators as a subword, and so the universal envelope is infinite dimensional. The first few dimensions of the homogeneous components of the associated graded algebra are as follows: 1, 4, 16, 44, 131, 344, 972, 2592, . . . . 10.2. Alternating sum. The original set of generators obtained from the structure constants consists of these 4 elements which form a Grobner basis for the ideal: cba − cab − bca + bac + acb − abc − b, dca − dac − cda + cad + adc − acd + d, dba − dab − bda + bad + adb − abd + a, dcb − dbc − cdb + cbd + bdc − bcd − c. There are infinitely many monomials in F ha, b, c, di which do not contain the leading monomial of one of these generators as a subword, and so in this case again, the universal associative envelope is infinite dimensional. The first few dimensions of the homogeneous components of the associated graded algebra are 1, 4, 16, 60, 225, 840, 3136, 11704, . . . . The On-line Encyclopedia of Integer Sequences [90], sequence A072335, suggests that the generating function for these dimensions is 1 (1 − x2)(1 − 4x + x2) . Since the generating function has such a simple form, it seems reasonable to ex- pect that the universal envelope has an interesting structure, and that the original 4-dimensional alternating triple system has a large class of finite dimensional irre- ducible representations. 32 MURRAY R. BREMNER Open Problem 10.1. Prove the last claim about the generating function for the dimensions of the homogeneous components of the associated graded algebra. Open Problem 10.2. Investigate the relationship between the universal envelopes for the symmetric and alternating sums and down-up algebras; see [8] and [46]. 10.3. Cyclic sum. The original set of generators obtained from the structure con- stants consists of these 24 elements, forming a self-reduced set: a3, ca2 + aca + a2c − a, ba2 + aba + a2b, b2a + bab + ab2, b3, cab + bca + abc, c2b + cbc + bc2, cba + bac + acb − b, c3, da2 + ada + a2d, dac + cda + acd − d, dba + bad + adb, dca + cad + adc, d2b + dbd + bd2 − d, dcb + cbd + bdc − c, d2c + dcd + cd2, cb2 + bcb + b2c, c2a + cac + ac2 − c, dab + bda + abd − a, db2 + bdb + b2d − b, dc2 + cdc + c2d, d3. dbc + cdb + bcd, d2a + dad + ad2, This is not a Grobner basis; there are 40 distinct nontrivial compositions of these generators, the most complicated of which has this normal form: bacda + bacad + acbda − acadb − abcad + abadc − 2abacd + a2cbd + a2bdc − 2a2bcd − 2bda − bad + adb − aca − abd − 2a2c + 2a. Combining the original 24 generators with the 40 compositions gives a set of 64 elements; applying self-reduction to this set produces a new generating set of 59 el- ements. This new generating set produces 724 distinct nontrivial compositions. The combined set of 783 generators self-reduces to 62 elements, which form a Grobner basis for the ideal: a3, abd − a2c, ab2, abc, bab, aba, a2d, ba2, bc2, cac + ac2 − c, bcd, bac + acb − b, bda + a2c − a, cad, cba, aca + a2c − a, b3, b2c, bdc − ac2, cbd + ac2 − c, bad, b2a, bdb − acb, cb2, cbc, adb, ada, b2d + acb − b, bd2 − acd, c2a, c2b, a2b, ad2, adc, bcb, cab, c2d, bca, ca2, c3, db2, dbc, a2cb − ab, cda, cdb, cdc, cd2, da2, dab, dbd + acd − d, a2cd − ad. dca, dcb, dc2, dac + acd − d, d2b, d2a, dcd, dad, d2c, dba, d3, Only finitely many monomials in F ha, b, c, di do not have a subword equal to the leading monomial of an element of this Grobner basis. The universal associative envelope has a basis consisting of the cosets of these 26 monomials: 1, a, b, c, d, cd, da, db, dc, a2, d2, ab, a2c, ac, ad, ba, b2, acb, ac2, acd, bc, a2c2. bd, ca, cb, c2, Exercise 10.3. Determine the radical of the universal envelope in this case, and the decomposition of the semisimple quotient into a direct sum of simple ideals. 10.4. Lie q = ∞. In this case we are studying a simple Lie triple system; see Lister [74]. The original set of generators obtained from the structure constants consists of 24 elements; after self-reduction, we are left with 20 elements: ca2 − 2aca + a2c + 2a, ba2 − 2aba + a2b, b2a − 2bab + ab2, GR OBNER BASES AND UNIVERSAL ENVELOPES 33 cab − bca + bac − acb + b, c2a − 2cac + ac2 + 2c, dab − bda + bad − adb + a, dba − bda − adb + abd + a, dca − cda − adc + acd, d2a − 2dad + ad2, cba − bca − acb + abc + b, cb2 − 2bcb + b2c, c2b − 2cbc + bc2, da2 − 2ada + a2d, dac − cda + cad − adc − d, db2 − 2bdb + b2d + 2b, dbc − cdb + cbd − bdc − c, dcb − cdb − bdc + bcd, dc2 − 2cdc + c2d, d2b − 2dbd + bd2 + 2d, d2c − 2dcd + cd2. There are 24 distinct compositions; the most complicated normal form is bdcda − bdadc − bcdad + badcd + adcdb − adbdc − acdbd + abdcd + bd2 + 3acd. Combining the original 20 generators with the 24 compositions and applying self- reduction gives a new generating set of 16 elements, which is a Grobner basis: dc − cd, ba − ab, cb2 − 2bcb + b2c, ca2 − 2aca + a2c + 2a, cab − bca − acb + abc + b, c2a − 2cac + ac2 + 2c, c2b − 2cbc + bc2, dab − bda − adb + abd + a, dbc − cdb + cbd − bcd − c, cbca − cacb − bcac + acbc + cb + bc, dac − cda + cad − acd − d, d2a − 2dad + ad2, d2b − 2dbd + bd2 + 2d, dbda − dadb − bdad + adbd − da − ad. da2 − 2ada + a2d, db2 − 2bdb + b2d + 2b, There are infinitely many monomials in F ha, b, c, di which do not contain the lead- ing monomial of one of these generators as a subword, and so in this case, the universal associative envelope is infinite dimensional. The generating function for the dimensions seems to be as follows; see [90], sequence A038164: 1 (1 − x)4(1 − x2)4 . The first few terms are 1, 4, 14, 36, 85, 176, 344, 624, 1086, 1800, 2892, 4488, . . . . Open Problem 10.4. Investigate the universal enveloping algebras of Lie triple systems and their representation theory. Every finite dimensional Lie triple system can be embedded into a finite dimensional Lie algebra as the odd subspace of a 2-grading on the Lie algebra. For recent work, see Hodge and Parshall [65]. 10.5. Lie q = 1 2 . In this case we are studying a simple anti-Lie triple system. The original set of generators obtained from the structure constants consists of 40 elements; after self-reduction, we are left with 20 elements: ba2 − a2b, b2a − ab2, ca2 − a2c, cba + bca − acb − abc + b, dab − bda − bad + adb + a, dba + bda − adb − abd − a, cb2 − b2c, dac − cda − cad + adc − d, db2 − b2d, dbc − cdb − cbd + bdc + c, cab − bca − bac + acb − b, c2b − bc2, c2a − ac2, da2 − a2d, dca + cda − adc − acd, d2c − cd2. d2b − bd2, dcb + cdb − bdc − bcd, dc2 − c2d, d2a − ad2, There are 26 distinct nontrivial compositions of these generators, the most compli- cated of which has normal form bdcda − bdadc − bcdad + badcd − adcdb + adbdc + acdbd − abdcd − bd2 − acd. 34 MURRAY R. BREMNER Combining the original 20 generators with the 26 compositions and applying self- reduction gives a new generating set of 12 elements, which is a Grobner basis: b2, c2, dc + cd, ba + ab, a2, dab − bda + adb + abd + a, dbc − cdb − cbd − bcd + c, dbda − dadb − bdad + adbd − da + ad. d2, cab − bca + acb + abc − b, dac − cda − cad − acd − d, cbca − cacb − bcac + acbc + cb − bc, There are infinitely many monomials in F ha, b, c, di which do not contain the leading monomial of one of these generators as a subword, and so again in this case, the universal associative envelope is infinite dimensional. The first few dimensions of the homogeneous components in the associated graded algebra are 1, 4, 10, 20, 35, 56, 84, 120, 165, 220, 286, 364, 455, 560, 680, 816, 969, . . . . According to [90], sequence A000292, these are the tetrahedral numbers; the generating function is ∞ Xn=1 (cid:18)n + 2 3 (cid:19)xn. Open Problem 10.5. Investigate the universal enveloping algebras of anti-Lie triple systems. Every finite dimensional anti-Lie triple system can be embedded into a finite dimensional Lie superalgebra as the odd subspace. See the recent monograph by Musson [88] on Lie superalgebras and their enveloping algebras. 10.6. Jordan q = ∞. In this case we are studying a simple Jordan triple system. The original set of 40 generators is already self-reduced: a3, bca + acb − b, b2a + ab2, ba2 + a2b, aca − a, ada, aba, bab, b3, bda + adb − a, bdb − b, ca2 + a2c, cab + bac, cac − c, cba + abc, bcb, cb2 + b2c, da2 + a2d, dbc + cbd − c, d2a + ad2, cbc, c2a + ac2, c2b + bc2, c3, cda + adc, dab + bad, dac + cad − d, dad, dba + abd, cdb + bdc, db2 + b2d, cdc, dbd − d, dca + acd, dcb + bcd, dc2 + c2d, dcd, d2b + bd2, d2c + cd2, d3. There are 32 distinct nontrivial compositions; their normal forms are a2, ba, ab, abd + a2c, bc2, abd, b2d + bac, b2, c2, ac2, bcd, cd, dc, adc, acd, bdc + ac2, d2, ad2, a2b, a2c, a2d, ab2, abc, bac, bad, b2c, b2d, bdc, bd2 + acd, bd2, c2d, cd2. The combined set of 72 elements self-reduces to 20, forming a Grobner basis: ba, b2, ab, bda + adb − a, a2, bcb, dbc + cbd − c, dbd − d. c2, cd, bdb − b, d2, dc, cac − c, aca − a, cbc, ada, bca + acb − b, dac + cad − d, dad, There are only 19 monomials which do not contain the leading monomial of an ele- ment of the Grobner basis, so the universal associative envelope is finite dimensional and has the cosets of the following monomials as a basis: 1, a, b, c, d, ac, ad, bc, bd, ca, cb, da, db, acb, adb, cad, cbd, acbd, cadb. GR OBNER BASES AND UNIVERSAL ENVELOPES 35 Exercise 10.6. Compute the Wedderburn decomposition of the universal associa- tive envelope of this Jordan triple system. In particular, prove or disprove that the envelope is isomorphic to F ⊕ M3(F ) ⊕ M3(F ). Exercise 10.7. Let T be a finite dimensional Jordan triple system. Prove that U (T ) is also finite dimensional. 10.7. Jordan q = 0. The original self-reduced set of generators has 40 elements. There are 20 distinct nontrivial compositions, and the combined set of 60 elements self-reduces to 27 elements. These 27 generators have 4 distinct nontrivial composi- tions, and the combined set of 31 elements self-reduces to 15 elements, which form a Grobner basis for the ideal: (7) (cid:26) a2, dc, ab, d2, ba, ad, aca − a, b2, bc, bd − ac, acb − b, cac − c, c2, dac − d. cd, The universal associative envelope has dimension 10 with basis consisting of the cosets of the elements 1, a, b, c, d, ac, ca, cb, da, db. See Exercise 10.8 below. 10.8. Jordan q = 1. The original self-reduced set of generators has 40 elements. There are 19 distinct nontrivial compositions, and the combined set of 59 elements self-reduces to 27 elements. These 27 generators have 6 distinct nontrivial composi- tions, and the combined set of 33 elements self-reduces to 15 elements, which form a Grobner basis for the ideal. This Grobner basis is the same as in the previous case, and so the universal envelopes are isomorphic. 10.9. Jordan q = 1 2 . The original self-reduced set of 40 generators has 94 distinct nontrivial compositions, and the combined set of 134 elements self-reduces to 15 elements, which form the Grobner basis (7). Exercise 10.8. Compute the Wedderburn decomposition of the universal envelope for the Grobner basis (7). Prove or disprove that it is isomorphic to F ⊕ M3(F ). 10.10. Anti-Jordan q = ∞. The original self-reduced set of 24 generators has 76 distinct nontrivial compositions, and the combined set of 100 generators self-reduces to 15 elements, which is the Grobner basis (7). 10.11. Anti-Jordan q = −1. The original self-reduced set of 24 generators has 37 distinct nontrivial compositions. The combined set of 61 elements self-reduces to 23 elements with 6 distinct nontrivial compositions. The combined set of 29 elements self-reduces to 15 elements, which is the Grobner basis (7). 10.12. Anti-Jordan q = 2. The original self-reduced set of 24 generators has 37 distinct nontrivial compositions. The combined set of 61 elements self-reduces to 23 elements with 4 distinct nontrivial compositions. The combined set of 27 elements self-reduces to 15 elements, which is the Grobner basis (7). 10.13. Anti-Jordan q = 1 2 . In this case we are studying a simple anti-Jordan triple system; see Faulkner and Ferrar [50] and the Ph.D. thesis of Bashir [6]. The original self-reduced set of 24 generators is as follows: ba2 − a2b, bda − adb − a, ca2 − a2c, cab − bac, cba − abc, da2 − a2d, b2a − ab2, cb2 − b2c, bca − acb + b, c2a − ac2, c2b − bc2, cda − adc, cdb − bdc, dab − bad, dac − cad − d, dba − abd, db2 − b2d, dbc − cbd + c, 36 MURRAY R. BREMNER dca − acd, dcb − bcd, dc2 − c2d, d2a − ad2, d2b − bd2, d2c − cd2. There are 32 distinct nontrivial compositions: d2, ad2, c2, ac2, acd, a2, b2, cd, dc, ba, ab, abd + a2c, bc2, abd, b2d + bac, bcd, bdc, adc, bdc + ac2, a2b, a2c, a2d, ab2, abc, bac, bad, b2c, b2d, bd2 + acd, bd2, c2d, cd2. The combined set of 56 elements self-reduces to a Grobner basis of 12 elements: ab, a2, ba, dac − cad − d, c2, b2, cd, dbc − cbd + c. dc, d2, bca − acb + b, bda − adb − a, The universal associative envelope is infinite dimensional; the dimensions of the homogeneous components of the associated graded algebra appear to be 1 2 (n + 1)(n + 3) (n odd), 1 2 (n + 2)2 (n even). Open Problem 10.9. Prove that the universal envelope is infinite dimensional, and that the dimensions of the homogeneous components are as stated. 10.14. Fourth family q = ∞. The original set of 52 generators self-reduces to 44 elements, which have 140 distinct nontrivial compositions. Self-reducing the combined set of 184 generators produces the Grobner basis (7). 10.15. Fourth family q = 0. The original set of 52 generators self-reduces to 44 el- ements, which have 88 distinct nontrivial compositions. Self-reducing the combined set of 132 generators produces the Grobner basis (7). 10.16. Fourth family q = 1. The original set of 52 generators self-reduces to 44 el- ements, which have 76 distinct nontrivial compositions. Self-reducing the combined set of 120 generators produces the Grobner basis (7). 10.17. Fourth family q = −1. The original set of 64 generators self-reduces to 44 elements, which have 209 distinct nontrivial compositions. Self-reducing the combined set of 253 generators produces the Grobner basis (7). 10.18. Fourth family q = 2. The original set of 64 generators self-reduces to 44 elements, which have 227 distinct nontrivial compositions. Self-reducing the combined set of 271 generators produces the Grobner basis (7). 10.19. Fourth family q = 1 2 . The original set of 64 generators self-reduces to 44 elements, which have 184 distinct nontrivial compositions. Self-reducing the combined set of 228 generators produces the Grobner basis (7). 10.20. Cyclic commutator. The original set of 60 generators self-reduces to 40 elements, which have 86 distinct nontrivial compositions. Self-reducing the com- bined set of 126 generators produces the Grobner basis (7). 10.21. Weakly commutative operation. The original set of 64 generators self- reduces to 60 elements, which have 15 distinct nontrivial compositions. Self- reducing the combined set of 75 generators produces the Grobner basis (7). 10.22. Weakly anticommutative operation. The original set of 60 generators self-reduces to 44 elements, which have 41 distinct nontrivial compositions. Self- reducing the combined set of 85 generators produces the Grobner basis (7). GR OBNER BASES AND UNIVERSAL ENVELOPES 37 operation U (Aω 1 ) U (Aω 2 ) U (Aω 3 ) 1,6,36,160,750,3240, . . . 1,6,36,196,1071,5796, . . . n unable to complete 1,6,30,110,360,1026, . . . 1,6,24,74,195,456, . . . 1,6,18,36,72,120, . . . 26 69 17 17 17 9 9 19 10 10 10 Alt sum Lie q = 1 Jor q = 1 1,2,4,4,5, . . . 1,2,4,4,5, . . . 1,2,4,6,9,12, . . . 1,2,4,6,9,12, . . . 1,2,4,6,9,12, . . . 1,4,10,20,35,56, . . . 2 n 6,4 4 5 5 Jor q = ∞ n 6,4 4 Jor q = 0 n 6 Jor q = 1 n 6 AJ q = ∞ n 2 AJ q = −1 n 2,2 4,2 4 Sym sum n 4 n 20 1,4,16,44,131,344, . . . n 56 n 0 1,2,4,8,16,32, . . . n 4 1,4,16,60,225,840, . . . n 20 n 24,40 59,724 62 Cyc sum n 4 n 20,24 16 1,4,14,36,85,176, . . . n 70,140 51 Lie q = ∞ n 2 n 20,26 12 n 70,147 39 2 n 2 n 40,32 20 n 126,107 54 n 40,20 27,4 15 n 126,97 71,9 32 n 40,19 27,6 15 n 126,93 71,18 32 n 40,94 15 n 126,542 32 n 24,76 15 n 90,513 32 n 24,37 23,6 15 n 90,135 62,18 32 n 24,32 12 n 90,107 36 n 24,37 23,4 15 n 90,137 62,9 32 n 40,140 15 n 146,1065 32 n 44,88 15 n 146,737 32 n 44,76 15 n 146,618 32 n 44,209 15 n 146,1432 32 n 44,227 15 n 146,1601 32 n 44,184 15 n 146,1347 32 n 40,86 15 n 140,396 32 n 60,15 15 n 196,58 32 n 44,41 15 n 160,124 32 AJ q = 2 n 2,2 4,2 4 4th q = ∞ n 6,4 4 4th q = 0 n 6 4th q = 1 n 6 4th q = −1 n 6,5 4 4th q = 2 n 6,5 4 2 n 6,4 4 Cyc com n 4,4 4 n 8,2 4 Weak AC n 4,4 4 2 n 2 1,4,8,12,18,24, . . . 1,2,4,6,9,12, . . . 5 5 9 9 5 5 5 5 5 5 4th q = 1 AJ q = 1 17 17 17 17 Weak C 17 17 10 10 10 10 10 10 10 17 17 10 10 10 10 17 17 17 10 17 5 Table 5. Universal associative envelopes of nonassociative triple systems 38 MURRAY R. BREMNER 10.23. Summary. Table 5 summarizes the results of Elgendy [45, 46] for U (Aω the results of this section for U (Aω entry in the table has the form 2 ), and further computations for U (Aω 1 ), 3 ). Each n algorithm dimension where "algorithm" describes the performance of the Grobner basis algorithm, and "dimension" gives the dimension of the universal associative envelope. The "al- gorithm" data consists of a sequence of pairs x, y corresponding to the iterations of the algorithm; x is the size of the self-reduced set of generators at the start of the iteration, and y is the number of distinct nontrivial compositions in normal form at the end of the iteration (if y = 0 it is omitted). The "dimension" data consists either of a single number (in the case where the universal envelope is fi- nite dimensional), or a sequence of numbers giving the first few dimensions of the homogeneous components of the associated graded algebra (in the case where the universal envelope is infinite dimensional). Dimensions in boldface indicate values that repeat indefinitely. For example, consider the entry in row "Lie q = ∞" and column "U (Aω 3 )", the Lie triple product on the 6-dimensional simple associative triple system: n 70, 140 51 1, 6, 30, 110, 360, 1026, . . . This means: (a) The algorithm terminated after two iterations: the original self-reduced set of 70 generators produced 140 nontrivial compositions; the combined set of 210 generators self-reduced to a Grobner basis of 51 elements. (b) The universal associative envelope is infinite dimensional, and the gener- ating function for the dimensions of the homogeneous components of the associated graded algebra begins with the terms 1 + 6z + 30z2 + 110z3 + 360z4 + 1026z5 + · · · In one case, U (Aω 3 ) for the cyclic sum, the computations were so complicated that Maple 14 on my MacBook Pro was unable to complete them in a reasonable time. This may be related to the fact that the polynomial identities satisfied by this operation are extremely complicated; see my paper with Peresi [30]. 10.24. Conclusions. The results of Table 5 suggest a slightly different classifica- tion of operations into "Lie type" and "Jordan type" from that of Elgendy [45, 46]. Two operations, the cyclic sum and the anti-Jordan q = ∞ operation, produce infi- nite dimensional envelopes for Aω 2 . It seems likely that Aω 1 is exceptional, owing to its small dimension, and that the universal associative envelopes will be finite dimensional when either of these operations is applied to a simple associative triple system of dimension > 2. If this is correct, then these two operations should be reclassified as having "Jordan type". 1 but finite dimensional envelopes for Aω Four operations produced a non-semisimple envelope for Aω 1 : Jordan q = 0, 1 and fourth family q = 0, 1. In these cases, the 9-dimensional envelope has a 4- dimensional radical and a 5-dimensional semisimple quotient which is isomorphic to F ⊕ M2(F ). For these operations it seems very likely that U (Aω n) (n = 2, 3) is semisimple and is isomorphic to F ⊕ Mn+1(F ). The reason is that the dimension of the envelope (10 for n = 2 and 17 for n = 3) is the sum of the squares of the GR OBNER BASES AND UNIVERSAL ENVELOPES 39 dimensions of the 1-dimensional trivial representation and the (n+1)-dimensional natural representation. This seems to hold for most of the operations: the universal envelopes are finite dimensional and the only irreducible representations are the trivial representation and the natural representation. Conjecture 10.10. Let Ap,q (p ≤ q) be the simple associative triple system con- sisting of (p+q) × (p+q) block matrices of the form (cid:20) 0 q × p p × q 0 (cid:21) . Let ω be one of the following trilinear operations from Table 4: Jordan (q = 0, 1, 1 2 ), anti-Jordan (q = ∞, −1, 2), fourth family (all cases), cyclic commutator, weakly commutative, weakly anticommutative. Then, with finitely many exceptions, U (Aω) is finite dimensional and semisimple and is isomorphic to F ⊕ Mp+q(F ). The operations not included in this conjecture are the first three operations (the symmetric, alternating, and cyclic sums), together with the four classical operations (the Lie, anti-Lie, Jordan, and anti-Jordan triple products). These seem likely to be the operations producing nonassociative triple systems with the most interesting representation theory. This is well-known for the four classical operations, owing to their close connection with Lie and Jordan algebras and superalgebras. On the other hand, very little is known about the representation theory of nonassociative triple systems arising from the first three operations. Open Problem 10.11. Study the structure of the universal associative envelopes, and classify the finite dimensional irreducible representations, for the nonassociative triple systems Aω p,q where ω is the symmetric, alternating, and cyclic sum. 11. Bibliographical Remarks The historical origins of the theory of Grobner bases are complex, with similar ideas discovered in different contexts at different times by different people. 11.1. The commutative case. The most famous branch of the theory, owing to its close connections with algebraic geometry, is that of commutative Grobner bases. Many of these ideas can be traced back to the work of Macaulay; his 1916 monograph on The Algebraic Theory of Modular Systems is available online [82]. The original work of Grobner most often cited as the origin of the theory of commu- tative Grobner bases is his 1939 paper on linear differential equations [61]; this has appeared in English translation [62] with commentary by the translator [1]. The modern form of the theory which emphasizes the algorithmic aspects originated in the 1965 Ph.D. thesis of Buchberger which has been translated into English [31] with commentary by the author [33]; see also his 1970 paper [32]. There are many textbooks on the theory of commutative Grobner bases and their applications; see Adams and Loustaunau [2], Becker and Weispfennig [7], Cox et al. [42], Ene and Herzog [48], and Froberg [51]. 11.2. The noncommutative case. The theory of noncommutative Grobner bases seems to have originated with the Russian school of nonassociative algebra; see the papers of Zhukov [104] and especially Shirshov [98, 99]. The first systematic state- ments of the Composition (Diamond) Lemma in the noncommutative case, and 40 MURRAY R. BREMNER its application to the proof of the PBW theorem, were published almost simul- taneously by Bokut [11] and Bergman [9]. The latter paper traces the origins of the theory to earlier work of Newman [89]. The computational complexity of al- gorithms for constructing noncommutative Grobner bases has been studied by F. Mora [86]. The Ph.D. thesis of Keller [71, 72] on noncommutative Grobner bases led to the software package Opal [59]. A more recent software package, with extensive online documentation, has been developed by Cohen and Gijsbers [41]. For some important papers on theory and algorithms for noncommutative Grobner bases, see Borges-Trenard et al. [60], and Kang et al. [70]. For a connection between commutative and noncommutative Grobner bases, see Eisenbud et al. [44]. For an extension to noncommutative power series, see Gerritzen and Holtkamp [55]. For textbooks on noncommutative Grobner bases, see Bokut and Kukin [21], Bueso et al. [34], and Li [73]. [23], Gerritzen [52], Green et al. 11.3. The nonassociative case. The most important branch of the nonassocia- tive theory deals with Grobner-Shirshov bases for free Lie algebras; see Bokut and Chibrikov [19] and Bokut and Chen [13]. A theory of Grobner-Shirshov bases in free nonassociative algebras has been developed by Gerritzen [53, 54] and Rajaee [95]. For related work on Sabinin algebras, see Shestakov and Umirbaev [97], P´erez- Izquierdo [92], and Chibrikov [39]. [17]. For dendriform algebras, see Bokut et al. 11.4. Loday algebras. An active area of current research is extending the Compo- sition (Diamond) Lemma from associative algebras to the dialgebras and dendriform algebras introduced by Loday [76, 77, 78]. For associative dialgebras, see Bokut et al. [15], Chen and Wang [38], as well as the papers on Rota-Baxter algebras by Bokut et al. [14, 18], Chen and Mo [37], Qiu [94], and Guo et al [63]. It is an open problem to extend these results further to the quadri-algebras of Aguiar and Loday [3], and to the Koszul dual of quadri-algebras introduced by Vallette [102, §5.6]. For Leibniz algebras, which are the analogues of Lie algebras in the setting of dialgebras, see Loday and Pirashvili [79], Aymon and Grivel [4], Casas et al. [36], Insua and Ladra [68]. For pre-Lie algebras, which are the analogue of Lie algebras in the setting of dendriform alge- bras, see Bokut et al. [16]. For L-dendriform algebras, which are the analogue of Lie algebras in the setting of quadri-algebras, see Bai et al. [5]. (For corresponding generalizations of Jordan algebras, see Hou et al. [66, 67].) 11.5. Survey papers. A survey of commutative and noncommutative Grobner bases from the point of view of theoretical computer science has been written by T. Mora [87]. For an introduction to noncommutative Grobner bases from the point of view of computer algebra, see Green [57, 58] and Ufnarovski [101]. A number of introductory surveys of Grobner-Shirshov bases in associative and nonassociative algebras have been written by Bokut and his co-authors: see Bokut [12], Bokut and Kolesnikov [20], and Bokut and Shum [22]. Acknowledgements I thank NSERC (Natural Sciences and Engineering Research Council of Canada) for financial support through a Discovery Grant, and the faculty and staff of CIMAT (Centro de Investigaci´on en Matem´aticas, Guanajuato, Mexico) for their hospitality during the Research School on Associative and Nonassociative Algebras and Dial- gebras: Theory and Algorithms - In Honour of Jean-Louis Loday (1946 -- 2012) from GR OBNER BASES AND UNIVERSAL ENVELOPES 41 February 17 to March 2, 2013 which was sponsored by CIMPA (Centre Interna- tional de Math´ematiques Pures et Appliqu´ees). I thank my former Ph.D. student Hader Elgendy for pointing out some errors in an earlier version of these notes. References [1] M. P. Abramson: Historical background to Grobner's paper. ACM Commun. Comput. Algebra 43 (2009) no. 1-2, 22 -- 23. [2] W. W. Adams, P. Loustaunau: An Introduction to Grobner Bases. Graduate Studies in Mathematics, 3. American Mathematical Society, Providence, RI, 1994. [3] M. Aguiar, J.-L. Loday: Quadri-algebras. J. Pure Appl. Algebra 191 (2004) no. 3, 205 -- 221. [4] M. Aymon, P.-P. Grivel: Un th´eor`eme de Poincar´e-Birkhoff-Witt pour les alg`ebres de Leibniz. Comm. Algebra 31 (2003) no. 2, 527 -- 544. [5] C. Bai, L. Liu, X. Ni: Some results on L-dendriform algebras. J. Geom. Phys. 60 (2010) no. 6-8, 940 -- 950. [6] S. Bashir: Automorphisms of Simple Anti-Jordan Pairs. Ph.D. Thesis, University of Ot- tawa, Canada, 2008. [7] T. Becker, V. Weispfenning: Grobner Bases: A Computational Approach to Commuta- tive Algebra. Graduate Texts in Mathematics, 141. Springer-Verlag, New York, 1993. [8] G. Benkart, T. Roby: Down-up algebras. J. Algebra 209 (1998) no. 1, 305 -- 344. Addendum: "Down-up algebras". J. Algebra 213 (1999), no. 1, 378. [9] G. M. Bergman: The diamond lemma for ring theory. Adv. in Math. 29 (1978) no. 2, 178 -- 218. [10] G. Birkhoff, P. M. Whitman: Representation of Jordan and Lie algebras. Trans. Amer. Math. Soc. 65 (1949) 116 -- 136. [11] L. A. Bokut: Imbeddings into simple associative algebras. Algebra i Logika 15 (1976) no. 2, 117 -- 142. [12] L. A. Bokut: The method of Grobner-Shirshov bases. Siberian Adv. Math. 9 (1999) no. 3, 1 -- 16. [13] L. A. Bokut, Y. Chen: Grobner-Shirshov bases for Lie algebras: after A. I. Shirshov. Southeast Asian Bull. Math. 31 (2007) no. 6, 1057 -- 1076. [14] L. A. Bokut, Y. Chen, X. Deng: Grobner-Shirshov bases for Rota-Baxter algebras. Sibirsk. Mat. Zh. 51 (2010) no. 6, 1237 -- 1250. [15] L. A. Bokut, Y. Chen, J. Huang: Grobner-Shirshov bases for L-algebras. Internat. J. Algebra Comput. (to appear). arXiv:1005.0118 [math.RA] [16] L. A. Bokut, Y. Chen, Y. Li: Grobner-Shirshov bases for Vinberg-Koszul-Gerstenhaber right-symmetric algebras. Fundam. Prikl. Mat. 14 (2008) no. 8, 55 -- 67. [17] L. A. Bokut, Y. Chen, C. Liu: Grobner-Shirshov bases for dialgebras. Internat. J. Algebra Comput. 20 (2010) no. 3, 391 -- 415. [18] L. A. Bokut, Y. Chen, X. Deng: Grobner-Shirshov bases for associative algebras with multiple operators and free Rota-Baxter algebras. J. Pure Appl. Algebra 214 (2010) no. 1, 89 -- 100. [19] L. A. Bokut, E. S. Chibrikov: Lyndon-Shirshov words, Grobner-Shirshov bases, and free Lie algebras. Non-associative Algebra and Its Applications, pages 17 -- 39. Lect. Notes Pure Appl. Math., 246, Chapman & Hall/CRC, Boca Raton, 2006. [20] L. A. Bokut, P. S. Kolesnikov: Grobner-Shirshov bases: from their incipiency to the present. J. Math. Sci. Vol. 116, No. 1, 2003. [21] L. A. Bokut, G. P. Kukin: Algorithmic and Combinatorial Algebra. Mathematics and its Applications, 255. Kluwer Academic Publishers Group, Dordrecht, 1994. [22] L. A. Bokut, K. P. Shum: Grobner and Grobner-Shirshov bases in algebra: an elementary approach. Southeast Asian Bull. Math. 29 (2005) no. 2, 227 -- 252. [23] M. A. Borges-Trenard, M. Borges-Quintana, T. Mora: Computing Grobner bases by FGLM techniques in a non-commutative setting. J. Symbolic Comput. 30 (2000) no. 4, 429 -- 449. [24] M. R. Bremner: How to compute the Wedderburn decomposition of a finite-dimensional associative algebra. Groups Complex. Cryptol. 3 (2011) no. 1, 47 -- 66. [25] M. R. Bremner: Algebras, dialgebras, and polynomial identities. Serdica Math. J. 38 (2012) 91 -- 136. 42 MURRAY R. BREMNER [26] M. R. Bremner, I. R. Hentzel: Identities for generalized Lie and Jordan products on totally associative triple systems. J. Algebra 231 (2000) no. 1, 387 -- 405. [27] M. R. Bremner, S. Madariaga: Polynomial identities for tangent algebras of monoasso- ciative loops. Comm. Algebra (to appear) [28] M. R. Bremner, L. A. Peresi: Classification of trilinear operations. Comm. Algebra 35 (2007) no. 9, 2932 -- 2959. [29] M. R. Bremner, L. A. Peresi: An application of lattice basis reduction to polynomial identities for algebraic structures. Linear Algebra Appl. 430 (2009) no. 2-3, 642 -- 659. [30] M. R. Bremner, L. A. Peresi: Polynomial identities for the ternary cyclic sum. Linear Multilinear Algebra 57 (2009) no. 6, 595 -- 608. [31] B. Buchberger: An Algorithm for Finding the Basis Elements of the Residue Class Ring of a Zero Dimensional Polynomial Ideal. Translated from the 1965 German original by Michael P. Abramson. J. Symbolic Comput. 41 (2006) no. 3-4, 475 -- 511. [32] B. Buchberger: Ein algorithmisches Kriterium fur die Losbarkeit eines algebraischen Gle- ichungssystems. Aequationes Math. 4 (1970) 374 -- 383. [33] B. Buchberger: Comments on the translation of my PhD thesis: "An Algorithm for Find- ing the Basis Elements of the Residue Class Ring of a Zero Dimensional Polynomial Ideal" J. Symbolic Comput. 41 (2006) no. 3-4, 471 -- 474. [34] J. Bueso, J. G´omez-Torrecillas, A. Verschoren: Algorithmic Methods in Noncommu- tative Algebra: Applications to Quantum Groups. Mathematical Modelling: Theory and Applications, 17. Kluwer Academic Publishers, Dordrecht, 2003. [35] R. Carlsson: n-ary algebras. Nagoya Math. J. 78 (1980) 45 -- 56. [36] J. M. Casas, M. A. Insua, M. Ladra: Poincar´e-Birkhoff-Witt theorem for Leibniz n- algebras. J. Symbolic Comput. 42 (2007) no. 11-12, 1052 -- 1065. [37] Y. Chen, Q. Mo: Embedding dendriform algebra into its universal enveloping Rota-Baxter algebra. Proc. Amer. Math. Soc. 139 (2011) no. 12, 4207 -- 4216. [38] Y. Chen, B. Wang: Grobner-Shirshov bases and Hilbert series of free dendriform algebras. Southeast Asian Bull. Math. 34 (2010) no. 4, 639 -- 650. [39] E. S. Chibrikov: On free Sabinin algebras. Comm. Algebra 39 (2011) no. 11, 4014 -- 4035. [40] J. M. Clifton: A simplification of the computation of the natural representation of the symmetric group Sn. Proc. Amer. Math. Soc. 83 (1981) no. 2, 248 -- 250. [41] A. M. Cohen, D. A. H. Gijsbers: Documentation on the GBNP Package. Available at: http://www.win.tue.nl/~amc/pub/grobner/doc.html (accessed 19 January 2013) [42] D. Cox, J. Little, D. O'Shea: Ideals, Varieties, and Algorithms: An Introduction to Com- putational Algebraic Geometry and Commutative Algebra. Undergraduate Texts in Mathe- matics. Springer-Verlag, New York, 1992. [43] W. A. de Graaf: Lie Algebras: Theory and Algorithms. North-Holland Mathematical Library, 56. North-Holland Publishing Co., Amsterdam, 2000. [44] D. Eisenbud, I. Peeva, B. Sturmfels: Non-commutative Grobner bases for commutative algebras. Proc. Amer. Math. Soc. 126 (1998) no. 3, 687 -- 691. [45] H. A. Elgendy: Polynomial Identities and Enveloping Algebras for n-ary Structures. Ph.D. thesis, University of Saskatchewan, Canada, 2012. [46] H. A. Elgendy: Universal associative envelopes of nonassociative triple systems. Comm. Algebra (to appear). arXiv:1211.4243 [math.RA] [47] H. A. Elgendy, M. R. Bremner: Universal associative envelopes of (n+1)-dimensional n-Lie algebras. Comm. Algebra 40 (2012) no. 5, 1827 -- 1842. [48] V. Ene, J. Herzog: Grobner Bases in Commutative Algebra. Graduate Studies in Mathe- matics, 130. American Mathematical Society, Providence, RI, 2012. [49] J. R. Faulkner: Identity classification in triple systems. J. Algebra 94 (1985) no. 2, 352 -- 363. [50] J. R. Faulkner, J. C. Ferrar: Simple anti-Jordan pairs. Comm. Algebra 8 (1980) no. 11, 993 -- 1013. [51] R. Froberg: An Introduction to Grobner Bases. Pure and Applied Mathematics (New York). John Wiley & Sons, Ltd., Chichester, 1997. [52] L. Gerritzen: On infinite Grobner bases in free algebras. Indag. Math. (N.S.) 9 (1998) no. 4, 491 -- 501. [53] L. Gerritzen: Hilbert series and non-associative Grobner bases. Manuscripta Math. 103 (2000) no. 2, 161 -- 167. GR OBNER BASES AND UNIVERSAL ENVELOPES 43 [54] L. Gerritzen: Tree polynomials and non-associative Grobner bases. J. Symbolic Comput. 41 (2006) no. 3-4, 297 -- 316. [55] L. Gerritzen, R. Holtkamp: On Grobner bases of noncommutative power series. Indag. Math. (N.S.) 9 (1998) no. 4, 503 -- 519. [56] C. M. Glennie: Some identities valid in special Jordan algebras but not valid in all Jordan algebras. Pacific J. Math. 16 (1966) 47 -- 59. [57] E. L. Green: An introduction to noncommutative Grobner bases. Computational Algebra, pp. 167 -- 190, Lecture Notes in Pure and Appl. Math., 151. Dekker, New York, 1994. [58] E. L. Green: Noncommutative Grobner bases, and projective resolutions. Computa- tional Methods for Representations of Groups and Algebras, pp. 29 -- 60. Progr. Math., 173. Birkhauser, Basel, 1999. [59] E. L. Green, L. S. Heath, B. J. Keller: Opal: a system for computing noncommuta- tive Grobner bases. Rewriting Techniques and Applications, pp. 331 -- 334. Lecture Notes in Computer Science, 1232. Springer, 1997. [60] E. L. Green, T. Mora, V. Ufnarovski: The non-commutative Grobner freaks. Sym- bolic Rewriting Techniques (Ascona, 1995), 93104, Progr. Comput. Sci. Appl. Logic, 15, Birkhauser, Basel, 1998. [61] W. Grobner: Uber die algebraischen Eigenschaften der Integrale von linearen Differential- gleichungen mit konstanten Koeffizienten. Monatsh. Math. Phys. 47 (1939) no. 1, 247 -- 284. [62] W. Grobner: On the algebraic properties of integrals of linear differential equations with constant coefficients. Translated from the German by Michael Abramson. ACM Commun. Comput. Algebra 43 (2009) no. 1-2, 24 -- 46. [63] L. Guo, W. Sit, R. Zhang: Differential type operators and Grobner-Shirshov bases. J. Symbolic Comput. 52 (2013) 97 -- 123. [64] M. R. Hestenes: A ternary algebra with applications to matrices and linear transforma- tions. Arch. Rational Mech. Anal. 11 (1962) 138 -- 194. [65] T. L. Hodge, B. J, Parshall: On the representation theory of Lie triple systems. Trans. Amer. Math. Soc. 354 (2002) no. 11, 4359 -- 4391. [66] D. Hou, C. Bai: J-dendriform algebras. Front. Math. China 7 (2012) no. 1, 29 -- 49. [67] D. Hou, X. Ni, C. Bai: Pre-Jordan algebras. Math. Scand. (to appear) [68] M. A. Insua, M. Ladra: Grobner bases in universal enveloping algebras of Leibniz algebras. J. Symbolic Comput. 44 (2009) no. 5, 517 -- 526. [69] N. Jacobson: Structure and Representations of Jordan Algebras. American Mathematical Society Colloquium Publications, Vol. XXXIX. American Mathematical Society, Providence, R.I., 1968. [70] S.-J. Kang, D.-I. Lee, K.-H. Lee, H. Park: Linear algebraic approach to Grobner-Shirshov basis theory. J. Algebra 313 (2007) no. 2, 988 -- 1004. [71] B. J. Keller: Algorithms and Orders for Finding Noncommutative Grobner Bases. Ph.D. Thesis, Virginia Polytechnic Institute and State University, 1997. [72] B. J. Keller: Alternatives in implementing noncommutative Grobner basis systems. Sym- bolic Rewriting Techniques, pp. 105 -- 126. Progr. Comput. Sci. Appl. Logic, 15. Birkhuser, Basel, 1998. [73] H. Li: Grobner Bases in Ring Theory. World Scientific Publishing Co. Pte. Ltd., Hacken- sack, NJ, 2012. [74] W. G. Lister: A structure theory of Lie triple systems. Trans. Amer. Math. Soc. 72 (1952) 217 -- 242. [75] W. G. Lister: Ternary rings. Trans. Amer. Math. Soc. 154 (1971) 37 -- 55. [76] J.-L. Loday: Une version non commutative des alg`ebres de Lie: les alg`ebres de Leibniz. Enseign. Math. (2) 39 (1993) no. 3-4, 269 -- 293. [77] J.-L. Loday: Alg`ebres ayant deux op´erations associatives (dig`ebres). C. R. Acad. Sci. Paris S´er. I Math. 321 (1995) no. 2, 141 -- 146. [78] J.-L. Loday: Dialgebras. Dialgebras and Related Operads, pp. 7 -- 66. Lecture Notes in Math., 1763, Springer, Berlin, 2001. [79] J.-L. Loday, T. Pirashvili: Universal enveloping algebras of Leibniz algebras and (co)homology. Math. Ann. 296 (1993) no. 1, 139 -- 158. [80] O. Loos: Lectures on Jordan Triples. The University of British Columbia, Vancouver, Canada, 1971. [81] O. Loos: Assoziative Tripelsysteme. Manuscripta Math. 7 (1972) 103 -- 112. 44 MURRAY R. BREMNER [82] F. S. Macaulay: of the 1916 reprint bridge Mathematical Library. Cambridge University Press, Cambridge, http://archive.org/details/algebraictheoryo00macauoft Systems. Revised original. With an introduction by Paul Roberts. Cam- 1994. The Algebraic Theory of Modular [83] K. McCrimmon: A Taste of Jordan Algebras. Universitext. Springer-Verlag, New York, 2004. [84] K. Meyberg: Lectures on Algebras and Triple Systems. Notes on a course of lectures given during the academic year 1971-1972. The University of Virginia, Charlottesville, 1972. [85] A. A. Mikhalev, A. A. Zolotykh: Standard Grobner-Shirshov bases of free algebras over rings. I. Free associative algebras. Internat. J. Algebra Comput. 8 (1998) no. 6, 689 -- 726. [86] F. Mora: Groebner bases for noncommutative polynomial rings. Algebraic Algorithms and Error-Correcting Codes, Lecture Notes in Computer Science, 229, pp. 353 -- 362. Springer, Berlin, 1986. [87] T. Mora: An introduction to commutative and noncommutative Grobner bases. Theoret. Comput. Sci. 134 (1994) 131 -- 173. [88] I. M. Musson: Lie Superalgebras and Enveloping Algebras. Graduate Studies in Mathemat- ics, 131. American Mathematical Society, Providence, 2012. [89] M. H. A. Newman: On theories with a combinatorial definition of "equivalence". Annals of Math. 43, 2 (1942) 223 -- 243. [90] On-line Encyclopedia of Integer Sequences: http://oeis.org/ [91] J.-M. P´erez-Izquierdo: An envelope for Bol algebras. J. Algebra 284 (2005) no. 2, 480 -- 493. [92] J.-M. P´erez-Izquierdo: Algebras, hyperalgebras, nonassociative bialgebras and loops. Adv. Math. 208 (2007) no. 2, 834 -- 876. [93] J.-M. P´erez-Izquierdo, I. P. Shestakov: An envelope for Malcev algebras. J. Algebra 272 (2004) no. 1, 379 -- 393. [94] J. Qiu: Grobner-Shirshov bases for commutative algebras with multiple operators and free commutative Rota-Baxter algebras. arXiv:1301.5018 [95] S. Rajaee: Non-associative Grobner bases. J. Symbolic Comput. 41 (2006) no. 8, 887 -- 904. [96] D. E. Rutherford: Substitutional Analysis. Edinburgh, at the University Press, 1948. [97] I. P. Shestakov, U. U, Umirbaev: Free Akivis algebras, primitive elements, and hyperal- gebras. J. Algebra 250 (2002) no. 2, 533 -- 548. [98] A. I. Shirshov: Some algorithmic problems for ǫ-algebras. Sibirsk. Mat. Zh. 3 (1962) 132 -- 137. [99] A. I. Shirshov: On a hypothesis in the theory of Lie algebras. Sibirsk. Mat. Zh. 3 (1962) 297 -- 301 (1962). [100] A. I. Shirshov: Selected Works of A. I. Shirshov. Translated from the Russian by M. R. Bremner and M. V. Kotchetov. Edited by L. A. Bokut, V. N. Latyshev, I. P. Shestakov and E. Zelmanov. Contemporary Mathematicians. Birkhauser Verlag, Basel, 2009. [101] V. S. Ufnarovski: Introduction to noncommutative Grobner bases theory. Grobner Bases and Applications, pp. 259 -- 280. London Math. Soc. Lecture Note Ser., 251. Cambridge Univ. Press, Cambridge, 1998. [102] B. Vallette: Manin products, Koszul duality, Loday algebras and Deligne conjecture. J. Reine Angew. Math. 620 (2008) 105 -- 164. [103] A. Young: The Collected Papers of Alfred Young (1873 -- 1940). With a foreword by G. de B. Robinson and a biography by H. W. Turnbull. Mathematical Expositions, No. 21. University of Toronto Press, 1977. [104] A. I. Zhukov: Reduced systems of defining relations in non-associative algebras. Mat. Sbornik N.S. 27(69) (1950) 267 -- 280. Department of Mathematics and Statistics, University of Saskatchewan, Canada E-mail address: [email protected] URL: math.usask.ca/~bremner
1910.09018	1	1910	2019-10-20T16:46:18	Point modules over regular graded skew Clifford algebras	[ "math.RA" ]	Results of Vancliff, Van Rompay and Willaert in 1998 prove that point modules over a regular graded Clifford algebra (GCA) are determined by (commutative) quadrics of rank at most two that belong to the quadric system associated to the GCA. In 2010, Cassidy and Vancliff generalized the notion of a GCA to that of a graded skew Clifford algebra (GSCA). The results in this article show that prior results may be extended, with suitable modification, to GSCAs. In particular, using the notion of {\mu}-rank introduced recently by the authors, the point modules over a regular GSCA are determined by (noncommutative) quadrics of {\mu}-rank at most two that belong to the noncommutative quadric system associated to the GSCA.	math.RA	math	POINT MODULES OVER REGULAR GRADED SKEW CLIFFORD ALGEBRAS Michaela Vancliff∗ Department of Mathematics, P.O. Box 19408 University of Texas at Arlington, Arlington, TX 76019-0408 [email protected] www.uta.edu/math/vancliff and Padmini P. Veerapen† Department of Mathematics, P.O. Box 19408 University of Texas at Arlington, Arlington, TX 76019-0408 [email protected] 9 1 0 2 t c O 0 2 ] . A R h t a m [ 1 v 8 1 0 9 0 . 0 1 9 1 : v i X r a Abstract. Results of Vancliff, Van Rompay and Willaert in 1998 ([VVW]) prove that point modules over a regular graded Clifford algebra (GCA) are determined by (commutative) quadrics of rank at most two that belong to the quadric system associated to the GCA. In 2010, in [CV], Cassidy and Vancliff generalized the notion of a GCA to that of a graded skew Clifford algebra (GSCA). The results in this article show that the results of [VVW] may be extended, with suitable modification, to GSCAs. In particular, using the notion of µ-rank introduced recently by the authors in [VV], the point modules over a regular GSCA are determined by (noncommutative) quadrics of µ-rank at most two that belong to the noncommutative quadric system associated to the GSCA. 2010 Mathematics Subject Classification. 16W50, 14A22, 16S36. Key words and phrases. Clifford algebra, quadratic form, skew polynomial ring, point module. ∗The first author was supported in part by NSF grants DMS-0457022 & DMS-0900239. †The second author was supported in part by NSF grant DGE-0841400 as a Graduate Teaching Fellow in U.T. Arlington's GK-12 MAVS Project. 2 POINT MODULES OVER REGULAR GRADED SKEW CLIFFORD ALGEBRAS Introduction The notion of a graded skew Clifford algebra (GSCA) was introduced in [CV], and it is an algebra that may be viewed as a quantized analog of a graded Clifford algebra (GCA). In [CV], it was shown that many of the results that hold for GCAs have analogous counterparts in the context of GSCAs. In particular, homological and algebraic properties of a GSCA, A, are determined by properties of a certain quadric system associated to A. The importance of GSCAs was highlighted in [NVZ], where they were shown to play a critical role in the classification of the quadratic AS-regular algebras of global dimension three. Hence, GSCAs are expected to play a critical role in the classification of the quadratic AS-regular algebras of global dimension four and greater. The reader is referred to [AS, ATV1, ATV2] for results concerning AS-regular algebras and their associated geometric data, and to [SV, VVW] for results concerning GCAs and their associated geometric data. Consequently, it is reasonable to attempt to extend the results in [VVW] concerning point modules over GCAs to point modules over GSCAs. As such, our main objective in this article is to generalize [VVW, Theorem 1.7]. That result states, in part, that if the number, N, of point modules over a regular GCA, C, is finite, then N = 2r2 + r1, where rj is the number of elements of rank j that belong to the projectivization of a certain quadric system associated to C (see Theorem 2.10 for the precise statement). We achieve our objective in Theorem 2.12, where the notion of µ-rank (introduced in [VV]) is used in place of the traditional notion of rank. However, [VVW, Theorem 1.7] also states, in part, that if N < ∞, then r1 ∈ {0, 1}. We present examples in the last section that demonstrate that this part of [VVW, Theorem 1.7] appears not to have an obvious counterpart in the setting of GSCAs. Although the flow of this article follows that of [VVW, §1], many of our results require methods of proof that differ substantially from those used in [VVW, §1], since the proofs in [VVW] make use of standard results concerning symmetric matrices and the general linear group. This article consists of two sections: in Section 1, notation and terminology are defined, while Section 2 is devoted to proving our main result, which is given in Theorem 2.12. 1. Graded Skew Clifford Algebras In this section, we define the notion of a graded skew Clifford algebra from [CV], and give the relevant results from [CV] needed in Section 2. Throughout the article, k denotes an algebraically closed field such that char(k) 6= 2, and M(n, k) denotes the vector space of n×n matrices with entries in k. For a graded k-algebra B, the span of the homogeneous elements in B of degree i will be denoted Bi, and T (V ) will POINT MODULES OVER REGULAR GRADED SKEW CLIFFORD ALGEBRAS 3 denote the tensor algebra on the vector space V . If C is a vector space, then C × will denote the nonzero elements in C, and C ∗ will denote the vector space dual of C. For {i, j} ⊂ {1, . . . , n}, let µij ∈ k× satisfy the property that µijµji = 1 for all i, j where i 6= j. We write µ = (µij) ∈ M(n, k). As in [CV], we write S for the quadratic k-algebra on generators z1, . . . , zn with defining relations zjzi = µijzizj for all i, j = 1, . . . , n, where µii = 1 for all i. We set U ⊂ T (S1)2 to be the span of the defining relations of S and write z = (z1, . . . , zn)T . Definition 1.1. [CV, §1.2] (a) With µ and S as above, a (noncommutative) quadratic form is defined to be any element of S2. (b) A matrix M ∈ M(n, k) is called µ-symmetric if Mij = µijMji for all i, j = 1, . . . , n. Henceforth, we assume µii = 1 for all i, and write M µ(n, k) for the vector space of µ- symmetric n × n matrices with entries in k. By [CV], there is a one-to-one correspondence between elements of M µ(n, k) and S2 via M 7→ zT Mz ∈ S. Notation 1.2. Let τ : P(M µ(n, k)) → P(S2) be defined by τ (M) = zT Mz. Remark 1.3. Henceforth, we fix M1, . . . , Mn ∈ M µ(n, k). For each k = 1, . . . , n, we fix representatives qk = τ (Mk). By [CV, Lemma 1.3], {qk}n k=1 is linearly independent in S if and only if {Mk}n k=1 is linearly independent. This correspondence mirrors the correspondence between symmetric matrices and commutative quadratic forms. Definition 1.4. [CV] A graded skew Clifford algebra A = A(µ, M1, . . . , Mn) associated to µ and M1, . . . , Mn is a graded k-algebra on degree-one generators x1, . . . , xn and on degree-two generators y1, . . . , yn with defining relations given by: n (a) xixj + µijxjxi = (Mk)ijyk for all i, j = 1, . . . , n, and Xk=1 (b) the existence of a normalizing sequence {y′ 1, . . . , y′ n} that spans ky1 + · · · + kyn. Remark 1.5. If A is a graded skew Clifford algebra (GSCA), then [CV, Lemma 1.13] implies that yi ∈ (A1)2 for all i = 1, . . . , n if and only if M1, . . . , Mn are linearly independent. Thus, hereafter, we assume that M1, . . . , Mn are linearly independent. By [CV], the degree of the defining relations of A and certain homological properties of A are intimately tied to certain geometric data associated to A as follows. Definition 1.6. [CV] (a) Let V(U) ⊂ P((S1)∗) × P((S1)∗) denote the zero locus of U. For any q ∈ S × 2 , we call the 4 POINT MODULES OVER REGULAR GRADED SKEW CLIFFORD ALGEBRAS zero locus of q in V(U) the quadric associated to q, and denote it by VU (q); in other words, VU (q) = V(kq + U) = V(q) ∩ V(U), where q is any lift of q to T (S1)2. The span of elements Q1, . . . , Qm ∈ S2 will be called the quadric system associated to Q1, . . . , Qm. (b) If a quadric system is given by a normalizing sequence in S, then it is called a normal- izing quadric system. (c) We call a point (a, b) ∈ V(U) a base point of the quadric system associated to Q1, . . . , Qm ∈ S2 if (a, b) ∈ VU (Qk) for all k = 1, . . . , m. We say such a quadric system is base-point free if Tm k=1 VU (Qk) is empty. Theorem 1.7. [CV] For all k = 1, . . . , n, let Mk and qk be as in Remark 1.3. A graded skew Clifford algebra A = A(µ, M1, . . . , Mn) is a quadratic, Auslander-regular algebra of global dimension n that satisfies the Cohen-Macaulay property with Hilbert series 1/(1 − t)n if and only if the quadric system associated to {q1, . . . , qn} is normalizing and base-point free; in this case, A is a noetherian Artin-Schelter regular domain and is unique up to isomorphism. Remark 1.8. (a) Henceforth, we assume that the quadric system associated to {q1, . . . , qn} is normalizing and base-point free. By Theorem 1.7, this assumption allows us to write A = T (V )/hW i, where V = (S1)∗ and W ⊆ T (V )2, and write the Koszul dual of A as T (S1)/hW ⊥i = S/hq1, . . . , qni. In this setting, {x1, . . . , xn} is the dual basis in V to the basis {z1, . . . , zn} of S1, and we write Pi,j αijm(xixj + µijxjxi) = 0 for the defining relations of A, where αijm ∈ k for all i, j, m, and 1 ≤ m ≤ n(n − 1)/2. (b) By [CV, Lemma 5.1] and its proof, the set of pure tensors in P(W ⊥), that is, {a ⊗ b ∈ P(W ⊥) : a, b ∈ S1}, is in one-to-one correspondence with the zero locus Γ, in P(S1) × P(S1), of W given by Γ = {(a, b) ∈ P(S1) × P(S1) : w(a, b) = 0 for all w ∈ W }. We will now make more precise the connection between points in the zero locus of W and certain quadratic forms. Lemma 1.9. If a, b ∈ S × 1 , then the quadratic form ab ∈ P(cid:0)Pn i=1 Proof . Suppose w((a, b)) = 0. By Remark 1.8(b), w(a ⊗ b) = 0 for all w ∈ W , and so a ⊗ b ∈ W ⊥. Since S is a domain, ab 6= 0 in S, so ab ∈ P(cid:0)Pn argument is reversible, so the converse holds. kqi(cid:1) if and only if (a, b) ∈ Γ. kqi(cid:1), as desired. This i=1 POINT MODULES OVER REGULAR GRADED SKEW CLIFFORD ALGEBRAS 5 2. Point Modules over Graded Skew Clifford Algebras In this section, we prove results that relate point modules over GSCAs to noncommutative quadrics in the sense of Definition 1.6. In particular, we use the notion of µ-rank introduced in [VV] to extend results in [VVW] about graded Clifford algebras (GCAs) to GSCAs, with our main result being Theorem 2.12. Although the overall approach and some of the proofs are influenced by those in [VVW, §1], many of the proofs involve new arguments. In [VV], a notion of µ-rank of a (noncommutative) quadratic form on n generators was defined, where n = 2 or 3. The results in [VV] suggest a notion of µ-rank at most two of a (noncommutative) quadratic form on n generators for any n ∈ N as follows. Definition 2.1. [VV] Let S be as in Definition 1.1, where n is an arbitrary positive integer, and let Q ∈ S2. (a) If Q = 0, we define µ-rank(Q) = 0. (b) If Q = L2 for some L ∈ S × (c) If Q 6= L2 for any L ∈ S × 1 , we define µ-rank(Q) = 1. 1 , we define µ-rank(Q) = 2. Moreover, if M ∈ P(M µ(n, k)) and if µ-rank(τ (M)) ≤ 2, where τ is given in Notation 1.2, then we define µ-rank(M) to be the µ-rank of τ (M). 1 , but Q = L1L2 where L1, L2 ∈ S × Remark 2.2. In contrast to the commutative setting, there exist noncommutative quadratic forms q where 0 6= q = L2 = L1L2, with L, L1, L2 ∈ S1 and L1, L2 linearly independent. For example, let n = 2 = µ12 and q = (z1 + 2z2)2 = (z1 + z2)(z1 + 4z2). We now define a function Φ that will play a role similar to that played by the function φ in [VVW, Section 1]. Definition 2.3. Let a, b ∈ Pn−1, with a = (a1, . . . , an), b = (b1, . . . , bn), where ai, bi ∈ k for all i. We define Φ : Pn−1 × Pn−1 → P(M µ(n, k)) by (a, b) 7→ (aibj + µijajbi) for all i, j = 1, . . . , n. Remark 2.4. With a, b as in Definition 2.3, let q ∈ S2 be the quadratic form q = n Xi=1 aizi! n Xi=1 bizi! ∈ P(S2), so µ-rank(q) ≤ 2. However, using the relations of S, we find q = n Xi=1 aibiz2 i + n Xi,j=1 i<j (aibj + µijajbi) zizj. 6 POINT MODULES OVER REGULAR GRADED SKEW CLIFFORD ALGEBRAS It follows that q = τ (M), where M = (aibj + µijajbi), so M ∈ k×Φ(a, b). Hence, µ- rank(Φ(a, b)) ≤ 2 for all a, b ∈ Pn−1. Proposition 2.5. Im(Φ) = {M ∈ P(M µ(n, k)) : µ-rank(M) ≤ 2}. Proof . By Remark 2.4, Im(Φ) ⊆ {M ∈ P(M µ(n, k)) : µ-rank(M) ≤ 2} = X. Conversely, let M ∈ X and write q = τ (M) ∈ P(S2). Since µ-rank(q) ≤ 2, we have q = ab for some a, b ∈ S × i=1 bizi, with ai, bi ∈ k for all i. By Remark 2.4, it follows that M = Φ((ai), (bj)). i=1 aizi and b =Pn 1 , where a =Pn Remark 2.6. Recall the notation in Remark 1.8, and suppose (a, b) ∈ P(S1) × P(S1). By our for all m, where a = (ai), b = (bj); that is, if and only if the µ-symmetric matrix Φ(a, b) is a assumption in Remark 1.8(a), the point (a, b) ∈ Γ if and only if Pi,j αijm(aibj + µijajbi) = 0 zero of Pi,j αijmXij for all m, where Xij is the ij'th coordinate function on M(n, k). Proposition 2.7. With the assumption in Remark 1.8(a), Im(ΦΓ) =(M ∈ P(cid:0) n Xk=1 kMk(cid:1) : µ-rank(M) ≤ 2) . Proof . Let H = {M ∈ P(cid:0)Pn µ-symmetric of µ-rank at most two, there exists (a, b) ∈ Pn−1 × Pn−1 such that Φ(a, b) = M, by Proposition 2.5. Thus, by Lemma 1.9, (a, b) ∈ Γ, so H ⊆ Im(ΦΓ). kMk(cid:1) : µ-rank(M) ≤ 2} and let M ∈ H. Since M is k=1 For the converse, our argument follows that of [VVW, Proposition 1.5]. Let M = (aij) ∈ i,j=1 αijmaij = 0 k=1 βkMk, where β1, . . . , βn ∈ k are defined as follows. i,j=1 γijkYij, where Yij = xixj + µijxjxi and γijk ∈ k for all i, j, k. For each k = 1, . . . , n, we define βk ∈ k by βk = i,j=1 αijmYij = 0 in A for all m, and, by Definition 1.4(a), k=1 Mkyk. Since the behavior of the Yij is mirrored by the aij, it follows that Im(ΦΓ). So, by Proposition 2.5, µ-rank(M) ≤ 2 and, by Remark 2.6, Pn for all m. We will prove M = Pn By Remark 1.5, for each k ∈ {1, . . . , n}, yk ∈ (A1)2, so yk = Pn Pn i,j=1 γijkaij. By Remark 1.8(a), Pn (Yij) = Pn (Yij)(β1,...,βn) = (aij) = M, since (Yij)(y1,...,yn) = (Yij). Hence, βkMk = n Xk=1 n Xk=1 Mkyk(β1,...,βn) = (Yij)(β1,...,βn) = M, as desired. It follows that M ∈ H and so Im(ΦΓ) ⊆ H. In order to use Φ to help count the point modules over a regular GSCA, we need to determine which (noncommutative) quadratic forms factor uniquely. To do this, we first prove, in Theorem 2.8, that a quadratic form can be factored in at most two distinct ways. POINT MODULES OVER REGULAR GRADED SKEW CLIFFORD ALGEBRAS 7 Theorem 2.8. A quadratic form can be factored in at most two distinct ways up to a nonzero scalar multiple. Proof. Let q ∈ S × assume 2 . If q cannot be factored, then the result is trivially true. Hence, we may q = n Xi=1 βizi! n Xi=1 β ′ izi! , where βi, β ′ suppose that n ≥ 3 and that the result holds for n − 1 generators. i ∈ k for all i. If n = 2, then the result follows from [VV, Lemma 2.2]. Hereafter, Case I. Suppose βiβ ′ i 6= 0 for some i. Without loss of generality, we may assume that i = n and that βn = 1 = β ′ n. Suppose q factors in the following three ways: q = (a + zn)(a′ + zn) = (b + zn)(b′ + zn) = (c + zn)(c′ + zn), where a, a′, b, b′, c, c′ ∈ Pn−1 kzk. Let ¯q denote the image of q in S/hzni; clearly, ¯q = aa′ = bb′ = cc′. The induction hypothesis implies that ¯q factors in at most two distinct ways up to a nonzero scalar multiple. Thus, without loss of generality, we may assume that c = b and c′ = b′. It follows that q factors in at most two distinct ways up to a nonzero scalar multiple. k=1 Case II. Suppose βiβ ′ i = 0 for all i, so q = Pi<j δijzizj where δij ∈ k for all i, j. We may assume, without loss of generality, that there exists k ∈ {1, . . . , n} such that βi = 0 for all i > k and β ′ i = 0 for all i ≤ k. By the induction hypothesis, we may also assume that βi 6= 0 for all i ≤ k and β ′ i 6= 0 for all i > k. i=1 where a, b ∈ Pn−1 If q ∈ hzii for some i, we may assume i = n and so k = n − 1. It follows that q = azn = znb, kzi. If q = znb′, where b′ ∈ S1, then b = b′ since S is a domain; similarly, if q = a′zn. Moreover, the image of q in the domain S/hzni is zero, so if also q = cd, where c, d ∈ S1, then c ∈ kzn or d ∈ kzn, so q factors in at most two distinct ways up to a nonzero scalar multiple. Suppose q /∈ hzii for all i = 1, . . . , n, and let ¯q denote the image of q in S/hzni. By the induction hypothesis, ¯q factors in at most two distinct ways up to a nonzero scalar multiple, izi. i=k+1 β ′ so we may assume ¯q = ab = cd, where c, d ∈Pn−1 Lifting to S, we have i=1 kzi and a =Pk i=1 βizi and b =Pn−1 q = a(b + β ′ nzn) and q = c(d + αzn) or (c + γzn)d, where α, γ ∈ k×, and these are the only ways q can factor in S. Hence, if q factors in three nα−1a, distinct ways in S, then β ′ since S is a domain, and b = β ′ nzn) is a nonzero scalar multiple of c(d + αzn) and γ has a unique solution. Thus, q factors in at most two distinct ways up to a nonzero scalar multiple. nazn = αczn = γznd, since ab = cd. It follows that c = β ′ nα−1d, since S/hzni is a domain, and so a(b + β ′ 8 POINT MODULES OVER REGULAR GRADED SKEW CLIFFORD ALGEBRAS Theorem 2.8 brings us close to our goal of generalizing (most of) [VVW, Theorem 1.7] from the setting of GCAs to the setting of GSCAs. We first require one last technical result. Lemma 2.9. Let ∆µ denote the points (a, b) ∈ Pn−1 × Pn−1 such that (τ ◦ Φ)(a, b) factors uniquely (up to nonzero scalar multiple). The restriction of τ ◦ Φ to (Pn−1 × Pn−1)\∆µ has degree two and is unramified, whereas τ ◦ Φ∆µ is one-to-one. Proof. The result is an immediate consequence of Theorem 2.8 and the definition of ∆µ. Our next result generalizes (most of) [VVW, Theorem 1.7], which we now state for com- parison. Theorem 2.10. [VVW, Theorem 1.7] Let C denote a GCA determined by symmetric matri- ces N1, . . . , Nn ∈ M(n, k) and let Q denote the corresponding quadric system in Pn−1. If Q has no base points, then the number of isomorphism classes of left (respectively, right) point modules over C is equal to 2r2 + r1 ∈ N ∪ {0, ∞}, where rj denotes the number of matrices in P(cid:0)Pn k=1 finite, then r1 ∈ {0, 1}. kNk(cid:1) that have rank j. If the number of left (respectively, right) point modules is Remark 2.11. In the setting of GCAs, if M is a symmetric matrix, then τ (M) is a commu- tative quadratic form where S, in this case, is commutative; thus, if a, b ∈ S × 1 are linearly independent, then we view q = ab = ba as two different ways to factor q in S. It follows that a symmetric matrix M has rank j, where j = 1 or 2, if and only if τ (M) factors in j distinct ways, up to a nonzero scalar multiple. With this in mind, our next result is clearly a generalization of the first part of Theorem 2.10. Theorem 2.12. If the quadric system {q1, . . . , qn} associated to the GSCA, A, is normalizing and base-point free, then the number of isomorphism classes of left (respectively, right) point modules over A is equal to 2f2 + f1 ∈ N ∪ {0, ∞}, where fj denotes the number of matrices M in P(cid:0)Pn k=1 to a nonzero scalar multiple). kMk(cid:1) such that µ-rank(M) ≤ 2 and such that τ (M) factors in j distinct ways (up Proof . Using the notation from Remark 1.8, by [ATV1], the hypotheses on A imply that the set of isomorphism classes of left (respectively, right) point modules over A is in bijection with Γ. Hence, the result follows from Lemma 1.9, Proposition 2.7 and Lemma 2.9. The last part of Theorem 2.10 appears not to extend to the setting of GSCAs. More precisely, the proof of the last part of Theorem 2.10 uses the correspondence between rank and factoring described in Remark 2.11. Given Remark 2.2, the obvious counterpart in the setting of GSCAs is either f1 ∈ {0, 1} or the number of elements of µ-rank one being at POINT MODULES OVER REGULAR GRADED SKEW CLIFFORD ALGEBRAS 9 most one. However, the following two examples demonstrate that both these properties are unsuitable for generalizing the last part of Theorem 2.10 to the setting of GSCAs. Example 2.13. Take n = 4 and let µ12 = µ13 = µ14 = −µ23 = µ24 = µ34 = 1, q1 = z2 4, q2 = z2z3, q3 = (z1 + z2)(z1 + z4), q4 = b2z2 1 − a2z2 2 + z2 3 + 2bz1z3, where a, b ∈ k× and a2 6= b2. Since the quadric system is normalizing and base-point free, the corresponding GSCA, A, is quadratic and regular of global dimension four (by Theorem 1.7), and is the k-algebra on generators x1, . . . , x4 with defining relations: x1x2 + x2x1 = x2 x1x4 + x4x1 = x2 x2x4 + x4x2 = x2 1 − b2x2 3, 1 − b2x2 3, 1 − b2x2 3, x1x3 + x3x1 = 2bx2 3, x3x4 + x4x3 = 0, 2 + a2x2 x2 3 = 0, and has exactly eleven point modules. In this example, A is a GCA, but the algebra S has been chosen to be noncommutative (via the choice of µ23). Here, P(cid:0)P4 elements that factor uniquely, namely k=1 kqk(cid:1) contains three q1, q4 + 2aq2 and q4 − 2aq2. (To see that q4 +2aq2 factors uniquely, we note that the only way it can factor is as q4 +2aq2 = (bz1 + αz2 + z3)(bz1 + βz2 + z3), for some α, β ∈ k, since its image factors uniquely in S/hz2i; solving for α, β yields only one solution: α = a, β = −a. Similarly, for q4 − 2aq2.) Hence, A has a finite number of point modules, yet r1 = 3 > 1. In the previous example, if, instead, one takes µ23 = 1, so that S is now commutative (as in [VVW]), then the quadric system contains only one element of rank one (up to nonzero scalar multiple), which agrees with Theorem 2.10. Example 2.14. For our second example, we consider a GSCA in [CV, §5.3] with n = 4, where q1 = z1z2, q2 = z2 3, µ23 = 1 = −µ34, 1 − z2z4, q3 = z2 (µ14)2 = µ24 = −1, q4 = z2 2 + z2 4 − z2z3, µ13 = −µ14, so the quadric system is normalizing and base-point free. By Theorem 1.7, the corresponding GSCA, A, is quadratic and regular of global dimension four, and is the k-algebra on generators x1, . . . , x4 with defining relations: 10 POINT MODULES OVER REGULAR GRADED SKEW CLIFFORD ALGEBRAS x1x3 = µ14x3x1, x3x4 = x4x3, x2x3 + x3x2 = −x2 4, x1x4 = −µ14x4x1, 4 = x2 x2 2, x2x4 − x4x2 = −x2 1, and has exactly five nonisomorphic point modules, two of which correspond to q1 = z1z2 = z2z1. The other three point modules correspond to two quadratic forms in P(cid:0)P4 have µ-rank one, namely k=1 kqk(cid:1) that q2 = z2 3 and q2 + 4q4 = (z2 − + z4)2 = (−z2 + + z4)2, z3 2 z3 2 where the latter quadratic form clearly factors in two distinct ways. Hence, A has a finite number of point modules even though two distinct elements of P(cid:0)P4 k=1 kqk(cid:1) have µ-rank one. References [AS] M. Artin and W. Schelter, Graded Algebras of Global Dimension 3, Adv. Math. 66 (1987), 171-216. [ATV1] M. Artin, J. Tate and M. Van den Bergh, Some Algebras Associated to Automorphisms of Elliptic Curves, The Grothendieck Festschrift 1, 33-85, Eds. P. Cartier et al., Birkhauser (Boston, 1990). [ATV2] M. Artin, J. Tate and M. Van den Bergh, Modules over Regular Algebras of Dimension 3, Invent. Math. 106 (1991), 335-388. [CV] T. Cassidy and M. Vancliff, Generalizations of Graded Clifford Algebras and of Complete Intersec- tions, J. Lond. Math. Soc. 81 (2010), 91-112. [NVZ] M. Nafari, M. Vancliff and Jun Zhang, Classifying Quadratic Quantum P2s by using Graded Skew Clifford Algebras, J. Algebra 346 No. 1 (2011), 152-164. [SV] D. R. Stephenson and M. Vancliff, Constructing Clifford Quantum P3s with Finitely Many Points, J. Algebra 312 No. 1 (2007), 86-110. [VVW] M. Vancliff, K. Van Rompay and L. Willaert, Some Quantum P3s with Finitely Many Points, Comm. Alg. 26 No. 4 (1998), 1193-1208. [VV] M. Vancliff and P. P. Veerapen, Generalizing the Notion of Rank to Noncommutative Quadratic Forms, preprint, August 2012.
1808.02308	1	1808	2018-08-07T11:34:56	Commutators and Anti-Commutators of Idempotents in Rings	[ "math.RA" ]	We show that a ring $\,R\,$ has two idempotents $\,e,e'\,$ with an invertible commutator $\,ee'-e'e\,$ if and only if $\,R \cong {\mathbb M}_2(S)\,$ for a ring $\,S\,$ in which $\,1\,$ is a sum of two units. In this case, the "anti-commutator" $\,ee'+e'e\,$ is automatically invertible, so we study also the broader class of rings having such an invertible anti-commutator. Simple artinian rings $\,R\,$ (along with other related classes of matrix rings) with one of the above properties are completely determined. In this study, we also arrive at various new criteria for {\it general\} $\,2\times 2\,$ matrix rings. For instance, $R\,$ is such a matrix ring if and only if it has an invertible commutator $\,er-re\,$ where $\,e^2=e$.	math.RA	math	Commutators and Anti-Commutators of Idempotents in Rings Dinesh Khurana and T. Y. Lam Abstract We show that a ring R has two idempotents e, e′ with an invertible commutator ee′ − e′e if and only if R ∼= M2(S) for a ring S in which 1 is a sum of two units. In this case, the "anti-commutator" ee′ + e′e is automatically invertible, so we study also the broader class of rings having such an invertible anti-commutator. Simple artinian rings R (along with other related classes of matrix rings) with one of the above properties are completely determined. In this study, we also arrive at various new criteria for general 2 × 2 matrix rings. For instance, R is such a matrix ring if and only if it has an invertible commutator er − re where e2 = e. §1. Introduction The work in this paper was inspired by an insightful exercise of Kaplansky in his 1968 book [Ka] on rings of operators. Given two idempotents e, e′ in a ring R with the property that their commutator ee′ − e′e is invertible, "Exercise 6" in [Ka: p. 25] asked the readers to show that the idempotents e, e′, 1 − e, 1 − e′ are pairwise isomorphic; in other words, the four principal right ideals generated by them are isomorphic as right R-modules. This exercise was intended for prospective use toward the proofs of Theorem 60 and Theorem 61 on Baer ∗ -rings in [Ka: pp. 91 -- 96]. For his Exercise 6 on p. 25, Kaplansky kindly offered his readers a "Hint". However, rendered with the author's trademark brevity, this "Hint" was itself no less than another substantial exercise. In all fairness, a full solution of Kaplansky's "Exercise 6" proved to be a considerable challenge by any yardstick. That was the situation about fifty years ago, in 1967 -- 68. Nowadays, with so much more known about idempotents and idempotent identities in rings, a rather natural new solution can be given for Kaplansky's "Exercise 6". More remarkably, the theme of this exercise can be further developed so as to give a full-fledged characterization theorem for the rings R that appeared in the exercise. To explain this from a more general point of view, we'll use the notation [x, y] for the commutator xy − yx, and the notation hx, yi for the anti-commutator xy + yx, for any two elements x, y in a ring R. For the main purposes of this paper, it is convenient to introduce the following two ring-theoretic properties, where idem (R) will henceforth denote the set of idempotents in the ring R, and U(R) will denote the group of units of R. Property K : There exist e, e′ ∈ idem (R) such that [ e, e′ ] ∈ U(R). Property K : There exist e, e′ ∈ idem (R) such that h e, e′ i ∈ U(R). 1 The easiest class of rings to deal with in the investigation of these properties is the class of abelian rings; that is, rings in which all idempotents are central. It is easy to see that such a ring R has Property K iff R = 0, while it has Property K iff 2 ∈ U(R). (In fact, in view of the fact that h 1, 1 i = 2, any ring with 2 ∈ U(R) has Property K.) Another easy observation we can make is that, if R → R′ is a (unital) ring homomorphism, then R having one of the two properties above implies that R′ has the same property, whereby any matrix ring Mn(R) will also have the same property. The first two main results in this paper are Theorem A and Theorem B below, where the former (to be proved in §2) explains our choice of the notation K since it has the obvious consequence that the class of rings with Property K is (properly) contained in the class of rings with Property K. Theorem A. For any e, e′ ∈ idem (R), (though not conversely). In particular, if R has Property K, then it has Property K. [ e, e′ ] ∈ U(R) implies that h e, e′ i ∈ U(R) Theorem B. A ring R has Property K iff R ∼= M2(S) for a ring S in which 1 ∈ U(S) + U(S). From our more general perspective, the solution of Kaplansky's Exercise 6 is just a part of the work needed for proving the "only if" part of Theorem B. In §3, this theorem is proved by using judiciously various idempotent identities of Nicholson [Ni], Kato [Kt], and Koliha-Rakocevi´c [KR1, KR2], which we first develop ad hoc in §2. The classical notion of Bott-Duffin invertibility (relative to an idempotent) introduced in [BD] turns out to play a significant role in proving Theorem B, so the basic ingredients of the Bott-Duffin theory are briefly recalled at the beginning of §2 as well. With Theorem A and Theorem B at our disposal, a natural question to ask is when would a matrix ring Mn(T ) (over a given type of base rings T ) have Property K or Property K. For instance, in the case where T is a division ring, this would amount to asking which simple artinian rings would have Property K or Property K. Indeed, taking T to be a local ring or a nonzero commutative ring, we have the following result in §3 on matrix rings Mn(T ) with Property K. Theorem C. Let R = Mn(T ), where T is a local ring or a nonzero commutative ring. Then R has Property K iff either n ∈ {4, 6, 8, . . . }, or n = 2 and 1T ∈ U(T ) + U(T ). (Note that in the case where (T, m) is local ring, the condition 1T ∈ U(T ) + U(T ) would amount to the simpler statement that T /m > 2.) As for Property K, the corresponding result (to be proved in §4) is the following. Theorem D. Let T be a local ring or a commutative ring. Then R = Mn(T ) has Property K iff we are in one of the following cases: (1) n ∈ {4, 6, 8, . . . }. (2) n = 2 and 1T ∈ U(T ) + U(T ). (3) n is odd and 2 ∈ U(T ). 2 As a spin-off of our investigations on invertible commutators of the form [ e, e′ ] where e, e′ ∈ idem (R), we also consider the case where e′ is allowed to be an arbitrary element of R, while e is replaced by an element with some specific property "comparable" to idempotency. Working with these more general commutators (and anti-commutators) and building on the work of Fuchs-Maxson-Pilz in [FMP: (III.2)], we obtain in §5 a number of new criteria for 2 × 2 matrix rings, three of which are summarized as follows. Theorem E. A ring R is a 2 × 2 matrix ring over some other ring iff there is an invertible commutator [ e, ∗ ] where e2 = e, iff there is an invertible commutator [ p, ∗ ] where p2 = 0, iff there is an invertible commutator [ u, ∗ ] where u2 = 1. The terminology and notations introduced so far in this Introduction will be used freely throughout the paper. For any ring R, rad (R) denotes the Jacobson radical of R, and the words "exchange ring" will be used in the sense of Warfield [Wa] and Nicholson [Ni]. By saying that a matrix M ∈ Mn(T ) is diagonalizable, we'll mean that M is similar to a diagonal matrix in Mn(T ). Other standard terminology and conventions in ring theory follow mainly those in [Go], [La1], and [La3]. §2. Idempotent Identities of Kato and Koliha-Rakocevi´c One key ingredient used in this beginning section is the basic notion of Bott-Duffin invertibility introduced in the early paper [BD]. In order to give a relatively self-contained exposition of our results, we'll start by recalling some definitions and facts from [BD]. For any idempotent e in a ring R, an element a ∈ R is said to be Bott-Duffin invertible relative to e if eae ∈ U(eRe). In this case, the inverse of eae in the corner ring eRe is said to be the Bott-Duffin inverse of a relative to e. For the sake of completeness, we state and prove the following classical characterization result for Bott-Duffin invertibility. Theorem 2.1. For a ∈ R and e ∈ idem (R), the following are equivalent: (1) a is Bott-Duffin invertible relative to e. (2) 1 − e + ae ∈ U(R). (3) 1 − e + ea ∈ U(R). (4) 1 − e + eae ∈ U(R). Proof. For any ring R, Jacobson's Lemma (see, e.g. [La2: Exercise 1.6]) states that, for any x, y ∈ R, 1 − xy ∈ U(R) iff 1 − yx ∈ U(R). Since 1 − e + ae = 1 − (1 − ae) e, this Lemma gives (2) ⇔ (4). By left-right symmetry, we have also (3) ⇔ (4). To see that (4) ⇔ (1), let f = 1 − e. With respect to the idempotent e, the element f + eae has a diagonal Peirce decomposition matrix (cid:18)eae 0 eae ∈ U(eR e), which is the defining condition for (1). This shows that (4) ⇔ (1). 0 f(cid:19). Such a matrix is invertible iff Next, we state the following key result from Kato's book [Kt: (I.4.34), (I.4.44)]. Proposition 2.2. (Kato's Identities) For any e, e′ ∈ idem (R), we have (2.3) which amounts to (e − e′)2 + (1 − e − e′)2 = 1, (2.4) (e − e′)2 = (e + e′) (2 − e − e′), 3 where the two factors on the RHS commute. Also, writing r = 1 − e + e′e and s = 1 − e′ + ee′, we have (2.5) rs = sr = (1 + e − e′) (1 − e + e′) = (1 − e − e′)2. Proof. The identity (2.3) is easily verified by a direct expansion of the LHS. By transpos- ing the term (1 −e−e′)2 to the RHS and using the factorization 1 −x2 = (1 −x) (1 + x), we get the identity (2.4). Similarly, by transposing the term (e − e′)2 in (2.3) to the RHS and using the factorization 1 − y2 = (1 + y) (1 − y), we get the last equality in (2.5). The first two equalities in (2.5) are easily verified by direct expansions. Corollary 2.6. Two idempotents e, e′ ∈ R are Bott-Duffin invertible relative to each other iff 1 − e − e′ ∈ U(R). Proof. This follows from (2.5), applied in conjunction with Theorem 2.1. Remark 2.7. In general, e′ being Bott-Duffin invertible relative to e alone does not imply that e is Bott-Duffin invertible relative to e′. For instance, if e′ = 1 6= e, then e′ is Bott-Duffin invertible relative to e, but e is not Bott-Duffin invertible relative to e′. In this example, we have 1 − e − e′ = −e /∈ U(R). In [KR1: Theorem 3.5], it was proved that, for any idempotents e, e′ in any ring, e − e′ ∈ U(R) iff e + e′ ∈ U(R) and 1 − ee′ ∈ U(R). (The hypothesis 2 ∈ U(R) was included in the statement of this theorem, but the result can be seen to be true without such a hypothesis.) Using here the crucial identity (2.4), we'll prove the following closely related result. Theorem 2.8. For any e, e′ ∈ idem (R), we have e − e′ ∈ U(R) iff e + e′ ∈ U(R) and f + f ′ ∈ U(R), where f = 1 − e and f ′ = 1 − e′. In this case, e is similar to 1 − e′. Proof. The "iff" statement follows from the identity (2.4) (since 2 − e − e′ = f + f ′). The last statement is a special case of a result of Nicholson [Ni: Proposition 1.8] on the clean representations of elements in a ring. To make our exposition self-contained, we recall Nicholson's proof (in our special case). If u := e − e′ ∈ U(R), then (1 − e′) u = (1 − e′) (e − e′) = (1 − e′) e = (e − e′) e = ue. Thus, e = u−1(1 − e′) u is similar to 1 − e′. Remark 2.9. In general, e + e′ ∈ U(R) alone need not imply that e − e′ ∈ U(R). For instance, in any nonzero ring R in which 2 ∈ U(R), the idempotents e = e′ = 1 have the property that e + e′ = 2 ∈ U(R), but e − e′ = 0 /∈ U(R). Next we give a systematic derivation for two useful identities on the commutator [ e, e′ ] = ee′ − e′e and the anti-commutator h e, e′ i = ee′ + e′e from [KR2: p. 103] and [KR1: p. 289]. 4 Proposition 2.10. (Koliha-Rakocevi´c Identities) For any e, e′ ∈ idem (R) and f = 1 − e, we have (2.11) (2.12) [ e, e′ ] = (e − e′) (e′ − f ) = (f − e′) (e − e′); h e, e′ i = (e + e′) (e′ − f ) = (e′ − f ) (e + e′). Proof. From (e + e′)2 = e + e′ + h e, e′ i, transposition of the term e + e′ gives the two equalities in (2.12). Similarly, from (e − e′) (e + e′) = e − e′ + [ e, e′ ], transposition of the term e − e′ gives the first equality in (2.11), and the second equality follows by working instead with (e + e′) (e − e′) = e − e′ − [ e, e′ ]. Kaplansky asserted in [Ka: p. 25, Exercise 6] that, under the assumption [ e, e′ ] ∈ U(R), the idempotents e, e′ ∈ R are Bott-Duffin invertible relative to each other. With the help of Proposition 2.10, one can prove the more precise result in (1) below, and by the same token, prove also its complete analogue in (2) for anti-commutators. Theorem 2.13. For any e, e′ ∈ idem (R) and f = 1 − e, the following hold. (1) [ e, e′ ] ∈ U(R) iff e − e′ ∈ U(R) and f − e′ ∈ U(R), iff e − e′ ∈ U(R) and e, e′ are Bott-Duffin invertible relative to each other. (2) h e, e′ i ∈ U(R) iff e + e′ ∈ U(R) and f − e′ ∈ U(R), iff e + e′ ∈ U(R) and e, e′ are Bott-Duffin invertible relative to each other. (3) [ e, e′ ] ∈ U(R) implies that h e, e′ i ∈ U(R). Proof. For (1), the first "iff" statement follows from (2.11); see [KR2: (3.6)]. The second "iff" statement then follows from Corollary 2.6. (2) is proved similarly by using (2.11) and Corollary 2.6 (although the first "iff" statement in (2) was stated in [KR1: Theorem 3.6] with an extra assumption that 2 ∈ U(R)). Finally, (3) follows from (1) and (2), since e − e′ ∈ U(R) ⇒ e + e′ ∈ U(R) according to Theorem 2.8. Recalling from §1 that "Property K" (respectively, "Property K ") on a ring R means the existence of e, e′ ∈ idem (R) such that [ e, e′ ] ∈ U(R) (respectively, h e, e′ i ∈ U(R)), we can draw the following conclusions from Theorem 2.13. Corollary 2.14. If a ring R has Property K, then it has Property K. However, the converse of this statement fails in general. Proof. The first statement here follows from part (3) of Theorem 2.13. The second statement follows from the observation (made in §1) that any ring R with 2 ∈ U(R) has Property K, while an abelian ring R 6= 0 cannot have Property K. For future reference, we record in the following a useful result on the Jacobson radical and an immediate consequence thereof. Proposition 2.15. Let J ⊆ rad (R) be an ideal of R such that idempotents lift modulo J. Then R has Property K iff R/J does. The same statement holds for Property K. 5 Proof. As we have observed in §1, the "only if" part holds without any assumptions on the ideal J. In the case where idempotents lift modulo J ⊆ rad (R), the "if" part follows from the observation that an element s ∈ R is a unit iff s is a unit in R = R/J. . Corollary 2.16. Let J be an ideal of R such that one of the following holds: (1) J is a nil ideal. (2) J ⊆ rad (R) and R is an exchange ring. Then the conclusions of Proposition 2.15 hold for the two rings R and R/J. Proof. (1) If J is a nil ideal, we have J ⊆ rad (R) by [La1: (4.11)]. On the other hand, by [La1: (21.28)], idempotents lift modulo J. Thus, Proposition 2.15 applies. (2) If R is an exchange ring, idempotents can be lifted modulo any ideal in R (according to [Ni]), so again Proposition 2.15 applies. We shall conclude the discussions in this section by using Corollary 2.16 to prove the following two easy results on upper triangular rings. Proposition 2.17. Let S, T be two rings, and let M be any (S, T )-bimodule. Then the formal triangular ring R = (cid:18)S M T(cid:19) has Property K (respectively, Property K ) iff 0 S and T both do. Proof. The "only if" part is clear, since S and T can both be thought of as factor rings of R. Conversely, assume that S and T have Property K. Then S × T also does. As J := (cid:18)0 M 0(cid:19) is a square-zero ideal in R, and R/J ∼= S × T , it follows from Case (1) of Corollary 2.16 that R also has Property K. The case where S and T both have Property K is similar. 0 In the case of n × n upper triangular matrices over a given ring S, the ideal J of strictly upper triangular matrices has the property that J n = 0. Thus, the same method of proof (using Case (1) of Corollary 2.16) yields the following similar result. Proposition 2.18. Let R = Tn(S) be the ring of n × n upper triangular matrices over a ring S. Then R has Property K (respectively, Property K ) iff S does. §3. Rings with Property K To study rings with the Property K, we begin by giving a streamlined proof for the rest of Kaplansky's "Exercise 6" from [Ka: p. 25]; see also Theorem 60 in [Ka: p. 96]. The proof to be given below is in some sense much easier to grasp than that given by Kaplansky since it makes full use of the efficient tools developed in §2. Theorem 3.1. If e, e′ ∈ idem (R) are such that [ e, e′ ] ∈ U(R), then the following hold. (1) The four idempotents e, e′, 1 − e, 1 − e′ are pairwise similar. (2) R = eR ⊕ e′R = R e ⊕ R e′. 6 Proof. (1) Let f = 1−e. By Theorem 2.13(1), we have e−e′ ∈ U(R) and f −e′ ∈ U(R). Therefore, by the last part of Theorem 2.8, e′ is similar to f as well as to e. By symmetry, it follows that e, e′, 1 − e, 1 − e′ are pairwise similar. (2) By symmetry, it suffices to prove that R = eR ⊕ e′R. Since e − e′ ∈ U(R), clearly eR + e′R = R. Also, for any x ∈ eR ∩ e′R, we have x = ex = e′x. Thus, (e − e′) x = 0, and hence x = 0. This proves that R = eR ⊕ e′R. (In conclusion, we note that (2) can also be deduced as a consequence of [KR2: Theorem 3.2].) In general, if two idempotents e, e′ Remark 3.2. in a ring R are such that e, e′, 1 − e, 1 − e′ are pairwise similar, the commutator [ e, e′ ] need not be a unit in R, and e′ need not be Bott-Duffin invertible relative to e. For instance, if R = M2(S) over a nonzero ring S and Eij are the matrix units, then e = E11 and e′ = E22 certainly have the similarity properties mentioned above, but [ e, e′ ] = 0, and ee′e = 0 (so e′ is not Bott-Duffin invertible relative to e). On the other hand, if we have only [ e, r] ∈ U(R) where r ∈ R (or even r ∈ U(R)), it also does not follow that r is Bott-Duffin invertible relative to e, as is shown by the example e = E11 and r = E12 + E21, for which [ e, r] = E12 − E21 ∈ U(R), but again ere = 0. Now we are in a position to prove Theorem B stated in the Introduction. The remarkable thing about this theorem is that a single commutator property on a ring turns out to be enough to determine the structure of the whole ring. Just for the record, we note that it was Kaplansky's "Exercise 6" in [Ka: p. 25] which had given us the impetus to prove the "only if" part of this theorem. Theorem 3.3. A ring R has Property K iff R ∼= M2(S) for a ring S in which 1 ∈ U(S) + U(S). Proof. First assume that 1 = a + b ∈ S where a, b ∈ U(S). In the matrix ring M2(S), we check easily that e = (cid:18)a b [ e, E11 ] = (cid:18)0 −b implication statement in Theorem 2.13(3), the anti-commutator he, E11 i =(cid:18)2 a b a b(cid:19) is an idempotent, and that the commutator 0(cid:19) is a unit. Thus, M2(S) has Property K. (In testimony to the 0(cid:19) is a a indeed a unit too, so the ring M2(S) also has Property K as expected.) Conversely, if a ring R has Property K, fix two idempotents e, e′ ∈ R such that [ e, e′ ] ∈ U(R). By Theorem 3.1, f := 1 − e is similar to e. This implies that f R ∼= eR, and so RR = eR ⊕ f R ∼= eR ⊕ eR. Taking endomorphism rings, we have R ∼= EndR(eR ⊕ eR) ∼= M2(S) where S := EndR(eR) can be identified with the corner ring eR e. By Theorem 2.13(1), ee′e ∈ U(S). Similarly, for f ′ := 1 − e′, the fact that [ e, f ′ ] = −[ e, e′ ] ∈ U(R) implies that ef ′e ∈ U(S) too. Thus, 1S = e = ee′e + ef ′e ∈ U(S) + U(S), as desired. In connection with the condition 1 ∈ U(S) + U(S) appearing in Theorem 3.3, we should point out that it is in fact equivalent to the following ostensibly stronger condition: 7 for any integer n ≥ 1, any unit in S is a sum of n units in S. In the case where S is a unit-regular ring, the condition 1S ∈ U(S) + U(S) can also be self-strengthened into S = U(S) + U(S), according to [GW: Theorem 3.8]. In the following, we will state two interesting consequences of Theorem 3.3. Corollary 3.4. A ring R has Property K with 2 ∈ U(R) iff R ∼= M2(S) for some ring S with 2 ∈ U(S). Proof. The "if" part follows from Theorem 3.3 since 2 ∈ U(S) implies that 1 = 2−1 + 2−1 ∈ U(S) + U(S), as well as 2 ∈ U(cid:0)M2(S)(cid:1). Conversely, if a ring R has Property K along with 2 ∈ U(R), Theorem 3.3 implies that R ∼= M2(S) for some ring S. Clearly, the fact that 2 ∈ U(R) implies that 2 ∈ U(S). A second application of the "if" part of Theorem 3.3 is that it gives a good supply of examples of matrix rings having the Property K, as follows. Corollary 3.5. If R = M2m(T ) (for any ring T ) where m ≥ 2, or R = EndD(V ) for some infinite-dimensional right vector space V over a division ring D, then R has Property K. Proof. First assume that R = M2m(T ) where m ≥ 2. By a result of Henriksen [He], the identity matrix Im is a sum of two units in Mm(T ). Thus, by Theorem 3.3, R ∼= M2(cid:0)Mm(T )(cid:1) has Property K. Finally, assume that R = End (VD) as in the statement of the Corollary. By [La3: Example 1.4], we have R ∼= Mn(R) for every n ≥ 1. Applying this for n = 4 (for instance), we are back to the case treated above. Example 3.6. In general, if T is a ring such that R := M2(T ) has Property K, Theorem 3.3 only says that R ∼= M2(S) for some ring S in which 1 is a sum of two units. Since T may not be isomorphic to S, this does not imply that 1T ∈ U(T )+U(T ). An explicit example to illustrate the possible failure of 1T ∈ U(T ) + U(T ) is as follows. For any field k, we construct after Leavitt [Le] and Cohn [Co1] a k-algebra T with a generic invertible 3 × 2 matrix. This matrix defines a right T -module isomorphism T 2 ∼= T 3, which induces another isomorphism T 2 ∼= T 4. Taking T -endomorphism rings gives a ring isomorphism R := M2(T ) ∼= M4(T ). Thus, it follows from Corollary 3.5 that R has Property K. By a recent result of G. Bergman [Be2] based on his earlier work [Be1], U(T ) = U(k). In particular, if we take k to be the field of two elements, we'll have U(T ) = {1}, in which case clearly 1T /∈ U(T ) + U(T ). Example 3.6 leads naturally to the following Question 3.7. If T is a ring such that M2(T ) has Property K, under what additional assumptions on T can we conclude that 1T ∈ U(T ) + U(T ) ? The difficulty in dealing with this question stems mainly from the fact that 1T ∈ U(T ) + U(T ) is not a Morita invariant property of rings. Nevertheless, there are many classes of rings T for which we can answer Question 3.7 in a satisfactory way. First, recall from [La3: §17C] that a ring T is said to be Mn-unique if, for any ring S, 8 Mn(T ) ∼= Mn(S) implies that T ∼= S. Taking stock in this definition, we do have the following partial positive answer to Question 3.7. Theorem 3.8. Let R = M2(T ) where T satisfies one of the following conditions: (1) T is M2-unique. (2) T is an abelian ring and every idempotent in R is diagonalizable. Then R has Property K iff 1T ∈ U(T ) + U(T ). Proof. The "if" part is true without any assumptions on T by Theorem 3.3. The "only if" part in the two cases (1) and (2) will be handled separately, as follows. Case (1). If R has Property K, Theorem 3.3 implies that R ∼= M2(S) for some ring S with 1S ∈ U(S) + U(S). Given the M2-unique assumption in Case (1), we have T ∼= S, and so 1T ∈ U(T ) + U(T ). Case (2). If R has Property K, fix a commutator [ e, e′ ] ∈ U(R) with e, e′ ∈ idem (R). Given the hypothesis in this case, we may assume (after a conjugation) that e′ = diag (s, t), where s, t ∈ T are necessarily central idempotents. Writing e = (cid:18)a b d(cid:19), we have [ e, e′ ] =(cid:18) 0 0 (cid:19) ∈ U(R), which implies that b, c ∈ U(T ). From the equation e2 = e, we get a2 + b c = a, so a, 1 − a are units in T , with sum 1. c b (t − s) c (s − t) Having proved Theorem 3.8, we will mention in (3.9) -- (3.13) below some of the more important classes of base rings T to which the theorem can be applied. (3.9) The most obvious class of examples of a ring T satisfying the hypothesis (2) of Theorem 3.8 is given by the projective-free rings defined by P. M. Cohn in [Co2, Co3]: a ring T is projective-free if every finitely generated projective right module over T is free of a unique rank. (This notion is known to be left-right symmetric; see [Co2].) Clearly, such a ring T has only trivial idempotents (so it is abelian), and any idempotent in Mn(T ) is similar to a diagonal matrix with 0's and 1's on the diagonal. (3.10) Another example of a ring satisfying the hypothesis (2) in Theorem 3.8 is an abelian (von Neumann) regular ring T . The fact that any idempotent matrix over a regular ring is diagonalizable can be shown in at least two ways: first by using the refine- ment theorem (for finitely generated projective modules) of Goodearl and Handelman in [GH: (3.8)], and second, by using [Go: (2.6)] in conjunction with [SG: Theorem 9]. (3.11) In the formulation of "Open Problem (47)" in Goodearl's book [Go: p. 349], there is an extensive list of different classes of regular rings T that are classically known to be Mn-unique for all n. Thus, Theorem 3.8 is applicable to all such regular rings, under the hypothesis (1). The two best known classes among those mentioned in [Go: p. 349] are right self-injective regular rings [Go: (10.35)], and all regular rings whose primitive factors are artinian [Go: (6.12)]. (The latter class includes all abelian regular rings, which we have already mentioned in (3.10).) But unfortunately, unit-regular rings are not known to be Mn-unique (even for n = 2), so the answer to Question 3.7 has so far remained unknown in the case where the base ring T is unit-regular. 9 (3.12) According to [La3: (17.26)], any semilocal ring T is Mn-unique for all n. Thus, Theorem 3.8 is applicable to T again under the hypothesis (1). In the special case where (T, m) is a local ring, it would be projective-free as well. In this case, as we have noted in the statement of Theorem B in §1, the conclusion 1T ∈ U(T ) + U(T ) in Theorem 3.8 would amount to the simpler statement that T /m > 2. (3.13) Theorem 3.8 is applicable to any commutative ring T (under hypothesis (1)) since T is also known to be Mn-unique (for all n) by [La3: (17.31)]. In this case, the conclusion of Theorem 3.8 is capable of another more concrete derivation, as follows. If e, e′ are idempotents in R = M2(T ) with [ e, e′ ] ∈ U(R), it is easy to show using [KLS: Formula (1.1)] that tr (ee′) and 1 − tr (ee′) are both units in T , with sum 1. Using Theorem 3.3 and Corollary 3.5, it is now easy to determine all matrix rings over local rings and nonzero commutative rings that have Property K. Theorem 3.14. Let T be a local ring or a nonzero commutative ring. Then R := Mn(T ) has Property K iff either n ∈ {4, 6, 8, . . . }, or n = 2 and 1T ∈ U(T ) + U(T ). Proof. The "if" part follows from Theorem 3.3 and Corollary 3.5. For the "only if" part, assume that R = Mn(T ) has Property K. Since T is a local ring or a nonzero commutative ring, there exists an ideal m ⊆ T such that T /m is a division ring. Then Mn(cid:0)T /m(cid:1) ∼= Mn(T )/Mn(m) also has Property K. Applying Theorem 3.3, we have Mn(cid:0)T /m(cid:1) ∼= M2(S) for some ring S with 1S ∈ U(S) + U(S). By the uniqueness part of Wedderburn's theorem, we see that n must be even. Finally, in the special case n = 2, R = M2(T ) having Property K would imply that 1T ∈ U(T ) + U(T ) by the remarks in (3.12) and (3.13). Example 3.15. Let R = Mn(D) be a typical simple artinian ring, where D is a division ring. According to Theorem 3.14, R has Property K except precisely when n is odd or when n = 2 and D = 2. Example 3.16. Let Z(p) denote the localization of Z at the prime ideal (p). If T = Z or Z(2), Mn(T ) has Property K iff n ∈ {4, 6, 8, . . . }. On the other hand, if T = Z(p) where p is an odd prime, then Mn(T ) has Property K iff n is even. Example 3.17. If the ring T in Theorem 3.14 is allowed to be noncommutative and not a local ring, the "only if" part in the theorem may no longer hold. For instance, we may take T to be any nonzero ring with Property K (so T is necessarily noncommutative). Then for any n ≥ 1 (odd or even), R = Mn(T ) also has Property K since T does. §4. Interplay Between Property K and Property K In this section, we turn our attention to rings with Property K. Our goal is to understand more precisely the relationship and interaction between Property K and Property K. We begin by recalling that any ring R with 2 ∈ U(R) has Property K, but not necessarily Property K. On the other hand, a 2 × 2 matrix ring such as R = M2(cid:0)F2(cid:1) does not have Property K (e.g. by Theorem 3.3), and hence also does not have Property K since char (R) = 2. From this example, it follows for instance that if 10 T is any ring with an ideal of index 2, then M2(T ) (with factor ring R ) does not have Property K, and hence also not Property K. Our first main result in this section is Theorem 4.1 below, where part (1) gives some necessary conditions on the invertibility of an anti-commutator h e, e′ i ∈ R, while part (2) offers a somewhat unexpected characterization for the rings with Property K without the use of products of idempotents. Theorem 4.1. (1) If e, e′ ∈ idem (R) are such that h e, e′ i ∈ U(R), then e′ is similar to e, and we have h e, w i ∈ U(R) for some w ∈ U(R). (2) A ring R has Property K iff there exists a unit u ∈ idem (R) + idem (R) such that 1 − u ∈ U(R). Proof. (1) According to Theorem 2.13(2), h e, e′ i ∈ U(R) amounts to two conditions: (4.2) e + e′ ∈ U(R), and 1 − e − e′ ∈ U(R). The latter implies (by Theorem 2.8) that e′ = wew−1 for some w ∈ U(R). Then e + wew−1 = e + e′ ∈ U(R) implies that h e, w i = ew + we = uw ∈ U(R). (2) If R has Property K, there exist e, e′ ∈ idem (R) satisfying (4.2). Adding the two elements in (4.2) shows that 1 is a sum of two units, the first one of which is a sum of two idempotents in R. Conversely, if 1 = u + u′ where u, u′ ∈ U(R) and u = e + e′ for some e, e′ ∈ idem (R), then 1 − e − e′ = 1 − u = u′ ∈ U(R). Thus, (4.2) is satisfied, so R has Property K. Theorem 4.1 for rings with Property K has a complete analogue for rings with Prop- erty K too, which we shall present below. One good reason we have chosen to prove Theorem 4.1 for Property K first is that the proof of its part (2) is easier and more intuitive, while it gives an impetus toward finding an analogous (but somewhat harder) characterization result in the case of Property K. (1) If e, e′ ∈ idem (R) are such that [ e, e′ ] ∈ U(R), then we have Theorem 4.3. [ e, v ] ∈ U(R) for some v ∈ U(R). (2) A ring R has Property K iff there exists a unit v ∈ idem (R) − idem (R) such that 1 ± v ∈ U(R) for both signs. Proof. (1) By Theorem 2.13(1), [ e, e′ ] ∈ U(R) ⇒ v := e − e′ ∈ U(R), and so (4.4) [ e, v ] = [ e, e − e′ ] = −[ e, e′ ] ∈ U(R). The fact that one can "replace" the idempotent e′ by a unit v (while still achieving the property [ e, v ] ∈ U(R)) reminds us somewhat of the theorem of de S´a as discussed in [Sa] and [KL], although our result here applies quite generally to any ring R, as long as e, e′ ∈ idem (R). (Needless to say, the converse of (1) fails in general. For instance, taking R = M2(T ) for any ring T , [e, v] is an invertible commutator for e = E11 ∈ idem (R) and v ∈ U(R). But for S = Z or F2 for instance, we cannot find any invertible [e1, e2] with e1, e2 ∈ idem (R).) 11 (2) First assume R has Property K and fix an invertible commutator [ e, e′ ] with e, e′ ∈ idem (R). We have (1 − e) − e′ ∈ U(R) (by (2.13)(1)), so Theorem 2.8 implies that (1 − e) + e′ ∈ U(R); that is, 1 − v ∈ U(R) for v := e − e′ ∈ U(R) as in the last paragraph. On the other hand, 1 + v = (1 − e′) + e ∈ U(R) too, again by Theorem 2.8 since (1 −e′) −e ∈ U(R). This proves the "only if" part of (2). For the converse, assume that there is a unit v = e − e′ for suitable e, e′ ∈ idem (R) such that 1± v ∈ U(R) for both signs. By Theorem 2.13(1), we will have Property K on R if we can show that 1 − e − e′ ∈ U(R). Appealing to Theorem 2.8 once more (with the idempotent e there replaced by 1 − e), this would follow if (1 − e) + e′ ∈ U(R) and e + (1 − e′) ∈ U(R). The former amounts to 1 − v ∈ U(R), while the latter amounts to 1 + v ∈ U(R). Since both of these conditions were given, we are done. Our next goal is to study the possible presence of Property K on n × n matrix rings over certain types of rings T . The first case we can treat without too much difficulty is when (T, m) is a local ring. Here, a judicious use of Corollary 2.16 enables us to (essentially) "replace" T by the division ring T /m. As we have pointed out before in (3.9), the condition 1T ∈ U(T ) + U(T ) in the statement (2) below can be more simply expressed by T /m > 2. However, we still prefer to use the former condition because of its more general nature. Similarly, in the statement (3) below, we prefer the condition "2 ∈ U(T )" to the equivalent condition "char(cid:0)T /m(cid:1) 6= 2 ". Theorem 4.5. Let (T, m) be a local ring. Then R = Mn(T ) has Property K iff we are in one of the following cases: (1) n ∈ {4, 6, 8, . . . }. (2) n = 2 and 1T ∈ U(T ) + U(T ). (3) n is odd and 2 ∈ U(T ). Proof. If (1) or (2) holds, we know from Corollary 3.5 and Theorem 3.3 respectively that R has Property K, so of course R has Property K. If (3) holds instead, then < In, In >= 2 In ∈ U(R) shows that R has Property K. Conversely, assume in the following that R has Property K. If n ∈ {4, 6, 8, . . . }, then (1) holds, so for the rest of the proof, we need only work with odd n and n = 2. Case A. T is a division ring. Fix an anti-commutator h e, e′ i ∈ U(R) with (necessarily nonzero) e, e′ ∈ idem (R). First assume that n = 2. If e = I2, h e, e′ i = 2 e′ ∈ U(R) implies that 2 ∈ U(T ), in which case 1T = 2−1 + 2−1 ∈ U(T ) + U(T ). If e 6= I2, c we may assume (after a conjugation) that e′ = E11. Writing e = (cid:18)a b h e, e′ i = (cid:18)2a b d(cid:19), we have 0(cid:19) ∈ U(R), so b, c ∈ U(T ). As e2 = e implies that a2 + b c = a, we see that a, 1 − a are in U(T ) too, with sum 1, so (2) holds. Finally, suppose n is odd. Here we may assume that e′ = diag (Ik, 0n−k) where 1 ≤ k ≤ n. If 2 /∈ U(T ), then 2 = 0 ∈ T , and so k < n (for otherwise h e, e′ i = 2 e′ = 0, which is impossible). c Writing e =(cid:18)A B C D(cid:19) with A ∈ Mk(T ) and D ∈ Mn−k(T ), we have h e, e′ i =(cid:18) 0 B C 0(cid:19). As B and C are non-square matrices, we can take a nonzero vector v such that either 12 B v = 0 or C v = 0. Then we'll have either h e, e′ i(cid:18)0 v(cid:19) = 0 or h e, e′ i(cid:18)v contradiction to the fact that h e, e′ i ∈ U(R). This completes the proof that 2 ∈ U(T ) when n is odd. 0(cid:19) = 0, in Case B. (T, m) is a local ring. Let T be the division ring T /m. By [La1: p. 57], rad (Mn(T )) = Mn(rad (T )) = Mn(m). Therefore, (4.6) Since Mn(T ) is an exchange ring (by [Wa] or [Ni]), Corollary 2.16 implies that it has Mn(T )/rad (Mn(T )) = Mn(T )/Mn(m) ∼= Mn(cid:0)T(cid:1). Property K iff Mn(cid:0)T(cid:1) does. Therefore, we are free to replace T by T to assume that T is a division ring, in which case we are fully covered by Case A above. In general, we do not know exactly when a general matrix ring Mn(T ) will have Property K if n is odd or n = 2. Aside from the case where T is local (as treated above in Theorem 4.5), a second manageable case is where T is a commutative ring. Working under this assumption, it turns out that the more substantial case to consider is n = 2. In this case, we were pleasantly surprised to find that Property K and Property K are equivalent on M2(T ) ! To prove this, we start by first working out some easy facts about determinants of 2 × 2 matrices over a commutative ring. Proposition 4.7. For A, B ∈ R = M2(T ) where T is a commutative ring, we have (4.8) det (A + B) + det (A − B) = 2 [ det (A) + det (B) ] ∈ T. Proof. Writing A =(cid:18)p q s(cid:19) and B =(cid:18)w x z(cid:19), a quick computation shows that y r det (A + B) = det (A) + det (B) + (pz + sw − qy − rx), and det (A − B) = det (A) + det (B) − (pz + sw − qy − rx). Adding these two formulas gives the desired equation (4.8). Corollary 4.9. In the notations of Proposition 4.7, if 2 [ det (A) + det (B) ] ∈ rad (T ), then A + B ∈ U(R) iff A − B ∈ U(R). In particular, this "iff " statement holds in case det (A) + det (B) = 0; or more specifically, if det (A) = det (B) = 0. Proof. The first "iff" statement holds on account of (4.8) since a matrix C ∈ R is invertible iff det (C) ∈ U(T ), while for any t, t′ ∈ T , t + t′ ∈ rad (T ) implies that t ∈ U(T ) iff t′ ∈ U(T ). The rest of the Corollary is clear. With the above Corollary providing a crucial link between the invertibility of A + B and A − B, we are now ready to prove the following result. Theorem 4.10. Let R = M2(T ) where T is a commutative ring. Then R has Property K iff it has Property K. Proof. Of course, only the "if" part is at stake. Before beginning its proof, we first point out that the "if" part does not mean that for any e, e′ ∈ R, h e, e′ i ∈ U(R) 13 implies that [ e, e′ ] ∈ U(R). Indeed, in the case where 2 ∈ U(T ), choosing e = e′ = I2 gives h e, e′ i = 2 I2 ∈ U(R), but [ e, e′ ] = 0 /∈ U(R) if T 6= 0. (Nevertheless, under the assumption that 2 ∈ U(T ), R does have Property K by taking a = b = 2−1 in the proof of Theorem 3.1.) In the following, we assume that R has Property K. We'll show that R has Property K in two steps. Step 1. The desired conclusion is true if T is a connected ring; that is, if idem (T ) = {0, 1}. Indeed, taking an invertible h e, e′ i (for suitable e, e′ ∈ idem (R)), det (e) is either 1 or 0. In the former case, e is invertible, so e = I2. Here, h e, e′ i = 2 e′ ∈ U(R) implies that 2 ∈ U(T ), so of course R has Property K. We may thus assume that det (e) = 0, in which case det (ee′) = det (e′e) = 0. As ee′ + e′e ∈ U(R), the last part of Corollary 4.9 implies that ee′ − e′e ∈ U(R), so R has Property K. Step 2. Assume that R does not have Property K. By applying Zorn's Lemma to the family F of ideals Ji ⊆ T such that M2(T /Ji) does not have Property K, we see that F has a maximal member (with respect to inclusion), say J. The maximal choice of J implies that T /J is a connected ring, for otherwise M2(T /J) would have been a direct product of a pair of 2 × 2 matrix rings each of which has Property K, in contradiction to the fact that J ∈ F . On the other hand, M2(T /J) is isomorphic to a factor ring of M2(T ), so it has Property K. As T /J is a connected ring, what we have done in Step 1 implies that M2(T /J) has Property K. This is a contradiction. With the help of Theorem 4.10, we can now decide exactly when a matrix ring Mn(T ) over a commutative ring T will have Property K. It is of interest to note that the conclusions in the following theorem happen to be identical to those of Theorem 4.5 over a (possibly noncommutative) local ring T , although the proofs are rather different in the key case where n = 2. Theorem 4.11. For any commutative ring T , R = Mn(T ) has Property K iff we are in one of the following cases: (1) n ∈ {4, 6, 8, . . . }. (2) n = 2 and 1T ∈ U(T ) + U(T ). (3) n is odd and 2 ∈ U(T ). Proof. If (1) or (2) holds, R would have Property K, and therefore Property K. If (3) holds, of course R has trivially Property K. Conversely, assume R has Property K. If n ∈ {4, 6, 8, . . . }, we are in Case (1). If n = 2, we know from Theorem 4.10 that R has Property K, so Theorem 3.14 implies that 1T ∈ U(T ) + U(T ). Finally, let n be odd. If 2 /∈ U(T ), then 2 lies in some maximal ideal m ⊆ T . Since Mn(cid:0)T /m(cid:1) is isomorphic to a factor ring of R, it also has Property K. As T /m is a field of characteristic 2, Mn(cid:0)T /m(cid:1) would also have Property K, which would contradict Theorem 3.14. Therefore, we must have 2 ∈ U(T ), so we are in Case (3). The result above suggests that the case of M2(T ) for commutative rings T is of special interest in the treatment of Property K (or equivalently, Property K by The- orem 4.10). We'll now conclude this paper by giving some more characterizations for these properties by using commutators and anti-commutators whose second entries are 14 diagonalizable matrices instead of idempotent matrices. Such a replacement is by no means automatic since in general these two classes of matrices are logically independent. Theorem 4.12. For R = M2(T ) where T is a commutative ring, the following state- ments are equivalent: (1) R has Property K (or equivalently, 1T ∈ U(T ) + U(T )). (2) [ e, δ ] ∈ U(R) for some e ∈ idem (R) and some diagonalizable matrix δ ∈ R. (3) h e, δ i ∈ U(R) for some e ∈ idem (R) and some diagonalizable matrix δ ∈ R. In the case where T is an exchange ring, these statements are also equivalent to: (4) T has no ideal of index 2. Proof. First assume (1) holds; say 1 = a + b for some a, b ∈ U(T ). Taking e = (cid:18)a b a b(cid:19) ∈ idem (R) and δ = diag (1, 0), we have [ e, δ ] ∈(cid:18) 0 (cid:18)2 a b −a 0(cid:19) ∈ U(R), and h e, δ i ∈ 0(cid:19) ∈ U(R), so (2) and (3) both hold. Next, (2) ⇒ (1) follows from the calculation in the proof of Case (2) in Theorem 3.8, after a reduction to the case where δ is diagonal. Now we come to the harder implication (3) ⇒ (1). Given e, δ as in (3), we may again a b assume that δ = diag (s, t) for some s, t ∈ T . Letting e = (cid:18)a b d(cid:19), the determinant of c h e, δ i is easily computed to be (4.13) 4 adst − b c (s + t)2 ∈ U(T ). Let g = det (e) = ad − bc, which is an idempotent in T . We have projection maps from T onto gT and onto (1 − g) T . Taken together, these define a natural ring isomorphism from T onto gT × (1 − g) T . Therefore, our job is reduced to showing that 1 is a sum of two units in both gT and (1 − g) T . Under the two projections, g projects to the identity and to zero respectively. Thus, it suffices to argue in the following two cases. Case 1. g = 1. In this case, e = I2, so h e, δ i = 2 δ ∈ U(R) implies that 2 ∈ U(R). Thus, 2 ∈ U(T ), and so 1 = 2−1 + 2−1 ∈ U(T ) + U(T ). Case 2. g = 0. idem (R) implies that a2 + b c = a, so a, 1 − a are units in T , with sum 1. In this case, ad = bc, so (4.13) shows that b c ∈ U(T ). But e ∈ Finally, (1) ⇒ (4) is always true, since the existence of an ideal I ⊆ T of index two would imply that R has a factor ring isomorphic to M2(cid:0)F2(cid:1), which does not have Property K. Conversely, assume that T is an exchange ring satisfying (4). To prove (1), we assume instead that 1T /∈ U(T ) + U(T ). Consider the nonempty family G of ideals Ii ⊆ T for which 1 is not a sum of two units in T /Ii. Applying Zorn's Lemma to the family G, we see that G has a maximal member I (with respect to inclusion). The maximal choice of I implies (as in the proof of (4.10)) that the commutative factor ring T /I is connected. Since T /I remains to be an exchange ring (and T /I 6= 0), it must be a local ring (by [Wa: Proposition 1]). As 1 is not a sum of two units in T /I, the local ring T /I must have residue field F2. This implies that T has a factor ring isomorphic to F2, which contradicts (4). 15 §5. Commutator Characterizations for 2 × 2 Matrix Rings According to Theorem 3.3, rings with Property K are precisely 2 × 2 matrix rings over base rings with a specific unit property. One may naturally ask: how about char- acterizations of the most general 2 × 2 matrix rings? In the literature, there is a simple criterion for these ((A) ⇔ (H) in Theorem 5.1), that is due to Fuchs, Maxson and Pilz [FMP: (III.2)]. To make our exposition more useful and yet completely self-contained, we'll reprove the Fuchs-Maxson-Pilz theorem in a much expanded form below, following the idea of using anti-commutators (as well as commutators) in [La4: p. 349] but not assuming any general matrix ring recognition theorems in [La3: §17]. In the following result, the statements (B) through (G) are expressed in terms of anti-commutators, while (I), (J) and (K) are expressed in terms of commutators. Theorem 5.1. For any ring R, the following statements are equivalent: (A) R ∼= M2(S) for some ring S. (B) There exist p, q ∈ R with p2 = q2 = 0 such that h p, q i = 1. (C) There exist p, q ∈ R with p2 = q2 = 0 such that h p, q i ∈ U(R). (D) There exist v ∈ U(R) and p ∈ R with p2 = 0 such that h p, v i = 1. (E) There exist v ∈ U(R) and p ∈ R with p2 = 0 such that h p, v i ∈ U(R). (F) There exist r, p ∈ R with p2 = 0 such that h p, r i = 1. (G) There exist r, p ∈ R with p2 = 0 such that h p, r i ∈ U(R). (H) There exist p, q ∈ R with p2 = q2 = 0 and p + q ∈ U(R). ( I ) There exist p, q ∈ R with p2 = q2 = 0 such that [ p, q ] ∈ U(R). (J ) There exist v ∈ U(R) and p ∈ R with p2 = 0 such that [ p, v ] ∈ U(R). (K) There exist r, p ∈ R with p2 = 0 such that [ p, r ] ∈ U(R). Proof. (A) ⇒ (B). If (say) R = M2(S), then (B) holds for p = E12 and q = E21. (B) ⇒ (C) is a tautology. (C) ⇒ (H). For p, q as in (C), we have (p + q)2 = h p, q i ∈ U(R), so p + q ∈ U(R). (H) ⇒ (D). For p, q ∈ R as in (H), let v be the inverse of u := p + q ∈ U(R). Left multiplying by q and right multiplying by p give qu = qp and up = qp respectively. Equating these gives vq = pv, so h p, v i = pv + vp = vq + vp = vu = 1. (D) ⇒ (G) is a tautology. (G) ⇒ (F). For r, p as in (G), let w = pr + rp ∈ U(R). Left multiplying by p gives pw = prp, and right multiplying by p gives wp = prp. Thus, pw = wp, so left multiplying by w−1 yields 1 = w−1pr + w−1rp = p (w−1r) + (w−1r) p, which verifies (F). (F) ⇒ (A). For r, p as in (F), left multiplying pr + rp = 1 by e := rp shows that e ∈ idem (R), with complementary idempotent f := pr. By [La1: (21.20)], e, f are isomorphic idempotents. Thus, we have RR = eR ⊕ f R ∼= eR ⊕ eR as right R-modules. Taking endomorphism rings gives R ∼= EndR(eR) ∼= M2(eRe). (D) ⇔ (E). We need only show that (E) ⇒ (D), so let w := pv + vp ∈ U(R) where p2 = 0 and v ∈ U(R). Applying the same proof for (G) ⇒ (F) here gives 1 = p (w−1v) + (w−1v) p. Since w−1v ∈ U(R), this proves (D). 16 (A) ⇒ {(I), (J) and (K)}. If (say) R = M2(S), then (I) holds for p = E12 and q = E21, and (J), (K) both hold for p = E12 and v = r = E12 + E21. (I) ⇒ (C). Given p, q in (I), let w := [ p, q ] ∈ U(R). Repeating the first steps in the proof of (G) ⇒ (F), we can show here that w anti-commutes with both p and q. Thus, (5.2) 1 = w−1pq − w−1qp = −p (w−1q) − (w−1q) p = −h p, w−1q i . Since (w−1q)2 = w−1qw−1q = −w−2q2 = 0, this verifies (C). (J) ⇒ (K) ⇒ (G). It suffices to prove the latter implication. Given w := pr −rp ∈ U(R) as in (K), we have again wp = −pw. Thus, by repeating the calculations in (5.2) (with q replacing r), we get 1 = −h p, w−1r i . Since p2 = 0, this verifies (G). Remark 5.3. It would be tempting to think that in the statements (B), (D) and (F), we can replace the anti-commutators by commutators and get three more equivalent statements in terms of commutators. However, this is not the case! In fact, given (A), we may not be able to get an equation [ p, q ] = 1 with p2 = q2 = 0. Indeed, if R = M2(S) for a commutative ring S, any commutator in R will have zero trace, while the identity matrix I2 has trace 2. Similarly, we must refrain from trying to change the anti-commutator into a commutator in the statements (D) and (F). As for statement (H) (the Fuchs-Maxson-Pilz characterization for (A)), it is easy to see that we also cannot change the requirement p + q ∈ U(R) into p + q = 1 ∈ R (or p + q being a central unit), since this would have implied that p = q = 0, and hence R = 0 ! Remark 5.4. As we have pointed out in the paragraph preceding Theorem 5.1, the proof of that theorem was designed to be entirely self-contained so that it can be read independently of the standard recognition theorems for matrix rings as developed, for instance, in [La3]. In the simplest case, according to [La3: (17.10)], a standard character- ization for a ring R to be a 2×2 matrix ring is that there exist p, r, r′ ∈ R with p2 = 0 such that pr + r′p = 1. This condition can thus be added to all the other equivalent conditions listed in Theorem 5.1. We note that it is ostensibly "weaker" (as a sufficient condition for (A)) than the three conditions (B), (D), and (F). At this point, it would be helpful to recall a key definition introduced in our earlier work [KL] on invertible commutators. In that paper, we defined a ring element a ∈ R to be completable if there exists r ∈ R such that [ a, r ] ∈ U(R). Using this terminology, the equivalence of (A) and (K) in Theorem 5.1 says precisely that R is a 2 × 2 matrix ring (over some other ring) iff R has a square-zero completable element. Inspired by this fact (as well as by our earlier Theorem 3.3), we stumbled upon the idea of looking for completable idempotents in R. It was a pleasant surprise to us that the existence of a completable idempotent in a ring R turns out to be also equivalent to R being a 2 × 2 matrix ring. This fact and some of its variations are collected in the new characterization theorem below for 2 × 2 matrix rings. Recall from Corollary 3.4 that, in the special case where 2 ∈ U(R), we did know that R is a 2 × 2 matrix ring iff R has Property K (that is, [ e, e′ ] ∈ U(R) for some e, e′ ∈ idem (R)). However, the following new equivalence result, like Theorem 5.1, is valid for all rings. Theorem 5.5. For any ring R, the following statements are equivalent: 17 (1) R ∼= M2(S) for some ring S. (2) There exist e ∈ idem (R) and r ∈ R with r2 = 1 such that [ e, r ] ∈ U(R). (3) There exist e ∈ idem (R) and r ∈ R with r2 = −1 such that [ e, r ] ∈ U(R). (4) There exist e ∈ idem (R) and r ∈ R with r3 = 1 such that [ e, r ] ∈ U(R). (5) There exist e ∈ idem (R) and r ∈ U(R) such that [ e, r ] ∈ U(R). (6) R has a completable idempotent. (7) There exist e ∈ idem (R) and r ∈ R with r2 = 0 such that [ e, r ] ∈ U(R). (8) There exist e ∈ idem (R) and a nilpotent r ∈ R such that [ e, r ] ∈ U(R). Proof. We first prove that (1) implies (2), (3), (4) and (7). Indeed, if R = M2(S) for some ring S, we can take e = (cid:18)1 0 and r7 = (cid:18) 1 −1 −1(cid:19). Then e2 = e, r2 easily that [ e, ri ] ∈ U(R) for all i. 1 2 = r3 4 = 1, r2 3 = −1, and r2 7 = 0, and we check 0 0(cid:19), r2 = (cid:18)0 1 1 0(cid:19), r3 = (cid:18) 0 −1 0(cid:19), r4 = (cid:18)−1 1 −1 0(cid:19), 1 (7) ⇒ (8) ⇒ (6) are tautologies. [We note on the side that (7) also trivially implies (K) (and hence (A)) in Theorem 5.1, although this information is not needed here.] (6) ⇒ (1). Let e = e2 ∈ R and r ∈ R be such that the following commutator is a unit: (5.6) [ e, r ] = er − re = er (1 − e) − (1 − e) re. Letting p = er (1−e) and q = −(1−e) re, we have clearly p2 = q2 = 0, and p+q ∈ U(R) according to (5.6). Thus, (H) ⇒ (A) in Theorem 5.1 gives (1). The proof is now complete, since {(2) or (3) or (4)} ⇒ (5) ⇒ (6) are tautologies. It is of interest to point out that, in the various criteria above, r can be chosen to be a unit (respectively, a square root of ±1 or a cubic root of 1), or a nilpotent element (respectively, a square-zero element), but not necessarily an idempotent. We note also that, in the criterion (5) above, the commutator [ e, r ] cannot be replaced by the anti-commutator h e, r i. Indeed, if we take R to be any nonzero commutative ring with 2 ∈ U(R), and choose e = r = 1, then h e, r i ∈ U(R). Here, R cannot possibly be a 2 × 2 matrix ring. Nevertheless, we do have the following result for anti-commutators. Corollary 5.7. h e, r i ∈ U(R), then R/2R ∼= M2(S) for some ring S. If R is a ring with an idempotent e and an element r such that Proof. In R = R/2R, we have er − re = er + re ∈ U(cid:0)R(cid:1), so we can apply (5) ⇒ (1) in Theorem 5.5 to the factor ring R. Emboldened by the fact that 2 × 2 matrix rings are characterized by the existence of completable idempotents as well as the existence of completable square-zero elements, we were led to the consideration of completable involutions as well. (By an involution, we simply mean an element u ∈ R with u2 = 1.) In ring theory, it is rather rare that idempotents, involutions and square-zero elements would play parallel roles in the treatment of a certain problem or property. But for the problem of characterizing 2 × 2 matrix rings, this does turn out to be the case, as the following result shows. 18 Theorem 5.8. A ring R is a 2 × 2 matrix ring iff it has a completable involution. Proof. If R = M2(S) for some ring S, we have already pointed out earlier that the involution E12 + E21 ∈ R is completable. Conversely, assume that there exist u, r ∈ R with u2 = 1 such that v := [ u, r ] ∈ U(R). Repeating the idea (in the proof of (5.1)) of left and right multiplying by u, and keeping in mind the equation u2 = 1, we see here that uv = −vu. Thus, (5.9) 1 = v−1ur − v−1ru = (1 − u) v−1r − v−1r (1 + u); that is, 1 = e + f , where e := (1 − u) v−1r and f := −v−1r (1 + u). Since f e = 0, we have f ∈ idem (R), with complementary idempotent e. It suffices to show that e and f are isomorphic idempotents, as that will show (as in (F) ⇒ (A) in the proof of Theorem 5.1) that R ∼= M2(eRe). Right multiplying (5.9) by 1 − u and left multiplying it by 1 + u give the following two "von Neumann regularity" equations: (1 − u) v−1r (1 − u) = 1 − u, and (1 + u) (−v−1r) (1 + u) = 1 + u. From the former, we have eR = (1 − u) R. From the latter, we have Rf = R (1 + u), which is well known to imply that f R ∼= (1 + u) R as right R-modules. (For a full proof of this implication, see [La2: Exercise 1.17].) Noting that (1 − u) R = (1 − u) vR = v (1 + u) R ∼= (1 + u) R, we conclude that eR ∼= f R, as desired. Remark 5.10. In view Theorem 5.8 (and parts of Theorem 5.5), it might be tempting to surmise that the existence of a completable element u ∈ R with un = 1 for some n ≥ 3 might also imply that R is a 2 × 2 matrix ring (over some other ring). However, this is not the case. For instance, let R be Hamilton's quaternion division algebra generated over R by i, j, with the relations i2 = j2 = −1 and ij = −ji. Let u = a + b i (a, b ∈ R) be a primitive n-th of unity in R [ i ] ∼= C. If n ≥ 3, then b 6= 0, and [u, j] = [a + b i, j] = b [i, j] = 2 b ij ∈ U(R). This shows that u ∈ R is completable. However, being a division algebra, R is not a 2 × 2 matrix ring over any ring. In a similar vein, if a ring R has a completable element w such that wn ∈ {0, w} for some n ≥ 3, R need not be a 2 × 2 matrix ring either. For instance, let R = M3(S) over a ring S. For the two matrices w = 1 1 0 0 −1! and w1 = 1 0 0 0 1 1 0 0 0 1 0 0 0!, we have [w, w1] = 0 −1 −2 −2 −1! ∈ U(R). 0 0 1 1 Thus, w is completable in R, and we can check easily that w3 = w. However, if we choose S to be a division ring (or S = Z), R is not a 2 × 2 matrix ring over any ring. The case where wn = 0 for some n ≥ 3 can be handled similarly by taking instead w = 0 1 0 0 0 0 0 1 0! and w2 = 0 1 1 0 0 0 0 1 0!, with w3 = 0 and [w, w2] = 1 1 1 −1 0 0 −1 0! ∈ U(R). 0 Acknowledgments. We are grateful to Professor S. K. Berberian who kindly commu- nicated to us his detailed personal notes on Kaplansky's "Exercise 6" from [Ka: p. 25]. 19 These notes helped us come to a better understanding of Kaplansky's proposed so- lution of his exercise, which eventually led to the present work. We thank Professor G. M. Bergman for his valuable contribution [Be2] toward the validation of Example 3.6, and we also thank the referee of this paper for his/her many thoughtful comments. References [Be1] G. M. Bergman: Modules over coproducts of rings. Trans. Amer. Math. Soc. 200 (1974), 1 -- 32. [Be2] G. M. Bergman: Email communication, July 3, 2017. [BD] R. Bott and R. J. Duffin: On the algebra of networks. Trans. Amer. Math. Soc. 74 (1953), 99 -- 109. [Co1] P. M. Cohn: Some remarks on the invariant basis property. Topology 5 (1966), 215 -- 228. [Co2] P. M. Cohn: Some remarks on projective-free rings. Algebra Universalis 49 (2003), 159 -- 164. [Co3] P. M. Cohn: Another criterion for a ring to be projective-free. Bull. London Math. Soc. 37 (2005), 857 -- 859. [FMP] P. R. Fuchs, C. J. Maxson and G. F. Pilz: On rings for which homogeneous maps are linear. Proc. Amer. Math. Soc. 112 (1991), 1 -- 7. [Go] K. R. Goodearl: Von Neumann Regular Rings. Second Edition, Robert E. Krieger Publishing Company, Malabar, Florida, 1991. [GH] K. R. Goodearl and D. Handelman: Simple self-injective rings. Comm. Alg. 3 (1975), 797 -- 834. [GW] H. Grover, Z. Wang, D. Khurana, J. Chen and T. Y. Lam: Sums of units in rings. J. Alg. Appl. 13 (2014), 1350072, 10 pp. [He] M. Henriksen: Two classes of rings generated by their units. J. Alg. 31 (1974), 182 -- 193. [Ka] I. Kaplansky: Rings of Operators. W. A. Benjamin, New York, 1968. [Kt] T. Kato: Perturbation Theory for Linear Operators. Second Ed., Classics in Mathematics, Springer-Verlag, Berlin-Heidelberg-New York, 1995. [KL] D. Khurana and T. Y. Lam: Invertible commutators in matrix rings. J. Algebra Appl. 10 (2011), 51 -- 71. [KLS] D. Khurana, T. Y. Lam and N. Shomron: A quantum-trace determinantal for- mula for matrix commutators, and applications. Lin. Alg. Appl. 436 (2012), 2380 -- 2397. 20 [KR1] J. J. Koliha and V. Rakocevi´c: Invertibility of the sum of idempotents. Lin. Mul- tilin. Algebra 50 (2002), 285 -- 292. [KR2] J. J. Koliha and V. Rakocevi´c: Invertibility of the difference of idempotents. Lin. Multilin. Algebra 51 (2003), 97 -- 110. [La1] T. Y. Lam: A First Course in Noncommutative Rings. Second Edition, Graduate Texts in Math., Vol. 131, Springer-Verlag, Berlin-Heidelberg-New York, 2001. [La2] T. Y. Lam: Exercises in Classical Ring Theory. Second Edition, Problem Books in Mathematics, Springer-Verlag, Berlin-Heidelberg-New York, 2003. [La3] T. Y. Lam: Lectures on Modules and Rings. Graduate Texts in Math., Vol. 189, Springer-Verlag, Berlin-Heidelberg-New York, 1999. [La4] T. Y. Lam: Exercises in Modules and Rings. Problem Books in Mathematics, Springer-Verlag, Berlin-Heidelberg-New York, 2007. [Le] W. G. Leavitt: Rings without invariant basis number. Proc. Amer. Math. Soc. 8 (1957), 322 -- 328. [Ni] W. K. Nicholson: Lifting idempotents and exchange rings. Trans. Amer. Math. Soc. 229 (1977), 269 -- 278. [Sa] E. M. de S´a: The rank of a difference of similar matrices. Portugal. Math. 46 (1989), 177 -- 187. [SG] G. Song and X. Guo: Diagonability of idempotent matrices over noncommutative rings. Lin. Multilin. Algebra 297 (1999), 1 -- 7. [Wa] R. B. Warfield: Exchange rings and decompositions of modules. Math. Ann. 199 (1972), 31 -- 36. Footnotes: 2010 AMS Subject Classification: 12E15, 15B33, 15B36, 16E50, 16N40, 16U60. Keywords: Units, idempotents, commutators, anti-commutators, Bott-Duffin inverse, matrix rings, simple artinian rings, exchange rings, commutative rings. Department of Mathematics Panjab University Chandigarh 160 014, India [email protected] Department of Mathematics University of California Berkeley, CA 94720, USA [email protected] 21
1505.06224	2	1505	2015-05-27T12:16:28	Algebras with Medial-like Functional Equations on Quasigroups	[ "math.RA" ]	We consider $14$ medial-like balanced functional equations with four object variables for a pair $(f, g)$ of binary quasigroup operations. Then, we prove that every algebra $(B; f, g)$ with quasigroup operations satisfying a medial-like balanced functional equation has a linear representation on an abelian group $(B; +)$.	math.RA	math	ALGEBRAS WITH MEDIAL-LIKE FUNCTIONAL EQUATIONS ON QUASIGROUPS AMIR EHSANI, ALEKSANDAR KRAPEZ, AND YURI MOVSISYAN Abstract. We consider 14 medial-like balanced functional equations with four object variables for a pair (f, g) of binary quasigroup operations. Then, we prove that every algebra (B; f, g) with quasigroup operations satisfying a medial-like balanced functional equation has a linear representation on an abelian group (B; +). 1. Introduction A binary algebra B is an ordered pair (B; F ), where B is a nonempty set and F is a family of binary operations f : B2 → B. The set B is called the universe (base, underlying set) of the algebra B = (B; F ). If F is finite, say F = {f1, . . . , fk}, we often write (B; f1, . . . , fk) for (B; F ), by [5]. The algebra B is a groupoid if it has only one binary operation. A binary quasigroup is usually defined to be a groupoid (B; f ) such that for any a, b ∈ B there are unique solutions x and y to the following equations: f (a, x) = b and f (y, a) = b, by [13]. If (B; f ) is quasigroup we say that f is a quasigroup operation. A loop is a quasigroup with unit (e) such that f (e, x) = f (x, e) = x. Groups are associative quasigroups, i.e. they satisfy: f (f (x, y), z) = f (x, f (y, z)) and they necessarily contain a unit. A quasigrouup is commutative if (1.1) f (x, y) = f (y, x). Commutative groups are known as abelian groups. A triple (α, β, γ) of bijections from a set B onto a set C is called an isotopy of a groupoid (B; f ) onto a groupoid (C; g) provided γf (x, y) = g(αx, βy) for all x, y ∈ B. (C; g) is then called an isotope of (B; f ), and groupoids (B; f ) and (C; g) are called isotopic to each other. An isotopy of (B; f ) onto (B; f ) is called an autotopy of (B; f ). Let α and β be permutations of B and let ι denote the identity map on B. Then (α, β, ι) is a principal isotopy of a groupoid (B; f ) onto a groupoid (B; g) means that (α, β, ι) is an isotopy of (B; f ) onto (B; g). Isotopy is a generalization of isomorphism. Isotopic image of a quasigroup is again a quasigroup. A loop isotopic to a group is isomorphic to it. Every quasigroup is isotopic to some loop i.e., it is a loop isotope. 2010 Mathematics Subject Classification. 20N05; 39B52; 08A05. Key words and phrases. Medial equation, Paramedial equation, Balanced equation, Beluosov equation, Quasigroup operation, Pair operation, Hyperidentity. 1 2 A. EHSANI, A. KRAPEZ, AND YU. M. MOVSISYAN A binary quasigroup (B; f ) is linear over an abelian group if f (x, y) = ϕx + a + ψy, where (B; +) is an abelian group, ϕ and ψ are automorphisms of (B; +) and a ∈ B is a fixed element. Quasigroup linear over an abelian group is also called a T - quasigroup. Quasigroups are important algebraic (combinatorial, geometric) structures which arise in various areas of mathematics and other disciplines. We mention just a few of their applications: in combinatorics (as latin squares, see [6]), in geometry (as nets/webs, see [3]), in statistics (see [8]), in special theory of relativity (see [16]), in coding theory and cryptography ([14]). 2. Functional equations on quasigroups We use (object) variables x, y, u, v, z and operation symbols (i.e. functional vari- ables) f, g. We assume that all operation symbols represent quasigroup operations. A functional equation is an equality s = t, where s and t are terms with symbols of unknown operations occurring in at least one of them. Definition 2.1. Functional equation s = t is balanced if every (object) variable appears exactly once in s and once in t. Example 2.2. The following are various functional equations: (2.1) (2.2) (2.3) (2.4) (2.5) (2.6) f (f (x, y), z) = f (x, f (y, z)), f (f (x, y), f (u, v)) = f (f (x, u), f (y, v)), f (f (x, y), f (u, v)) = f (f (f (x, u), y), v), f (x, f (y, z)) = f (f (x, y), f (x, z)), f (f (x, y), f (y, z)) = f (x, z), f (x, x) = x. Associativity (Eq.(2.1)), mediality (Eq.(2.2)) and pseudomediality (Eq.(2.3)) are balanced, transitivity (Eq.(2.5)), left distributivity (Eq.(2.4)) and idempotency (Eq.(2.6)) are not. We can define 4! = 24 balanced functional equations with four object variables on a quasigroup (B; f ): (2.7) (2.8) (2.9) (2.10) (2.11) (2.12) (2.13) (2.14) (2.15) (2.16) (2.17) (2.18) (2.19) (2.20) f (f (x, y), f (u, v)) = f (f (x, y), f (u, v)) f (f (x, y), f (u, v)) = f (f (x, y), f (v, u)) f (f (x, y), f (u, v)) = f (f (x, u), f (y, v)) f (f (x, y), f (u, v)) = f (f (x, u), f (v, y)) f (f (x, y), f (u, v)) = f (f (x, v), f (y, u)) f (f (x, y), f (u, v)) = f (f (x, v), f (u, y)) f (f (x, y), f (u, v)) = f (f (y, x), f (u, v)) f (f (x, y), f (u, v)) = f (f (y, x), f (v, u)) f (f (x, y), f (u, v)) = f (f (y, u), f (x, v)) f (f (x, y), f (u, v)) = f (f (y, u), f (v, x)) f (f (x, y), f (u, v)) = f (f (y, v), f (x, u)) f (f (x, y), f (u, v)) = f (f (y, v), f (u, x)) f (f (x, y), f (u, v)) = f (f (u, x), f (y, v)) f (f (x, y), f (u, v)) = f (f (u, x), f (v, y)) ALGEBRAS WITH MEDIAL-LIKE FUNCTIONAL EQUATIONS ON QUASIGROUPS 3 (2.21) (2.22) (2.23) (2.24) (2.25) (2.26) (2.27) (2.28) (2.29) (2.30) f (f (x, y), f (u, v)) = f (f (u, y), f (x, v)) f (f (x, y), f (u, v)) = f (f (u, y), f (v, x)) f (f (x, y), f (u, v)) = f (f (u, v), f (x, y)) f (f (x, y), f (u, v)) = f (f (u, v), f (y, x)) f (f (x, y), f (u, v)) = f (f (v, x), f (y, u)) f (f (x, y), f (u, v)) = f (f (v, x), f (u, y)) f (f (x, y), f (u, v)) = f (f (v, y), f (x, u)) f (f (x, y), f (u, v)) = f (f (v, y), f (u, x)) f (f (x, y), f (u, v)) = f (f (v, u), f (x, y)) f (f (x, y), f (u, v)) = f (f (v, u), f (y, x)) The equation (2.7) is trivial i.e., all quasigroups are solutions of this equation. The equations (2.8), (2.13), (2.14), (2.23), (2.24) and (2.29) are all equivalent to the equation (1.1); solutions are commutative quasigroups. Definition 2.3. The functional equation (2.9) is called medial identity and every quasigroup satisfying medial identity is called medial quasigroup. Theorem 2.4. If (B; f ) is a medial quasigroup then there exists an abelian group (B; +), such that where ϕ, ψ ∈ Aut(B; +), ϕψ = ψϕ and c ∈ Q, by [15]. f (x, y) = ϕ(x) + c + ψ(y), Definition 2.5. The functional equation (2.28) is called paramedial identity and every quasigroup satisfying medial identity is called paramedial quasigroup. Theorem 2.6. If (B; f ) is a paramedial quasigroup then there exists an abelian group (B; +), such that where ϕ, ψ ∈ Aut(B; +), ϕϕ = ψψ and c ∈ Q, by [12]. f (x, y) = ϕ(x) + c + ψ(y), Definition 2.7. A balanced equation s = t is Belousov if for every subterm p of s (t) there is a subterm q of t (s) such that p and q have exactly the same variables. Example 2.8. Functional equations (1.1), (2.30) and the following are Belousov equations: f (x, y) = f (x, y), f (x, f (y, z)) = f (f (z, y), x). The equations (2.1) − (2.3) are non-Belousov. By [9], we have the following results: Theorem 2.9. The solutions of the equation (2.30) belong to the variety of 4- palindromic quasigroups. Theorem 2.10. The equations (2.10) − (2.12), (2.15) − (2.22) and (2.25) − (2.27) are equivalent to commutative (para)mediality; solutions constitute the variety of commutative T -quasigroups (i.e., with ϕ = ψ). 4 A. EHSANI, A. KRAPEZ, AND YU. M. MOVSISYAN 3. Algebras with Medial-like functional equations As a generalization of the functional equations (2.7) − (2.30), let us consider the following balanced functional equations: (3.1) (3.2) (3.3) (3.4) (3.5) (3.6) (3.7) (3.8) (3.9) (3.10) (3.11) (3.12) (3.13) (3.14) (3.15) (3.16) (3.17) (3.18) (3.19) (3.20) (3.21) (3.22) (3.23) (3.24) f (g(x, y), g(u, v)) = g(f (x, y), f (u, v)) f (g(x, y), g(u, v)) = g(f (x, y), f (v, u)) f (g(x, y), g(u, v)) = g(f (x, u), f (y, v)) f (g(x, y), g(u, v)) = g(f (x, u), f (v, y)) f (g(x, y), g(u, v)) = g(f (x, v), f (y, u)) f (g(x, y), g(u, v)) = g(f (x, v), f (u, y)) f (g(x, y), g(u, v)) = g(f (y, x), f (u, v)) f (g(x, y), g(u, v)) = g(f (y, x), f (v, u)) f (g(x, y), g(u, v)) = g(f (y, u), f (x, v)) f (g(x, y), g(u, v)) = g(f (y, u), f (v, x)) f (g(x, y), g(u, v)) = g(f (y, v), f (x, u)) f (g(x, y), g(u, v)) = g(f (y, v), f (u, x)) f (g(x, y), g(u, v)) = g(f (u, x), f (y, v)) f (g(x, y), g(u, v)) = g(f (u, x), f (v, y)) f (g(x, y), g(u, v)) = g(f (u, y), f (x, v)) f (g(x, y), g(u, v)) = g(f (u, y), f (v, x)) f (g(x, y), g(u, v)) = g(f (u, v), f (x, y)) f (g(x, y), g(u, v)) = g(f (u, v), f (y, x)) f (g(x, y), g(u, v)) = g(f (v, x), f (y, u)) f (g(x, y), g(u, v)) = g(f (v, x), f (u, y)) f (g(x, y), g(u, v)) = g(f (v, y), f (x, u)) f (g(x, y), g(u, v)) = g(f (v, y), f (u, x)) f (g(x, y), g(u, v)) = g(f (v, u), f (x, y)) f (g(x, y), g(u, v)) = g(f (v, u), f (y, x)) Definition 3.1. A pair (f, g) of binary operations is called: − medial pair of operations, if the algebra (B; f, g) satisfies the equation (3.3). − paramedial pair of operations, if the algebra (B; f, g) satisfies the equation (3.22). Definition 3.2. A binary algebra B = (B; F ) is called: − medial algebra, if every pair of operations of the algebra B is medial (or, the algebra B satisfying medial hyperidentity). − paramedial algebra, if every pair of operations of the algebra B is paramedial (or, the algebra B satisfying paramedial hyperidentity). The following results are obtained in [11] and [7] respectively. Theorem 3.3. Let the set B, forms a quasigroup under the binary operations f and g. If the pair of binary operations (f, g) is medial, then there exists a binary operation,+, under which B forms an abelian group and for arbitrary elements x, y ∈ B we have: f (x, y) = ϕ1(x) + ψ1(y) + c1, ALGEBRAS WITH MEDIAL-LIKE FUNCTIONAL EQUATIONS ON QUASIGROUPS 5 g(x, y) = ϕ2(x) + ψ2(y) + c2, where c1, c2 are fixed elements of B, and ϕi, ψi ∈ Aut(B; +) for i = 1, 2, such that: ϕ1ψ2 = ψ2ϕ1, ϕ2ψ1 = ψ1ϕ2, ψ1ψ2 = ψ2ψ1 and ϕ1ϕ2 = ϕ2ϕ1. The group (B; +), is unique up to isomorphisms. Theorem 3.4. Let the set B, forms a quasigroup under the binary operations f and g. If the pair of binary operations (f, g) is paramedial, then there exists a binary operation,+, under which B forms an abelian group and for arbitrary elements x, y ∈ B we have: f (x, y) = ϕ1(x) + ψ1(y) + c1, g(x, y) = ϕ2(x) + ψ2(y) + c2, where c1, c2 are fixed elements of B, and ϕi and ψi are automorphisms on the abelian group (B; +) for i = 1, 2, such that: ϕ1ϕ2 = ψ2ψ1, ϕ2ϕ1 = ψ1ψ2, ϕ1ψ2 = ϕ2ψ1 and ψ1ϕ2 = ψ2ϕ1. The group (B; +), is unique up to isomorphisms. Example 3.5. Let B = Z2 ×Z2. We denote the elements of the group B as follows: and the set of all automorphisms of B as follow: {(0; 0), (1; 0), (0; 1), (1; 1)} AutB = {ε, ϕ2, ϕ3, ϕ4, ϕ5, ϕ6}, 0 1 (cid:19), ϕ3 = (cid:18) 1 0 1 1 (cid:19), ϕ4 = (cid:18) 0 1 1 0 (cid:19), ϕ5 = where ε = (cid:18) 1 0 (cid:18) 1 1 0 1 (cid:19), ϕ2 = (cid:18) 1 1 (cid:19). 1 0 (cid:19) and ϕ6 = (cid:18) 0 If ϕ = (cid:18) a b 1 1 1 d (cid:19) and X = (cid:0) x1 x2 (cid:1) then c ϕ(X) = (cid:0) x1 x2 (cid:1)(cid:18) a b c d (cid:19) . Put f1(x, y) = ϕ2(x) + ϕ3(y) + c, f2(x, y) = ϕ3(x) + ϕ2(y) + c, f3(x, y) = ε(x) + ϕ5(y), f4(x, y) = ε(x) + ϕ6(y), where c ∈ B, then the algebra (B; f1, f2) is a paramedial algebra with quasigroup operations which is not medial and (B; f3, f4) is a medial algebra with quasigroup operations which is not paramedial. The balanced functional equations (3.1), (3.2), (3.7), (3.8), (3.17), (3.18), (3.23) and (3.24) are Belousov, while the balanced functional equations (3.3) − (3.6), (3.9) − (3.16) and (3.19) − (3.22) are non-Belousov. Definition 3.6. If (B; ·) is a group, then the bijection, α : B → B, is called a holomorphism of (B; ·) if α(x · y−1 · z) = αx · (αy)−1 · αz, for every x, y, z ∈ B. Note that this concept is equivalent to the concept of quasiautomorphism of groups, by [2]. The set of all holomorphisms of (B; ·) is denoted by Hol(B; ·) and it is a group under the superposition of mappings: (α · β)x = β(αx), for every x ∈ B. The following properties of holomorphisms were proved for Muofang loops in [10]. 6 A. EHSANI, A. KRAPEZ, AND YU. M. MOVSISYAN Lemma 3.7. Let for bijections α1, α2, α3 on the group (B; ·), the following identity is satisfied: Then α1, α2, α3 ∈ Hol(B; ·). α1(x · y) = α2(x) · α3(y). Lemma 3.8. Every holomorphism α of the group (B; ·) has the following form: where, ϕ ∈ Aut(B; ·) and k ∈ B. αx = ϕx · k Theorem 3.9. Let the set B forms a quasigroup under the binary operations f and g. If the pair (f, g) of binary operations satisfies one of the balanced non- Belousov functional equations ((3.4) − (3.6), (3.9) − (3.16) or (3.19) − (3.21)), then there exists a binary operation '+' under which B forms an abelian group and for arbitrary elements x, y ∈ B we have: f (x, y) = ϕ1(x) + ψ1(y) + c1, g(x, y) = ϕ2(x) + ψ2(y) + c2, where c1, c2 are fixed elements of B and ϕi, ψi ∈ Aut(B; +) for i = 1, 2, such that: −: ϕ1ϕ2 = ϕ2ϕ1, ϕ1ψ2 = ψ2ψ1, ψ1ϕ2 = ϕ2ψ1 and ψ1ψ2 = ψ2ϕ1 for the equation (3.4). −: ϕ1ϕ2 = ϕ2ϕ1, ϕ1ψ2 = ψ2ϕ1, ψ1ϕ2 = ψ2ψ1 and ψ1ψ2 = ϕ2ψ1 for the equation (3.5). −: ϕ1ϕ2 = ϕ2ϕ1, ϕ1ψ2 = ψ2ψ1, ψ1ϕ2 = ψ2ϕ1 and ψ1ψ2 = ϕ2ψ1 for the equation (3.6). −: ϕ1ϕ2 = ψ2ϕ1, ϕ1ψ2 = ϕ2ϕ1, ψ1ϕ2 = ϕ2ψ1 and ψ1ψ2 = ψ2ψ1 for the equation (3.9). −: ϕ1ϕ2 = ψ2ψ1, ϕ1ψ2 = ϕ2ϕ1, ψ1ϕ2 = ϕ2ψ1 and ψ1ψ2 = ψ2ϕ1 for the equation (3.10). −: ϕ1ϕ2 = ψ2ϕ1, ϕ1ψ2 = ϕ2ϕ1, ψ1ϕ2 = ψ2ψ1 and ψ1ψ2 = ϕ2ψ1 for the equation (3.11). −: ϕ1ϕ2 = ψ2ψ1, ϕ1ψ2 = ϕ2ϕ1, ψ1ϕ2 = ψ2ϕ1 and ψ1ψ2 = ϕ2ψ1 for the equation (3.12). −: ϕ1ϕ2 = ϕ2ψ1, ϕ1ψ2 = ψ2ϕ1, ψ1ϕ2 = ϕ2ϕ1 and ψ1ψ2 = ψ2ψ1 for the equation (3.13). −: ϕ1ϕ2 = ϕ2ψ1, ϕ1ψ2 = ψ2ψ1, ψ1ϕ2 = ϕ2ϕ1 and ψ1ψ2 = ψ2ϕ1 for the equation (3.14). −: ϕ1ϕ2 = ψ2ϕ1, ϕ1ψ2 = ϕ2ψ1, ψ1ϕ2 = ϕ2ϕ1 and ψ1ψ2 = ψ2ψ1 for the equation (3.15). −: ϕ1ϕ2 = ψ2ψ1, ϕ1ψ2 = ϕ2ψ1, ψ1ϕ2 = ϕ2ϕ1 and ψ1ψ2 = ψ2ϕ1 for the equation (3.16). −: ϕ1ϕ2 = ϕ2ψ1, ϕ1ψ2 = ψ2ϕ1, ψ1ϕ2 = ψ2ψ1 and ψ1ψ2 = ϕ2ϕ1 for the equation (3.19). −: ϕ1ϕ2 = ϕ2ψ1, ϕ1ψ2 = ψ2ψ1, ψ1ϕ2 = ψ2ϕ1 and ψ1ψ2 = ϕ2ϕ1 for the equation (3.20). −: ϕ1ϕ2 = ψ2ϕ1, ϕ1ψ2 = ϕ2ψ1, ψ1ϕ2 = ψ2ψ1 and ψ1ψ2 = ϕ2ϕ1 for the equation (3.21). The group (B; +) is unique up to isomorphisms. Proof. Let (B; f, g) be an algebra with the pair (f, g) of binary quasigroup opera- tions satisfies the non-Belousov functional equation (3.4). We use the main result of [1]. Let the set B forms a quasigroup under six operations Ai(x, y) (for i = 1, . . . , 6). If these operations satisfy the following equation: (3.25) A1(A2(x, y), A3(u, v)) = A4(A5(x, u), A6(y, v)), ALGEBRAS WITH MEDIAL-LIKE FUNCTIONAL EQUATIONS ON QUASIGROUPS 7 for all elements x, y, u and v of the set B then there exists an operation '+' under which B forms an abelian group isotopic to all these six quasigroups. And there exists eight one-to-one mappings; α, β, γ, δ, ǫ, ψ, ϕ and χ of B onto itself such that: A1(x, y) = δx + ϕy, A2(x, y) = δ−1(αx + βy), A3(x, y) = ϕ−1(χx + γy), A4(x, y) = ψx + ǫy, A5(x, y) = ψ−1(αx + χy), A6(x, y) = ǫ−1(βx + γy). Now, let A∗ i (x, y) = Ai(y, x) then, using this equality in (3.25) we have: (3.26) and so, A1(A2(x, y), A3(u, v)) = A4(A5(x, u), A∗ 6(y, v)), A∗ 6(x, y) = A6(y, x) = ǫ−1(βy + γx) = ǫ−1(γx + βy), as (B; +) is an abelian group. So, let A1 = A5 = A∗ 6 = f, A2 = A3 = A4 = g. With these assumptions, from the equation (3.26) we can reach to the equation (3.4). Since A1 = A5, we have: δx + ϕy = ψ−1(χx + αy) ⇒ ψ(δx + ϕy) = χx + αy ⇒ ψ(x + y) = χ(δ−1x) + α(ϕ−1y) ⇒ ψ ∈ Hol(B; +) by, Lemma 3.7. Similarly, since A1 = A∗ Lemma 3.8 there exist ϕ2, ψ2 ∈ Aut(B; +) such that: 6, we have: ǫ ∈ Hol(A; +). Therefore, by ψx = ϕ2x + a, ǫx = b + ψ2x, where a, b are fixed elements in B. Hence, g(x, y) = A4(x, y) = ψx + ǫy = ϕ2x + a + b + ψ2x = ϕ2x + c2 + ψ2x, where c2 = a + b is a fixed element in B. By the same manner, we can show that: δ, ϕ ∈ Hol(B; +), since A2 = A∗ 4. So, there exist ϕ1, ψ1 ∈ Aut(B; +) such that: 4 and A3 = A∗ δx = ϕ1 + d, ϕx = e + ψ1, where d and e are fixed elements in B. Hence, f (x, y) = A1(x, y) = δx + ϕy = ϕ1x + c1 + ψ1y, where c1 = d + e is a fixed element in B. Now, put: f (x, y) = ϕ1(x) + ψ1(y) + c1, g(x, y) = ϕ2(x) + ψ2(y) + c2, in the equation (3.4) then, it is easy to check that ϕ1ϕ2 = ϕ2ϕ1, ϕ1ψ2 = ψ2ψ1, ψ1ϕ2 = ϕ2ψ1 and ψ1ψ2 = ψ2ϕ1. The uniqueness of the group (B; +) follows from the Albert theorem: if two groups are isotopic, then they are isomorphic, by [4]. 8 A. EHSANI, A. KRAPEZ, AND YU. M. MOVSISYAN The proof is similar for the rest of non-Belousov functional equations ((3.5), (cid:3) (3.6), (3.9) − (3.16) and (3.19) − (3.21)). Corollary 3.10. Let (B; F ) be a binary algebra with quasigroup operations which satisfies one of the non-Belousov functional equations ((3.4)−(3.6), (3.9)−(3.16) or (3.19) − (3.21)) then there exists an abelian group (B; +) such that every operation fi ∈ F is represented by the following rule: fi(x, y) = ϕi(x) + ci + ϕi(y), where ci ∈ B and ϕi ∈ Aut(B; +) such that ϕiϕj = ϕjϕi, for every 1 6 i, j 6 F . The group (B; +) is unique up to isomorphisms. Proof. Let (B; F ) satisfies the non-Belousov functional equation (3.4). If f0 ∈ F is a fixed operation then, by Theorem 2.10, f is principally isotopic to the abelian group operation '+' on B. Now, if fi ∈ F is any operation, then the pair of binary operations (f0, fi) satisfies the non-Belousov functional equation (3.4). Hence, f0 and fi are principally isotopic to another abelian group operation '∗' on B. Thus, by transitivity of isotopy, any operation fi is principally isotopic to the same abelian group operation '+'. Hence, according to the proof of previous theorem we have: fi(x, y) = ϕi(x) + ψi(y) + ci, where ci ∈ B and ϕi, ψi ∈ Aut(B; +) such that ϕiϕj = ϕj ϕi, ϕiψj = ψjψi, ψiϕj = ϕjψi and ψiψj = ψjϕi, for every 1 6 i, j 6 F . If i = j then So, ψ2 i = ψiϕi ⇒ ψi = ϕi. fi(x, y) = ϕi(x) + ci + ϕi(y), for 1 6 i 6 F . By putting fj(x, y) = ϕj(x) + cj + ϕj(y), fi(x, y) = ϕi(x) + ci + ϕi(y), in (3.4) we can obtain ϕiϕj = ϕjϕi, for every 1 6 i, j 6 F . The proof is similar for the rest of non-Belousov functional equations (3.5), (3.6), (3.9) − (3.16) and (3.19) − (3.21). (cid:3) The Theorems 3.3 − 3.9 enable us to give a short proof of Theorems 2.4, 2.6 and 2.10: Corollary 3.11. Every quasigroup (B; f ) satisfying one of the balanced non-Belousov functional equations (2.9) − (2.12), (2.15) − (2.22) or (2.25) − (2.28) has a linear representation on an abelian group (B; +): f (x, y) = ϕx + c + ψy, where c ∈ B is fixed element and ϕ and ψ are automorphisms on the abelian group (B; +) such that: −: ϕψ = ψϕ for the equation (2.9). −: ϕ = ψ for the equations (2.10) − (2.12), (2.15) − (2.22) and (2.25) − (2.27). −: ϕ2 = ψ2 for the equation (2.28). Proof. If (B; f ) satisfy the equation (2.9) then by putting f = g and using the Theorem 3.3, the proof is obvious. If (B; f ) satisfy one of the equations (2.10) − (2.12), (2.15) − (2.22) or (2.25) − (2.27), then by putting f = g and using the Theorem 3.9, the proof is obvious. If (B; f ) satisfy the equation (2.28) then by putting f = g and using the Theorem 3.4, the proof is obvious. (cid:3) ALGEBRAS WITH MEDIAL-LIKE FUNCTIONAL EQUATIONS ON QUASIGROUPS 9 The first author acknowledges the Mahshahr Branch, Islamic Azad University for financial assistance of the research project no. 12345. The second author is supported by the Ministry of Education, Science and Technological Development of Serbia through projects ON 174008 and ON 174026. References [1] Acz´el J, Belousov VD, Hossz´u M (1960) Generalized associativity and bisymmetry on quasi- groups. Acta Math. Sci. Hung. 11, 127-136. [2] Belousov VD (1967) Foundations of the theory of quasigroups and loops Nauka, Moscow [in Russian]. [3] Belousov VD (1979) Configurations in algebraic nets Shtiinca, Kishinev, [in Russian]. [4] Bruck RH (1944) Some results in the theory of quasigroups. Trans. American Math. Soc. 55 19-52. [5] Burris S, Sankappanavar HP (1981) A course in Universal Algebra Graduate Texts in Math- ematics, vol. 78, Springer-Verlag, Berlin-Heidelberg-New York. [6] D´enes J, Keedwell AD (1974) Latin squares and their applications Acadmiai Kiad´o, Budapest. [7] Ehsani A, Movsisyan YuM (2013) Linear representation of medial-like algebras. Comm. Al- gebra 41, no. 9 3429-3444. [8] Fisher RA (1966) The design of experiments (8th edition) Oliver & Boyd, Edinburgh. [9] Forg-Rob W, Krapez A (2005) Equations which preserve the height of variables. Aequat. Math. 70 63-76. [10] Movsisyan YuM (1986) Introduction to the theory of algebras with hyperidentities Yerevan State Univ. Press, Yerevan [in Russian]. [11] Nazari E, Movsisyan YuM (2011) Transitive modes. Demonstratio Math. 44, no. 3 511-522. [12] Nemec P, Kepka T (1971) T-quasigroups I. Acta Univ. Carolin. Math. Phys. 12, No. 1 39-49. [13] Pflugfelder HO (1990) Quasigroups and loops: introduction Sigma Series in Pure Mathemat- ics, Heldermann Verlag, Berlin. [14] Shcherbacov VA (2009) Quasigroups in cryptology. Comp Sci J Moldova 17, 2(50) 193 -- 228. [15] Toyoda K (1941) On Axioms of linear functions. Proc. Imp. Acad. Tokyo Conf. 17 211-237. [16] Ungar A (2001) Beyond the Einstein Addition Law and its Gyroscopic Thomas Precession - The Theory of Gyrogroups and Gyrovector Spaces Kluwer Academic Publishers, Dordrecht, Boston, London. Department of Mathematics, Mahshahr Branch, Islamic Azad University, Mahshahr, Iran. E-mail address: [email protected] Mathematical Institute of the Serbian Academy of Sciences and Arts, Knez Mi- hailova 36, 11001 Belgrade, Serbia. E-mail address: [email protected] Department of Mathematics and Mechanics, Yerevan State University, Alex Manoogian 1, Yerevan 0025, Armenia. E-mail address: [email protected]
1307.2540	3	1307	2013-07-31T17:44:38	Unified products for Leibniz algebras. Applications	[ "math.RA", "math.DG" ]	Let $\mathfrak{g}$ be a Leibniz algebra and $E$ a vector space containing $\mathfrak{g}$ as a subspace. All Leibniz algebra structures on $E$ containing $\mathfrak{g}$ as a subalgebra are explicitly described and classified by two non-abelian cohomological type objects: ${\mathcal H}{\mathcal L}^{2}_{\mathfrak{g}} \, (V, \, \mathfrak{g})$ provides the classification up to an isomorphism that stabilizes $\mathfrak{g}$ and ${\mathcal H}{\mathcal L}^{2} \, (V, \, \mathfrak{g})$ will classify all such structures from the view point of the extension problem - here $V$ is a complement of $\mathfrak{g}$ in $E$. A general product, called the unified product, is introduced as a tool for our approach. The crossed (resp. bicrossed) products between two Leibniz algebras are introduced as special cases of the unified product: the first one is responsible for the extension problem while the bicrossed product is responsible for the factorization problem. The description and the classification of all complements of a given extension $\mathfrak{g} \subseteq \mathfrak{E} $ of Leibniz algebras are given as a converse of the factorization problem. They are classified by another cohomological object denoted by ${\mathcal H}{\mathcal A}^{2}(\mathfrak{h}, \mathfrak{g} \, \| \, (\triangleright, \triangleleft, \leftharpoonup, \rightharpoonup))$, where $(\triangleright, \triangleleft, \leftharpoonup, \rightharpoonup)$ is the canonical matched pair associated to a given complement $\mathfrak{h}$. Several examples are worked out in details.	math.RA	math	UNIFIED PRODUCTS FOR LEIBNIZ ALGEBRAS. APPLICATIONS A. L. AGORE AND G. MILITARU Abstract. Let g be a Leibniz algebra and E a vector space containing g as a subspace. All Leibniz algebra structures on E containing g as a subalgebra are explicitly described and classified by two non-abelian cohomological type objects: HL2 g (V, g) provides the classification up to an isomorphism that stabilizes g and HL2 (V, g) will classify all such structures from the view point of the extension problem - here V is a complement of g in E. A general product, called the unified product, is introduced as a tool for our approach. The crossed (resp. bicrossed) products between two Leibniz algebras are introduced as special cases of the unified product: the first one is responsible for the extension problem while the bicrossed product is responsible for the factorization problem. The description and the classification of all complements of a given extension g ⊆ E of Leibniz algebras are given as a converse of the factorization problem. They are classified by another cohomological object denoted by HA2(h, g (⊲, ⊳, ↼, ⇀)), where (⊲, ⊳, ↼, ⇀) is the canonical matched pair associated to a given complement h. Several examples are worked out in details. Introduction Leibniz algebras were introduced by Bloh [8] under the name of D-algebras and redis- covered later on by Loday [20] as non-commutative generalizations of Lie algebras. A systematic study of Leibniz algebras was initiated in [21], [14]. Since then, Leibniz al- gebras generated a lot of interest and became a field of study in its own right. Several classical theorems known in the context of Lie algebras were extended to Leibniz alge- bras, there exists a (co)homology theory for them, the classification of certain types of Leibniz algebras of a given (small) dimension was recently performed, their interaction with vertex operator algebras, the Godbillon-Vey invariants for foliations or differential geometry was highlighted. For more details and motivations we refer to [5], [6], [7], [9], [10], [11], [12], [13], [15], [16], [17], [18], [20], [21], [25], [26], [27] and the references therein. The starting point of this paper is the following question: Extending structures problem. Let g be a Leibniz algebra and E a vector space containing g as a subspace. Describe and classify the set of all Leibniz algebra structures [−, −] that can be defined on E such that g becomes a Leibniz subalgebra of (E, [−, −]). 2010 Mathematics Subject Classification. 17A32, 17B05, 17B56. Key words and phrases. The extension and the factorization problem, non-abelian cohomology for Leibniz algebras, complements. A.L. Agore is research fellow "Aspirant" of FWO-Vlaanderen. This work was supported by a grant of the Romanian National Authority for Scientific Research, CNCS-UEFISCDI, grant no. 88/05.10.2011. 1 2 A. L. AGORE AND G. MILITARU In this paper, we will provide an answer to the above problem as follows: first we will describe explicitly all Leibniz algebra structures on E which contain g as a subalgebra; then we will classify them up to a Leibniz algebra isomorphism ϕ : E → E that stabilizes g, that is ϕ acts as the identity on g. The extending structures (ES) problem was formulated at the level of groups in [3] and for arbitrary categories in [1] where it was studied for quantum groups; recently we approached the problem for Lie algebras [4]. The ES problem is a very difficult one. If g = {0}, then the ES problem asks for the classification of all Leibniz algebra structures on an arbitrary vector space E, which is of course a hopeless problem for vector spaces of large dimension: the classification of all 3-dimensional (resp. 4-dimensional) Leibniz algebras was finished only recently in [13] (resp. [11]). For this reason, from now on we will assume that g 6= {0}. Even though the ES problem is a difficult one, we can still provide detailed answers to it in certain special cases which depend on the choice of the Leibniz algebra g and mainly on the codimension of g in E. It generalizes and unifies the extension problem and the factorization problem. The extension problem asks for the classification of all extensions of h by g and it was first studied in [21] for g abelian; in this case all such extensions are classified by the second cohomology group HL2(h, g) [21, Proposition 1.9]. The fact that g is abelian is essential in proving this classification result. However, a classification result can still be proved in the non-abelian case and the classification object denoted by HL2(h, g) will generalize the second cohomology group HL2(h, g) and it will be explicitly constructed as a special case of the ES problem. The main drawback of this construction is the fact that HL2(h, g) does not arise as a cohomology group of a certain complex, it will be constructed using the theory of crossed products for Leibniz algebras. To conclude, the extension problem appears as a special case of the ES problem as follows: if in the ES problem we replace the condition "g is a Leibniz subalgebra of (E, [−, −])" by a more restrictive one, namely "g is a two sided ideal of E and the quotient E/g is isomorphic to a given Leibniz algebra h", then what we obtain is in fact the extension problem. On the other hand, if in the ES problem we add the additional hypothesis "the comple- ment of g in E is isomorphic to a given Leibniz algebra h" we obtain the factorization problem for Leibniz algebras which can be explicitly formulated as follows: describe and classify all Leibniz algebras Ξ that factorize through two given Leibniz algebras g and h, i.e. Ξ contains g and h as Leibniz subalgebras such that Ξ = g + h and g ∩ h = {0}. Ex- actly as in the case of Lie algebras [23, Theorem 4.1], [22, Theorem 3.9] we will introduce the concept of a matched pair of Leibniz algebras and we will associate to it a bicrossed product which will be responsible for the factorization problem. However, in this case the definition of the concept of a matched pair for Leibniz algebras (Definition 4.5) is a lot more elaborated and difficult then the one for Lie algebras. The last section is de- voted to the converse of the factorization problem. It consists of the following question introduced in [2] in the context of Hopf algebras and Lie algebras: Classifying complements problem. Let g ⊆ E be an extension of Leibniz algebras. If a complement of g in E exists, describe explicitly and classify all complements of g in E, i.e. Leibniz subalgebras h of E such that E = g + h and g ∩ h = {0}. The paper is organized as follows: in Section 2 we introduce the abstract construction it is associated to a Leibniz algebra of the unified product g ⋉ V for Leibniz algebras: UNIFIED PRODUCTS FOR LEIBNIZ ALGEBRAS 3 g, a vector space V and a system of data Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) called an i.e. is a unified product. In this case, Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) will be called a extending datum of g through V . Theorem 2.3 establishes the set of axioms that has to be satisfied by Ω(g, V ) such that g ⋉ V with a given bracket becomes a Leibniz algebra, Leibniz extending structure of g through V . Now let g be a Leibniz algebra, E a vector space containing g as a subspace and V a given complement of g in E. Theorem 2.5 provides the answer to the description part of the ES problem: there exists a Leibniz algebra structure [−, −] on E such that g is a subalgebra of (E, [−, −]) if and only if there exists an isomorphism of Leibniz algebras (E, [−, −]) ∼= g ⋉ V , for some Leibniz extending structure Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) of g through V . The theoretical answer to the classification part of the ES problem is given in Theorem 2.9: we will construct explicitly a relative cohomology 'group', denoted by HL2 g (V, g), which will be the classifying object of all extending structures of the Leibniz algebra g to E - the classification is given up to an isomorphism of Leibniz algebras which stabilizes g. The construction of the second classifying object, denoted by HL2 (V, g) is also given and it parameterizes all extending structures of g to E up to an isomorphism which simultaneously stabilizes g and co-stabilizes V - i.e. this classification is given from the point of view of the extension problem. In Section 3 we give some explicit examples of computing HL2 g (V, g) and HL2 (V, g) in the case of flag extending structures as defined in Definition 3.1. The main result of this section is Theorem 3.6: several special cases of it are discussed and explicit examples are given. Section 4 deals with two main special cases of the unified product. The crossed product of two Leibniz algebras is introduced as a special case of the unified product. Corollary 4.1 shows that any extension of a given Leibniz algebra h by a Leibniz algebra g is equivalent to a crossed product extension and the classifying object HL2(h, g) of all extensions of h by g is constructed in Corollary 4.2 as a generalization of the second Loday-Pirashvili cohomology group HL2(h, g) [21]. The concept of matched pair of Leibniz algebras is introduced in Definition 4.5 generalizing the one for Lie algebras [22, Theorem 3.9], [23, Theorem 4.1]. To any matched pair of Leibniz algebras the bicrossed product is constructed as the tool responsible for the factorization problem (Corollary 4.6). Finally, Section 5 gives the full answer to the classifying complements problem for Leibniz algebras. The main result of the section is Theorem 5.5: if g is a Leibniz subalgebra of E and h is a fixed complement of g in E, then the isomorphism classes of all complements of g in E are parameterized by a certain cohomological object denoted by HA2(h, g (⊲, ⊳, ↼, ⇀)) which is explicitly constructed, where (⊲, ⊳, ↼, ⇀) is the canonical matched pair associated to the factorization E = g+h. The key points in proving this result are Theorem 5.2 and Theorem 5.3. 1. Preliminaries All vector spaces, linear or bilinear maps are over an arbitrary field k. A map f : V → W between two vector spaces is called the trivial map if f (v) = 0, for all v ∈ V . Let g ≤ E be a subspace in a vector space E; a subspace V of E such that E = g + V and V ∩ g = 0 is called a complement of g in E. Such a complement is unique up to an isomorphism and its dimension is called the codimension of g in E. A Leibniz algebra is a vector space 4 A. L. AGORE AND G. MILITARU g, together with a bilinear map [−, −] : g × g → g satisfying the Leibniz identity, that is: [g, [h, l] ] = [ [g, h], l] − [ [g, l], h] (1) for all g, h, l ∈ g. Any Lie algebra is a Leibniz algebra, and a Leibniz algebra satisfying [g, g] = 0, for all g ∈ g is a Lie algebra. The typical example of a Leibniz algebra is the following [20]: let g be a Lie algebra, (M, ⊳) a right g-module and µ : M → g a g-equivariant map, i.e. µ(m ⊳ g) = [µ(m), g], for all m ∈ M and g ∈ g. Then M is a Leibniz algebra with the bracket [m, n](⊳, µ) := m ⊳ µ(n), for all m, n ∈ M . Another important example was constructed in [19]: if g is a Lie algebra, then g ⊗ g is a Leibniz algebra with the bracket given by [x ⊗ y, a ⊗ b] := [x, [a, b]] ⊗ y + x ⊗ [y, [a, b]], for all x, y, a, b ∈ g. For other interesting examples of Leibniz algebras we refer to [21]. Let g be a Leibniz algebra. A subspace I ≤ g is called a two-sided ideal of g if [x, g] ∈ I and [g, x] ∈ I, for all x ∈ I and g ∈ g. g is called perfect if [g, g] = g and abelian if [g, g] = 0. By Z(g) we shall denote the center of g, that is the two-sided ideal consisting of all g ∈ g such that [g, x] = [x, g] = 0, for all x ∈ g. Der(g) stands for the space of all derivations of g, that is, all linear maps ∆ : g → g such that for any g, h ∈ g we have ∆([g, h]) = [∆(g), h] + [g, ∆(h)] A key role in the construction of the different non-abelian cohomological objects arising from the classification part of the ES problem will be played both by the classical space of derivations and by the space of anti-derivations as defined below. Definition 1.1. An anti-derivation of a Leibniz algebra g is a linear map D : g → g such that D([g, h]) = [D(g), h] − [D(h), g] (2) for all g, h ∈ g. We denote by ADer(g) the space of all anti-derivations of g. Example 1.2. For a Lie algebra g we have that ADer(g) = Der(g) but in the case of Leibniz algebras the two spaces are, in general, not equal. The next example illustrates this. Let g be the 3-dimensional Leibniz algebra with the basis {e1, e2, e3} and the bracket defined by: [e1, e3] = e2, [e3, e3] = e1. A straightforward computation shows that the set Der(g) (resp. ADer(g)) coincides with the set of all arrays ∆ (resp. D) of the form: ∆ =   2b1 b2 0 0 3b1 0 b2 b3 b1   (resp. D =   0 0 d1 0 0 d2 0 0 d3   ) (3) for all b1, b2, b3, d1, d2, d3 ∈ k. The following new concept will play an important role in the construction of the two non-abelian cohomological objects introduced in Section 3 and Section 4. Definition 1.3. Let g be a Leibniz algebra. A pointed double derivation of g is a triple (g0, D, ∆), where g0 ∈ g and D, ∆ : g → g are linear maps satisfying the following UNIFIED PRODUCTS FOR LEIBNIZ ALGEBRAS compatibilities for any g, h ∈ g: D(g0) = [g, g0] = [g, D(h) + ∆(h)] = D2(g) + D(cid:0)∆(g)(cid:1)= 0 D2(g) + ∆(cid:0)D(g)(cid:1)= [g0, g] ∆(cid:0)[g, h](cid:1)= [∆(g), h] + [g, ∆(h)] D(cid:0)[g, h](cid:1)= [D(g), h] − [D(h), g] 5 (4) (5) (6) (7) We denote by D(g) the space of all pointed double derivations. The compatibility conditions (6)-(7) show that ∆ (resp. D) is a derivation (resp. an antiderivation) of g, hence D (g) ⊆ g×ADer(g)×Der(g). If g is a Lie algebra then we can easily see that the space D(g) coincides with the set of all pairs (g0, D) ∈ Z(g) × Der(g) such that D(g0) = 0. An example of computing the space D(g) is given in Example 4.4. Let g be a Leibniz algebra, E a vector space such that g is a subspace of E and V a complement of g in E, i.e. V is a subspace of E such that E = g + V and V ∩ g = 0. For a linear map ϕ : E → E we consider the diagram: Id g g i i E π / V ϕ Id / E π / / V (8) where π : E → V is the canonical projection of E = g + V on V and i : g → E is the inclusion map. We say that ϕ : E → E stabilizes g (resp. co-stabilizes V ) if the left square (resp. the right square) of the diagram (8) is commutative. Two Leibniz algebra structures {−, −} and {−, −}′ on E containing g as a Leibniz sub- algebra are called equivalent and we denote this by (E, {−, −}) ≡ (E, {−, −}′), if there exists a Leibniz algebra isomorphism ϕ : (E, {−, −}) → (E, {−, −}′) which stabilizes g. {−, −} and {−, −}′ are called cohomologous, and we denote this by (E, {−, −}) ≈ (E, {−, −}′), if there exists a Leibniz algebra isomorphism ϕ : (E, {−, −}) → (E, {−, −}′) which stabilizes g and co-stabilizes V , i.e. the diagram (8) is commutative. ≡ and ≈ are both equivalence relations on the set of all Leibniz algebras structures on E containing g as subalgebra and we denote by ExtdL (E, g) (resp. ExtdL′ (E, g)) the set of all equivalence classes via ≡ (resp. ≈). ExtdL (E, g) is the classifying object of the ES problem: by explicitly computing ExtdL (E, g) we obtain a parametrization of the set of all isomorphism classes of Leibniz algebra structures on E that stabilize g. ExtdL′ (E, g) gives a classification of the ES problem from the point of view of the extension problem. Any two cohomologous brackets on E are of course equivalent, hence there exists a canonical projection: ExtdL′ (E, g) ։ ExtdL (E, g) For two sets X and Y we shall denote by X ⊔ Y the coproduct in the category of sets of X and Y , i.e. X ⊔ Y is the disjoint union of X and Y . / / / / 6 A. L. AGORE AND G. MILITARU 2. Unified products for Leibniz algebras In this section we shall give the theoretical answer to the ES-problem by constructing two cohomological type objects which will parameterize ExtdL (E, g) and ExtdL′ (E, g). First we introduce the following: Definition 2.1. Let g be a Leibniz algebra and V a vector space. An extending datum of g through V is a system Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) consisting of six bilinear maps ⊳ : V × g → V, ⊲ : V × g → g, ↼ : g × V → g, ⇀ : g × V → V f : V × V → g, {−, −} : V × V → V Let Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) be an extending datum. We denote by g ⋉Ω(g,V ) V = g ⋉ V the vector space g × V together with the bilinear map [−, −] : (g × V ) × (g × V ) → g × V defined by: [(g, x), (h, y)] := (cid:0)[g, h] + x ⊲ h + g ↼ y + f (x, y), {x, y} + x ⊳ h + g ⇀ y(cid:1) (9) for all g, h ∈ g and x, y ∈ V . The object g ⋉ V is called the unified product of g and Ω(g, V ) if it is a Leibniz algebra with the bracket given by (9). In this case the extending datum Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) is called a Leibniz extending structure of g through V . The maps ⊳, ⊲, ⇀ and ↼ are called the actions of Ω(g, V ) and f is called the cocycle of Ω(g, V ). Example 2.2. The unified product is a very general construction: in particular, the hemisemidirect product introduced in differential geometry [18, Example 2.2] is a special case of it. Let g be a Lie algebra and (V, ⊳) be a right g-module. Then g × V is a Leibniz algebra with the bracket [(g, x), (h, y)] := (cid:0)[g, h], x ⊳ h(cid:1), for all g, h ∈ g, x, y ∈ V called the hemisemidirect product of g and V . This Leibniz algebra is not a Lie algebra if g acts nontrivially on V . The hemisemidirect product is a special case of the unified product if we let ⊲, ↼, ⇀, f and {−, −} to be the trivial maps and g to be a Lie algebra. Let Ω(g, V ) be an extending datum of g through V . The bracket defined by (9) has a rather complicated formula; however, for some specific elements we obtain easier forms which will be very useful for future computations. More precisely, the following relations hold in g ⋉ V for any g, h ∈ g, x, y ∈ V : [(g, 0), (h, 0)] = (cid:0)[g, h] , 0(cid:1), [(0, x), (h, 0)] = (cid:0)x ⊲ h, x ⊳ h(cid:1), [(g, 0), (0, y)] = (cid:0)g ↼ y, g ⇀ y(cid:1) [(0, x), (0, y)] = (cid:0)f (x, y), {x, y}(cid:1) (10) (11) The next theorem provides the set of axioms that need to be fulfilled by an extending datum Ω(g, V ) such that g ⋉ V is a unified product. Theorem 2.3. Let g be a Leibniz algebra, V a vector space and Ω(g, V ) an extending datum of g by V . Then g ⋉V is a unified product if and only if the following compatibility conditions hold for any g, h ∈ g, x, y, z ∈ V : (L1) (V, ⊳) is a right g-module, i.e. x ⊳ [g, h] = (x ⊳ g) ⊳ h − (x ⊳ h) ⊳ g (L2) x ⊲ [g, h] = [x ⊲ g, h] − [x ⊲ h, g] + (x ⊳ g) ⊲ h − (x ⊳ h) ⊲ g (L3) [g, h] ⇀ x = g ⇀ (h ⇀ x) + (g ⇀ x) ⊳ h UNIFIED PRODUCTS FOR LEIBNIZ ALGEBRAS 7 (L4) [g, h] ↼ x = [g, h ↼ x] + [g ↼ x, h] + g ↼ (h ⇀ x) + (g ⇀ x) ⊲ h (L5) x⊲f (y, z) = f (x, y) ↼ z −f (x, z) ↼ y+f ({x, y}, z)−f ({x, z}, y)−f (x, {y, z}) (L6) x ⊳ f (y, z) = f (x, y) ⇀ z − f (x, z) ⇀ y + {{x, y}, z} − {{x, z}, y} − {x, {y, z}} (L7) {x, y} ⊲ g = x ⊲ (y ⊲ g) + (x ⊲ g) ↼ y + f (x, y ⊳ g) + f (x ⊳ g, y) − [f (x, y), g] (L8) {x, y} ⊳ g = x ⊳ (y ⊲ g) + (x ⊲ g) ⇀ y + {x, y ⊳ g} + {x ⊳ g, y} (L9) g ⇀ {x, y} = (g ↼ x) ⇀ y − (g ↼ y) ⇀ x + {g ⇀ x, y} − {g ⇀ y, x} (L10) g ↼ {x, y} = (g ↼ x) ↼ y − (g ↼ y) ↼ x + f (g ⇀ x, y) − f (g ⇀ y, x) − [g, f (x, y)] (L11) [g, h ↼ x] + [g, x ⊲ h] + g ↼ (h ⇀ x) + g ↼ (x ⊳ h) = 0 (L12) x ⊲ (y ⊲ g) + x ⊲ (g ↼ y) + f (x, y ⊳ g) + f (x, g ⇀ y) = 0 (L13) x ⊳ (y ⊲ g) + x ⊳ (g ↼ y) + {x, y ⊳ g} + {x, g ⇀ y} = 0 (L14) g ⇀ (h ⇀ x) + g ⇀ (x ⊳ h) = 0 Proof. The proof relies on a detailed analysis of the Leibniz identity for the bracket given by (9), similar to the one provided in the proof of [4, Theorem 2.2] corresponding to the Lie algebra case. There are, however, two significant differences which require some extra care, namely: the bracket on the Leibniz algebra g is not anti-symmetric and the Leibniz identity is not invariant under circular permutations. As the computations are rather long but straightforward we will only indicate the essential steps of the proof, the details being left to the reader. To start with, we note that g ⋉ V is a Leibniz algebra if and only if Leibniz's identity holds, i.e.: (cid:2)(g, x), [(h, y), (l, z)](cid:3)= (cid:2) [(g, x), (h, y)], (l, z)(cid:3)−(cid:2) [(g, x), (l, z)], (h, y)(cid:3) (12) for all g, h, l ∈ g and x, y, z ∈ V . Since in g ⋉ V we have (g, x) = (g, 0) + (0, x) it follows that (12) holds if and only if it holds for all generators of g ⋉ V , i.e. the set {(g, 0) g ∈ g} ∪ {(0, x) x ∈ V }. Hence we are left to deal with eight cases which are necessary and sufficient for testing the compatibility condition (12). First, we should notice that (12) holds for the triple (g, 0), (h, 0), (l, 0), since in g ⋉ V we have that [(g, 0), (h, 0)] = ([g, h], 0). Now, taking into account (10), we obtain that (12) holds for (g, 0), (h, 0), (0, x) if and only if (cid:0)[g, h] ↼ x − [g ↼ x, h] − (g ⇀ x) ⊲ h, [g, h] ⇀ x − (g ⇀ x) ⊳ h(cid:1)= (cid:0)[g, h ↼ x] + g ↼ (h ⇀ x), g ⇀ (h ⇀ x)(cid:1) i.e. if and only if (L3) and (L4) hold. A similar computation proves that the compatibility condition (12) is fulfilled for the triple (g, 0), (0, x), (h, 0) if and only if: [g, h] ↼ x = [g ↼ x, h] − [g, x ⊲ h] + (g ⇀ x) ⊲ h − g ↼ (x ⊳ h) [g, h] ⇀ x = (g ⇀ x) ⊳ h − g ⇀ (x ⊳ h) Taking into account that we are looking for a minimal and independent set of axioms we obtain, assuming that (L3) and (L4) hold, that (12) is fulfilled for the triple (g, 0), (0, x), (h, 0) if and only if (L11) and (L14) hold. Next, it is straightforward to see that (12) holds for (g, 0), (0, x), (0, y) if and only if (L9) and (L10) hold. In a similar manner, one can show that (12) holds for (0, x), (0, y), (0, z) if and only if axiom (L5) and (L6) are fulfilled. We are left with three more cases to study. First, observe that (12) holds 8 A. L. AGORE AND G. MILITARU for (0, x), (g, 0), (h, 0) if and only if (cid:0)x⊲[g, h], x⊳[g, h](cid:1)= (cid:0)[x⊲g, h]−[x⊲h, g]+(x⊳g)⊲h−(x⊳h)⊲g, (x⊳g)⊳h−(x⊳h)⊳g(cid:1) which is equivalent to the fact that axioms (L1) and (L2) hold. Analogously, we can show that (12) holds for (0, x), (0, y), (g, 0) if and only if (L7) and (L8) hold. Finally, it is straightforward to see that (12) holds for (0, x), (g, 0), (0, y) if and only if the following two compatibilities are fulfilled: {x, y} ⊲ g = (x ⊲ g) ↼ y + f (x ⊳ g, y) − [f (x, y), g] − x ⊲ (g ↼ y) − f (x, g ⇀ y) {x, y} ⊳ g = {x ⊳ g, y} + (x ⊲ g) ⇀ y − x ⊳ (g ↼ y) − {x, g ⇀ y} Since we are looking for a minimal set of axioms we obtain, assuming that (L7) and (L8) hold, that (12) is fulfilled for the triple (0, x), (g, 0), (0, y) if and only if (L12) and (L13) hold. The proof is now finished. (cid:3) From now on, a Leibniz extending structure of g through V will be viewed as a system Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) satisfying the compatibility conditions (L1) − (L14). We denote by LZ(g, V ) the set of all Leibniz extending structures of g through V . Example 2.4. We provide the first example of a Leibniz extending structure and the corresponding unified product. More examples will be given in Section 3 and Section 4. Let Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) be an extending datum of a Leibniz algebra g g ⇀ x = g ↼ x = 0, for all x ∈ V and g ∈ g. Then, Ω(g, V ) = (cid:0)f, {−, −}(cid:1) is a Leibniz through a vector space V such that its actions are all the trivial maps, i.e. x ⊳ g = x ⊲ g = extending structure of g through V if and only if (V, {−, −}) is a Leibniz algebra and f : V × V → g is an abelian 2-cocycle, that is [g, f (x, y)] = [f (x, y), g] = 0, f(cid:0)x, {y, z}(cid:1)−f(cid:0){x, y}, z(cid:1)+f(cid:0){x, z}, y(cid:1)= 0 (13) for all g ∈ g, x, y and z ∈ V . The first part of (13) shows that the image of f is contained in the center of g, while the second part is a 2-cocycle condition (see [21, Section 1.7]) which follows from the axiom (L5). In this case, the associated unified Ω(g,V ) V will be called the twisted product of the Leibniz algebras g and V . product g ⋉ Let Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, ·(cid:1) ∈ LZ(g, V ) be a Leibniz extending structure and g ⋉ V the associated unified product. Then the canonical inclusion ig : g → g ⋉ V, ig(g) = (g, 0) is an injective Leibniz algebra map. Therefore, we can see g as a subalgebra of g ⋉ V through the identification g ∼= ig(g) ∼= g × {0}. Conversely, we have the following result which provides an answer to the description part of the ES problem: Theorem 2.5. Let g be a Leibniz algebra, E a vector space containing g as a subspace and [−, −] a Leibniz algebra structure on E such that g is a Leibniz subalgebra in (E, [−, −]). Then there exists a Leibniz extending structure Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, [−, −](cid:1) of g through a subspace V of E and an isomorphism of Leibniz algebras (E, [−, −]) ∼= g ⋉ V that stabilizes g and co-stabilizes V . UNIFIED PRODUCTS FOR LEIBNIZ ALGEBRAS 9 by the following formulas: Proof. Since k is a field, there exists a linear map p : E → g such that p(g) = g, for all g ∈ g. Then V := Ker(p) is a complement of g in E. We define the extending datum Ω(g, V ) = (cid:0)⊳ = ⊳p, ⊲ = ⊲p, ↼=↼p, ⇀=⇀p, f = fp, [−, −] = [−, −]p(cid:1) of g through V x ⊳ g = [x, g] − p(cid:0)[x, g](cid:1) g ⇀ x = [g, x] − p(cid:0)[g, x](cid:1) { , } : V × V → V, {x, y} = [x, y] − p(cid:0)[x, y](cid:1) x ⊲ g = p(cid:0)[x, g](cid:1), g ↼ x = p(cid:0)[g, x](cid:1), ⇀ : g × V → V, f : V × V → g, f (x, y) = p(cid:0)[x, y](cid:1), ⊳ : V × g → V, ⊲ : V × g → g, ↼ : g × V → g, for all g ∈ g, x, y ∈ V . Now, the map ϕ : g × V → E, ϕ(g, x) := g + x, is a linear isomorphism between the direct product of vector spaces g × V and the Leibniz algebra (E, [−, −]) with the inverse given by ϕ−1(y) := (cid:0)p(y), y − p(y)(cid:1), for all y ∈ E. Hence, there exists a unique Leibniz algebra structure on g × V such that ϕ is an isomorphism of Leibniz algebras and this unique bracket is given by: [(g, x), (h, y)] := ϕ−1(cid:0)[ϕ(g, x), ϕ(h, y)](cid:1) for all g, h ∈ g and x, y ∈ V . Using Theorem 2.3, the proof will be finished if we prove that this bracket is the one defined by (9) associated to the system (cid:0)⊳p, ⊲p, ↼p , ⇀p, fp, {−, −}p(cid:1) constructed above. Indeed, for any g, h ∈ g, x, y ∈ V we have: [(g, x), (h, y)] = ϕ−1(cid:0)[ϕ(g, x), ϕ(h, y)](cid:1) = ϕ−1(cid:0)[g + x, h + y](cid:1) = ϕ−1(cid:0)[g, h] + [g, y] + [x, h] + [x, y](cid:1) = (cid:16)[g, h] + p([g, y]) + p([x, h]) + p([x, y]), [g, y] + [x, h] + [x, y] − p([g, y]) − p([x, h]) − p([x, y])(cid:17) = (cid:0)[g, h] + x ⊲ h + g ↼ y + f (x, y), {x, y} + x ⊳ h + g ⇀ y(cid:1) as needed. Thus, ϕ : g ⋉ V → E is an isomorphism of Leibniz algebras and the following diagram is commutative Id g g i / g ⋉ V ϕ i / E q π V / V Id where π : E → V is the projection of E = A+V on the vector space V and q : A⋉V → V , q(g, x) := x is the canonical projection. (cid:3) Based on Theorem 2.5, the classification of all Leibniz algebra structures on E that contain g as a Leibniz subalgebra reduces to the classification of all unified products g ⋉ V , associated to all Leibniz extending structures Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1), for a given complement V of g in E. In order to construct the cohomological objects HL2 g (V, g) and HL2 (V, g) which will parameterize the classifying sets ExtdL (E, g) and respectively ExtdL′ (E, g) we need the following technical result: / / / / / 10 A. L. AGORE AND G. MILITARU Lemma 2.6. Let Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) and Ω′(g, V ) = (cid:0)⊳′, ⊲′, ↼′, ⇀′ , f ′, {−, −}′(cid:1) be two Leibniz extending structures of g trough V and g ⋉ V , g ⋉′ V the as- sociated unified products. Then there exists a bijection between the set of all morphisms of Leibniz algebras ψ : g ⋉ V → g ⋉′ V which stabilizes g and the set of pairs (r, v), where r : V → g, v : V → V are two linear maps satisfying the following compatibility conditions for any g ∈ g, x, y ∈ V : (ML1) v(g ⇀ x) = g ⇀′ v(x); (ML2) v(x ⊳ g) = v(x) ⊳′ g; (ML3) x ⊲ g + r(x ⊳ g) = [r(x), g] + v(x) ⊲′ g; (ML4) g ↼ x + r(g ⇀ x) = [g, r(x)] + g ↼′ v(x); (ML5) v({x, y}) = r(x) ⇀′ v(y) + v(x) ⊳′ r(y) + {v(x), v(y)}′; (ML6) f (x, y) + r({x, y}) = [r(x), r(y)] + r(x) ↼′ v(y) + v(x) ⊲′ r(y) + f ′(v(x), v(y)) Under the above bijection the morphism of Leibniz algebras ψ = ψ(r,v) : g ⋉ V → g ⋉′ V corresponding to (r, v) is given for any g ∈ g and x ∈ V by: ψ(g, x) = (g + r(x), v(x)) Moreover, ψ = ψ(r,v) is an isomorphism if and only if v : V → V is an isomorphism and ψ = ψ(r,v) co-stabilizes V if and only if v = IdV . Proof. A linear map ψ : g ⋉ V → g ⋉′ V which stabilizes g is uniquely determined by two linear maps r : V → g, v : V → V such that ψ(g, x) = (g + r(x), v(x)), for all g ∈ g, and x ∈ V . Indeed, by denoting ψ(0, x) = (r(x), v(x)) ∈ g × V for all x ∈ V , we obtain: ψ(g, x) = ψ(cid:0)(g, 0) + ψ(0, x)(cid:1)= ψ(g, 0) + ψ(0, x) = (cid:0)g + r(x), v(x)(cid:1) Let ψ = ψ(r,v) be such a linear map, i.e. ψ(g, x) = (g + r(x), v(x)), for some linear maps r : V → g, v : V → V . We will prove that ψ is a morphism of Leibniz algebras if and only if the compatibility conditions (M L1) − (M L6) hold. It is enough to prove that the compatibility ψ(cid:0)[(g, x), (h, y)](cid:1)= [ψ(g, x), ψ(h, y)] (14) holds for all generators of g ⋉ V . First of all, it is easy to see that (14) holds for the pair (g, 0), (h, 0), for all g, h ∈ g. Now we prove that (14) holds for the pair (g, 0), (0, x) if and only if (M L1) and (M L4) hold. Indeed, ψ(cid:0)[(g, 0), (0, x)](cid:1)= [ψ(g, 0), ψ(0, x)] it is equivalent to ψ(g ↼ x, g ⇀ x) = [(g, 0), (r(x), v(x))] and hence to (g ↼ x + r(g ⇀ x), v(g ⇀ x)) = ([g, r(x)] + g ↼′ v(x), g ⇀′ v(x)), i.e. to the fact that (M L1) and (M L4) hold. Next we prove that (14) holds for the pair (0, x), (g, 0) if and only if (M L2) and (M L3) hold. Indeed, ψ(cid:0)[(0, x), (g, 0)](cid:1)= [ψ(0, x), ψ(g, 0)] it is equivalent to ψ(x ⊲ g, x ⊳ g) = [(r(x), v(x)), (g, 0)] and therefore to (x ⊲ g + r(x ⊳ g), v(x ⊳ g)) = ([r(x), g] + v(x) ⊲′ g, v(x) ⊳′ g), i.e. to the fact that (M L2) and (M L3) hold. To this end, we prove that (14) holds for the pair (0, x), (0, y) if and only if (M L5) and (M L6) hold. Indeed, ψ(cid:0)[(0, x), (0, y)](cid:1)= [ψ(0, x), ψ(0, y)] it is equivalent to ψ(f (x, y), {x, y}) = [(r(x), v(x)), (r(y), v(y))]; thus it is equivalent to: (cid:0)f (x, y)+r({x, y}), v({x, y})(cid:1) UNIFIED PRODUCTS FOR LEIBNIZ ALGEBRAS 11 = (cid:0)[r(x), r(y)] + r(x) ↼′ v(y) + v(x) ⊲′ r(y) + f ′(v(x), v(y)), r(x) ⇀′ v(y) + v(x) ⊳′ r(y) + {v(x), v(y)}′(cid:1), i.e. to the fact that (M L5) and (M L6) hold. Assume now that v : V → V is bijective. Then ψ(r,v) is an isomorphism of Leibniz algebras with the inverse given by ψ−1 y ∈ V . Conversely, assume that ψ(r,v) is bijective. It follows easily that v is surjective. Thus, we are left to prove that v is injective. Indeed, let x ∈ V such that v(x) = 0. We have ψ(r,v)(0, 0) = (0, 0) = (0, v(x)) = ψ(r,v)(−r(x), x), and hence we obtain x = 0, i.e. v is a bijection. The last assertion is trivial and the proof is now finished. (cid:3) (r,v)(h, y) = (cid:0)h − r(v−1(y)), v−1(y)(cid:1), for all h ∈ g and Definition 2.7. Two Leibniz extending structures Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) and Ω′(g, V ) = (cid:0)⊳′, ⊲′, ↼′, ⇀′, f ′, {−, −}′(cid:1) are called equivalent and we denote this and v ∈ Autk(V ) such that (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) is implemented from (cid:0)⊳′, ⊲′, ↼′, ⇀′ , f ′, {−, −}′(cid:1) using (r, v) via: by Ω(g, V ) ≡ Ω′(g, V ), if there exists a pair (r, v) of linear maps, where r : V → g x ⊳ g = v−1(cid:0)v(x) ⊳′ g(cid:1) g ⇀ x = v−1(cid:0)g ⇀′ v(x)(cid:1) x ⊲ g = [r(x), g] + v(x) ⊲′ g − r ◦ v−1(cid:0)v(x) ⊳′ g(cid:1) g ↼ x = [g, r(x)] + g ↼′ v(x) − r ◦ v−1(cid:0)g ⇀′ v(x)(cid:1) {x, y} = v−1(cid:0)r(x) ⇀′ v(y) + v(x) ⊳′ r(y) + {v(x), v(y)}′(cid:1) f (x, y) = [r(x), r(y)] + r(x) ↼′ v(y) + v(x) ⊲′ r(y) + f ′(cid:0)v(x), v(y)(cid:1)− r ◦ v−1(cid:0)r(x) ⇀′ v(y) + v(x) ⊳′ r(y) + {v(x), v(y)}′(cid:1) for all g ∈ g, x, y ∈ V . Using Lemma 2.6, we obtain that Ω(g, V ) ≡ Ω′(g, V ) if and only if there exists ψ : g ⋉ V → g ⋉′ V an isomorphism of Leibniz algebras that stabilizes g, where g ⋉ V and g ⋉′ V are the corresponding unified products. On the other hand, the isomorphisms between two unified products that stabilize g and co-stabilize V are decoded by the following: Definition 2.8. Two Leibniz extending structures Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) and Ω′(g, V ) = (cid:0)⊳′, ⊲′, ↼′, ⇀′, f ′, {−, −}′(cid:1) are called cohomologous and we denote this by Ω(g, V ) ≈ Ω′(g, V ) if and only if ⊳ = ⊳′, ⇀ = ⇀′ and there exists a linear map r : V → g such that for any g ∈ g, x, y ∈ V : x ⊲ g = [r(x), g] + x ⊲′ g − r(x ⊳′ g) g ↼ x = [g, r(x)] + g ↼′ x − r(g ⇀′ x) {x, y} = r(x) ⇀′ y + x ⊳′ r(y) + {x, y}′ f (x, y) = [r(x), r(y)] + r(x) ↼′ y + x ⊲′ r(y) + f ′(x, y) − r(cid:0)r(x) ⇀′ y + x ⊳′ r(y) + {x, y}′(cid:1) As a conclusion of this section, the theoretical answer to the ES-problem follows: 12 A. L. AGORE AND G. MILITARU Theorem 2.9. Let g be a Leibniz algebra, E a vector space that contains g as a subspace and V a complement of g in E. Then: (1) ≡ is an equivalence relation on the set LZ(g, V ) of all Leibniz extending structures of g through V . If we denote HL2 g (V, g) := LZ(g, V )/ ≡, then the map HL2 g (V, g) → ExtdL (E, g), (⊳, ⊲, ↼, ⇀, f, {−, −}) 7→ (cid:0)g ⋉ V, [−, −](cid:1) is bijective, where (⊳, ⊲, ↼, ⇀, f, {−, −}) is the equivalence class of (⊳, ⊲, ↼, ⇀, f, {−, −}) via ≡. (2) ≈ is an equivalence relation on the set LZ(g, V ) of all Leibniz extending structures of g through V . If we denote HL2 (V, g) := LZ(g, V )/ ≈, then the map HL2 (V, g) → ExtdL′ (E, g), (⊳, ⊲, ↼, ⇀, f, {−, −}) 7→ (cid:0)g ⋉ V, [−, −](cid:1) is bijective, where (⊳, ⊲, ↼, ⇀, f, {−, −}) is the equivalence class of (⊳, ⊲, ↼, ⇀, f, {−, −}) via ≈. 3. Flag extending structures of Leibniz algebras. Examples After we have provided a theoretical answer to the ES problem in Theorem 2.9, we are left to compute the classifying object HL2 g (V, g) for a given Leibniz algebra g that is a subspace in a vector space E with a complement V and then to describe all Leibniz algebra structures on E which extend the one of g. In this section we propose an algorithm to tackle the problem for a large class of such structures. Definition 3.1. Let g be a Leibniz algebra and E a vector space containing g as a subspace. A Leibniz algebra structure on E is called a flag extending structure of g if there exists a finite chain of Leibniz subalgebras of E g = E0 ⊂ E1 ⊂ · · · ⊂ Em = E (15) such that Ei has codimension 1 in Ei+1, for all i = 0, · · · , m − 1. As an easy consequence of Definition 3.1 we have that dimk(V ) = m, where V is the complement of g in E. In what follows we will provide a way of describing all flag extending structures of g to E in a recursive manner which relies on the first step, namely m = 1. Therefore, we start by describing and classifying all unified products g ⋉ V1, for a 1-dimensional vector space V1. This procedure can be iterated by replacing the initial Leibniz algebra g with a unified product g ⋉ V1 obtained in the previous step. After m steps we arrive at the description of all flag extending structures of g to E. We start by introducing the following two concepts: Definition 3.2. Let g be a Leibniz algebra. A flag datum of the first kind of g is a 5-tuple (g0, α, λ, D, ∆), where g0 ∈ g, α ∈ k, λ : g → k, D and ∆ : g → g are linear maps satisfying the following compatibilities for any g, h ∈ g: λ(cid:0)∆(g)(cid:1)= 0; (F1) λ(cid:0)[g, h](cid:1)= 0, (F2) D(g0) = − α g0, λ(g0) = − α2, α ∆(g) = −[g, g0]; (F3) [g, ∆(h)] + [g, D(h)] = −λ(h) ∆(g); λ(cid:0)D(g)(cid:1)+ α λ(g) = 0, UNIFIED PRODUCTS FOR LEIBNIZ ALGEBRAS 13 (F4) D2(g) + D(cid:0)∆(g)(cid:1) = −λ(g) g0; (F5) D2(g) + ∆(cid:0)D(g)(cid:1) = α D(g) + [g0, g] − 2 λ(g) g0; (F6) ∆(cid:0)[g, h](cid:1) = [∆(g), h] + [g, ∆(h)]; (F7) D(cid:0)[g, h](cid:1) = [D(g), h] − [D(h), g] + λ(g)D(h) − λ(h)D(g) We denote by F1 (g) the set of all flag datums of the first kind of g. Definition 3.3. Let g be a Leibniz algebra. A flag datum of the second kind of g is a quadruple (g0, ν, D, ∆), where g0 ∈ g, ν : g → k, ν 6= 0 is a non-trivial map, D and ∆ : g → g are linear maps satisfying the following compatibilities for any g, h ∈ g: [g, g0] = 0, D(g0) = 0, ν(g0) = 0; (G2) [g, ∆(h)] + [g, D(h)] = 0, (G1) ν(cid:0)[g, h](cid:1)= 0, (G3) D2(g) + D(cid:0)∆(g)(cid:1) = 0, D2(g) + ∆(cid:0)D(g)(cid:1) = [g0, g] + 2 ν(g) g0; (G4) ∆(cid:0)[g, h](cid:1) = [∆(g), h] + [g, ∆(h)] + ν(h)∆(g) + ν(g)D(h); (G5) D(cid:0)[g, h](cid:1) = [D(g), h] − [D(h), g] − ν(g)D(h) + ν(h)D(g); ν(cid:0)D(g)(cid:1)+ ν(cid:0)∆(g)(cid:1)= 0; We denote by F2 (g) the set of all flag datums of the second kind of g and by F (g) := F1 (g) ⊔ F2 (g), the disjoint union of the two sets. The elements of F (g) will be called flag datums of g. F (g) contains the space of pointed double derivations D (g) of g via the canonical embedding: D(g) ֒→ F1 (g), (g0, D, ∆) 7→ (g0, 0, 0, D, ∆). The next proposition shows that the space F (g) is the counterpart for Leibniz algebras of what we have called in [4, Definition 4.2] the space of twisted derivations of a Lie algebra: Proposition 3.4. Let g be a Leibniz algebra and V a vector space of dimension 1 with a basis {x}. Then there exists a bijection between the set LZ (g, V ) of all Leibniz extending structures of g through V and F (g) = F1 (g) ⊔ F2 (g). Under the above bijective correspondence the Leibniz extending structure Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) corresponding to (g0, α, λ, D, ∆) ∈ F1 (g) is given by: x ⊳ g = λ(g)x, g ↼ x = ∆(g), x ⊲ g = D(g), g ⇀ x = 0, f (x, x) = g0 {x, x} = α x (16) (17) while the Leibniz extending structure Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) corresponding to (g0, ν, D, ∆) ∈ F2 (g) is given by: x ⊳ g = −ν(g)x, g ↼ x = ∆(g), x ⊲ g = D(g), g ⇀ x = ν(g)x, f (x, x) = g0 {x, x} = 0 (18) (19) for all g ∈ g. Proof. We have to compute the set of all bilinear maps ⊳ : V × g → V , ⊲ : V × g → g, ↼: g × V → g, ⇀: g × V → V , f : V × V → g and {−, −} : V × V → V satisfying the compatibility conditions (L1) − (L14) of Theorem 2.3. Since V has dimension 1 there exists a bijection between the set of all bilinear maps ⊳ : V × g → V and the set of all linear maps λ : g → k and the bijection is given such that the action ⊳ : V × g → V associated to λ is given by the formula: x ⊳ g := λ(g)x, for all g ∈ g. In the same manner, the action ⇀: g × V → V is uniquely determined by a linear map ν : g → k such that 14 A. L. AGORE AND G. MILITARU g ⇀ x = ν(g)x, for all g ∈ g. Similarly, the bilinear maps ⊲ : V × g → g, ↼: g × V → g are uniquely implemented by linear maps D = D⊲ : g → g respectively ∆ = ∆↼ : g → g via the formulas: x ⊲ g := D(g) and g ↼ x := ∆(g), for all g ∈ g. Finally, any bilinear map f : V × V → g is uniquely implemented by an element g0 ∈ g such that f (x, x) = g0 and any bracket {−, −} : V × V → V is uniquely determined by a scalar α ∈ k such that {x, x} = αx. Now, the compatibility condition (L9) is equivalent to α ν(g) = 0, for all g ∈ g, while If ν = 0, the trivial map on g, then (L9) and (L14) are trivially fulfilled and the rest of the compatibility conditions of Theorem 2.3 came down to (F 1)-(F 7) from the definition of F1 (g). This can be proved (L14) gives ν(g)(cid:0)ν(h) + λ(h)(cid:1)= 0, for all g, h ∈ g. by a routinely computation: for instance, axiom (L1) is equivalent to λ(cid:0)[g, h](cid:1)= 0 while axiom (L2) is equivalent to (F 7). Otherwise, if ν 6= 0 implies that α = 0 and λ = −ν. Based on this, it is straightforward to see that the compatibility conditions (L1) − (L14) of Theorem 2.3 are equivalent to (G1) − (G5) from the definition of F2 (g). (cid:3) Let (g0, α, λ, D, ∆) ∈ F1 (g). The unified product g ⋉(g0, α, λ, D, ∆) V associated to the Leibniz extending structure given by (16) - (17) will be denoted by g1 (x (g0, α, λ, D, ∆)) and has the bracket defined by: [(g, 0), (h, 0)] = ([g, h], 0), [(0, x), (0, x)] = (g0, α x), [(g, 0), (0, x)] = (∆(g), 0) [(0, x), (g, 0)] = (D(g), λ(g)x) (20) (21) for all g, h ∈ g. On the other hand, for (g0, ν, D, ∆) ∈ F2 (g), the unified product g ⋉(g0, ν, D, ∆) V associated to the Leibniz extending structure given by (18)-(19) will be denoted by g2 (x (g0, ν, D, ∆)) and has the bracket defined by: [(g, 0), (h, 0)] = ([g, h], 0), [(0, x), (0, x)] = (g0, 0), [(g, 0), (0, x)] = (∆(g), ν(g)x) [(0, x), (g, 0)] = (D(g), −ν(g)x) (22) (23) for all g, h ∈ g. Thus, we have obtained the following: Corollary 3.5. Any Leibniz algebra that contains a given Leibniz algebra g as a subalge- bra of codimension 1 is isomorphic to a Leibniz algebra of type g1 (x (g0, α, λ, D, ∆)), for some (g0, α, λ, D, ∆) ∈ F1 (g) or to a Leibniz algebra of type g2 (x (g0, ν, D, ∆)), for some (g0, ν, D, ∆) ∈ F2 (g). Now, we shall classify these Leibniz algebras up to an isomorphism that stabilizes g, i.e. we give the first explicit classification result for the ES-problem. This is the key step in the classification of all flag extending structures of g. Theorem 3.6. Let g be a Leibniz algebra of codimension 1 in the vector space E. Then: ExtdL (E, g) ∼= HL2 g(k, g) ∼= (F1 (g)/ ≡1) ⊔ (F2 (g)/ ≡2), where : ≡1 is the equivalence relation on the set F1 (g) defined as follows: (g0, α, λ, D, ∆) ≡1 0, α′, λ′, D′, ∆′) if and only if λ = λ′ and there exists a pair (q, G) ∈ k∗ × g such that (g′ UNIFIED PRODUCTS FOR LEIBNIZ ALGEBRAS for any g ∈ g: g0 = q2 g′ α = q α′ + λ′(G) 0 + [G, G] + q D′(G) + q E ′(G) − q α′G − λ′(G)G D(g) = q D′(g) + [G, g] − λ′(g)G ∆(g) = q ∆′(g) + [g, G] 15 (24) (25) (26) (27) ≡2 is the equivalence relation on F2 (g) given by: (g0, ν, D, ∆) ≡2 (g′ only if ν = ν ′ and there exists a pair (q, G) ∈ k∗ × g such that for any g ∈ g: 0, ν ′, D′, ∆′) if and g0 = q2 g′ 0 + [G, G] + q D′(G) + q E ′(G) D(g) = q D′(g) + [G, g] + ν ′(g) G ∆(g) = q ∆′(g) + [g, G] − ν ′(g) G (28) (29) (30) The bijection between (F1 (g)/ ≡1) ⊔ (F2 (g)/ ≡2) and ExtdL (E, g) is given by: 1 (g0, α, λ, D, ∆) 1 2 7→ g1 (x (g0, α, λ, D, ∆)) and (g0, ν, D, ∆) 7→ g2 (x (g0, ν, D, ∆)) Proof. The proof relies on Proposition 3.4 and Theorem 2.9. Let V be a complement of g in E having {x} as a basis. Since dimk(V ) = 1, any linear map r : V → g is uniquely determined by an element G ∈ g such that r(x) = G, where {x} is a basis in V . On the other hand, any automorphism v of V is uniquely determined by a non- zero scalar q ∈ k∗ such such v(x) = qx. Based on these facts, a little computation shows that the compatibility conditions from Definition 2.7, imposed for the Leibniz extending structures (16)-(17) and respectively (18)-(19), take precisely the form given in the statement of the theorem. We should mention here that a Leibniz extending structure given by (16) - (17) is never equivalent in the sense of Definition 2.7 to a Leibniz extending structure given by (18)-(19), thanks to the compatibility condition (ML1) of Lemma 2.6. Therefore, we obtain the disjoint union from the statement and the proof is finished. (cid:3) Remark 3.7. In the context of Theorem 3.6 we also have that ExtdL′ (E, g) ∼= HL2(k, g) ∼= (F1 (g)/ ≈1) ⊔ (F2 (g)/ ≈2), where : ≈i is the following relation on Fi (g): (g0, α, λ, D, ∆) ≈1 (g′ 0, α′, λ′, D′, ∆′) if and only if λ = λ′ and there exists G ∈ g such that relations (24)-(27) hold for q = 1 and respectively (g0, ν, D, ∆) ≈2 (g′ 0, ν ′, D′, ∆′) if and only if ν = ν ′ and there exists G ∈ g such that (28)-(30) hold for q = 1. Theorem 3.6 takes a simplified form for perfect Leibniz algebras. Indeed, let g be a perfect Leibniz algebra, i.e. g is generated as a vector space by all brackets [x, y]. Then (G1) shows that F2 (g) is the empty set since, by definition, an element ν of a quadruple (g0, ν, D, ∆) ∈ F2 (g) is a non-trivial map. Thus, we have that F (g) = F1 (g). Let now (g0, α, λ, D, ∆) ∈ F1 (g); it follows from (F1) and (F2) that λ = 0, the trivial map, and α = 0. Furthermore, we can easily see that for a perfect Leibniz algebra g, F (g) identifies with the set of all triples (g0, D, ∆), where g0 ∈ g, D, ∆ : g → g are linear 1As usual we denote by yi the equivalence class of y via the relation ≡i, i = 1, 2. 16 A. L. AGORE AND G. MILITARU maps satisfying the compatibilities (4)-(7), that is F (g) ∼= D(g), where D(g) is the space of all pointed double derivations of g as defined in Definition 1.3. Two pointed double derivations (g0, D, ∆) and (g′ (g0, D, ∆) ≡ (g′ 0, D′, ∆′) if and only if there exists a pair (q, G) ∈ k∗ × g such that: 0, D′, ∆′) are equivalent and we write g0 = q2 g′ 0 + [G, G] + q D′(G) + q ∆′(G) D − q D′ = [G, −], ∆ − q ∆′ = [−, G] (31) (32) On the other hand, two pointed double derivations (g0, D, ∆) and (g′ homologous and we write (g0, D, ∆) ≈ (g′ that: 0, D′, ∆′) are co- 0, D′, ∆′) if and only if there exists G ∈ g such g0 = g′ 0 + [G, G] + D′(G) + ∆′(G) ∆ − ∆′ = [−, G] D − D′ = [G, −], (33) (34) Taking into account the unified product defined by (20)-(21) we obtain: Corollary 3.8. Let g be a perfect Leibniz algebra having {ei i ∈ I} as a basis. Then any Leibniz algebra E containing g as a subalgebra of codimension 1 has the bracket [−, −]E defined on the basis {x, ei i ∈ I} by: [ei, ej]E := [ei, ej], [x, x]E := g0, [ei, x]E := ∆(ei), [x, ei]E := D(ei) for all (g0, D, ∆) ∈ D(g). Furthermore, HL2 D (g)/ ≈, where ≡ (resp. ≈) is the relation defined by (31)-(32) (resp. (33)-(34)). g(k, g) ∼= D (g)/ ≡ and HL2(k, g) ∼= On the other hand, we have the following result for abelian Leibniz algebras: Example 3.9. Let g be a vector space with {ei i ∈ I} as a basis viewed as an abelian Leibniz algebra. Then, there exist three families of Leibniz algebras that contain g as a subalgebra of codimension 1. They have {x, ei i ∈ I} as a basis and the bracket given for any i ∈ I as follows: g (g0, D, ∆) 11 : [ei, ej] = 0, [ei, x] = ∆(ei), [x, x] = g0, [x, ei] = D(ei) for all triples (g0, D, ∆) ∈ g × Homk(g, g)2 such that g0 ∈ Ker(D) and D ◦ ∆ = ∆ ◦ D = −D2. The Leibniz algebra g is the unified product associated to the flag datum of the first kind (g0, α, λ, D, ∆) for which α := 0 and λ := 0. The second family of Leibniz algebras has the bracket given as follows: (g0, D, ∆) 11 g (u, h0, λ) 12 : [ei, ej] = 0, [x, x] = −u2 λ(h0) h0 − u λ(h0) x [ei, x] = 0, [x, ei] = u λ(ei) h0 + λ(ei) x for all triples (u, h0, λ) ∈ k∗ × g × Homk(g, k) such that λ 6= 0. The Leibniz al- is the unified product associated to the flag datum of the first kind gebra g (g0, α, λ, D, ∆) for which g0 := u2λ(h0) h0, α := −uλ(h0), ∆ := 0 and D(g) := uλ(g) h0, for all g ∈ g. (u, h0, λ) 12 Finally, for the last family of Leibniz algebras the bracket is given as follows: g (u, g0, h0, ν) 2 : [ei, ej] = 0, [ei, x] = − [x, ei] = −u ν(ei) h0 + ν(ei) x, [x, x] = g0 UNIFIED PRODUCTS FOR LEIBNIZ ALGEBRAS 17 for all (u, g0, h0, ν) ∈ k∗ × g2 × Homk(g, k) such that: 2g0 = 0, ν(g0) = 0 and ν 6= 0. The Leibniz algebra g is the unified product associated to the flag datum of the second kind (g0, ν, D, ∆) for which D(g) := u ν(g)h0 and ∆(g) := −u ν(g)h0, for all g ∈ g. (u, g0, h0, ν) 2 The above results are obtained by a straightforward computation which relies on the explicit description of the set F (g) for an abelian Leibniz algebra. For instance, axiom (F3) from the flag datum of first kind, for an abelian Leibniz algebra, takes the form λ(h)∆(g) = 0, for all g, h ∈ g. Therefore, we have to consider two cases in order to describe the set F1 (g): the first one corresponds to λ = 0, while for the second one we have λ 6= 0. The corresponding unified products are the ones given by the first two families of Leibniz algebras. In order to describe the set F2 (g) we mention that the condition 2g0 = 0 is derived from axiom (G2). In this case the corresponding unified product is the one given by the last family of Leibniz algebras. Next we provide some explicit examples. First, we prove that any Leibniz algebra which contains a semisimple Lie algebra g as a subalgebra of codimension 1 is in fact a Lie algebra and the classifying objects H2 g(k, g) and H2(k, g) are both singletons. Example 3.10. Let g be a semisimple Lie algebra of codimension 1 in the vector space E and {ei i = 1, · · · , n} a basis of g. Then any Leibniz algebra structure on E that contains g as a subalgebra is isomorphic to the Lie algebra having {x, ei i = 1, · · · , n} as a basis and the bracket [−, −]E defined by for any i = 1, · · · , n by: [ei, ej]E := [ei, ej], [x, ei]E = −[ei, x]E := [h0, ei], g(k, g) = HL2(k, g) = 0. [x, x] = 0 for some h0 ∈ g. Furthermore, HL2 Indeed, we apply Corollary 3.8 taking into account that any semisimple Lie algebra g is perfect, Inn(g) = Der(g) and Z(g) = 0. Let (g0, D, ∆) ∈ F (g) be a flag datum of g; then, since g has a trivial center we obtain from (4) that g0 = 0. Moreover, as g is perfect it follows again from (4) that ∆ = −D. Thus, F (g) = Der(g). Since g is semisimple any derivation D ∈ Der(g) is inner, i.e. there exists h0 ∈ g such that D = [h0, −]. Thus, the Leibniz algebra g1 (x (g0, α, λ, D, ∆)) = g1 (x D) defined by (20) and (21) takes the form given in the statement. Moreover, two derivations D = [h0, −] and D′ = [h′ 0, −] are equivalent in the sense of (34) if and only if there exists G ∈ g such that h0 = h′ 0 + G, i.e. any two derivations are cohomologous. This shows that HL2(k, g) is a singleton having only 0 as an element and so is HL2 g(k, g) being a quotient of it. Now we will provide an explicit example which highlights the efficiency of Theorem 3.6. More precisely, we will describe all 4-dimensional Leibniz algebras that contain a given non-perfect 3-dimensional Leibniz algebra g as a subalgebra. Then we will be able to compute the classifying object ExtdL (k4, g) ∼= HL2 g(k, g). The detailed computations are rather long but straightforward and can be provided upon request. Example 3.11. Let g be the 3-dimensional Leibniz algebra with the basis {e1, e2, e3} and the bracket defined by: [e1, e3] = e2, [e3, e3] = e1. Then, there exist four families of 4-dimensional Leibniz algebras which contain g as a subalgebra: they have {e1, e2, e3, x} as a basis and the bracket is given as follows (the 18 A. L. AGORE AND G. MILITARU first three families of Leibniz algebras can be defined over any field k while in case of the fourth family we need to distinguish between fields of characteristic 2 and those of characteristic different than 2): (1) If char(k) 6= 2 then the four families of Leibniz algebras that contain g as a subalgebra are the following: g (b1, b2, c, d1, d2) 11 : [e3, e3] = e1, [e1, e3] = e2, [e1, x] = b1 e2, [x, x] = b1d1 e1 + c e2, [e3, x] = b1 e1 + b2 e2, [x, e3] = d1 e1 + d2 e2 for all b1, b2, c, d1, d2 ∈ k. The Leibniz algebra g is the unified product associated to the flag datum of the first kind (g0, α, λ, D, ∆) defined as follows: α := 0, λ := 0, g0 := b1d1 e1 + c e2 and D, ∆ are given by (b1, b2, c, d1, d2) 11 D :=   0 0 d1 0 0 d2 0 0 0   ∆ :=   0 0 b1 b1 0 b2 0 0 0   The second family of Leibniz algebras has the bracket given by: g (b1, b2, b3, c, d) 12 : [e1, e3] = e2, [e2, x] = 3b1 e2, [x, x] = (2b1d + b2 [e3, e3] = e1, [e1, x] = 2b1 e1 + b2 e2, [e3, x] = b2 e1 + b3 e2 + b1 e3, 2 − b1b3) e1 + c e2, [x, e3] = b2 e1 + d e2 − b1 e3 for all b1 ∈ k∗ and b2, b3, c, d ∈ k. The Leibniz algebra g is the unified product associated to the flag datum of the first kind (g0, α, λ, D, ∆) defined as follows: α := 0, λ := 0, g0 := (2b1d + b2 2 − b1b3) e1 + c e2 and D, ∆ are given by: (b1, b2, b3, c, d) 12 D :=   0 0 b2 0 0 d 0 0 −b1   ∆ :=   2b1 b2 0 0 3b1 0   b2 b3 b1 The third family of Leibniz algebras has the bracket given by: g (α, λ0, d1, d2) 13 : [e3, e3] = e1, [e1, e3] = e2, [e3, x] = α λ−1 [x, x] = α λ−1 0 [x, e3] = d1 e1 + d2 e2 − α e3 + λ0 x 0 e1, [e1, x] = α λ−1 0 e2, (d1 e1 + d2 e2 − α e3) + α x, for all λ0 ∈ k∗ and α, d1, d2 ∈ k. The Leibniz algebra g is the unified product associated to the flag datum of the first kind (g0, α, λ, D, ∆) defined as follows: λ(e1) = λ(e2) := 0, λ(e3) := λ0 6= 0, g0 := α λ−1 0 e3 and D, ∆ are given by 0 d2 e2 − α2 λ−1 0 d1 e1 + α λ−1 (α, λ0, d1, d2) 13 D :=   0 0 d1 0 0 d2 0 0 −α   ∆ :=   0 α λ−1 0 0 0 α λ−1 0 0 0 0 0   UNIFIED PRODUCTS FOR LEIBNIZ ALGEBRAS 19 Finally, the last family of Leibniz algebras has the bracket defined as follows: g (ν0, d1, d2,,d3) 21 : [e1, x] = ν −1 [e3, e3] = e1, [e1, e3] = e2, [e3, x] = (−d1 + 2ν −1 [x, x] = ν −2 0 d1d3 − ν −3 [x, e3] = d1 e1 + d2 e2 + d3 e3 − ν0 x 0 d3)e1 − (d2 − ν −1 0 d2 3 e1 + (ν −2 0 d2 0 d3 e2, 0 d1 + ν −2 3) e2, 0 d3)e2 − d3e3 + ν0x, for all ν0 ∈ k∗ and d1, d2, d3 ∈ k. The Leibniz algebra g is the unified product associated to the flag datum of the second kind (g0, ν, D, ∆) defined as follows: ν(e1) = ν(e2) := 0, ν(e3) := ν0 6= 0, g0 := ν −2 3) e2 and D, ∆ are given by 0 d1d3 − ν −3 3 e1 + (ν −2 0 d2 0 d2 (ν0, d1, d2, d3) 21 D :=   0 0 d1 0 0 d2 0 0 d3   ∆ :=   0 −d1 + 2ν −1 0 ν −1 0 d3 0 −d2 + ν −1 0 0 d3 0 d1 − ν −2 −d3 0 0 d3   (35) (2) If char(k) = 2, then the four families of Leibniz algebras that contain g as a subalgebra are the following: g , g defined above together with the family of Leibniz algebras defines as follows: (b1, b2, c, d1, d2) 11 (b1, b2, b3, c, d) 12 (α, λ0, d1, d2) 13 , g g (c, ν0, d1, d2,,d3) 22 : [e3, e3] = e1, [e1, e3] = e2, [e3, x] = −d1 e1 − (d2 − ν −1 [x, x] = ν −2 3 e1 + c e2, 0 d2 [e1, x] = ν −1 0 d3 e2, 0 d1 + ν −2 [x, e3] = d1 e1 + d2 e2 + d3 e3 − ν0 x 0 d3)e2 − d3e3 + ν0x, for all ν0 ∈ k∗ and c, d1, d2, d3 ∈ k. The Leibniz algebra g is the unified product associated to the flag datum of the second kind (g0, ν, D, ∆) defined as follows: ν(e1) = ν(e2) := 0, ν(e3) := ν0 6= 0, g0 := ν −2 The proof is a purely computational one and we will only indicate the main steps. We start by computing F1 (g). First, notice that a linear map λ : g → k satisfies the first compatibility of (F1), i.e. λ([g, h]) = 0 if and only if λ is given by λ(e1) = λ(e2) = 0 and λ(e3) = λ0, for some λ0 ∈ k. For such a λ we can easily show that a pair (D, ∆) satisfies the compatibilities (F6) and (F7) if and only if we have: 3 e1 + c e2 and D, ∆ are given by (35). 0 d2 (c, ν0, d1, d2, d3) 22 D =   0 0 d1 0 0 d2 0 0 d3   ∆ =   2b1 b2 0 0 3b1 0 b2 b3 b1   for some d1, d2, d3, b1, b2, b3 ∈ k. Let now α ∈ k and consider g0 = c1e1 + c2e2 + c3e3, for some c1, c2, c3 ∈ k. We can easily see that the 5-tuple (g0, α, λ, D, ∆) satisfies the compatibilities (F1) - (F7), i.e. it is a flag datum of the first kind if and only if coincides with one of the three flag datums described in (1). For instance, the compatibility (F1) is fulfilled if and only if λ0(α + d3) = λ0 b1 = 0. This last equality leads us to consider two cases, namely λ0 = 0 or λ0 6= 0. It is now straightforward to describe F1 (g) (without depending on the characteristic of k). Analogously, we can describe F2 (g). A non-trivial map ν : g → k satisfies the first compatibility of (G1) if and only if ν is given by ν(e1) = ν(e2) = 0 and ν(e3) = ν0, 20 A. L. AGORE AND G. MILITARU for some ν0 ∈ k∗. For such a map ν, we can easily show that (D, ∆) satisfies the compatibilities (G4) and (G5) if and only if D and ∆ are given by (35). By considering again g0 = c1e1 + c2e2 + c3e3, for some c1, c2, c3 ∈ k we see that the compatibility (G1) is fulfilled if and only if c3 = 0. The last compatibility of (G3) is equivalent to: 2 c1 = 2 ν −2 0 d2 3, c1 + 2 ν0c2 = 2 ν −1 0 d1d3 − ν −1 0 d2 3 The above two compatibilities are the ones that lead us to the description of F2 (g) depending on the characteristic of k. Moreover, these computations provide also the description of the classifying object HL2 g(k, g). If char(k) 6= 2 then HL2 g(k, g) ∼= (k5/ ≡11) ⊔ ((k∗ × k4)/ ≡12) ⊔ ((k∗ × k3)/ ≡13) ⊔ ((k∗ × k3)/ ≡2) (36) where ≡1i are the equivalence relations (24)-(27) while ≡2 is the equivalence relation (28)-(30). In the case when char(k) = 2, then the last term of (36) is replaced by (k∗ × k4)/ ≡2. 4. Special cases of unified products In this section we deal with two special cases of the unified product namely the crossed (resp. bicrossed) product of two Leibniz algebras. We emphasize the problem for which each of these products is responsible. We use the following convention: if one of the maps ⊳, ⊲, ↼, ⇀, f or {−, −} of an extending datum Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) is trivial then we will omit it from the 6-tuple (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1). Crossed products and the extension problem for Leibniz algebras. We shall highlight a first special case of the unified product, namely the crossed product of Leibniz algebras that is the key player in the study of the extension problem in its full generality. Let Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) be an extending datum of a Leibniz algebra g g ∈ g. Then, it follows from Theorem 2.3 that Ω(g, V ) = (cid:0)⊲, ↼, f, {−, −}(cid:1) is a Leibniz through V such that ⊳ and ⇀ are both trivial, i.e. x ⊳ g = g ⇀ x = 0, for all x ∈ V and extending structure of g through V if and only if (V, {−, −}) is a Leibniz algebra and (g, V, ⊲, ↼, f ) is a crossed system of Leibniz algebras, i.e. the following compatibilities hold for any g, h ∈ g and x, y, z ∈ V : (CS1) [g, h] ↼ x = [g, h ↼ x] + [g ↼ x, h]; (CS2) g ↼ {x, y} = (g ↼ x) ↼ y − (g ↼ y) ↼ x − [g, f (x, y)]; (CS3) x⊲f (y, z) = f (x, y) ↼ z−f (x, z) ↼ y+f ({x, y}, z)−f ({x, z}, y)−f (x, {y, z}); (CS4) x ⊲ [g, h] = [x ⊲ g, h] − [x ⊲ h, g]; (CS5) {x, y} ⊲ g = x ⊲ (y ⊲ g) + (x ⊲ g) ↼ y − [f (x, y), g]; (CS6) [g, h ↼ x] + [g, x ⊲ h] = 0; (CS7) x ⊲ (y ⊲ g) + x ⊲ (g ↼ y) = 0. In this case, the associated unified product g ⋉ ⊲,↼ V and we shall call it the crossed product of the Leibniz algebras g and V . Hence, the crossed product associated to the crossed system (g, V, ⊲, ↼, f ) is the Leibniz algebra defined as follows: g#f ⊲,↼ V = g × V with the bracket given for any g, h ∈ g and x, y ∈ V by: Ω(g,V ) V will be denoted by g#f UNIFIED PRODUCTS FOR LEIBNIZ ALGEBRAS [(g, x), (h, y)] := (cid:0)[g, h] + x ⊲ h + g ↼ y + f (x, y), {x, y}(cid:1) 21 (37) The crossed product of Leibniz algebras is the object responsible for answering the following special case of the ES problem, which is a generalization of the extension problem: Let g be a Leibniz algebra, E a vector space containing g as a subspace. Describe and classify all Leibniz algebra structures on E such that g is a two-sided ideal of E. The classical extension problem initiated in [21] is a special case of this question if we require the additional assumption on the quotient E/g to be isomorphic to a given Leibniz algebra h. Indeed, let (g, V, ⊲, ↼, f ) be a crossed system of two Leibniz algebras. Then, g ∼= g × {0} ⊲,↼ V since [(g, 0), (h, y)] := (cid:0)[g, h] + g ↼ is a two-sided ideal in the crossed product g#f y, 0(cid:1) and [(g, x), (h, 0)] := (cid:0)[g, h] + x ⊲ h, 0(cid:1). Conversely, we have: Corollary 4.1. Let g be a Leibniz algebra, E a vector space containing g as a subspace. Then any Leibniz algebra structure on E that contains g as a two-sided ideal is isomorphic to a crossed product of Leibniz algebras g#f ⊲,↼ V and the isomorphism can be chosen to stabilize g and co-stabilize V . Proof. Let [−, −] be a Leibniz algebra structure on E such that g is a two-sided ideal in E. In particular, g is a subalgebra of E and hence we can apply Theorem 2.5. In this case the actions ⊳ = ⊳p and ⇀=⇀p of the Leibniz extending structure Ω(g, V ) = (cid:0)⊳p, ⊲p, fp, {−, −}p(cid:1) constructed in the proof of Theorem 2.5 are both trivial since for Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) constructed in the proof of Theorem 2.5 is precisely a any x ∈ V and g ∈ g we have that [x, g], [g, x] ∈ g and hence p([x, g]) = [x, g] and p([g, x]) = [g, x]. Thus, x ⊳p g = g ⇀p x = 0 and hence the Leibniz extending structure crossed system of Leibniz algebras and the unified product g ⋉ crossed product of g and V = Ker(p). Ω(g,V ) V = g#f ⊲,↼ V is the (cid:3) Let g and h be two given Leibniz algebras. The extension problem asks for the classi- fication of all extensions of h by g, i.e. of all Leibniz algebras E that fit into an exact sequence 0 / g i / E π / h / 0 (38) The classification is up to an isomorphism of Leibniz algebras that stabilizes g and co-stabilizes h and we denote by EP(h, g) the isomorphism classes of all extensions of If g is abelian, then EP(h, g) ∼= HL2(h, g), h by g up to this equivalence relation. where HL2(h, g) is the the second cohomology group [21, Proposition 1.9]. The crossed product is the tool to approach the extension problem in its full generality, leaving aside the abelian case. Let us explain this briefly. Consider g and h be two Leibniz algebras and we denote by CS (h, g) the set of all triples (⊲, ↼, f ) such that (g, h, ⊲, ↼, f ) is a crossed system of Leibniz algebras. First we remark that, if (g, h, ⊲, ↼, f ) is a crossed system, then the crossed product g#f ⊲,↼ h is an extension of h by g via 0 / g ig / g#f ⊲,↼ h πh / h / 0 (39) / / / / / / / / 22 A. L. AGORE AND G. MILITARU where ig(g) = (g, 0) and πh(g, h) = h are the canonical maps. Conversely, Corollary 4.1 shows that any extension E of h by g is equivalent to a crossed product extension of the form (39). Thus, the classification of all extensions of h by g reduces to the classification of all crossed products g#f ⊲,↼ h associated to all crossed systems of Leibniz algebras (g, h, ⊲, ↼, f ). Definition 2.8, in the special case of crossed systems, takes the following simplified form: two triples (⊲, ↼, f ) and (⊲′, ↼′, f ′) of CS (h, g) are cohomologous and we denote this by (⊲, ↼, f ) ≈ (⊲′, ↼′, f ′) if there exists a linear map r : h → g such that: x ⊲ g = x ⊲′ g + [r(x), g] g ↼ x = g ↼′ x + [g, r(x)] f (x, y) = f ′(x, y) + [r(x), r(y)] − r(cid:0){x, y}(cid:1)+ r(x) ↼′ y + x ⊲′ r(y) for all g ∈ g, x, y ∈ h. Then, as we mentioned before Definition 2.8, (⊲, ↼, f ) ≈ (⊲′, ↼′, f ′) if and only if there exists ψ : g#f ⊲′,↼′ h an isomorphism of Leibniz algebras that stabilizes g and co-stabilizes h. As a special case of Theorem 2.9, we obtain the theoretical answer to the extension problem in the general (non-abelian) case: ⊲,↼ h → g#f ′ Corollary 4.2. Let g and h be two arbitrary Leibniz algebras. Then ≈ is an equivalence relation on the set CS (h, g) of all crossed systems and the map HL2(h, g) := CS (h, g)/ ≈ −→ EP(h, g), (⊲, ↼, f ) 7→ g#f ⊲,↼ h is a bijection between sets, where (⊲, ↼, f ) is the equivalence class of (⊲, ↼, f ) via ≈. If g is an abelian Leibniz algebra, then HL2(h, g) coincides with the second cohomology group HL2(h, g) constructed in [21]. The explicit answer to the extension problem for two given Leibniz algebras g and h will be given once we compute the non-abelian cohomological object HL2(h, g) which in general is a highly non-trivial problem. A detailed study of this object for various Leibniz algebras will be given elsewhere; here we give only one example that corresponds to the case when h := k, the abelian Leibniz algebra of dimension 1, as this is a special case of Theorem 3.6 and Remark 3.7. Corollary 4.3. Let g be a Leibniz algebra with {ei i ∈ I} as a basis. Then HL2(k, g) ∼= D(g)/ ≈ where D(g) is the space of all pointed double derivations of g and ≈ is the equivalence relation defined by (33)-(34). In particular, any extension of k by g is isomorphic to the Leibniz algebra having {x, ei i ∈ I} as a basis and the bracket [−, −](g0, D, ∆) defined for any i ∈ I by: [ei, ej](g0, D, ∆) [ei, x](g0, D, ∆) := [ei, ej] , := ∆(ei), [x, x](g0, D, ∆) := g0 [x, ei](g0, D, ∆) := D(ei) for some (g0, D, ∆) ∈ D(g). Proof. Follows from Theorem 3.6 since the set of crossed systems CS (k, g) is precisely the set Ω(g, k) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) of all Leibniz extending structures of g through k having the actions ⊳ and ⇀ both trivial. Moreover, any extension E of k by g is a UNIFIED PRODUCTS FOR LEIBNIZ ALGEBRAS 23 Leibniz algebra containing g as a subalgebra of codimension 1. In this context, the compatibility conditions (F1)-(F7) that define a flag datum of the first kind collapses to (4)-(7). The fact that ⊳ is the trivial action implies that λ = 0. The Leibniz algebra from the statement is the unified (crossed) product defined by (20)-(21). (cid:3) In the next example we compute explicitly the object HL2(k, g) for a certain Leibniz algebra g. Example 4.4. Let g be the 3-dimensional Leibniz algebra with the basis {e1, e2, e3} and the bracket defined by: [e1, e3] = e2, [e3, e3] = e1. A little computation, similar to one performed in Example 3.11, shows that the set D(g) identifies with the set of all 6-tuples (c, b1, b2, b3, d1, d2) ∈ k6 which satisfy: b1 (d1 − b2) = 0 The bijection is defined such that (g0, D, ∆) ∈ D(g) corresponding to (c, b1, b2, b3, d1, d2) is given by g0 := (2 b1d2 + b2d1 − b1b3) e1 + c e2, D :=   0 0 d1 0 0 d2 0 0 −b1   ∆ :=   2b1 b2 0 0 3b1 0 b2 b3 b1   The compatibility condition b1 (d1 − b2) = 0 imposes a discussion on whether b1 = 0 or b1 6= 0. This leads to the description of L2(k, g) as the following coproduct of sets: HL2(k, g) ∼= (k5/ ≈1) ⊔ (k∗ × k4/ ≈2), where : 2, b′ 3, d′ 1, d′ 2) if and only if ≈1 is the following relation on k5: (c, b2, b3, d1, d2) ≈1 (c′, b′ there exist u, v ∈ k such that b2 = b′ c = c′ + uv, d1 = d′ b3 = b′ 2 + v, 3 + v, 1 + v, d2 = d′ 2 + u and ≈2 is the relation on k∗ × k4 defined by: (b1, c, b3, d1, d2) ≈2 (b′ and only if b1 = b′ 2 and there exist v, w ∈ k such that 3, d2 = d′ 1, b3 = b′ 1, c′, b′ 3, d′ 1, d′ 2) if c = c′ + vd′ 1 + 3wb′ 1 + (d1 − d′ 1) (v + d′ 2 + b′ 3) Bicrossed products and the factorization problem for Leibniz algebras. The concept of a matched pair of Lie algebras was introduced in [23, Theorem 4.1] and independently in [22, Theorem 3.9]. For any such matched pair of Lie algebras a new Lie algebra, called the bicrossed product is constructed under the name of bicrossproduct in [23, Theorem 4.1], double cross sum in [24, Proposition 8.3.2], double Lie algebra [22, Definition 3.3]. Now we shall introduce the concept of a matched pair of Leibniz algebras. As we will see, in this case the definition is a lot more laborious. Definition 4.5. A matched pair of Leibniz algebras is a system (g, h, ⊳, ⊲, ↼, ⇀) con- sisting of two Leibniz algebras (g, [−, −]), (h, {−, −}) and four bilinear maps ⊳ : h × g → h, ⊲ : h × g → g, ↼: g × h → g, ⇀: g × h → h satisfying the following compatibilities for any g, h ∈ g, x, y ∈ h: (MP1) (h, ⊳) is a right g-module, i.e. x ⊳ [g, h] = (x ⊳ g) ⊳ h − (x ⊳ h) ⊳ g; (MP2) (g, ↼) is a right h-module, i.e. g ↼ {x, y} = (g ↼ x) ↼ y − (g ↼ y) ↼ x; (MP3) x ⊲ [g, h] = [x ⊲ g, h] − [x ⊲ h, g] + (x ⊳ g) ⊲ h − (x ⊳ h) ⊲ g; 24 A. L. AGORE AND G. MILITARU (MP4) {x, y} ⊳ g = x ⊳ (y ⊲ g) + (x ⊲ g) ⇀ y + {x, y ⊳ g} + {x ⊳ g, y}; (MP5) {x, y} ⊲ g = x ⊲ (y ⊲ g) + (x ⊲ g) ↼ y; (MP6) [g, h] ↼ x = [g, h ↼ x] + [g ↼ x, h] + g ↼ (h ⇀ x) + (g ⇀ x) ⊲ h; (MP7) [g, h] ⇀ x = g ⇀ (h ⇀ x) + (g ⇀ x) ⊳ h; (MP8) g ⇀ {x, y} = (g ↼ x) ⇀ y − (g ↼ y) ⇀ x + {g ⇀ x, y} − {g ⇀ y, x}; (MP9) [g, h ↼ x] + [g, x ⊲ h] + g ↼ (h ⇀ x) + g ↼ (x ⊳ h) = 0; (MP10) x ⊲ (y ⊲ g) + x ⊲ (g ↼ y) = 0; (MP11) x ⊳ (y ⊲ g) + x ⊳ (g ↼ y) + {x, y ⊳ g} + {x, g ⇀ y} = 0; (MP12) g ⇀ (h ⇀ x) + g ⇀ (x ⊳ h) = 0. Let (g, h, ⊳, ⊲, ↼, ⇀) be a matched pair of Leibniz algebras. Then g ⊲⊳ h := g × h, as a vector space, with the bracket defined for any g, h ∈ g and x, y ∈ h by [(g, x), (h, y)] := (cid:0)[g, h] + x ⊲ h + g ↼ y, {x, y} + x ⊳ h + g ⇀ y(cid:1) (40) is a Leibniz algebra called the bicrossed product associated to the matched pair of Leibniz algebras (g, h, ⊳, ⊲, ↼, ⇀). This fact can be proved directly, but it can also be derived as a special case of Theorem 2.3. Indeed, let g be a Leibniz algebra and Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) an extending datum of g through V such that f is the trivial map, i.e. f (x, y) = 0, for all x, y ∈ V . Then, we can easily see that Ω(g, V ) = (cid:0)⊳, ⊲, ↼ , ⇀, {−, −}(cid:1) is a Leibniz extending structure of g through V if and only if (V, {−, −}) is a Leibniz algebra and (g, V, ⊳, ⊲, ↼, ⇀) is a matched pair of Leibniz algebras in the sense of Definition 4.5. In this case, the associated unified product g ⋉Ω(g,V ) V = g ⊲⊳ V is the bicrossed product of the matched pair (g, V, ⊳, ⊲, ↼, ⇀) as defined by (40). The bicrossed product of two Leibniz algebras is the construction responsible for the factorization problem, which is a special case of the ES problem and can be stated as follows: Let g and h be two given Leibniz algebras. Describe and classify all Leibniz algebras Ξ that factorize through g and h, i.e. Ξ contains g and h as Leibniz subalgebras such that Ξ = g + h and g ∩ h = {0}. Indeed, using Theorem 2.5 we can prove the following: Corollary 4.6. A Leibniz algebra Ξ factorizes through g and h if and only if there exists a matched pair of Leibniz algebras (g, h, ⊳, ⊲, ↼, ⇀) such that Ξ ∼= g ⊲⊳ h. Proof. To start with, notice that any bicrossed product g ⊲⊳ h factorizes through g ∼= g × {0} and h ∼= {0}×h. Conversely, assume that Ξ factorizes through g and h. Let p : Ξ → g be the k-linear projection of Ξ on g, i.e. p(g + x) := g, for all g ∈ g and x ∈ h. Now, we apply Theorem 2.5 for V := Ker(p) = h. Since V is a Leibniz subalgebra of E := Ξ, the map f = fp constructed in the proof of Theorem 2.5 is the trivial map as [x, y] ∈ V = Ker(p). Thus, the Leibniz extending structure Ω(g, V ) = (cid:0)⊳, ⊲, ↼, ⇀, f, {−, −}(cid:1) constructed in the proof of Theorem 2.5 is precisely a matched pair of Leibniz algebra and the unified product g ⋉ Ω(g,V ) V = g ⊲⊳ V is the bicrossed product of the matched pair (g, V, ⊳, ⊲, ↼, ⇀). Explicitly, the matched pair (g, h, ⊳ = ⊳p, ⊲ = ⊲p, ↼=↼p, ⇀=⇀p) is given by: x ⊲ g := p(cid:0)[x, g](cid:1), g ↼ x := p(cid:0)[g, x](cid:1), x ⊳ g := [x, g] − p(cid:0)[x, g](cid:1) g ⇀ x := [g, x] − p(cid:0)[g, x](cid:1) (41) (42) UNIFIED PRODUCTS FOR LEIBNIZ ALGEBRAS for all x ∈ h and g ∈ g. 25 (cid:3) From now on the matched pair constructed in (41) and (42) will be called the canonical matched pair associated to the factorization Ξ = g + h of Ξ through g and h. Based on Corollary 4.6 the factorization problem can be restated in a computational manner as follows: Let g and h be two given Leibniz algebras. Describe the set of all matched pairs (g, h, ⊳, ⊲, ↼, ⇀) and classify up to an isomorphism all bicrossed products g ⊲⊳ h. A detailed study of this problem will be given somewhere else. Example 4.7. Let g be the 3-dimensional Leibniz algebra considered in Example 3.11 and k be the 1-dimensional (abelian) Leibniz algebra. Then all bicrossed products g ⊲⊳ k can be explicitly described as a special case of Example 3.11. To this end we need to consider g0 := 0 and α := 0 in the unified products associated to all flag datums of the first kind provided in Example 3.11 and g0 := 0 in the unified products associated to all flag datums of the second kind. For instance, by taking d3 = 0 in the Leibniz algebra g of Example 3.11 we obtain the bicrossed product g ⊲⊳ k which is a 4-dimensional Leibniz algebra with the basis {e1, e2, e3, x} and the bracket given by: (ν0, d1, d2,,d3) 21 [e1, e3] = e2, [e3, e3] = e1, [x, e3] = d1 e1 + d2 e2 − ν0 x [e3, x] = −d1e1 − (d2 − ν −1 0 d1)e2 + ν0x, for all ν0 ∈ k∗ and d1, d2 ∈ k. This Leibniz algebra is the bicrossed product associated to the following matched pair (g, k, ⊳, ⊲, ↼, ⇀): x ⊳ e3 := −ν0 x, e3 ⇀ x := ν0x, x ⊲ e3 := d1e1 + d2e2, e3 ↼ x := −d1e1 + (−d2 + ν −1 0 d1)e2 where the undefined actions are zero and x is a basis of k. Another example of a matched pair of Leibniz algebras and the corresponding bicrossed product will be given in Example 5.6. 5. Classifying complements for extensions of Leibniz algebras This section is devoted to the classifying complements (CC) problem whose statement was given in the Introduction. Let g ⊆ Ξ be a Leibniz subalgebra of Ξ. A Leibniz subalgebra h of Ξ is called a complement of g in Ξ (or a g-complement of Ξ) if Ξ = g + h and g ∩ h = {0}. If h is a complement of g in Ξ, Corollary 4.6 shows that Ξ ∼= g ⊲⊳ h, where g ⊲⊳ h is the bicrossed product associated to the canonical matched pair of the factorization Ξ = g + h as constructed in (41) and (42). We denote by F(g, Ξ) the (possibly empty) isomorphism classes of all g-complements of Ξ. The factorization index of g in Ξ is defined by [Ξ : g]f := F(g, Ξ) as a numerical measure of the (CC) problem. Definition 5.1. Let (g, h, ⊲, ⊳, ↼, ⇀) be a matched pair of Leibniz algebras. A linear map r : h → g is called a deformation map of the matched pair (g, h, ⊲, ⊳, ↼, ⇀) if the following compatibility holds for any x, y ∈ h: r(cid:0)[x, y](cid:1) −(cid:2)r(x), r(y)(cid:3)= x ⊲ r(y) + r(x) ↼ y − r(cid:0)x ⊳ r(y) + r(x) ⇀ y(cid:1) (43) 26 A. L. AGORE AND G. MILITARU We denote by DM (h, g (⊲, ⊳, ↼, ⇀)) the set of all deformation maps of the matched pair (g, h, ⊲, ⊳, ↼, ⇀). The trivial map r(x) = 0, for all x ∈ h, is of course a deformation map. The right hand side of (43) measures how far r : h → g is from being a Leibniz algebra map. Using this concept which will play a key role in solving the (CC) problem, we introduce the following deformation of a Leibniz algebra: Theorem 5.2. Let g be a Leibniz subalgebra of Ξ, h a given g-complement of Ξ and r : h → g a deformation map of the associated canonical matched pair (g, h, ⊲, ⊳, ↼, ⇀). (1) Let fr : h → Ξ = g ⊲⊳ h be the k-linear map defined for any x ∈ h by: fr(x) = (r(x), x) Then eh := Im(fr) is a g-complement of Ξ. (2) hr := h, as a vector space, with the new bracket defined for any x, y ∈ h by: [x, y]r := [x, y] + x ⊳ r(y) + r(x) ⇀ y (44) algebras. (40) (43) of g ⊲⊳ h = Ξ. Indeed, for all x, y ∈ h we have: is a Leibniz algebra called the r-deformation of h. Furthermore, hr ∼= eh, as Leibniz Proof. (1) To start with, we will prove thateh = {(cid:0)r(x), x(cid:1) x ∈ h} is a Leibniz subalgebra = (cid:16)(cid:2)r(x), r(y)(cid:3)+x ⊲ r(y) + r(x) ↼ y, [x, y] + x ⊳ r(y) + r(x) ⇀ y(cid:17) (cid:2)(r(x), x), (r(y), y)(cid:3) = (cid:16)r([x, y] + x ⊳ r(y) + r(x) ⇀ y), [x, y] + x ⊳ r(y) + r(x) ⇀ y(cid:17) i.e. (cid:2)(r(x), x), (r(y), y)(cid:3)∈ eh. Moreover, it is straightforward to see that g ∩ eh = {0} and (g, x) = (cid:0)g − r(x), 0(cid:1)+(cid:0)r(x), x(cid:1)∈ g +eh for all g ∈ g, x ∈ h. Here, we view g ∼= g × {0} as a subalgebra of g ⊲⊳ h. Therefore, eh is a g-complement of Ξ = g ⊲⊳ h. (2) We denote by efr : h → eh the linear isomorphism induced by fr. We will prove that efr is also a Leibniz algebra map if we consider on h the bracket given by (44). Indeed, efr(cid:0)[x, y]r(cid:1) (44) = efr(cid:0)[x, y] + x ⊳ r(y) + r(x) ⇀ y(cid:1) = (cid:16)r(cid:0)[x, y] + x ⊳ r(y) + r(x) ⇀ y(cid:1), [x, y] + x ⊳ r(y) + r(x) ⇀ y(cid:17) = (cid:16)[r(x), r(y)] + x ⊲ r(y) + r((x) ↼ y, [x, y] + x ⊳ r(y) + r(x) ⇀ y(cid:17) = [(r(x), x), (r(y), y)] = [efr(x), efr(y)] for any x, y ∈ h we have: (43) (40) Therefore, hr is a Leibniz algebra and the proof is now finished. (cid:3) The following is the converse of Theorem 5.2: it proves that all g-complements of Ξ are r-deformations of a given complement. UNIFIED PRODUCTS FOR LEIBNIZ ALGEBRAS 27 Theorem 5.3. Let g be a Leibniz subalgebra of Ξ, h a given g-complement of Ξ with the associated canonical matched pair of Leibniz algebras (g, h, ⊲, ⊳, , ↼, ⇀). Then h is a g-complement of Ξ if and only if there exists an isomorphism of Leibniz algebras h ∼= hr, for some deformation map r : h → g of the matched pair (g, h, ⊲, ⊳, ↼, ⇀). Proof. Let h be an arbitrary g-complement of Ξ. Since Ξ = g ⊕ h = g ⊕ h we can find four k-linear maps: u : h → g, v : h → h, t : h → g, w : h → h such that for all x ∈ h and y ∈ h we have: x = u(x) ⊕ v(x), y = t(y) ⊕ w(y) (45) By an easy computation it follows that v : h → h is a linear isomorphism of vector spaces. We denote by v : h → g ⊲⊳ h the composition: v : h v−→ h i ֒→ Ξ = g ⊲⊳ h = (cid:0)−u(x), x(cid:1), for all x ∈ h. Then we shall prove that r := −u Therefore, we have v(x) is a deformation map and h ∼= hr. Indeed, h = Im(v) = Im(v) is a Leibniz subalgebra of Ξ = g ⊲⊳ h and we have: (45) [(cid:0)r(x), x(cid:1), (cid:0)r(y), y(cid:1)] (40) = (cid:0)[r(x), r(y)] + x ⊲ r(y) + r(x) ↼ y, [x, y] + x ⊳ r(y) + r(x) ⇀ y(cid:1) = (cid:0)r(z), z(cid:1) for some z ∈ h. Thus, we obtain: r(z) = [r(x), r(y)] + x ⊲ r(y) + r(x) ↼ y, z = [x, y] + x ⊳ r(y) + r(x) ⇀ y (46) By applying r to the second part of (46) it follows that r is a deformation map of the matched pair (g, h, ⊲, ⊳, ↼, ⇀). Furthermore, (46) and (44) show that v : hr → h is also a Leibniz algebra map which finishes the proof. (cid:3) In order to provide the classification of all complements we introduce the following: Definition 5.4. Let (g, h, ⊲, ⊳, ↼, ⇀) be a matched pair of Leibniz algebras. Two de- formation maps r, R : h → g are called equivalent and we denote this by r ∼ R if there exists σ : h → h a k-linear automorphism of h such that for any x, y ∈ h: σ(cid:0)[x, y](cid:1)−(cid:2)σ(x), σ(y)(cid:3)= σ(x) ⊳ R(cid:0)σ(y)(cid:1)+R(cid:0)σ(x)(cid:1)⇀ σ(y) − σ(cid:0)x ⊳ r(y)(cid:1)−σ(cid:0)r(x) ⇀ y(cid:1) To conclude this section, the following result provides the answer to the (CC) problem for Leibniz algebras: Theorem 5.5. Let g be a Leibniz subalgebra of Ξ, h a g-complement of Ξ and (g, h, ⊲, ⊳, ↼ , ⇀) the associated canonical matched pair. Then ∼ is an equivalence relation on the set DM (h, g (⊲, ⊳, ↼, ⇀)) and the map HA2(h, g (⊲, ⊳, ↼, ⇀)) := DM (h, g (⊲, ⊳, ↼, ⇀))/ ∼ −→ F(g, Ξ), r 7→ hr 28 A. L. AGORE AND G. MILITARU is a bijection between HA2(h, g (⊲, ⊳, ↼, ⇀)) and the isomorphism classes of all g- complements of Ξ. In particular, the factorization index of g in Ξ is computed by the formula: [Ξ : g]f = HA2(h, g (⊲, ⊳, ↼, ⇀)) Proof. Follows from Theorem 5.3 taking into account the fact that two deformation maps r and R are equivalent in the sense of Definition 5.4 if and only if the corresponding Leibniz algebras hr and hR are isomorphic. (cid:3) Example 5.6. Let h be the abelian Lie algebra of dimension 2 with basis {f1, f2} and g the Lie algebra with basis {e1, e2} and the bracket: [e2, e1] = −[e1, e2] = e2. Then there exists a matched pair of Leibniz algebras (g, h, ⊳, ⊲, ↼, ⇀), where the non-zero values of the actions are given as follows: f1 ⊳ e1 := f1, f2 ⊳ e1 := f2, f1 ⊲ e1 := e2, e1 ↼ f1 := −e2, e1 ⇀ f1 := −f1 The bicrossed product Ξ = g ⊲⊳ h associated to this matched pair is the following 4- dimensional Leibniz algebra having {e1, e2, f1, f2} as a basis and the bracket given by: [e2, e1] = −[e1, e2] = e2, [f1, e1] = f1 + e2, [e1, f1] = −f1 − e2, [f2, e1] = f2 Furthermore, the deformation maps associated with the above matched pair of Leibniz algebras are given as follows: r(γ, δ) r(α, β) : h → g, : h → g, r(f1) = γe2, r(f1) = αe2 + βe1, r(f2) = δe2 r(f2) = 0 for some scalars α, β, γ, δ ∈ k. One can easily see that hr(γ, δ) coincides with the Lie algebra h for all γ, δ ∈ k while hr(α, β) has the bracket given by: [f1, f2]r(α, β) = [f1, f1]r(α, β) = [f2, f2]r(α, β) = 0, [f2, f1]r(α, β) = βf2 Therefore, if β = 0 then hr(α, β) again coincides with h. If β 6= 0, then for any α ∈ k and β ∈ k∗, the Leibniz algebra hr(α, β) is isomorphic to the Leibniz algebra k with basis {F1, F2} and the bracket given by: [F1, F1] = F2, [F2, F1] = F2. The isomorphism is given by: ψ(F1) := f1 + f2, ψ(F2) := βf2. Since obviously k is ψ : k → hr(α, β) not isomorphic to the abelian Lie algebra h we obtain that the extension g ⊆ Ξ has factorization index [Ξ : g]f = 2. References [1] Agore, A.L. and Militaru, G. - Extending structures II: the quantum version, J. Algebra 336 (2011), 321 -- 341. [2] Agore, A.L. and Militaru, G. - Classifying complements for Hopf algebras and Lie algebras, J. Algebra 391 (2013), 193 -- 208. http://dx.doi.org/10.1016/j.jalgebra.2013.06.012. [3] Agore, A.L. and Militaru, G. - Extending structures I: the level of groups, to appear in Algebr. Represent. Theory, http://dx.doi.org/10.1007/s10468-013-9420-4, in press, arXiv:1011.1633. [4] Agore, A.L. and Militaru, G. - Extending structures for Lie algebras, to appear in Monatshefte fur Mathematik, http://dx.doi.org/10.1007/s00605-013-0537-7, in press, arXiv:1301.5442. [5] Albeverio, S., Omirov, B.A and Rakhimov, I.S. - Varieties of nilpotent complex Leibniz algebras of dimension less than five, Comm. Algebra 33 (2005) 1575-1585. UNIFIED PRODUCTS FOR LEIBNIZ ALGEBRAS 29 [6] Albeverio, S., Ayupov, Sh.A., Omirov, B.A. and Khudoyberdiyev, A.Kh. - n-Dimensional filiform Leibniz algebras of length (n − 1) and their derivations, J. Algebra, 319 (2008) 2471-2488. [7] Barnes, D. - On Levis Theorem for Leibniz algebras, Bull. Australian Math. Soc., 86 (2012), 184-185. [8] Bloh, A.M. - On a generalization of the concept of Lie algebra, Dokl. Akad. Nauk SSSR, 165(1965), 471-473. [9] Calderon Martin, A, J. and Sanchez Delgado, J. M. - On split Leibniz algebras, Linear Algebra Appl., 436 (2012), 1651 -- 1663. [10] Camacho, L.M., Canetea, E.M., G´omez, J.R. and Redjepov, Sh.B. - Leibniz algebras of nilindex n − 3 with characteristic sequence (n − 3, 2, 1), Linear Algebra Appl., 438 (2013), 1832 -- 1851. [11] Canete, E.M and Khudoyberdiyev, A. Kh. - The classification of 4-dimensional Leibniz algebras, arXiv:1301.7665. [12] Casas, J.M. and Corral, N. - On universal central extensions of Leibniz algebras, Comm. Algebra, 37 (2009) 2104-2120. [13] Casas, J.M., Insua, M.A., Ladra, M. and Ladra S. - An algorithm for the classification of 3- dimensional complex Leibniz algebras, Linear Algebra Appl., 436 (2012), 3747 -- 3756. [14] Cuvier, C. - Alg`ebres de Leibniz: d´efinitions, propri´et´es, Ann. Scient. Ec. Norm. Sup., 27 (1994), 1 -- 45. [15] Fialowski, M. and Mandal, A. - Leibniz algebra deformations of a Lie algebra, J. Math. Physics, 49(2008), 093512, 10 pages. [16] Gorbatsevich, V. V. - On some basic properties of Leibniz algebras, arXiv:1302.3345. [17] Hu, N., Pei, Y. and Liu, D. - A cohomological characterization of Leibniz central extensions of Lie algebras, Proc. AMS, 136(2008), 437-447. [18] Kinyon, M. K. and Weinstein, A. - Leibniz algebras, Courant algebroids, and multiplications on reductive homogeneous spaces, Amer. J. Math., 123(2001), 525 -- 550. [19] Kurdiani, R. and Pirashvilli, T. - A Leibniz algebra structure on the second tensor power, J. of Lie Theory, 12(2002), 583 -- 596. [20] Loday, J.-L. - Une version non commutative des alg`ebres de Lie: les alg`ebres de Leibniz, L'Enseignement Math., 39 (1993), 269 -- 293. [21] Loday, J.-L. and Pirashvili, T. - Universal enveloping algebras of Leibniz algebras and (co)homology, Math. Ann., 296(1993), 139 -- 158. [22] Lu, J.H. and Weinstein, A. - Poisson Lie groups, dressing transformations and Bruhat decomposi- tions, J. Differential Geom., 31(1990), 501 -- 526. [23] Majid, S. - Physics for algebraists: non-commutative and non-cocommutative Hopf algebras by a bicrossproduct construction, J. Algebra, 130 (1990), 17 -- 64. [24] Majid, S. - Foundations of quantum groups theory, Cambridge University Press, 1995. [25] Mason, G. and Yamskulna, G. - Leibniz algebras and Lie algebras, arXiv:1201.5071. [26] Rakhimov, I.S. and Hassan, M.A. - On isomorphism criteria for Leibniz central extensions of a linear deformation of µn, Internat. J. Algebra Comput., 21 (2011) 715-729. [27] Rakhimov, I.S. and Said Husain, S.K. - Classification of a subclass of nilpotent Leibniz algebras, Linear and Multilinear Algebra, 59 (2011) 339-354. Faculty of Engineering, Vrije Universiteit Brussel, Pleinlaan 2, B-1050 Brussels, Belgium E-mail address: [email protected] and [email protected] Faculty of Mathematics and Computer Science, University of Bucharest, Str. Academiei 14, RO-010014 Bucharest 1, Romania E-mail address: [email protected] and [email protected]
1805.04267	2	1805	2019-07-08T16:06:59	Commutative post-Lie algebra structures on Kac--Moody algebras	[ "math.RA" ]	We determine commutative post-Lie algebra structures on some infinite-dimensional Lie algebras. We show that all commutative post-Lie algebra structures on loop algebras are trivial. This extends the results for finite-dimensional perfect Lie algebras. Furthermore we show that all commutative post-Lie algebra structures on affine Kac--Moody Lie algebras are "almost trivial".	math.RA	math	COMMUTATIVE POST-LIE ALGEBRA STRUCTURES ON KAC -- MOODY ALGEBRAS DIETRICH BURDE AND PASHA ZUSMANOVICH ABSTRACT. We determine commutative post-Lie algebra structures on some infinite-dimensional Lie al- gebras. We show that all commutative post-Lie algebra structures on loop algebras are trivial. This extends the results for finite-dimensional perfect Lie algebras. Furthermore we show that all commutative post-Lie algebra structures on affine Kac -- Moody Lie algebras are "almost trivial". 1. INTRODUCTION Recently there is a surge of interest in so-called post-Lie algebras and post-Lie algebra structures. One origin comes from the study of geometric structures on Lie groups, where post-Lie algebras arise as a common generalization of pre-Lie algebras [13, 14, 19, 2, 3, 4] and LR-algebras [5, 6]. Here pre- Lie algebras, also called left-symmetric algebras, Vinberg algebras, or Koszul -- Vinberg algebras, have been studied intensively before. For a survey, see [4]. On the other hand, post-Lie algebras have been introduced by Vallette [22] in 2007 in connection with the homology of partition posets and the study of Koszul operads. Then they were studied by several authors in various contexts, e.g., for algebraic op- erad triples [17], in connection with modified Yang -- Baxter equations, Rota -- Baxter operators, universal enveloping algebras, double Lie algebras, R-matrices, isospectral flows, Lie -- Butcher series and many other topics [1, 10, 12]. Concerning post-Lie algebra structures on pairs of Lie algebras (g, n), the existence question and In [7] we the classification is of particular interest. There have been many results obtained so far. introduced a special class of post-Lie algebra structures, namely commutative ones. In this case, the two Lie algebras g and n coincide, and we obtain a bilinear commutative product satisfying a certain compatibility condition with the Lie bracket, which can be considered as a generalization of the left- symmetric identity (for a precise definition, see the introductory section below). Commutative post-Lie algebra structures, CPA-structures in short, are much more tractable, and we have obtained several existence and classification results [7, 8, 9]. Among other things we proved in [8] that any commutative post-Lie algebra structure on a finite-dimensional perfect Lie algebra over field of characteristic zero is trivial. Moreover we classified CPA-structures on certain classes of nilpotent Lie algebras. It is natural to study CPA-structures also for infinite-dimensional Lie algebras. In [20] and [21] this has been done already for the two-sided infinite-dimensional Witt algebra and some of its generalizations. We want to continue these investigations in this paper. We will prove that CPA-structures on loop algebras are trivial, and "almost trivial" on Kac -- Moody algebras; see Theorem 1 and Theorem 3 for exact formulations. 2. DEFINITIONS, NOTATIONS AND CONVENTIONS Let A be a nonassociative algebra over a field K in the sense of Schafer [18], with K-bilinear product A × A → A, (a, b) 7→ ab. We will assume that K is an arbitrary field of characteristic different from 2, if not said otherwise. Consider bilinear maps ϕ : A × A → A such that for any a ∈ A, the linear map Date: First written May 10, 2018; last minor revision May 6, 2019. 2010 Mathematics Subject Classification. 17B67. Key words and phrases. Commutative post-Lie algebra; loop algebra; Kac -- Moody algebra. Comm. Algebra, to appear; arXiv:1805.04267. 1 COMMUTATIVE POST-LIE ALGEBRA STRUCTURES ON KAC -- MOODY ALGEBRAS 2 ϕ(a, ·) : A → A is a derivation of A. In other words, ϕ(a, bc) = ϕ(a, b)c + bϕ(a, c) for any a, b, c ∈ A. The set of such bilinear maps forms a vector space which will be denoted by D(A), and the subspace of such symmetric maps will be denoted by Dcomm(A). Recall that a commutative post-Lie algebra structure (or, CPA-structure) on a Lie algebra L is a new binary multiplication · on L which lies in Dcomm(L), i.e., satisfying and x · y = y · x x · [y, z] = [x · y, z] − [x · z, y] and which, additionally, satisfies the condition [x, y] · z = x · (y · z) − y · (x · z) for any x, y, z ∈ L. When evaluating CPA-structures on various classes of Lie algebras, it will be conve- nient to write this new multiplication as a bilinear map ϕ : L × L → L, i.e., the previous three conditions are written in the form (1) (2) (3) for any x, y, z ∈ L. ϕ(x, y) = ϕ(y, x) ϕ(x, [y, z]) = [ϕ(x, y), z] − [ϕ(x, z), y] ϕ([x, y], z) = ϕ(x,ϕ(y, z)) − ϕ(y,ϕ(x, z)) For a Lie algebra L, Z(L) denotes the center of L. If Z(L) = 0, then L is called centerless. If [L, L] = L, then L is called perfect. All unadorned tensor products are over the base field K. The symbol ⊕ denotes the direct sum in the category of vector spaces. 3. TWISTED LOOP ALGEBRAS Given a Lie algebra L and a commutative associative algebra A, the current Lie algebra L ⊗ A carries a multiplication uniquely defined by the formula where x, y ∈ L, and a, b ∈ A. In the particular case A = K[t, t−1], the Laurent polynomial algebra, we speak of the (untwisted) loop Lie algebra associated with L. [x ⊗ a, y ⊗ b] = [x, y] ⊗ ab, Now let (4) L = Mi∈Z/nZ Li be a Z/nZ-graded Lie algebra, and consider the Lie algebra (5) (Li(mod n) ⊗ t i). bL =Mi∈Z This subalgebra of the loop Lie algebra L ⊗ K[t, t−1] will be called a twisted loop Lie algebra associated to the graded Lie algebra L. The untwisted case is formally included in the twisted one, when n = 1 and the grading (4) consists of the single zero component. The direct sum (5) is a Z-grading, what will be crucial in what follows. More generally, let L = Lg∈G Lg be a Lie algebra graded by an abelian group G. Then both vector spaces D(L) and Dcomm(L) inherit a G-grading from L. Indeed, let us say that a bilinear map ϕ : L × L → L has degree g ∈ G, or, symbolically, degϕ = g, if ϕ(Lh, L f ) ⊆ Lh+ f +g for any h, f ∈ G. Then: COMMUTATIVE POST-LIE ALGEBRA STRUCTURES ON KAC -- MOODY ALGEBRAS 3 Proposition 1. For an arbitrary G-graded Lie algebra L, we have D(L) =Mg∈G Dcomm(L) =Mg∈G {ϕ ∈ D(L) degϕ = g}, {ϕ ∈ Dcomm(L) degϕ = g}. The proof is standard, and follows almost verbatim the proof of a similar statement for derivations. See, for example, Proposition 1.1 in [11]. In the sequel, it will be convenient to make use of the following auxiliary definition. Let us say that the Lie algebra L satisfies the condition () if any ϕ ∈ Dcomm(L) such that ϕ(x,ϕ(y, z)) = ϕ(y,ϕ(x, z)) for any x, y, z ∈ L, vanishes. Proposition 2. Let L be a Z/nZ-graded Lie algebra satisfying the condition (), and bL the twisted loop algebra associated to L. Then there is a bijection between the sets of CPA-structures on bL and CPA-structures on L. Namely, any CPA-structure onbL is of the form (x ⊗ t i, y ⊗ t j) 7→ ϕ(x, y) ⊗ t i+ j, (6) for any x ∈ Li(mod n), y ∈ L j(mod n) and i, j ∈ Z, where ϕ is a CPA-structure on L. (7) Proof. Let Φ be a CPA-structure onbL. By Proposition 1, where Φℓ is an element of Dcomm(bL) of degree ℓ, i.e. Φ = ∑ ℓ Φℓ, Φℓ(x ⊗ t i, y ⊗ t j) = ϕℓ(x, y) ⊗ t i+ j+ℓ for any x ∈ Li(mod n), y ∈ L j(mod n), and some bilinear map ϕℓ : L × L → L of degree ℓ(mod n). The commutativity of Φℓ is equivalent to the commutativity of ϕℓ, and the condition (2) for Φℓ is equivalent to the same condition for ϕℓ, whence ϕℓ ∈ Dcomm(L) for any ℓ. The condition (3) for Φ is equivalent to ϕℓ(x, [y, z]) ⊗ t i+ j+k+ℓ = ∑ (8) ∑ ℓ ∑ s (cid:0)ϕm(x,ϕs(y, z)) − ϕs(y,ϕm(x, z))(cid:1) ⊗ t i+ j+k+m+s m for any x ∈ Li(mod n), y ∈ L j(mod n), z ∈ Lk(mod n), and i, j, k ∈ Z. Assume the sum (7) contains elements of positive degree, and let N be the largest such degree. The maximal possible degree of a summand at the left-hand side of the equality (8) is i + j + k + N, while at the right-hand side the summand (9) ϕN(x,ϕN(y, z)) − ϕN(y,ϕN(x, z)) has degree i + j + k + 2N, and all other summands have a smaller degree. Consequently, the expression (9) vanishes for any x, y, z belonging to arbitrary homogeneous components of L, and hence vanishes for any x, y, z ∈ L. But then ϕN vanishes, a contradiction. The same reasoning shows that the sum (7) does not contain summands of negative degree, and hence Φ is of degree zero, i.e. of the form (6). In this situation, the condition (3) for Φ is equivalent to the condition (3) for ϕ. (cid:3) Remark. An analogous result can be obtained by replacing the Laurent polynomial algebra by the (ordinary) polynomial algebra K[t], or any similar graded polynomial-like algebra. Lemma 1. Let L be a Lie algebra such that: (i) L is centerless; (ii) all derivations of L are inner; (iii) any linear map ω from L to an abelian subalgebra of L, satisfying the condition for any x, y ∈ L, vanishes. [ω(x), y] + [x,ω(y)] = 0 COMMUTATIVE POST-LIE ALGEBRA STRUCTURES ON KAC -- MOODY ALGEBRAS 4 Then L satisfies the condition (). Proof. Since all derivations of L are inner, any map ϕ ∈ Dcomm(L) has the form ϕ(x, y) = [y,ω(x)] for x, y ∈ L and some linear map ω : L → L. Since ϕ is symmetric, we have [ω(x), y] + [x,ω(y)] = 0. The condition ϕ(x,ϕ(y, z)) = ϕ(y,ϕ(x, z)) is equivalent then to [[z,ω(y)],ω(x)] = [[z,ω(x)],ω(y)] what, together with the Jacobi identity and the fact that L is centerless, implies [ω(x),ω(y)] = 0, i.e. ω(L) is an abelian subalgebra in L. Now (iii) implies that ω, and hence ϕ, vanishes, i.e. L satisfies the condition (). (cid:3) Now we can prove one of our main results. Theorem 1. Let g = Mi∈Z/nZ gi be a Z/nZ-graded simple finite-dimensional complex Lie algebra. Then any CPA-structure on the asso- ciated twisted loop algebrabg vanishes. Proof. The Lie algebra g satisfies the conditions of Lemma 1. Indeed, (i) is evident and (ii) is well- known. According to Lemma 6.1 in [16], any linear map ω : g → g satisfying [ω(x), y]+[x,ω(y)] = 0 for any x, y ∈ g, vanishes, thus (iii) is satisfied. Therefore, g satisfies the condition (), and by Proposition 2 the set of CPA-structures onbg is in bijection with the set of CPA-structures on g. But any CPA-structure on g vanishes according to Proposition 5.4 of [7], or Proposition 3.1 of [8]. (cid:3) Remarks. (i) On practice, only the cases n = 1 (the untwisted case) and n = 2, 3 (the twisted case) may occur; see Chapter 8 in [15]. (ii) According to Theorem 3.3 of [8], CPA-structures vanish not only on g, but on any perfect finite- dimensional Lie algebra over a field of characteristic zero. The twisted loop algebrabg is perfect, but not finite-dimensional, so Theorem 1 is not covered, at least in a straightforward way, by that result. 4. DIGRESSION: GRADED LIE ALGEBRAS, WITT ALGEBRAS, AND CURRENT ALGEBRAS This section contains some comments on and alternative approaches to the results of the previous section. The proofs are omitted, and the results stated here will be not used in the next section. In the previous section we took advantage of the graded structure of twisted loop algebras, which, being coupled with the nonlinear condition (3), implies a strong restriction on CPA-structures. One of the possible generalizations along these line is: Proposition 3. Let L be a Z-graded Lie algebra L =Li∈Z Li satisfying the condition (). Then any CPA-structure ϕ on L is of degree 0: ϕ(Li, L j) ⊆ Li+ j for any i, j ∈ Z. The proof is straightforward and is similar to the proof of Proposition 2. In the situation of twisted loop algebras it was more convenient to use a somewhat more specific Proposition 2, and not this general result. However, Proposition 3 can be used to establish the vanishing of CPA-structures on Witt algebras in a somewhat different way than this was done in [20]. Recall that the Witt algebras are defined as Lie algebras over a field K of characteristic zero, having a set of basis vectors {ei} with multiplication [ei, e j] = ( j − i)ei+ j. Depending on whether the indices run over all integers, or over integers ≥ −1, we get the two-sided or one-sided Witt algebra, respectively. Theorem 2. Any CPA-structure on a Witt algebra (one- or two-sided) vanishes. The proof is obtained by an easy combination of reasonings as in the proof of Corollary 1, the facts that all derivations of a Witt algebra are inner, and all abelian subalgebras are one-dimensional, and Proposition 3. COMMUTATIVE POST-LIE ALGEBRA STRUCTURES ON KAC -- MOODY ALGEBRAS 5 Picking up another thread in §3, let us outline an alternative approach to the proof of Theorem 1; it takes advantage of the fact that one of the defining conditions of the post-Lie algebra is linear, and employs a linear-algebraic technique from [23], used earlier in [23] and [24] to describe other kinds of linear structures on current Lie algebras, such as derivations, low-degree cohomology, Poisson struc- tures, etc., in terms of some invariants of the tensor factors. Before we formulate the corresponding statement, a few definitions are in order. Recall that the centroid of a Lie algebra L is the space of linear maps ϕ : L → L commuting with inner derivations of L, i.e. ϕ([x, y]) = [ϕ(x), y] for any x, y ∈ L. A Lie algebra is called central, if its centroid coincides with multiplications by the elements of the base field. The set of all bilinear maps ϕ : L × L → L such that for any x ∈ L, the linear map ϕ(x, ·) : L → L belongs to the centroid of L, i.e., for any x, y, z ∈ L, forms a vector space denoted by C(L). For a vector space V , End(V ) denotes the set of all linear maps V → V . ϕ(x, [y, z]) = [ϕ(x, y), z] Proposition 4. Let L be a centerless Lie algebra, A an associative commutative algebra with unit, and one of L, A is finite-dimensional. Then (10) D(L ⊗ A) ≃(cid:16)D(L) ⊗ End(A)(cid:17) ⊕(cid:16)C(L) ⊗ D(A)(cid:17). Each element of D(L ⊗ A) can be written as a sum of maps of the form ϕ⊗α, where ϕ : L × L → L and α : A × A → A are bilinear maps, of the two kinds: (i) ϕ ∈ D(L), and α(a, b) = β(a)b for any a, b ∈ A and some linear map β : A → A; (ii) ϕ ∈ C(L), and α ∈ D(A). The proof is very similar to the proof of the formula from Corollary 2.2 in [23] expressing derivations of a current Lie algebra in terms of its tensor factors. The condition that L is centerless is not crucial and can be removed at the expense of more laborious computations and cumbersome formulas. Similarly, we have: Proposition 5. In the setup of Proposition 4, assume additionally that L is central. Then Dcomm(L ⊗ A) ≃ Dcomm(L) ⊗ A. Each element of Dcomm(L ⊗ A) can be written as the sum of maps of the form ϕ ⊗ α, where ϕ ∈ Dcomm(L), and α : A × A → A is a bilinear map of the form α(a, b) = abu for some u ∈ A. Then, using Proposition 5, we may impose on Dcomm(L ⊗ A) the additional condition (3) to try to get an analogous formula for the set of CPA-structures on L ⊗ A. However, the nonlinearity of (3) makes the task much more difficult, and we seemingly have to abandon the generality of Propositions 4 and 5, and assume A to be a more or less concrete algebra. This provides a somewhat alternative way to the results of §3, at least in the nontwisted case. We use the well-known realization of untwisted and twisted affine Kac -- Moody algebras as extensions of current Lie algebras by a derivation and a central element: 5. KAC -- MOODY ALGEBRAS (11) d d t ⊕ Cz , bg ⊕ Ct wherebg denotes, as previously, the twisted loop algebra associated to a Z/nZ-graded simple finite-di- dt acts on the current Lie algebra g ⊗ C[t, t−1], mensional complex Lie algebra g. The Euler derivation t d COMMUTATIVE POST-LIE ALGEBRA STRUCTURES ON KAC -- MOODY ALGEBRAS 6 and, by restriction, onbg ⊆ g ⊗ C[t, t−1], as the derivation of the second tensor factor, i.e. where x ∈ L and i ∈ Z. The element z is central, and multiplication onbg is twisted by the Cz-valued "Kac -- Moody cocycle" (whose concrete form is not important for us here). d d t [x ⊗ t i, t ] = ix ⊗ t i, Lemma 2. Let L be a perfect centerless Lie algebra such that any CPA-structure on L vanishes, D an outer derivation of L, and L = L ⊕ KD be the semidirect sum with D acting on L. Then any CPA- structure on L vanishes. Proof. Let Φ be a CPA-structure on L . We may write Φ(x, y) = ϕ(x, y) + λ(x, y)D Φ(x, D) = ψ(x) + µ(x)D Φ(D, D) = a + ηD where x, y ∈ L, for some (bi)linear maps ϕ : L × L → L, λ : L × L → K, ψ : L → L, µ : L → K, and a ∈ L, η ∈ K. Then the symmetricity of Φ implies the symmetricity of ϕ, and the condition (2) for Φ, written for an arbitrary triple x, y, z ∈ L, is equivalent to ϕ(x, [y, z]) = [ϕ(x, y), z] − [ϕ(x, z), y] + λ(x, y)D(z) − λ(x, z)D(y), and to the condition λ(L, [L, L]) = 0. But since L is perfect, λ vanishes, and hence ϕ ∈ Dcomm(L). Imposing on Φ the condition (3) leads to the conclusion that the restriction ϕ = ΦL is a CPA-structure on L, and hence vanishes. The condition (2) for Φ, written for the triple D, x, y, where x, y ∈ L, implies µ([L, L]) = 0, whence µ vanishes. Then writing the same condition for the triple x, D, y, and taking into account all the vanishing conditions obtained so far, we get ψ(L) ⊆ Z(L) = 0, whence ψ vanishes. Finally, the condition (2) for Φ, written for the triple D, D, x ∈ L, yields ηD = ad a. But since D is an (cid:3) outer derivation, η = 0, a ∈ Z(L) = 0, and Φ vanishes identically. Lemma 3. Let L be a centerless Lie algebra such that any CPA-structure on L vanishes, and L = L⊕Kz a nontrivial one-dimensional central extension of L. Then the set of CPA-structures on L consists of Kz-valued symmetric bilinear maps vanishing whenever one of the arguments belongs to [L, L] ⊕ Kz. Proof. Write the Lie bracket on L as {x, y} = [x, y] + ξ(x, y)z, where x, y ∈ L, [ · , · ] is a Lie bracket on L, and ξ is a (nontrivial) 2-cocycle on L. Let Φ be a CPA-structure on L . Similarly with the proof of Lemma 2, we may write Φ(x, y) = ϕ(x, y) + λ(x, y)z Φ(x, z) = ψ(x) + µ(x)z Φ(z, z) = a + ηz where x, y ∈ L, and all the maps and elements occurring at the right-hand side have the same meaning as in the proof of Lemma 2. The symmetricity of Φ implies the symmetricity of ϕ and λ. The condition (2) for Φ, written for the triple x, y, z, where x, y ∈ L, yields ψ(L) ⊆ Z(L) = 0, whence ψ vanishes. Then the same condition written for the triple z, x, y, yields a = 0 and ηξ(x, y) = −µ([x, y]). But since ξ is a nontrivial (i.e., not equal to a coboundary) cocycle, the latter equality implies η = 0. Further, the condition (2) for Φ, written for the triple x, y, t ∈ L, yields ϕ(x, [y, t]) = [ϕ(x, y), t] − [ϕ(x, t), y], i.e., ϕ ∈ Dcomm(L), and (12) λ(x, [y, t]) + ξ(y, t)µ(x) = ξ(ϕ(x, y), t) − ξ(ϕ(x, t), y). COMMUTATIVE POST-LIE ALGEBRA STRUCTURES ON KAC -- MOODY ALGEBRAS 7 The condition (3) for Φ, written for the triple x, y, t ∈ L, yields ϕ([x, y], t) = ϕ(x,ϕ(y, t)) − ϕ(y,ϕ(x, t)), i.e., ϕ is a CPA-structure on L, whence ϕ vanishes. The condition (3) for Φ, written for the triple z, x, x, where x ∈ L, yields µ(x)2 = 0, whence µ vanishes. To summarize: the only nonzero values of Φ are given by Φ(x, y) = λ(x, y)z, where x, y ∈ L. More- over, (12) now implies λ(L, [L, L]) = 0. These are exactly the maps as specified in the statement of the lemma. It is straightforward to check that conversely, any such map is a CPA-structure on L . (cid:3) Theorem 3. CPA-structures on an affine Kac -- Moody algebra written in the form (11), form a one- dimensional vector space, spanned by the map: (cid:18)t d d t(cid:19) 7→ z d d t , t all other pairs of basis vectors 7→ 0. Proof. As by Theorem 1 any CPA-structure on bg vanishes, and the Euler derivation t d nonzero derivation of the Laurent polynomials extended to the loop algebrabg) is an outer derivation of bg, Lemma 2 implies that any CPA-structure on the semidirect sumbg ⊕ Ct d turn, to apply Lemma 3 to L =bg ⊕ Ct d dt . Since [L, L] =bg, Lemma 3 yields the desired result. dt vanishes. This allows, in its (cid:3) dt (in fact, any ACKNOWLEDGEMENT Thanks are due to the referee for helpful remarks which improved the presentation. Dietrich Burde is supported by the Austrian Science Foundation FWF, grant P28079 and grant I3248. REFERENCES [1] C. Bai, L. Guo, X. Ni: Nonabelian generalized Lax pairs, the classical Yang-Baxter equation and PostLie algebras. Comm. Math. Phys. 297 (2010), no. 2, 553 -- 596; arXiv:0910.3262. [2] D. Burde: Affine structures on nilmanifolds. Intern. J. Math. 7 (1996), no. 5, 599 -- 616. [3] D. Burde, K. Dekimpe, S. Deschamps: The Auslander conjecture for NIL-affine crystallographic groups. Math. Ann. 332 (2005), no. 1, 161 -- 176; arXiv:math/0409476 [4] D. Burde: Left-symmetric algebras, or pre-Lie algebras in geometry and physics. Centr. Eur. J. Math. 4 (2006), no. 3, 323 -- 357; arXiv:math-ph/0509016. [5] D. Burde, K. Dekimpe, S. Deschamps: LR-algebras. New Developments in Lie Theory and Geometry (ed. C. S. Gordon et al.), Contemp. Math. 491 (2009), 125 -- 140; arXiv:0801.1280. [6] D. Burde, K. Dekimpe, K. Vercammen: Complete LR-structures on solvable Lie algebras. J. Group Theory 13 (2010), no. 5, 703 -- 719; arXiv:0906.1151. [7] D. Burde, K. Dekimpe: Post-Lie algebra structures on pairs of Lie algebras. J. Algebra 464 (2016), 226 -- 245; arXiv:1505.00955. [8] D. Burde, W. A. Moens: Commutative post-Lie algebra structures on Lie algebras. J. Algebra 467 (2016), 183 -- 201; arXiv:1512.05096. [9] D. Burde, K. Dekimpe, W. A. Moens: Commutative post-Lie algebra structures and linear equations for nilpotent Lie algebras. J. Algebra 526 (2019), 12 -- 29; arXiv:1711.01964. [10] K. Ebrahimi-Fard, A. Lundervold, I. Mencattini, H. Z. Munthe-Kaas: Post-Lie algebras and isospectral flows. SIGMA 11 (2015), Paper 093, 16 pp.; arXiv:1505.02436. [11] R. Farnsteiner: Derivations and central extensions of finitely generated graded Lie algebras. J. Algebra 118 (1988), no. 1, 33 -- 45. [12] V. Gubarev: Universal enveloping Lie Rota-Baxter algebra of pre-Lie and post-Lie algebras. J. Algebra 516 (2018), 298 -- 328; arXiv:1708.06747. [13] J. Helmstetter: Radical d'une alg`ebre sym´etrique a gauche. Ann. Inst. Fourier 29 (1979), no. 4, 17 -- 35. [14] H. Kim: Complete left-invariant affine structures on nilpotent Lie groups. J. Diff. Geom. 24 (1986), no. 3, 373 -- 394. [15] V. G. Kac: Infinite-dimensional Lie algebras. Third edition, Cambridge University Press, Cambridge, 1990, 400 pp. [16] G. F. Leger, E. M. Luks: Generalized derivations of Lie algebras. J. Algebra 228 (2000), no. 1, 165 -- 203. [17] J.-L. Loday: Generalized bialgebras and triples of operads. Ast´erisque 320 (2008), 116 pp.; arXiv:math/0611885. [18] R. D. Schafer: A introduction to nonassociative algebras. Dover Publications, Inc., New York, 1995, 166 pp. [19] D. Segal: The structure of complete left-symmetric algebras. Math. Ann. 293 (1992), no. 3, 569 -- 578. COMMUTATIVE POST-LIE ALGEBRA STRUCTURES ON KAC -- MOODY ALGEBRAS 8 [20] X. Tang: Biderivations, linear commuting maps and commutative post-Lie algebra structures on W-algebras. Comm. Algebra 45 (2017), no. 12, 5252 -- 5261. [21] X. Tang, Y. Yang: Biderivations of the higher rank Witt algebra without anti-symmetric condition. Open Math. 16 (2018), no. 1, 447 -- 452. [22] B. Vallette: Homology of generalized partition posets. J. Pure Appl. Algebra 208 (2007), no. 2, 699 -- 725. [23] P. Zusmanovich: Low-dimensional cohomology of current Lie algebras and analogs of the Riemann tensor for loop manifolds. Lin. Algebra Appl. 407 (2005), 71 -- 104; arXiv:math/0302334. [24] P. Zusmanovich: A compendium of Lie structures on tensor products. Zapiski Nauchnykh Seminarov POMI 414 (2013) (N.A. Vavilov Festschrift), 40 -- 81; reprinted in J. Math. Sci. 199 (2014), no. 3, 266 -- 288; arXiv:1303.3231. UNIVERSIT AT WIEN, WIEN, AUSTRIA E-mail address: [email protected] UNIVERSITY OF OSTRAVA, OSTRAVA, CZECH REPUBLIC E-mail address: [email protected]
1612.08244	1	1612	2016-12-25T07:39:57	Derived brackets for fat Leibniz algebras	[ "math.RA" ]	Given a Leibniz algebra L with left center Z, we work on C(L,Z,S(Z)), the Z-standard complex of L with coefficients in S(Z). We construct the derived bracket for a fat Leibniz algebra in terms of a certain 3-cocycle and a Poisson algebra structure on the space of so-called "representable cochains".	math.RA	math	DERIVED BRACKETS FOR FAT LEIBNIZ ALGEBRAS XIONGWEI CAI AND ZHANGJU LIU Abstract. Given a Leibniz algebra L with left center Z, we work on C(L, Z, S •(Z)), the Z- standard complex of L with coefficients in S •(Z). We construct the derived bracket for a fat Leibniz algebra in terms of a certain 3-cocycle and a Poisson algebra structure on the space of so-called "representable cochains". 6 1 0 2 c e D 5 2 ] . A R h t a m [ 1 v 4 4 2 8 0 . 2 1 6 1 : v i X r a 1. Introduction Leibniz algebras, objects that first appeared in Bloh's work [3] and named by Loday [9], can be viewed as the noncommutative analogue of Lie algebras. Some theorems and properties of Lie algebras have been proved to be still valid for Leibniz algebras, while many other questions are still open. Courant algebroids, first introduced by Liu, Weinstein and Xu in [8], can be viewed as the geometric realization of Leibniz algebras. The algebraization of Courant algebroids, Courant- Dorfman algebras, are special examples of Leibniz algebras. The derived bracket for a Lie algebra with an ad-invariant inner product is constructed by Lecomte-Roger [7] and Kosmann-Schwarzbach [5], in order to study the homological algebra of Lie bialgebras and quasi-Lie bialgebras, respectively. While the construction of derived bracket for a Courant algebroid was given by Kosmann-Schwarzbach [6], Royternberg [11] and Alekseev-Xu [1]. It is a natural question to ask whether there is a derived bracket construction for Leibniz algebras. In this paper, we succeed to give a positive answer for fat Leibniz algebras. By a fat Leibniz algebra, we mean a Leibniz algebra whose naturally defined symmetric product is non-degenerate. Note that this is a different notion from a quadratic Leibniz algebra, defined by Benayadi-Hidri [2]. Given a Leibniz algebra L with left center Z, we will work on the H-standard complex (see Cai [4]) of L in the particular case when H = Z, V = S•(Z). We will define a canonical 3-cocycle Θ and prove that the subcomplex consisting of the so-called "representable cochains" is a graded Poisson algebra. Finally we show that the Leibniz bracket of a fat Leibniz algebra can be represented by a derived bracket. Acknowledgements. This paper is based on the PhD dissertation of the first author, which is funded by the University of Luxembourg. The first author would like to thank his advisors, Prof. Martin Schlichenmaier and Prof. Ping Xu, for their continual encouragement and support. 2. Standard complex In this section, we recall the definition of H-standard complex of a Leibniz algebra L with coefficients in V ([4]), and consider a 3-cocycle in the particular case when H = Z, V = S•(Z). Given a Leibniz algebra L with left center Z, let H ⊇ Z be an isotropic ideal in L, and (V, τ ) be an H-trivial representation of L (i.e. a left representation of L on which H acts trivially). Key words and phrases. Leibniz algebras, representable cochains, derived brackets. 1 DERIVED BRACKETS FOR FAT LEIBNIZ ALGEBRAS 2 Denote by C n(L, H, V ) the space of all sequences ω = (ω0, · · · , ω[ n 2 ]), where ωk is a linear map from (⊗n−2kL) ⊗ (⊙kH) to V , ∀k, and is weakly skew-symmetric in arguments of L up to ωk+1: ωk(e1, · · · ei, ei+1, · · · en−2k; h1, · · · hk) + ωk(e1, · · · ei+1, ei, · · · en−2k; h1, · · · hk) = −ωk+1(· · · bei, dei+1, · · · ; (ei, ei+1), · · · ) ∀e ∈ L, h ∈ H C(L, H, V ) , Ln C n(L, H, V ) becomes a cochain complex under the coboundary map d = d0 + δ + d′, called the H-standard complex of L with coefficients in V , where d0, δ, d′ are defined for any ω ∈ C n(L, H, V ) respectively by: (d0ω)k(e1, · · · , en+1−2k; h1, · · · hk) , X +X (δω)k(e1, · · · , en+1−2k; h1, · · · hk) , X (d′ω)k(e1, · · · en+1−2k; h1, · · · hk) , X a<b a j (−1)a+1ρ(ea)ωk(· · · bea, · · · ; · · · ) (−1)aωk(· · · bea, · · · ea ◦ eb, · · · ; · · · ) ωk−1(αj , e1, · · · en+1−2k; · · · bhj, · · · ) (−1)a+1ωk(· · · bea, · · · ; · · · bhj, hj ◦ ea, · · · ). a,j The Leibniz bracket of L induces a left action ρ of L on Z: ρ(e)f , e ◦ f , ∀e ∈ L, f ∈ Z. And it can be extended by Leibniz rule to a left action of L on the symmetric tensor S•(Z), still denoted by ρ. (S•(Z), ρ) is obviously a Z-trivial representation of L, so we have the Z-standard complex (C(L, Z, S•(Z)), d). Note that d′ is 0, so d = d0 + δ in this case. Definition 2.1. (C(L, Z, S•(Z)), d = d0 + δ) is called the standard complex of L. For simplicity, we will denote C(L, Z, S•(Z)) by C(L) from now on. Proposition 2.2. C(L) is a differential graded commutative algebra, with the multiplication map defined for ω ∈ C n(L), η ∈ Cm(L) by: (ω · η)k(e1, · · · , en+m−2k; f1, · · · , fk) (2.1) , X (−1)σωi(eσ(1) · · · eσ(n−2i); fµ(1) · · · fµ(i))ηj (eσ(n−2i+1) · · · ; fµ(i+1) · · · ), i+j=k σ∈sh(n−2i,m−2j) µ∈sh(i,j) ∀e ∈ L, f ∈ Z, where sh( , ) means the shuffle permutation. Proof. The multiplication map above is obviously graded commutative, i.e. ω · η = (−1)nmη · ω. We give the proof in 3 steps. Step 1: DERIVED BRACKETS FOR FAT LEIBNIZ ALGEBRAS 3 C(L) is closed under the multiplication, i.e. ω · η ∈ C n+m(L): (ω · η)k(· · · ea, ea+1, · · · ; f1, · · · fk) + (ω · η)k(· · · ea+1, ea, · · · ; f1, · · · fk) = X X (−1)σ(cid:0)ωi(· · · ea, ea+1, · · · ; · · · ) + ωi(· · · ea+1, ea, · · · ; · · · )(cid:1)ηj(· · · ) i+j=k τ ∈sh(i,j) + X σ∈sh(n−2i,m−2j) σ−1 (a),σ−1 (a+1)≤n−2i X i+j=k σ∈sh(n−2i,m−2j) τ ∈sh(i,j) + X σ−1 (a),σ−1 (a+1)>n−2i X i+j=k σ∈sh(n−2i,m−2j) τ ∈sh(i,j) + X σ−1 (a)≤n−2i<σ−1 (a+1) X i+j=k τ ∈sh(i,j) σ∈sh(n−2i,m−2j) σ−1 (a+1)≤n−2i<σ−1 (a) (−1)σωi(· · · )(cid:0)ηj(· · · ea, ea+1, · · · ; · · · ) + ηi(· · · ea+1, ea, · · · ; · · · )(cid:1) (−1)σ(cid:0)ωi(· · · ea · · · )ηj(· · · ea+1 · · · ) + ωi(· · · ea+1 · · · )ηj (· · · ea · · · )(cid:1) (−1)σ(cid:0)ωi(· · · ea+1 · · · )ηj(· · · ea · · · ) + ωi(· · · ea · · · )ηj (· · · ea+1 · · · )(cid:1) (note that the same sequence (· · · ea, · · · ea+1, · · · ) viewed as permutations of (· · · ea, ea+1, · · · ) and (· · · ea+1, ea, · · · ) have opposite signs) = X X i+j=k τ ∈sh(i,j) + X σ∈sh(n−2i,m−2j) σ−1 (a),σ−1 (a+1)≤n−2i X i+j=k σ∈sh(n−2i,m−2j) σ−1 (a),σ−1 (a+1)>n−2i τ ∈sh(i,j) X = X (−1)σ+1ωi+1(· · · , bea, dea+1, · · · ; (ea, ea+1), · · · )ηj(· · · ) (−1)σ+1ωi(· · · )ηj+1(· · · , bea, dea+1, · · · ; (ea, ea+1), · · · ) (−1)σ+1ωl(· · · ; (ea, ea+1), · · · )ηj(· · · ) l+j=k+1 τ ∈sh(l,j) σ∈sh(n−2l,m−2j) + X τ −1((ea ,ea+1))≤l X i+l=k+1 τ ∈sh(i,l) σ∈sh(n−2i,m−2l) τ −1((ea ,ea+1 ))>i (−1)σ+1ωi(· · · )ηl(· · · ; (ea, ea+1), · · · ) = −(ω · η)k+1(e1, · · · , bea, dea+1, · · · ; (ea, ea+1), · · · ) Step 2: The multiplication is associative: ∀ω ∈ C n(L), η ∈ Cm(L), λ ∈ C l(L), by definition it is an easy calculation that, ((ω · η) · λ)k(e1, · · · , en+m+l−2k; f1, · · · , fk) and (ω · (η · λ))k(e1, · · · , en+m+l−2k; f1, · · · , fk) both equal to: X a+b+c=k σ∈sh(n−2a,m−2b,l−2c) τ ∈sh(a,b,c) (−1)σωa(· · · )ηb(· · · )λc(· · · ) Step 3: The differential d is a graded derivation: d(ω · η) = (dω) · η + (−1)nω · (dη), ∀ω ∈ C n(L), η ∈ Cm(L). Since d = d0 + δ, it suffices to prove the equation for d0, δ respectively. For d0, we only give the proof for the case of degree 0 here, since the proof is almost the same for cases of higher degrees (the only difference is that the sum should be taken over permutations of the arguments in Z as well). DERIVED BRACKETS FOR FAT LEIBNIZ ALGEBRAS 4 (d0(ω · η))0(e1, · · · , en+m+1) a = X = X +X (−1)a+1ρ(ea)(ω · η)0(· · · bea · · · ) + X (−1)a+1ρ(ea)(cid:0) (−1)a X σ∈sh(n,m){··· ,a,··· } X a<b a (−1)a(ω · η)0(· · · bea · · · beb, ea ◦ eb · · · ) (−1)σω0(eσ(1) · · · eσ(n))η0(eσ(n+1) · · · eσ(n+m))(cid:1) a<b σ∈sh(n,m){···a,··· } σ−1 (b)<n+1 +X (−1)a X a<b σ∈sh(n,m){···a,··· } σ−1(b)>n (−1)σω0(eσ(1), · · · beb, ea ◦ eb, · · · eσ(n))η0(eσ(n+1), · · · eσ(n+m+1)) (−1)σω0(eσ(1), · · · eσ(n))η0(eσ(n+1), · · · beb, ea ◦ eb, · · · eσ(n+m+1)) (let σ1, σ2 be the permutations adding a to σ in front and at back respectively) = X X (−1)a+1(−1)σ1+σ−1 1 (a)−a 1 (a)+1) · · · eσ1(n+1))(cid:1)η0(eσ1(n+2) · · · eσ1(n+m+1)) a σ1∈sh(n+1,m) X (cid:0)ρ(ea)ω0(eσ1(1) · · · bea, eσ1(σ−1 +X ω0(eσ2(1) · · · eσ2(n))(cid:0)ρ(ea)η0(eσ2(n+1) · · · bea, eσ2(σ−1 +X (−1)a+1(−1)σ2+σ−1 (−1)a(−1)σ1+σ−1 σ2∈sh(n,m+1) X 1 (a)−a 2 (a)−a a 2 (a)+1) · · · eσ2(n+m+1))(cid:1) a<b σ1∈sh(n+1,m) (b)<n+2 σ −1 1 ω0(eσ1(1) · · · bea, eσ1(σ−1 +X X (−1)a(−1)σ2+σ−1 2 (a)−a 1 (a)+1) · · · beb, ea ◦ eb · · · )η0(eσ1(n+2) · · · eσ1(n+m+1)) a<b σ2∈sh(n,m+1) (b)>n+1 σ −1 2 = X σ1 2 (a)+1) · · · beb, ea ◦ eb · · · eσ2(n+m+1)) ω0(eσ2(1) · · · )η0(eσ2(n+1) · · · bea, eσ2(σ−1 (−1)σ1 X (cid:0)ρ(eσ1(a1))ω0(eσ1(1) · · · \eσ1(a1) · · · eσ1(n+1))(cid:1)η0(eσ1(n+2) · · · eσ1(n+m+1)) +X (−1)a1+1 1 (a))<n+2 (−1)σ1 (−1)a1 X (a1,σ−1 σ1 (a1,σ−1 1 (a))<(b1,σ−1 1 (b))<n+2 ω0(eσ1(1) · · · \eσ1(a1) · · · \eσ1(b1), eσ1(a1) ◦ eσ1(b1) · · · )η0(eσ1(n+2) · · · eσ1(n+m+1)) +X (−1)σ2+nω0(eσ2(1), · · · , eσ2(n)) · X σ2 a2,σ−1 2 (a) (−1)a2−n+1ρ(eσ2(a2))η0(eσ2(n+1), · · · , \eσ2(a2), · · · , eσ2(n+m+1)) +X (−1)σ2+nω0(eσ2(1), · · · , eσ2(n)) · X σ2 n<(a2,σ−1 2 (a))<(b2,σ−1 2 (b)) (−1)a2−nη0(eσ2(n+1) · · · \eσ2(a2) · · · \eσ2(b2), eσ2(a2) ◦ eσ2(b2) · · · eσ2(n+m+1)) DERIVED BRACKETS FOR FAT LEIBNIZ ALGEBRAS 5 = (cid:0)(d0ω) · η + (−1)nω · (d0η)(cid:1)0(e1, · · · , en+m+1) For δ, = X = X i (δ(ω · η))k(e1, · · · , en+m+1−2k; f1, · · · , fk) (ω · η)k−1(fi, e1, · · · , en+m+1−2k; · · · , fi, · · · ) X X (−1)σωa(fi, · · · ; · · · fi · · · )ηb(· · · ) i a+b=k−1 τ ∈sh(a,b) +X X σ∈sh(n−2a,m−2b) σ−1 (fi )≤n−2a X i a+b=k−1 τ ∈sh(a,b) σ∈sh(n−2a,m−2b) σ−1 (fi )>n−2a (−1)σωa(· · · )ηb(fi, · · · ; · · · fi · · · ) (removing fi from σ, adding fi to τ in front and at back respectively) = X X (−1)σωa−1(fi, eσ(1), · · · ; · · · , \fτ (τ −1(i)), · · · )ηb(· · · ) a+b=k σ∈sh(n+1−2a,m−2b) + X a+b=k σ∈sh(n+1−2a,m−2b) τ ∈sh(a,b) τ −1 (i)>a τ ∈sh(a,b) τ −1(i)≤a X (−1)σ+nωa(· · · )ηb−1(fi, eσ(n−2a+1), · · · ; · · · , \fτ (τ −1(i)), · · · ) = ((δω) · η)k(· · · ) + (−1)n(ω · (δη))k(· · · ) The proof is finished. Remark 2.3. In [4], we construct a Courant-Dorfman algebra structure on S•(Z)⊗L for any Leibniz algebra L with left center Z, and prove an isomorphism between H-standard complexes of them. So it is a direct conclusion that C(L) is isomorphic to the standard complex of the Courant-Dorfman algebra S•(Z) ⊗ L. Next we consider a 3-cochain in C(L). Let Θ0 : L ⊗ L ⊗ L → S•(Z) and Θ1 : L ⊗ Z → S•(Z) be defined as: Θ0(e1, e2, e3) = (e1 ◦ e2, e3) (2.2) Θ1(e; f ) = −(e, f ). We can prove that Θ = (Θ0, Θ1) is a 3-cocycle by a direct calculation, but actually we have the following: Proposition 2.4. Θ = dζ is a 3-coboundary, where ζ = (ζ0, ζ1) ∈ C2(L) is defined by: ζ0(e1, e2) , (e1, e2), ζ1(f ) , −2f. Proof. Since ζ0(e1, e2) + ζ0(e2, e1) = 2(e1, e2) = −ζ1((e1, e2)), DERIVED BRACKETS FOR FAT LEIBNIZ ALGEBRAS 6 ζ = (ζ0, ζ1) is a 2-cochain in C2(L). By definition, (dζ)0(e1, e2, e3) = ρ(e1)ζ0(e2, e3) − ρ(e2)ζ0(e1, e3) + ρ(e3)ζ0(e1, e2) −ζ0(e1 ◦ e2, e3) − ζ0(e2, e1 ◦ e3) + ζ0(e1, e2 ◦ e3) = ρ(e1)(e2, e3) − (e1 ◦ e2, e3) − (e2, e1 ◦ e3) −ρ(e2)(e1, e3) + (e1, e2 ◦ e3) + ρ(e3)(e1, e2) = −(e2 ◦ e1, e3) + ρ(e3)ζ0(e1, e2) = (e1 ◦ e2, e3) (dζ)1(e; f ) = ρ(e)ζ1(f ) + ζ0(f, e) = −2(e, f ) + (e, f ) = −(e, f ) so Θ = dζ is a 3-coboundary. Actually Θ is exactly the restriction of the canonical 3-cocycle of the Courant-Dorfman algebra S•(Z) ⊗ L (see Remark 2.3). We will call Θ the canonical 3-cocycle of L. 3. Poisson structure on a subcomplex In this section, we consider a subcomplex, denoted by C(L), consisting of the so-called "repre- sentable cochains", and construct a Poisson algebra structure on C(L). Let L∨ , Hom(L, S•(Z)). In this paper, Hom always means k-linear homomorphisms. ∀ω ∈ C n(L), ωk gives rise to a map ¯ωk : L⊗n−2k−1 ⊗ Sk(Z) → L∨ : ¯ωk(e1, · · · , en−2k−1; f1, · · · fk)(e) , (ιfk · · · ιf1 ιen−2k−1 · · · ιe1 ωk)(e) = ωk(e1, · · · , en−2k−1, e; f1, · · · , fk). The symmetric product (·, ·) of L can be S•(Z)-linearly extended to a symmetric product on S•(Z) ⊗ L, thus inducing a map φ , (·, ) : S•(Z) ⊗ L → L∨. Definition 3.1. Given any ω ∈ C n(L), if Im(¯ωk) ⊆ Im(φ), ∀k, we call ω a "representable cochain". The graded subspace of C(L) consisting of all representable cochains is denoted by C(L). By definition, e♭ , (e, ) : L → Z is obviously a representable cochain. Given ω ∈ C n(L), ωk induces a k-linear map ωk : L⊗n−2k−1 → Hom(Sk(Z), S•(Z) ⊗ L), which is defined by ωk(e1, · · · , en−2k−1)(f1, · · · , fk) , φ−1(¯ωk(e1, · · · , en−2k−1; f1, · · · , fk)). Note that, to determine ωk, we only need to choose the preimage of ¯ωk for given basis of L and Z, and then take the k-linear extension. So ωk depends on the choices, it is not uniquely determined unless φ is injective (i.e. the bilinear product of L is non-degenerate). Proposition 3.2. C(L) is a subcomplex of C(L). DERIVED BRACKETS FOR FAT LEIBNIZ ALGEBRAS 7 Proof. ∀ω ∈ C n(L), we need to prove that dω ∈ C n+1(L): (dω)k(e1, · · · en+1−2k; f1, · · · fk) a = X (−1)a+1ρ(ea)ωk(· · · bea, · · · ; · · · ) + X +X ωk−1(fi, e1, · · · ; · · · bfi, · · · ) = X (−1)a+1ρ(ea)(ωk(e1, · · · bea, · · · en−2k)(f1, · · · fk), en+1−2k) a<b i (−1)aωk(· · · bea, · · · ea ◦ eb, · · · ; · · · ) a<b≤n−2k a≤n−2k +(−1)nρ(en+1−2k)ωk(e1, · · · en−2k; · · · ) + X + X +X (−1)a(ωk(· · · bea, · · · ea ◦ eb, · · · en−2k)(f1, · · · fk), en+1−2k) (−1)aωk(· · · bea, · · · en−2k, ea ◦ en+1−2k; · · · ) (ωk−1(fi, e1, · · · en−2k)(· · · bfi, · · · ), en+1−2k) a≤n−2k i = (•, en+1−2k). The proof is finished. Next, we will define a graded bracket on C(L). ∀α ∈ Hom(Sk(Z), S•(Z)⊗L), β ∈ Hom(Sl(Z), S•(Z)⊗L), define hα·βi ∈ Hom(Sk+l(Z), S•(Z)) as hα · βi(f1, · · · , fk+l) , X σ∈sh(k,l) (α(fσ(1), · · · , fσ(k)), β(fσ(k+1), · · · , fσ(k+l))). ∀γ ∈ Hom(Sk(Z), S•(Z)), δ ∈ Hom(Sl(Z), S•(Z)), define γ ◦ δ ∈ Hom(Sk+l−1(Z), S•(Z)) as γ ◦ δ(f1, · · · , fk+l−1) , X γ(δ(fσ(1), · · · , fσ(l)), fσ(l+1), · · · , fσ(l+k−1)), σ∈sh(l,k−1) where γ : S•(Z) ⊗ Sk−1(Z) → S•(Z) is extended from γ by Leibniz rule in the first argument. Now given ω ∈ C n(L), η ∈ Cm(L), we define the bracket {ω, η} as follows: (3.1) {ω, η} , ω • η + ω ⋄ η − (−1)nmη ⋄ ω, where ω • η = ((ω • η)0, (ω • η)1, · · · ), with (ω • η)k : ⊗n+m−2−2kL → Hom(Sk(Z), S•(Z)) defined by (ω • η)k(e1, · · · , en+m−2−2k) X , (−1)m−1 (−1)σhωi(eσ(1), · · · eσ(n−2i−1)) · ηj(eσ(n−2i), · · · eσ(n+m−2−2k))i, i+j=k σ∈sh(n−2i−1,m−2j−1) (obviously the value does not depend on the choices of ωi and ηj , so it is well-defined) and ω ⋄ η = ((ω ⋄ η)0, (ω ⋄ η)1, · · · ), with (ω ⋄ η)k : ⊗n+m−2−2kL → Hom(Sk(Z), S•(Z)) defined by , (ω ⋄ η)k(e1, · · · , en+m−2−2k) X i+j=k σ∈sh(n−2i−2,m−2j) (−1)σωi+1(eσ(1) · · · eσ(n−2i−2)) ◦ ηj(eσ(n−2i−1) · · · eσ(n+m−2−2k)). DERIVED BRACKETS FOR FAT LEIBNIZ ALGEBRAS 8 The following is the main theorem of this section: Theorem 3.3. ( C(L), {·, ·}) is a graded Poisson algebra. Before the proof of this theorem, we prove the following two lemmas first. Lemma 3.4. C(L) is a subalgebra of C(L) with the multiplication map defined in 2.1. Proof. Given η ∈ Cm(L), λ ∈ C l(L), we need to prove that ηλ ∈ Cm+l(L): (ηλ)k(e1, · · · , em+l−2k; f1, · · · , fk) = X (−1)σηi(eσ(1) · · · ; fτ (1) · · · fτ (i))λj (eσ(m−2i+1) · · · eσ(m+l−2k); fτ (i+1) · · · fτ (k)) i+j=k σ∈sh(m−2i,l−2j) τ ∈sh(i,j) = X X (−1)σ( ηi(· · · , \em+l−2k; · · · ), em+l−2k)λj (· · · ; · · · ) i+j=k τ ∈sh(i,j) + X σ∈sh(m−2i,l−2j) σ−1 (m+l−2k)=m−2i X i+j=k σ∈sh(m−2i,l−2j) (−1)σηi(· · · ; · · · )( λj (· · · , \em+l−2k; · · · ), em+l−2k) τ ∈sh(i,j) = ({ X i+j=k τ ∈sh(i,j) + X i+j=k τ ∈sh(i,j) σ−1 (m+l−2k)=m+l−2k X (−1)¯σ+l ηi(e¯σ(1), · · · ; fτ (1), · · · )λj (e¯σ(m−2i) · · · ; fτ (i+1), · · · ) ¯σ∈sh(m−2i−1,l−2j) X ¯σ∈sh(m−2i,l−2j−1) (−1)¯σηi(e¯σ(1) · · · ; fτ (1), · · · ) λj(e¯σ(m−2i+1) · · · ; fτ (i+1), · · · )}, em+l−2k) The lemma is proved. Lemma 3.5. ω • η, ω ⋄ η, {ω, η} are all cochains in C n+m−2(L). Proof. ω • η is a cochain in C n+m−2(L) because: (ω • η)k(e1 · · · ea, ea+1 · · · en+m−2−2k; f1 · · · fk) + (ω • η)k(· · · ea+1, ea · · · ; · · · ) { X (−1)σh ωi(eσ(1) · · · ea · · · ) · ηj(eσ(n−2i) · · · ea+1 · · · eσ(n+m−2−2k))i σ,a∈ω,a+1∈η (−1)σh ωi(eσ(1) · · · ea · · · ) · ηj(eσ(n−2i) · · · ea+1 · · · eσ(n+m−2−2k))i} { X (−1)σh ωi(eσ(1) · · · ea+1, · · · ) · ηj(eσ(n−2i) · · · , ea · · · eσ(n+m−2−2k))i i+j=k σ,a∈η,a+1∈ω σ,a∈η,a+1∈ω i+j=k = (−1)m−1 X + X +(−1)m−1 X + X +(−1)m−1 X + X +(−1)m−1 X + X σ,a∈ω,a+1∈η σ,a∈ω,a+1∈ω σ,a∈η,a+1∈η (−1)σh ωi(eσ(1) · · · ea+1 · · · ) · ηj(eσ(n−2i) · · · ea · · · eσ(n+m−2−2k))i} { X (−1)σh ωi(eσ(1) · · · ea, ea+1 · · · ) · ηj(eσ(n−2i) · · · eσ(n+m−2−2k))i i+j=k σ,a∈ω,a+1∈ω (−1)σh ωi(eσ(1) · · · ea+1, ea · · · ) · ηj (eσ(n−2i) · · · eσ(n+m−2−2k))i} { X (−1)σh ωi(eσ(1) · · · ) · ηj(eσ(n−2i) · · · ea, ea+1 · · · eσ(n+m−2−2k))i i+j=k σ,a∈η,a+1∈η (−1)σh ωi(eσ(1) · · · ) · ηj(eσ(n−2i) · · · ea+1, ea · · · )i} DERIVED BRACKETS FOR FAT LEIBNIZ ALGEBRAS 9 = (−1)m−1 X X (−1)σ(−1) i+j=k σ∈sh(n−2i−1,m−2j−1) a∈ω,a+1∈ω h ωi+1(eσ(1) · · · bea, dea+1 · · · eσ(n−2i−1); (ea, ea+1)) · ηj(eσ(n−2i) · · · )i +(−1)m−1 X (−1)σ(−1) X i+j=k σ∈sh(n−2i−1,m−2j−1) a∈ω,a+1∈ω = (−1)m X h ωi(eσ(1) · · · ) · ηj+1(eσ(n−2i) · · · bea, dea+1 · · · eσ(n+m−2−2k); (ea, ea+1))i X (−1)σ′ X i′+j=k+1 σ′∈sh(n−2i′−1,m−2j−1) τ ∈sh(i′ ,j) (ea ,ea+1 )∈ω ( ωi′ (eσ′(1) · · · )((ea, ea+1), fτ (1) · · · ), ηj(eσ′(n−2i′) · · · )(fτ (i′) · · · fτ (k))) +(−1)m X (−1)σ′ X X i+j′=k+1 σ′∈sh(n−2i−1,m−2j′−1) τ ∈sh(i′ ,j) (ea ,ea+1)∈η ( ωi(eσ′(1) · · · )(fτ (1) · · · ), ηj′ (eσ′(n−2i) · · · )((ea, ea+1), fτ (i+1) · · · fτ (k))) = −(ω • η)k+1(e1, · · · , bea, dea+1, · · · , en+m−2−2k; (ea, ea+1), f1, · · · , fk). ω ⋄ η is a cochain in C n+m−2(L) because: (ω ⋄ η)k(· · · , ea, ea+1 · · · ; f1 · · · fk) + (ω ⋄ η)k(· · · ea+1, ea · · · ; · · · ) (−1)σωi+1(eσ(1) · · · ea · · · ) ◦ ηj(eσ(n−2i−1) · · · ea+1 · · · ) = X (−1)σωi+1(eσ(1) · · · ea · · · ) ◦ ηj(eσ(n−2i−1) · · · ea+1 · · · )} (−1)σωi+1(eσ(1) · · · ea+1 · · · ) ◦ ηj(eσ(n−2i−1) · · · ea · · · ) σ,a∈η,a+1∈ω { X σ,a∈η,a+1∈ω i+j=k σ,a∈ω,a+1∈η i+j=k { X + X + X + X + X + X + X + X i+j=k σ,a∈ω,a+1∈ω { X σ,a∈ω,a+1∈η { X i+j=k σ,a∈η,a+1∈η σ,a∈ω,a+1∈ω (−1)σωi+1(eσ(1) · · · ea+1 · · · ) ◦ ηj(eσ(n−2i−1) · · · ea · · · )} (−1)σωi+1(eσ(1) · · · ea, ea+1 · · · ) ◦ ηj(eσ(n−2i−1) · · · ) (−1)σωi+1(eσ(1) · · · ea+1, ea · · · ) ◦ ηj (eσ(n−2i−1) · · · )} (−1)σωi+1(eσ(1) · · · ) ◦ ηj(eσ(n−2i−1) · · · ea, ea+1 · · · ) (−1)σωi+1(eσ(1) · · · ) ◦ ηj(eσ(n−2i−1) · · · ea+1, ea · · · )} = X σ,a∈η,a+1∈η X i+j=k σ∈sh(n−2i−2,m−2j) a∈ω,a+1∈ω (−1)σ(−1) ωi+2(eσ(1) · · · bea, dea+1 · · · eσ(n−2i−2); (ea, ea+1)) ◦ ηj(eσ(n−2i−1) · · · ) + X (−1)σ(−1) X i+j=k σ∈sh(n−2i−2,m−2j) a∈η,a+1∈η DERIVED BRACKETS FOR FAT LEIBNIZ ALGEBRAS 10 ωi+1(eσ(1) · · · eσ(n−2i−2)) ◦ ηj+1(eσ(n−2i−1) · · · dea+1, bea · · · ; (ea, ea+1)) = X X (−1)σ′ X i′+j=k+1 σ′∈sh(n−2i′−2,m−2j) τ ∈sh(j,i′ −1) (ea ,ea+1)∈ω ωi′+1(eσ′(1) · · · ; ηj(eσ′(n−2i′−1) · · · ; fτ (1) · · · ), (ea, ea+1), fτ (j+1) · · · fτ (k)) + X (−1)σ′ X X i+j′=k+1 σ′∈sh(n−2i−2,m−2j′) τ ∈sh(j,i′ −1) (ea ,ea+1 )∈η ωi+1(eσ′(1) · · · ; ηj′ (eσ′(n−2i−1) · · · ; (ea, ea+1), fτ (1) · · · ), fτ (j′) · · · fτ (k)) = −(ω ⋄ η)k+1(e1, · · · , bea, dea+1, · · · , en+m−2−2k; (ea, ea+1), f1, · · · , fk). So {ω, η} = ω • η + ω ⋄ η − (−1)nmη ⋄ ω is also a cochain in C n+m−2(L). Proof of theorem 3.3: Proof. 1) By the lemmas above, in order for C(L) to be a graded Poisson algebra, we need to prove the following: (1). For any two representable cochains ω, η, {ω, η} = −(−1)nm{η, ω}, (2). For any representable cochains ω, η, λ, {ω, ηλ} = {ω, η}λ + (−1)nmη{ω, λ}, (3). The bracket of any two representable cochains is still a representable cochain, and {ω, {η, λ} = {{ω, η}, λ} + (−1)nm{η, {ω, λ}}. For (1), it suffices to prove ω • η = −(−1)nmη • ω. ∀σ ∈ sh(n − 2i − 1, m − 2j − 1), switching the first n − 2i − 1 arguments with the last m − 2j − 1 arguments results in a sign difference (−1)(n−1)(m−1), so by definition there is merely a sign difference between ω • η and η • ω of (−1)n−m+(n−1)(m−1) = (−1)nm+1. Thus (1) is proved. For (2), we need to prove that {ω, •} is a graded derivative. {ω, ηλ}k(e1, · · · , en+m+l−2−2k; f1, · · · , fk) = (ω • ηλ)k(· · · ) + (ω ⋄ ηλ)k(· · · ) + (−1)n(m+l)+1(ηλ ⋄ ω)k(· · · ) We calculate the three parts above respectively: DERIVED BRACKETS FOR FAT LEIBNIZ ALGEBRAS 11 (ω • ηλ)k(e1, · · · , en+m+l−2−2k; f1, · · · , fk) = (−1)m+l+1 = (−1)m+l+1 X a+b=k σ∈sh(n−2a−1,m+l−2b−1) τ ∈sh(a,b) X a+b+c=k ( ωa(eσ(1), · · · ; fτ (1), · · · ), ](ηλ)b(eσ(n−2a), · · · ; fτ (a+1), · · · )) (−1)σ+l (ωa(eσ(1) · · · ), ηb(eσ(n−2a) · · · )λc(eσ(n+m−2a−2b−1) · · · )) σ∈sh(n−2a−1,m−2b−1,l−2c) +(−1)m+l+1 τ ∈sh(a,b,c) X (−1)σ( ωa(eσ(1) · · · ), ηb(eσ(n−2a) · · · ) λc(eσ(n+m−2a−2b) · · · )) = X a+c=k a+b+c=k σ∈sh(n−2a−1,m−2b,l−2c−1) τ ∈sh(a,b,c) (−1)σ(ω • η)a(eσ(1), · · · )λc(eσ(n+m−2a−1), · · · ) σ∈sh(n+m−2a−2,l−2c) τ ∈sh(a,c) X + b+a=k σ∈sh(m−2b,n+l−2a−2) (−1)σ+(n−1)m(−1)mηb(eσ(1), · · · )(ω • λ)a(eσ(m−2b+1), · · · ) τ ∈sh(b,a) = (cid:0)(ω • η) · λ(cid:1)k(· · · ) + (−1)nm(cid:0)η · (ω • λ)(cid:1)k(· · · ) (ω ⋄ ηλ)k(e1, · · · , en+m+l−2−2k; f1, · · · , fk) X = = = = (−1)σωa+1(eσ(1) · · · eσ(n−2a−2); (ηλ)b(eσ(n−2a−1) · · · ; fτ (1) · · · ), fτ (b+1) · · · ) a+b=k σ∈sh(n−2a−2,m+l−2b) τ ∈sh(b,a) X a+b+c=k σ∈sh(n−2a−2,m−2b,l−2c) τ ∈sh(b,c,a) X (−1)σωa+1(eσ(1) · · · ; ηb(eσ(n−2a−1) · · · )λc(eσ(n+m−2a−2b−1) · · · ), · · · ) (−1)σωa+1(eσ(1) · · · ; ηb(eσ(n−2a−1) · · · ), · · · )λc(eσ(n+m−2a−2b−1) · · · ) a+b+c=k σ∈sh(n−2a−2,m−2b,l−2c) τ ∈sh(b,a,c) X + a+b+c=k σ∈sh(m−2b,n−2a−2,l−2c) (−1)σ+nmηb(eσ(1) · · · )ωa+1(eσ(m−2b+1) · · · ; λc(eσ(n+m−2a−2b−1) · · · ), · · · ) τ ∈sh(b,c,a) X a+c=k σ∈sh(n+m−2a−2,l−2c) (−1)σ(ω ⋄ η)a(eσ(1), · · · ; fτ (1), · · · )λc(eσ(n+m−2a−1), · · · ; fτ (a+1), · · · ) τ ∈sh(a,c) +(−1)nm X a+b=k (−1)σηb(eσ(1), · · · ; fτ (1), · · · )(ω ⋄ λ)a(eσ(m−2b+1), · · · ; fτ (b+1), · · · ) σ∈sh(m−2b,n+l−2a−2) = (cid:0)(ω ⋄ η) · λ(cid:1)k(· · · ) + (−1)nm(cid:0)η · (ω ⋄ λ)(cid:1)k(· · · ) τ ∈sh(b,a) DERIVED BRACKETS FOR FAT LEIBNIZ ALGEBRAS 12 = = = (ηλ ⋄ ω)k(e1, · · · , en+m+l−2−2k; f1, · · · , fk) X a+c=k σ∈sh(m+l−2a−2,n−2c) τ ∈sh(c,a) X a+b+c=k (−1)σ(ηλ)a+1(eσ(1), · · · ; ωc(eσ(m+l−2a−1), · · · ), · · · ) (−1)σ+nlηa+1(eσ(1) · · · ; ωc(eσ(m−2a−1) · · · ), · · · )λb(eσ(n+m−2a−2c−1) · · · ) σ∈sh(m−2a−2,n−2c,l−2b) τ ∈sh(c,a,b) X + a+b+c=k σ∈sh(m−2a,l−2b−2,n−2c) (−1)σηa(eσ(1) · · · )λb+1(eσ(m−2a+1) · · · ; ωc(eσ(m+l−2a−2b−1) · · · ), · · · ) τ ∈sh(a,c,b) X (−1)σ(−1)nl(η ⋄ ω)a(eσ(1), · · · ; fτ (1), · · · )λb(eσ(n+m−2a−1), · · · ; fτ (a+1), · · · ) a+b=k σ∈sh(m+n−2a−2,l−2b) τ ∈sh(a,b) X + a+b=k σ∈sh(m−2a,n+l−2b−2) (−1)σηa(eσ(1), · · · ; fτ (1), · · · ) · (λ ⋄ ω)b(eσ(m−2a+1), · · · ; fτ (a+1), · · · ) τ ∈sh(a,b) = (−1)nl(cid:0)(η ⋄ ω) · λ(cid:1)k(· · · ) + (cid:0)η · (λ ⋄ ω)(cid:1)k(· · · ) So {ω, ηλ}k(e1, · · · , en+m+l−2−2k; f1, · · · , fk) = (cid:0)(ω • η) · λ(cid:1)k(· · · ) + (−1)nm(cid:0)η · (ω • λ)(cid:1)k(· · · ) +(cid:0)(ω ⋄ η) · λ(cid:1)k(· · · ) + (−1)nm(cid:0)η · (ω ⋄ λ)(cid:1)k(· · · ) +(−1)nm+1(cid:0)(η ⋄ ω) · λ(cid:1)k(· · · ) + (−1)nm(−1)nl+1(cid:0)η · (λ ⋄ ω)(cid:1)k(· · · ) = ({ω, η} · λ)k(· · · ) + (−1)nm(η · {ω, λ})k(· · · ) {ω, •} is a graded derivative, (2) is proved. For (3), in order for {ω, η} to be a representable cochain, we need to prove {ω, η}k(e1, · · · , en+m−2k; f1, · · · , fk) = (en+m−2k, •), ∀k. {ω, η}k(e1, · · · , en+m−2−2k; f1, · · · , fk) X = (−1)m+1 (−1)σ( ωa(eσ(1), · · · ; fτ (1), · · · ), ηb(eσ(n−2a), · · · ; fτ (a+1), · · · )) a+b=k σ∈sh(n−2a−1,m−2b−1) + X a+b=k τ ∈sh(a,b) (−1)σωa+1(eσ(1), · · · ; ηb(eσ(n−2a−1), · · · ; fτ (1), · · · ), fτ (b+1), · · · ) σ∈sh(n−2a−2,m−2b) τ ∈sh(b,a) +(−1)nm+1 X a+b=k σ∈sh(m−2a−2,n−2b) (−1)σηa+1(eσ(1), · · · ; ωb(eσ(m−2a−1), · · · ; fτ (1), · · · ), fτ (b+1), · · · ) = (−1)m+1 τ ∈sh(b,a) X a+b=k σ∈sh(n−2a−2,m−2b−1) τ ∈sh(b,a) (−1)σ+m+1 ωa(eσ(1) · · · eσ(n−2a−2), en+m−2−2k, ηb(eσ(n−2a−1) · · · ; fτ (1) · · · ); fτ (b+1) · · · ) DERIVED BRACKETS FOR FAT LEIBNIZ ALGEBRAS 13 +(−1)m+1 X a+b=k σ∈sh(m−2b−2,n−2a−1) τ ∈sh(a,b) (−1)σ+(n−1)m ηb(eσ(1) · · · eσ(m−2b−2), en+m−2−2k, ωa(eσ(m−2b−1) · · · ; fτ (1) · · · ); fτ (a+1) · · · ) +{(en+m−2−2k, •) + X (−1)σ a+b=k σ∈sh(n−2a−2,m−2b−1) τ ∈sh(b,a) ωa+1(eσ(1), · · · ; ( ηb(eσ(n−2a−1), · · · ; fτ (1), · · · ), en+m−2−2k), fτ (b+1), · · · )} +{(en+m−2−2k, •) + (−1)nm+1 X (−1)σ ηa+1(eσ(1), · · · ; ( ωb(eσ(m−2a−1), · · · ; fτ (1), · · · ), en+m−2−2k), fτ (b+1), · · · )} a+b=k σ∈sh(m−2a−2,n−2b−1) τ ∈sh(b,a) (−1)σωa(· · · , en+m−2−2k, ηb(· · · ); · · · ) (−1)σ+nm+1ηb(· · · , en+m−2−2k, ωa(· · · ); · · · ) (−1)σ+1 = (en+m−2−2k, •) X a+b=k σ∈sh(n−2a−2,m−2b−1) τ ∈sh(b,a) X a+b=k σ∈sh(m−2b−2,n−2a−1) τ ∈sh(a,b) X a+b=k σ∈sh(n−2a−2,m−2b−1) τ ∈sh(b,a) + + + + {ωa(· · · , en+m−2−2k, ηb(· · · ); · · · ) + ωa(· · · , ηb(· · · ), en+m−2−2k; · · · )} X (−1)σ+nm a+b=k σ∈sh(m−2a−2,n−2b−1) τ ∈sh(b,a) {ηa(· · · , en+m−2−2k, ωb(· · · ); · · · ) + ηa(· · · , ωb(· · · ), en+m−2−2k; · · · )} = (en+m−2−2k, •) Thus (3) is proved. If φ is an isomorphism(i.e. the symmetric product of S•(Z) ⊗ L is strongly non-degenerate), any ω ∈ C(L) is a representable cochain, so C(L) = C(L) is a graded commutative Poisson algebra. In this section we prove that the Leibniz bracket of a fat Leibniz algebra is a derived bracket. First we prove the following: 4. Derived brackets Proposition 4.1. {Θ, η} = −dη, ∀η ∈ C(L). Proof. It is obvious that Θ is a representable cochain. DERIVED BRACKETS FOR FAT LEIBNIZ ALGEBRAS 14 (Θ • η)k(e1, · · · , em+1−2k; f1, · · · , fk) = (−1)m−1 X +(−1)m−1 X σ∈sh(2,m−2k−1) τ ∈sh(1,k−1) (−1)σ( Θ0(eσ(1), eσ(2)), ηk(eσ(3) · · · eσ(m+1−2k); · · · )) ( Θ1(fτ (1)), ηk−1(e1, · · · , em+1−2k; fτ (2), · · · , fτ (k))) (−1)a+b+1(ea ◦ eb, ηk(e1, · · · , ea, · · · , eb, · · · em+1−2k; f1, · · · fk)) (−1)(fi, ηk−1(e1, · · · , em+1−2k; f1, · · · , fi, · · · , fk)) (−1)a+bηk(e1, · · · , ea, · · · , eb, · · · , em+1−2k, ea ◦ eb; f1, · · · , fk) = (−1)m−1X +(−1)m−1X a<b i = (−1)m X +(−1)m X a<b i ηk−1(e1, · · · , em+1−2k, fi; f1, · · · , fi, · · · , fk) (Θ ⋄ η)k(e1, · · · , em+1−2k; f1, · · · , fk) (−1)a+1Θ1(ea) ◦ ηk(e1, · · · , ea, · · · , em+1−2k) (−1)a+1(−1)(ea, ηk(e1, · · · , ea, · · · em+1−2k; f1, · · · , fk)) (−1)aρ(ea)ηk(e1, · · · , ea, · · · , em+1−2k; f1, · · · , fk) a = X = X = X a a (−1)m+1(η ⋄ Θ)k(e1, · · · , em+1−2k; f1, · · · , fk) = (−1)m+1 X (−1)σ σ∈sh(m−2k−2,3) ηk+1(eσ(1), · · · , eσ(m−2k−2)) ◦ Θ0(eσ(m−2k−1), eσ(m−2k), eσ(m−2k+1)) +(−1)m+1X (−1)a+m+1ηk(e1, · · · ea, · · · , em+1−2k) ◦ Θ1(ea) a a<b<c = X +X = X a +X = X a<b (−1)a+b+c+1ηk+1(e1 · · · ea · · · eb · · · ec · · · em+1−2k; (ea ◦ eb, ec), f1 · · · fk) (−1)aX (−1)a+b X ηk(e1, · · · , ea, · · · , em+1−2k; −(ea, fi), f1, · · · , fi, · · · , fk) (−1)c{ηk(· · · , ea, · · · , eb, · · · , ec−1, ea ◦ eb, ec, · · · ; · · · ) i a<b b<c<m+2−2k +ηk(· · · , ea, · · · , eb, · · · , ec−1, ec, ea ◦ eb, · · · ; · · · )} (−1)a{ηk−1(· · · ea−1, fi, ea, · · · ; fi · · · ) + ηk−1(· · · ea−1, ea, fi, · · · ; fi · · · )} i,a (−1)a+1{ηk(· · · ea · · · ea ◦ eb, · · · ; · · · ) + (−1)b+mηk(· · · ea · · · eb · · · ea ◦ eb; · · · )} DERIVED BRACKETS FOR FAT LEIBNIZ ALGEBRAS 15 −X ηk−1(fi, e1, · · · ; · · · fi · · · ) + (−1)m+1X i i ηk−1(e1 · · · em+1−2k, fi; · · · fi · · · ) The sum of the equations above is {Θ, η}k(e1, · · · , em+1−2k; f1, · · · , fk) (−1)aρ(ea)ηk(e1, · · · , ea, · · · , em+1−2k; f1, · · · , fk) a = X +X −X a<b (−1)a+1ηk(· · · , ea, · · · , ea ◦ eb, · · · ; · · · ) ηk−1(fi, e1, · · · , em+1−2k; · · · , fi, · · · ) i = −(dη)k(e1, · · · , em+1−2k; f1, · · · , fk) The proof is finished. Next, we give the definition of fat Leibniz algebras: Definition 4.2. Given a Leibniz algebra L, if the symmetric product (·, ·) is non-degenerate, we call L a fat Leibniz algebra. The omni Lie algebra ol(V ) = gl(V ) ⊕ V is obviously a fat Leibniz algebra. And the space of sections of any Courant algebroid is also a fat Leibniz algebra. Actually given any Leibniz algebra L with trivial center, there is associated a fat Leibniz algebra L: Proposition 4.3. Suppose L is a Leibniz algebra with trivial center, then L , L/K is a fat Leibniz algebra, where K is the kernel of the bilinear product of L, i.e. K = {k ∈ L(k, e) = 0, ∀e ∈ L}. Proof. Since the product of L is invariant: τ (e1)(k, e2) = (e1 ◦ k, e2) + (k, e1 ◦ e2), ∀e1, e2 ∈ L, ∀k ∈ K, it follows that (e1 ◦ k, e2) = 0, ∀e2 ∈ L, so e1 ◦ k ∈ K. Furthermore since e1 ◦ k + k ◦ e1 = (k, e1) = 0, so k ◦ e1 = −e1 ◦ k is also in K. Thus K is an ideal of L. The Leibniz bracket of L naturally induces a bracket on L/K: ¯e1 ◦ ¯e2 , e1 ◦ e2, where ¯e is the equivalent class of e ∈ L in L/K. Suppose there exists ¯k ∈ L/K such that (¯k, ¯e) = 0, ∀¯e ∈ L/K, i.e. (k, e) ∈ K, ∀e ∈ L. Since (k, e) is in the left center of L, (k, e) ∈ K implies that (k, e) is also in the right center of L. So (k, e) = 0 by the assumption that the center of L is trivial. It follows that k itself is in K, ¯k = 0 ∈ L/K. As a result, the bilinear product on L/K is non-degenerate. Finally we give the main theorem of this section: Theorem 4.4. With the above notations, we have (e1 ◦ e2)♭ = −{{Θ, e♭ 1}, e♭ 2}. In particular, if L is a fat Leibniz algebra, then the Leibniz bracket can be represented as a derived bracket: e1 ◦ e2 = −{{Θ, e♭ 1}, e♭ 2}♯, DERIVED BRACKETS FOR FAT LEIBNIZ ALGEBRAS 16 where (•)♯ : Im((•)♭) → L is the (partial) inverse map of (•)♭, i.e. ((φ)♯)♭ , φ, ∀φ ∈ Im((•)♭). Proof. {Θ, e♭ 1} is a 2 cochain: {Θ, e♭ 1i + Θ1(e2) ◦ e♭ 1}0(e2, e3) = h Θ0(e2, e3), e♭ = (e2 ◦ e3, e1) − (e2, (e1, e3)) + (e3, (e1, e2)) = −(e2 ◦ e1, e3) + (e3, e1 ◦ e2 + e2 ◦ e1) = (e1 ◦ e2, e3) 1(e3) − Θ1(e3) ◦ e♭ 1(e2) {Θ, e♭ 1}1(f ) = h Θ1(f ), e♭ 1i = −(e1, f ) We see that {Θ, e♭ {{Θ, e♭ 1}, e♭ 2} is a 1 cochain: 1} is a representable cochain. {{Θ, e♭ 1}1 ◦ e♭ 2(e3) 1}, e♭ 2}0(e3) 1}0(e3), e♭ {Θ, e♭ 2i + {Θ, e♭ = h = (e1 ◦ e3, e2) − (e1, (e2, e3)) = −(e1 ◦ e2, e3) 1}, e♭ 2}(e3) = (e1 ◦ e2, e3). So (e1 ◦ e2)♭(e3) = −{{Θ, e♭ The proof is finished. Remark 4.5. As mentioned in Remark 2.3, C(L) is isomorphic to the standard complex of the Courant-Dorfman algebra S•(Z) ⊗ L. Actually theorem 4.4 is true for e1, e2 ∈ S•(Z) ⊗ L. In [10], Roytenberg proved that the Dorfman bracket of a non-degenerate Courant-Dorfman algebra (i.e. the symmetric product is strongly non-degenerate) is a derived bracket. Our theorem 4.4 can be viewed as a generalization of his result, since the symmetric product of S•(Z) ⊗ L is only non-degenerate, but not strongly non-degenerate. References [1] Alekseev, Anton and Xu, Ping, Derived brackets and Courant algebroids, Unpublished manuscript, available at https://www.math.psu.edu/ping/papers.html (2001). [2] Benayadi, Saïd and Hidri, Samiha, Quadratic Leibniz algebras, Journal of Lie Theory 24.3 (2014): 737-759. [3] Bloh, A., On a generalization of the concept of Lie algebra, Dokl. Akad. Nauk SSSR. Vol. 165. No. 3 (1965), 471-473. [4] Cai, Xiongwei, H-standard cohomology for Courant-Dorfman algebras and Leibniz algebras, arXiv preprint math.RA/1612.05297. [5] Kosmann-Schwarzbach, Yvette, Jacobian quasi-bialgebras and quasi-Poisson Lie groups, Mathematical aspects of classical field theory, pages 459 -- 489. Contemp. Math., 132, Amer. Math. Soc., 1992. [6] Kosmann-Schwarzbach, Yvette, Derived brackets, Letters in Mathematical Physics 69.1 (2004): 61-87. [7] Lecomte, P. and Roger, C., Modules et cohomologie des bigébres de Lie, Comptes rendus Acad. Sci. Paris, 310:405 -- 410, 1990. [8] Liu, Zhang-Ju, Weinstein, Alan and Xu, Ping, Manin triples for Lie bialgebroids, J. Differential Geom 45.3 (1997): 547-574. [9] Loday, J.L., Une version non commutative des algèbres de Lie, L'Ens. Math. (2), 39 (1993), 269-293. [10] Roytenberg, Dmitry, Courant -- Dorfman algebras and their cohomology, Letters in Mathematical Physics 90.1-3 (2009), 311-351. [11] Roytenberg, Dmitry, On the structure of graded symplectic supermanifolds and Courant algebroids, Contem- porary Mathematics 315 (2002): 169-186. DERIVED BRACKETS FOR FAT LEIBNIZ ALGEBRAS 17 Mathematics Research Unit, FSTC, University of Luxembourg, Luxembourg E-mail address: [email protected] Department of Mathematics, Peking University, Beijing E-mail address: [email protected]
1810.10580	2	1810	2018-10-30T18:13:39	Ideals of etale groupoid algebras and Exel's Effros-Hahn conjecture	[ "math.RA", "math.OA" ]	We extend to arbitrary commutative base rings a recent result of Demeneghi that every ideal of an ample groupoid algebra over a field is an intersection of kernels of induced representations from isotropy groups, with a much shorter proof, by using the author's Disintegration Theorem for groupoid representations. We also prove that every primitive ideal is the kernel of an induced representation from an isotropy group; however, we are unable to show, in general, that it is the kernel of an irreducible induced representation. If each isotropy group is finite (e.g., if the groupoid is principal) and if the base ring is Artinian (e.g., a field), then we can show that every primitive ideal is the kernel of an irreducible representation induced from isotropy.	math.RA	math	IDEALS OF ´ETALE GROUPOID ALGEBRAS AND EXEL'S EFFROS-HAHN CONJECTURE BENJAMIN STEINBERG Abstract. We extend to arbitrary commutative base rings a recent re- sult of Demeneghi that every ideal of an ample groupoid algebra over a field is an intersection of kernels of induced representations from isotropy groups, with a much shorter proof, by using the author's Disintegration Theorem for groupoid representations. We also prove that every prim- itive ideal is the kernel of an induced representation from an isotropy group; however, we are unable to show, in general, that it is the kernel of an irreducible induced representation. If each isotropy group is finite (e.g., if the groupoid is principal) and if the base ring is Artinian (e.g., a field), then we can show that every primitive ideal is the kernel of an irreducible representation induced from isotropy. 1. Introduction The original Effros-Hahn conjecture [4, 5] suggested that every primitive ideal of a crossed product of an amenable locally compact group with a com- mutative C ∗-algebra should be induced from a primitive ideal of an isotropy group. The result was proved by Sauvageot [17] for discrete groups and a more general result than the original conjecture was proved by Gootman and Rosenberg in [7]. Crossed products of the above form are special cases of groupoid C ∗-algebras and analogues of the Effros-Hahn conjecture in the groupoid setting were achieved by Renault [15] and Ionescu and Williams [9]. In [18], the author initiated the study of convolution algebras of ample groupoids over commutative rings with unit; see also [1]. R. Exel conjec- tured at the PARS meeting in Gramado, 2014 (and perhaps earlier) that an analogue of the Effros-Hahn conjecture should hold in this context. The author had developed in [18] a theory of induction from isotropy groups in this setting and had proven that inducing an irreducible representation from an isotropy group results in an irreducible representation of the groupoid al- gebra. Date: November 1, 2018. 2010 Mathematics Subject Classification. 20M18, 20M25, 16S99,16S36, 22A22, 18F20. Key words and phrases. ´etale groupoids, groupoid algebras, Effros-Hahn conjecture. This work was supported in part by a PSC-CUNY grant and by the Fulbright Com- mission, which supported the author's visit to Brazil where much of this work was done. He thanks the Universidade Federal de Santa Catarina for its hospitality. 1 2 BENJAMIN STEINBERG In [3], Dokuchaev and Exel showed that if a discrete group G acts partially on a locally compact and totally disconnected space X, then every ideal of the partial crossed product Cc(X, k) ⋊ G, where Cc(X, k) is the ring of locally constant, compactly supported functions from X to the field k, is an intersection of ideals induced from isotropy. Note that such partial crossed products are ample groupoid convolution algebras. Since in a C ∗-algebra, every closed ideal is an intersection of primitive ideals, this result can be viewed as an analogue of Effros-Hahn for partial crossed products. Demeneghi [2] extended the result of Dokuchaev and Exel to arbitrary ample groupoid algebras over a field. Namely, he showed that each ideal is an intersection of kernels of induced representations from isotropy sub- groups. His proof is rather indirect. First he develops a theory of induced representations for crossed products of the form Cc(X, k) ⋊ S where S is an inverse semigroup acting on a locally compact and totally disconnected space X. Then he proves the result for such crossed products. Finally, he proves that groupoid convolution algebras are such crossed products us- ing the full strength of his theory (and the converse is essentially true as well) and he shows that crossed product induction corresponds to groupoid induction under the isomorphism. His paper is around 50 pages in all. In this paper, we prove that over an arbitrary base commutative ring each ideal of an ample groupoid convolution algebra is an intersection of kernels of induced representations from isotropy groups. Moreover, our proof is direct -- circumventing entirely the crossed product machinery -- and short. It relies on the author's Disintegration Theorem [19], which shows that modules for ample groupoid convolution algebras come from sheaves on the groupoid. This machinery is not very cumbersome to develop and is quite useful for analyzing irreducible representations, as was done in [20]. In future work, it will be shown that the Disintegration Theorem can be used to establish the isomorphism between inverse semigroup crossed products and groupoid algebras directly, without using induced representations. We also obtain some new progress on Exel's original conjecture on the structure of primitive ideals for groupoid algebras. Namely, we show that every primitive ideal is the kernel of a single representation induced from an isotropy group (rather than an infinite intersection of such kernels). We are, unfortunately, not able to show in general that it is the kernel of an irreducible representation induced from an isotropy group. We are, however, able to prove Exel's version of the Effros-Hahn conjecture on primitive ideals if the base ring R is Artinian and each isotropy group is either finite, or locally finite abelian with orders of elements invertible in R/J(R) where J(R) is the Jacobson radical of R (e.g., if R has characteristic zero). 2. Preliminaries This section summarizes definitions and results from [18] and [19] that we use throughout. There are no new results in this section. IDEALS OF ´ETALE GROUPOID ALGEBRAS 3 2.1. Groupoids. Following Bourbaki, compactness will include the Haus- dorff axiom throughout this paper. However, we do not require locally compact spaces to be Hausdorff. A topological groupoid G = (G (0), G (1)) is ´etale if its domain map d (or, equivalently, its range map r) is a local homeomorphism. In this case, identifying objects with identity arrows, we have that G (0) is an open subspace of G (1) and the multiplication map is a local homeomorphism. See, for example, [6, 13, 16]. Following [13], an ´etale groupoid is called ample if its unit space G (0) is locally compact Hausdorff with a basis of compact open subsets. We shall say that an ample groupoid G is Hausdorff if G (1) is Hausdorff. A local bisection of an ´etale groupoid G is an open subset U ⊆ G (1) such that both d U and r U are homeomorphisms with their images. The local bisections form a basis for the topology on G (1) [6]. The set Γ(G ) of local bisections is an inverse monoid (cf. [11]) under the binary operation U V = {γη γ ∈ U, η ∈ V, d(γ) = r(η)}. The semigroup inverse is given by U −1 = {γ−1 γ ∈ U }. The set Γc(G ) of compact local bisections is an inverse subsemigroup of Γ(G ) [13]. Note that G is ample if and only if Γc(G ) is a basis for the topology on G (1) [6, 13]. If u ∈ G (0), then the orbit Ou of u consists of all v ∈ G (0) such that there is an arrow γ with d(γ) = u and r(γ) = v. The orbits form a partition of G (0). A subset X ⊆ G (0) is invariant if it is a union of orbits. If u ∈ G (0), the isotropy group of G at u is Gu = {γ ∈ G (1) d(γ) = u = r(γ)}. Isotropy groups of elements in the same orbit are isomorphic. 2.2. Ample groupoid algebras. Fix a commutative ring with unit R. The author [18] associated an R-algebra RG to each ample groupoid G as follows. We define RG to be the R-span in RG (1) of the characteristic functions 1U of compact open subsets U of G (1). It is shown in [18, Proposition 4.3] that RG is spanned by the elements 1U with U ∈ Γc(G ). If G (1) is Hausdorff, then RG consists of the locally constant R-valued functions on G (1) with compact support. Convolution is defined on RG by f ∗ g(γ) = Xd(η)=d(γ) f (γη−1)g(η) = Xαβ=γ f (α)g(β). The finiteness of the sums is proved in [18]. The fact that the convolution belongs to RG comes from the computation 1U ∗ 1V = 1U V for U, V ∈ Γc(G ) [18]. The algebra RG is unital if and only if G (0) is compact, but it always has local units (i.e., is a directed union of unital subrings) [18, 19]. A module M over a ring S with local units is termed unitary if SM = M . A module M over a unital ring is unitary if and only if 1m = m for all m ∈ M . The category of unitary S-modules is denoted S-mod. Notice that every simple 4 BENJAMIN STEINBERG module is unitary; M is simple if SM 6= 0 and M has no proper, non-zero submodules. 2.3. Induced modules. We recall from [18] the induction functor Indu : RGu-mod −→ RG -mod. For u ∈ G(0), let G u = d−1(u) denote the set of all arrows starting at u. Then Gu acts freely on the right of G u by multiplication. Hence RG u is a free right RGu-module. A basis can be obtained by choosing, for each v ∈ Ou, an arrow γv : u −→ v. We normally choose γu = u. There is a left RG -module structure on RG u given by f α = Xd(γ)=r(α) f (γ)γα (1) for α ∈ G u and f ∈ RG . The RG -action commutes with the RGu-action by associativity and so RG u is an RG -RGu-bimodule, unitary under both actions (as is easily checked). The functor Indu is defined by Indu(M ) = RG u ⊗RGu M. This functor is exact by freeness of RG u as a right RGu-module and there is, in fact, an R-module direct sum decomposition Indu(M ) = Mv∈Ou γv ⊗ M. The action of f ∈ RG in these coordinates is given by f (γv ⊗ m) = Xw∈Ou γw ⊗ Xγ : v−→w f (γ)(γ−1 w γγv)m, (2) as is easily checked. An immediate corollary of (2) is the following (see also [2]). Proposition 1. Let u ∈ G (0) and M an RGu-module. Fix γu : u −→ v for all v ∈ Ou. Then the equality Ann(Indu(M )) =(f ∈ RG ∀v, w ∈ Ou, Xγ : v−→w f (γ)(γ−1 w γγv) ∈ Ann(M )) holds. A crucial result is that induction preserves simplicity. Theorem 2 (Steinberg, Prop. 7.19, Prop. 7.20 of [18]). Let M be a simple RGu-module with u ∈ G (0). Then Indu(M ) is a simple RG -module. More- over, the functor Indu reflects isomorphism and Indu(M ) ∼= Indv(N ) implies Ou = Ov. IDEALS OF ´ETALE GROUPOID ALGEBRAS 5 The Effros-Hahn conjecture for ample groupoids, first stated to the best of our knowledge by Exel at the PARS2014 conference in Gramado (see also [3]), says that each primitive ideal of RG is of the form Ann(Indu(M )) for some u ∈ G (0) and simple RGu-module M . 2.4. Modules over ample groupoid algebras. The key ingredient to our approach is the author's analogue [19] of Renault's Disintegration The- orem [14] for modules over groupoid algebras. Here R will be a fixed commu- tative ring with unit and G an ample groupoid (not necessarily Hausdorff). A G -sheaf E = (E, p) consists of a space E, a local homeomorphism p : E −→ G (0) and an action map G (1) ×G (0) E −→ E (where the fiber product is with respect to d and p), denoted (γ, e) 7→ γe, satisfying the following axioms: • p(e)e = e for all e ∈ E; • p(γe) = r(γ) whenever p(e) = d(γ); • γ(ηe) = (γη)e whenever p(e) = d(η) and d(γ) = r(η). A G -sheaf of R-modules is a G -sheaf E = (E, p) together with an R- module structure on each stalk Eu = p−1(u) such that: • the zero section, u 7→ 0u (the zero of Eu), is continuous; • addition E ×G (0) E −→ E is continuous; • scalar multiplication R × E −→ E is continuous; • for each γ ∈ G (1), the map Ed(γ) −→ Er(γ) given by e 7→ γe is R-linear; where R has the discrete topology in the third item. Note that the first three conditions are equivalent to (E, p) being a sheaf of R-modules over G (0). Crucial to this paper is that Eu is an RGu-module for each u ∈ G (0). Note that the zero subspace 0 = {0u u ∈ G(0)} is an open subspace of E, being the image of a section of a local homeomor- phism. The support of E is supp(E) = {u ∈ G (0) Eu 6= {0u}}. Note that supp(E) is an invariant subset of G0 but it need not be closed. A (global) section of E is a continuous mapping s : G (0) −→ E such that p ◦ s = 1G (0). Note that if s : G (0) −→ E is a section, then its support supp(s) = s−1(E \ 0) is closed. We denote by Γc(E) the set of (global) sections with compact sup- port. Note that Γc(E) is an R-module with respect to pointwise operations and it be comes a unitary left RG -module under the operation (f s)(u) = Xr(γ)=u f (γ)γs(d(γ)) = Xv∈Ou Xγ : v−→u f (γ)γs(v). (3) 6 BENJAMIN STEINBERG See [19] for details (where right actions and right modules are used). If e ∈ Eu, then there is always a global section s with compact support such that s(u) = e. Indeed, we can choose a neighborhood U of e such that pU : U −→ p(U ) is a homeomorphism with p(U ) open. Then we can find a compact open neighborhood V of u with V ⊆ U and define s to be the restriction of (pU )−1 on V and 0, elsewhere. Then s is continuous, s(u) = e and the support of s is closed and contained in V , whence compact. Conversely, if M is a unitary left RG -module, we can define a G -sheaf of 1U M (with the direct limit R-modules Sh(M ) = (E, p) where Eu = lim over all compact open neighborhoods U of u in G (0)) and if γ : u −→ v and [m]u is the class of m at u, then γ[m]u = [U m]v where U is any compact −→u∈U local bisection containing γ. Here E = `u∈G (0) Eu has the germ topology. See [19] for details. Theorem 3 (Steinberg [19]). The functor E 7→ Γc(E) is an equivalence between the category BGR of G -sheaves of R-modules and RG -mod with quasi-inverse M 7→ Sh(M ). Remark 4. As a consequence of Theorem 3, M ∼= Γc(Sh(M )) for any unitary module M and so we have an equality of annihilator ideals Ann(M ) = Ann(Γc(Sh(M ))). In particular, we have I = Ann(Γc(Sh(RG /I))). Thus we can describe the ideal structure of RG in terms of the annihilators of modules of the form Γc(E). 3. The ideal structure of ample groupoid algebras In this section, we show how the Disintegration Theorem (Theorem 3) provides information about the ideal structure of RG . 3.1. General ideals. Our main goal in this subsection is to prove the fol- lowing theorem, generalizing a result of Demeneghi [2] that was originally proved over fields. Demeneghi's proof is indirect, via crossed products, and therefore quite long. Our proof is direct, using the Disintegration Theorem, and shorter, even including the 11 pages of [19]. Theorem 5. Let G be an ample groupoid and R a ring. Let E = (E, p) be a G -sheaf of R-modules. Then the equality Ann(Γc(E)) = \u∈G (0) Ann(Indu(Eu)) holds. Consequently, every ideal I ⊳ RG is an intersection of annihilators of induced modules. Proof. The final statement follows from the first by the equivalence of cat- egories in Theorem 3 (cf. Remark 4). So we prove the first statement. Let I = Ann(Γc(E)) and put Ju = Ann(Indu(Eu)) for u ∈ G (0). Set J =Tu∈G (0) Ju. Then we want to prove that I = J. IDEALS OF ´ETALE GROUPOID ALGEBRAS 7 Fix u ∈ G (0) and, for each v ∈ Ou, choose γv : u −→ v. To show I ⊆ Ju, it suffices, by Proposition 1, to show that if f ∈ I, then, for each v, w ∈ Ou, we have that f (γ)(γ−1 w γγv) ∈ Ann(Eu). Xγ : v−→w So let e ∈ Eu and let s ∈ Γc(E) with s(u) = e. Since G is ample, the set r−1(w) ∩ supp(f ) is finite. Since G (0) is Hausdorff, we can find U ⊆ G (0) compact open with v ∈ U and U ∩ d(r−1(w) ∩ supp(f )) ⊆ {v}. Let Uv be a compact local bisection containing γv and Uw a compact local bisection containing γw. Replacing Uv by U Uv, we may assume that r(Uv) ∩ d(r−1(w) ∩ supp(f )) ⊆ {v}. By construction, the only elements in the support of 1U −1 u are of the form γ−1 we have by (3) that w γγv with γ : v −→ w and f (γ) 6= 0. As 1U −1 w w ∗f ∗1Uv with range ∗f ∗1Uv ∈ I, 0 = (1U −1 w ∗ f ∗ 1Uv s)(u) (1U −1 w ∗ f ∗ 1Uv )(α)αs(d(α)) = Xr(α)=u = Xγ : v−→w = Xγ : v−→w f (γ)(γ−1 w γγv)s(u) f (γ)(γ−1 w γγv)e as required. Thus I ⊆ Ju for all u ∈ G (0). Suppose now that f ∈ J and let s ∈ Γc(E). Then (f s)(v) = Xu∈Ov Xγ : u−→v f (γ)γs(u). (4) Let us fix u ∈ Ov and let γu = u and γv : u −→ v be arbitrary. Then, since f ∈ J ⊆ Ju, we have by Proposition 1 that Multiplying on the left by γv yields that, for each u ∈ Ov, f (γ)(γ−1 v γ)s(u) = 0. f (γ)γs(u) = 0 Xγ : u−→v Xγ : u−→v and so the right hand side of (4) is 0. Thus f ∈ I. This completes the proof. (cid:3) Remark 6. More concretely, if I ⊳ RG is an ideal, then following the con- struction of the proof we see that Ann Indu lim −→ u∈U 1U (RG /I)!! I = \u∈G (0) 8 BENJAMIN STEINBERG where lim −→u∈U 1U (RG /I) has the RGu-module structure γ[1U f + I] = [1V 1U f + I] for V any compact local bisection containing γ. 3.2. Primitive ideals. Recall that an ideal I of a ring is primitive if it is the annihilator of a simple module. (Technically, we should talk about left primitive ideals since we are using left modules, but because groupoid alge- bras admit an involution, it doesn't matter.) We prove that each primitive ideal is the annihilator of a single induced representation (rather than an intersection of such annihilators, as in Theorem 5). Unfortunately, we are not yet able to show, in general, that the induced representation is simple. Still, this is new progress toward Exel's Effros-Hahn conjecture. Theorem 7. Let R be a commutative ring with unit and G an ample group- oid. Let I ⊳ RG be a primitive ideal. Then I = Ann(Indu(M )) for some u ∈ G (0) and RGu-module M . Proof. By Theorem 3 we may assume that our simple module with annihi- lator I is of the form Γc(E) for some G -sheaf E = (E, p) of R-modules. Let u ∈ supp(E) and put Ju = Ann(Indu(Eu)). We claim that I = Ju. We know that I ⊆ Ju by Theorem 5; we must prove the converse. Suppose that Ju does not annihilate Γc(E). Then we can find a section s with Jus 6= {0}. As Ju is an ideal, Jus is a submodule and so Jus = Γc(E). Let 0u 6= e ∈ Eu. Then there is a section t ∈ Γc(E) such that t(u) = e. Let f ∈ Ju with f s = t. Then we have that e = t(u) = (f s)(u) = Xv∈Ou Xγ : v−→u f (γ)γs(v). (5) Let us fix v ∈ Ou and fix γv : u −→ v; put γu = u. Then by the assumption f ∈ Ju and Proposition 1, we have (since γ−1 v s(v) ∈ Eu) that 0 = Xγ : v−→u f (γ)(γγv)(γ−1 v s(v)) = Xγ : v−→u f (γ)γs(v). Thus the right hand side of (5) is 0, contradicting that e 6= 0. We conclude that Ju ⊆ I. This completes the proof. (cid:3) We now show that the RGu-module M above can be chosen to be simple under some strong hypotheses on the base ring and isotropy groups. Let J(S) denote the Jacobson radical of a ring S. A ring S is called a left max ring if each non-zero left S-module has a maximal (proper) submodule. For example, any Artinian ring S is a left max ring. Indeed, if M 6= 0, then J(S)M 6= M by nilpotency of the Jacobson radical. But M/J(S)M is then a non-zero S/J(S)-module and every non-zero module over a semisimple ring is a direct sum of simple modules and hence has a simple quotient. Thus M has a maximal proper submodule. A result of Hamsher [8] says that if S is commutative, then S is a left max ring if and only if J(S) is T -nilpotent (e.g., if J(S) is nilpotent) and S/J(S) is von Neumann regular ring. We IDEALS OF ´ETALE GROUPOID ALGEBRAS 9 now establish the Effros-Hahn conjecture in the case that all isotropy group rings are left max rings. Theorem 8. Let R be a commutative ring and G an ample groupoid such that RGu is a left max ring for all u ∈ G (0). Then the primitive ideals of RG are exactly the ideals of the form Ann(Indu(M )) where M is a simple RGu-module. Proof. By Theorem 2 it suffices to show that any primitive ideal I is of the form Ann(Indu(M )) where M is a simple RGu-module. By Theorem 3 we may assume that our simple module with annihilator I is of the form Γc(E) for some G -sheaf E = (E, p) of R-modules. Let u ∈ supp(E). We already know from the proof of Theorem 7 that I = Ann(Indu(Eu)). Let N be a maximal submodule of Eu (which exists by assumption on RGu) and let J = Ann(Indu(Eu/N )). Since Eu/N is simple, it suffices to show that J = I. Clearly, I ⊆ J (by Proposition 1) since Ann(Eu) ⊆ Ann(Eu/N ). So it suffices to show that J annihilates Γc(E). Suppose that this is not the case. Then there exists s ∈ Γc(E) with Js 6= 0. Since J is an ideal and Γc(E) is simple, we deduce Js = Γc(E). Let e ∈ Eu \ N (using that N is a proper submodule) and let t ∈ Γc(E) with t(u) = e. Then t = f s with f ∈ J. Let us compute e = t(u) = (f s)(u) = Xv∈Ou Xγ : v−→u f (γ)γs(v). (6) Let us fix v ∈ Ou, fix γv : u −→ v and set γu = u. Then by the assumption f ∈ J and Proposition 1, we have (since γ−1 v s(v) ∈ Eu) that Xγ : v−→u f (γ)γs(v) = Xγ : v−→u f (γ)(γγv)(γ−1 v s(v)) ∈ N. We deduce from (6) that e ∈ N , which is a contradiction. It follows that J annihilates Γc(E) and so I = J. (cid:3) We obtain as a consequence a special case of Exel's Effros-Hahn conjec- ture, which will include the algebras of principal groupoids (or, more gen- erally, groupoids with finite isotropy) over a field. Recall that a groupoid is called principal if all its isotropy groups are trivial. It is well known that a group ring RG is Artinian if and only if R is Artinian and G is finite [12]. A result of Villamayor [21] says that, for a group G and commutative ring R, one has that RG is von Neumann regular if and only if R is von Neumann regular, G is locally finite and the order of any element of G is invertible in R. The reader should recall that any Artinian semisimple ring is von Neumann regular and also that any von Neumann regular ring has a zero Jacobson radical [10, Cor. 4.24]. Corollary 9. Let R be a commutative Artinian ring and G an ample group- oid such that each isotropy group of G is either finite or locally finite abelian with elements having order invertible in R/J(R). Then the primitive ideals 10 BENJAMIN STEINBERG of RG are precisely the annihilators of modules induced from simple modules of isotropy group rings. Proof. By Theorem 8 it suffices to show that RGu is a left max ring for each u ∈ G (0). If Gu is finite, then RGu is Artinian and hence a left max ring. Suppose that Gu is locally finite abelian with each element of order invertible in R/J(R). In particular, RGu is commutative and it suffices by Hamsher's theorem to show that J(RGu) is nilpotent and RGu/J(RGu) is von Neumann regular. First note that since J(R) is nilpo- tent, we have that J(R)RGu is a nilpotent ideal of RGu and hence contained ∼= (R/J(R))Gu is von Neu- J(RGu). On the other hand, RGu/J(R)RGu mann regular by [21] and the hypotheses, and hence has a zero radical. Thus J(RGu) = J(R)RGu, and hence is nilpotent, and RGu/J(RGu) is von Neumann regular. Therefore, RGu is a left max ring. (cid:3) If R is commutative Artinian of characteristic zero, then the order of any element of a locally finite group is invertible in R/J(R) (which is a product of fields of characteristic zero) and so Corollary 9 applies if each isotropy group is either finite or locally finite abelian. Acknowledgments. The author would like to thank Enrique Pardo for pointing out Villamayor's result [21] and that regular rings have zero Ja- cobson radical, leading to the current formulation of Corollary 9, which improves on a previous version. References [1] L. O. Clark, C. Farthing, A. Sims, and M. Tomforde. A groupoid generalisation of Leavitt path algebras. Semigroup Forum, 89(3):501 -- 517, 2014. [2] P. Demeneghi. The ideal structure of Steinberg algebras. ArXiv e-prints, Oct. 2017. [3] M. Dokuchaev and R. Exel. The ideal structure of algebraic partial crossed products. Proc. Lond. Math. Soc. (3), 115(1):91 -- 134, 2017. [4] E. G. Effros and F. Hahn. Locally compact transformation groups and C ∗- algebras. Memoirs of the American Mathematical Society, No. 75. American Mathematical Society, Providence, R.I., 1967. [5] E. G. Effros and F. Hahn. Locally compact transformation groups and C ∗-algebras. Bull. Amer. Math. Soc., 73:222 -- 226, 1967. [6] R. Exel. Inverse semigroups and combinatorial C ∗-algebras. Bull. Braz. Math. Soc. (N.S.), 39(2):191 -- 313, 2008. [7] E. C. Gootman and J. Rosenberg. The structure of crossed product C ∗-algebras: a proof of the generalized Effros-Hahn conjecture. Invent. Math., 52(3):283 -- 298, 1979. [8] R. M. Hamsher. Commutative rings over which every module has a maximal sub- module. Proc. Amer. Math. Soc., 18:1133 -- 1137, 1967. [9] M. Ionescu and D. Williams. The generalized Effros-Hahn conjecture for groupoids. Indiana Univ. Math. J., 58(6):2489 -- 2508, 2009. [10] T. Y. Lam. A first course in noncommutative rings, volume 131 of Graduate Texts in Mathematics. Springer-Verlag, New York, 1991. [11] M. V. Lawson. Inverse semigroups. World Scientific Publishing Co. Inc., River Edge, NJ, 1998. The theory of partial symmetries. [12] D. S. Passman. The algebraic structure of group rings. Pure and Applied Mathematics. Wiley-Interscience [John Wiley & Sons], New York-London-Sydney, 1977. IDEALS OF ´ETALE GROUPOID ALGEBRAS 11 [13] A. L. T. Paterson. Groupoids, inverse semigroups, and their operator algebras, volume 170 of Progress in Mathematics. Birkhauser Boston Inc., Boston, MA, 1999. [14] J. Renault. Repr´esentation des produits crois´es d'alg`ebres de groupoıdes. J. Operator Theory, 18(1):67 -- 97, 1987. [15] J. Renault. The ideal structure of groupoid crossed product C ∗-algebras. J. Operator Theory, 25(1):3 -- 36, 1991. With an appendix by Georges Skandalis. [16] P. Resende. ´Etale groupoids and their quantales. Adv. Math., 208(1):147 -- 209, 2007. [17] J.-L. Sauvageot. Id´eaux primitifs induits dans les produits crois´es. J. Funct. Anal., 32(3):381 -- 392, 1979. [18] B. Steinberg. A groupoid approach to discrete inverse semigroup algebras. Adv. Math., 223(2):689 -- 727, 2010. [19] B. Steinberg. Modules over ´etale groupoid algebras as sheaves. J. Aust. Math. Soc., 97(3):418 -- 429, 2014. [20] B. Steinberg. Simplicity, primitivity and semiprimitivity of ´etale groupoid algebras with applications to inverse semigroup algebras. J. Pure Appl. Algebra, 220(3):1035 -- 1054, 2016. [21] O. E. Villamayor. On weak dimension of algebras. Pacific J. Math., 9:941 -- 951, 1959. (B. Steinberg) Department of Mathematics, City College of New York, Con- vent Avenue at 138th Street, New York, New York 10031, USA E-mail address: [email protected]
1301.0731	2	1301	2013-06-11T22:28:45	The direct limit closure of perfect complexes	[ "math.RA", "math.AC" ]	Every projective module is flat. Conversely, every flat module is a direct limit of finitely generated free modules; this was proved independently by Govorov and Lazard in the 1960s. In this paper we prove an analogous result for complexes of modules, and as applications we reprove some results due to Enochs and Garc\'ia Rozas and to Neeman.	math.RA	math	THE DIRECT LIMIT CLOSURE OF PERFECT COMPLEXES LARS WINTHER CHRISTENSEN AND HENRIK HOLM To Hans-Bjørn Foxby -- our teacher, colleague, and friend Abstract. Every projective module is flat. Conversely, every flat module is a direct limit of finitely generated free modules; this was proved independently by Govorov and Lazard in the 1960s. In this paper we prove an analogous result for complexes of modules, and as applications we reprove some results due to Enochs and Garc´ıa Rozas and to Neeman. ] . A R h t a m [ 2 v 1 3 7 0 . 1 0 3 1 : v i X r a 1. Introduction Let R be a ring. In contrast to the projective objects in the category of R-modules, i.e. the projective R-modules, the projective objects in the category of R-complexes are not of much utility; indeed, they are nothing but contractible (split) complexes of projective R-modules. In the category of complexes, the relevant alternative to projectivity -- from the homological point of view, at least -- is semi-projectivity. A complex P is called semi-projective (or DG-projective) if the the total Hom functor Hom(P, −) preserves surjective quasi-isomorphisms, i.e. surjective morphisms that induce isomorphisms in homology. The semi-projective complexes are exactly the cofibrant objects in the standard model structure on the category of complexes; see Hovey [11, §2.3]. Alternatively, a complex is semi-projective if and only if it consists of projective modules and it is K-projective in the sense of Spaltenstein [18]. The notion of semi-projectivity in the category of complexes extends the notion of projectivity in the category of modules in a natural and useful way: A module is projective if and only if it is semi-projective when viewed as a complex. Similarly, a complex F is semi-flat if the total tensor product functor − ⊗ F preserves injective quasi-isomorphisms; equivalently, F is a complex of flat modules and K-flat in the sense of [18]. A module is flat if and only if it is semi-flat when viewed as a complex. Every semi-projective complex is semi-flat, and simple ex- amples of semi-projective complexes are bounded complexes of finitely generated projective modules, also known as perfect complexes. The class of semi-flat com- plexes is closed under direct limits, and our main result, Theorem 1.1 below, shows that every semi-flat complex is a direct limit of perfect complexes. For modules, the theorem specializes to a classic result, proved independently by Govorov [9] and Lazard [13]: Every flat module is a direct limit of finitely generated free modules. Date: 11 June 2013. 2010 Mathematics Subject Classification. Primary 16E05. Secondary 13D02; 16E35. Key words and phrases. Direct limit; finitely presented complex; perfect complex; purity; semi-flat complex; semi-projective complex. This work was partly supported by NSA grant H98230-11-0214 (L.W.C.). 1 2 L. W. CHRISTENSEN AND H. HOLM 1.1 Theorem. For an R-complex F the following conditions are equivalent. (i) F is semi-flat. (ii) Every morphism of R-complexes ϕ : N → F with N bounded and degreewise finitely presented admits a factorization, ϕ N ❄❄❄❄❄❄ κ L , / F ?⑧⑧⑧⑧⑧⑧ λ where L is a bounded complex of finitely generated free R-modules. (iii) There exists a set {Lu}u∈U of bounded complexes of finitely generated free R-modules and a pure epimorphism `u∈U Lu → F . (iv ) F is isomorphic to a filtered colimit of bounded complexes of finitely generated free R-modules. (v ) F is isomorphic to a direct limit of bounded complexes of finitely generated free R-modules. The theorem is proved in Section 5. The terminology used in the statement is clar- ified in the sections leading up to the proof. In Section 4 we show that the finitely presented objects in the category of complexes are exactly the bounded complexes of finitely presented modules. Results of Breitsprecher [5] and Crawley-Boevey [6] show that the category of complexes is locally finitely presented, see Remark 4.7. Therefore, the equivalence of (ii), (iii), and (iv ) follows from [6, (4.1)]. Further- more, a result by Ad´amek and Rosick´y [1, thm. 1.5] shows that (iv ) and (v ) are equivalent for quite general reasons; thus our task is to prove that the equivalent conditions (ii) -- (v ) are also equivalent to (i). The characterization of semi-flat complexes in Theorem 1.1 opens to a study of the interplay between semi-flatness and purity in the category of complexes; this is the topic of Section 6. We show, for example, that a complex F is semi-flat if and only if every surjective quasi-isomorphism M → F is a pure epimorphism. This compares to Lazard's [13, cor. 1.3] which states that a module F is flat if and only if every surjective homomorphism M → F is a pure epimorphism. In the final Section 7, we use Theorem 1.1 to reprove a few results due to Enochs and Garc´ıa Rozas [7] and to Neeman [17]; our proofs are substantially different from the originals. In Theorem 7.3 we show that an acyclic semi-flat complex is a direct limit of contractible perfect complexes. Combined with a result of Benson and Goodearl [4] this enables us to show in Theorem 7.8 that a semi-flat complex of projective modules is semi-projective. 2. Complexes In this paper R is a ring, and the default action on modules is on the left. Thus, R-modules are left R-modules, while right R-modules are considered to be (left) modules over the opposite ring R◦. The definitions and results listed in this section are standard and more details can be found in textbooks, such as Weibel's [19], and in the paper [2] by Avramov and Foxby. An R-complex M is a graded R-module M = `v∈Z Mv equipped with a dif- ferential, that is, an R-linear map ∂M : M → M that satisfies ∂M ∂M = 0 and / ? THE DIRECT LIMIT CLOSURE OF PERFECT COMPLEXES 3 ∂M (Mv) ⊆ Mv−1 for every v ∈ Z. The homomorphism Mv → Mv−1 induced by ∂M is denoted ∂M v . Thus, an R-complex M can be visualized as follows, M = · · · −→ Mv+1 ∂M v+1−−−→ Mv ∂M v−−→ Mv−1 −→ · · · . The category of R-complexes is denoted C(R). We identify the category of graded R-modules with the full subcategory of C(R) whose objects are R-complexes with zero differential. For an R-complex M with differential ∂M , set Z(M ) = Ker ∂M , B(M ) = Im ∂M , C(M ) = Coker ∂M , and H(M ) = Z(M )/ B(M ); they are sub-, quotient, and sub- quotient complexes of M . Furthermore, Z(−), B(−), H(−) and C(−) are additive endofunctors on C(R). A complex M with H(M ) = 0 is called acyclic. The shift of M is the complex ΣM with (ΣM )v = Mv−1 and ∂ ΣM v−1. A morphism α : M → N of com- plexes is called a quasi-isomorphism if H(α) : H(M ) → H(N ) is an isomorphism. v = −∂M 2.1. To a morphism α : M → N of R-complexes one associates a complex Cone α, called the mapping cone of α; it fits into a degreewise split exact sequence, The morphism α is a quasi-isomorphism if and only if Cone α is acyclic. 0 −→ N −→ Cone α −→ ΣM −→ 0 . 2.2. Let M and N be R-complexes. The total Hom complex, written HomR(M, N ), yields a functor HomR(−, −) : C(R)op × C(R) −→ C(Z) . The functor HomR(M, −) commutes with mapping cones, that is, for every mor- phism α of R-complexes there is an isomorphism of Z-complexes, Cone HomR(M, α) ∼= HomR(M, Cone α) . There is an equality of abelian groups, where the right-hand side is the hom-set in the category C(R). Z0(HomR(M, N )) = C(R)(M, N ) , 2.3. Let M be an R◦-complex and let N be an R-complex. The total tensor product complex, written M ⊗R N , yields a functor − ⊗R − : C(R◦) × C(R) −→ C(Z) . The functor M ⊗R − commutes with mapping cones, that is, for every morphism α of R-complexes there is an isomorphism of Z-complexes, Cone(M ⊗R α) ∼= M ⊗R Cone α . For a homogeneous element m in a graded module (or a complex) M , we write m for its degree. 2.4. Let M be an R-complex. The biduality morphism δM : M −→ HomR◦ (HomR(M, R), R) is given by for homogeneous elements m ∈ M and ψ ∈ HomR(M, R). δM (m)(ψ) = (−1)ψmψ(m) 4 L. W. CHRISTENSEN AND H. HOLM The morphism δM of R-complexes is an isomorphism if M is a complex of finitely generated projective R-modules. 2.5. Let M and N be R-complexes and let X be a complex of R -- R◦-bimodules. The tensor evaluation morphism ωM XN : HomR(M, X) ⊗R N −→ HomR(M, X ⊗R N ) is given by ωM XN (ψ ⊗ n)(m) = (−1)mnψ(m) ⊗ n for homogeneous elements ψ ∈ HomR(M, X), n ∈ N , and m ∈ M . The morphism ωM XN of Z-complexes is an isomorphism if M is a bounded complex of finitely generated projective R-modules and X = R. 2.6. Let M be an R-complex, let N be an R◦-complex, and let X be a complex of R -- R◦-bimodules. The homomorphism evaluation morphism θXN M : HomR◦ (X, N ) ⊗R M −→ HomR◦ (HomR(M, X), N ) is given by θXN M (ψ ⊗ m)(ϕ) = (−1)ϕmψϕ(m) for homogeneous elements ψ ∈ HomR◦ (X, N ), m ∈ M , and ϕ ∈ HomR(M, X). The morphism θXN M of Z-complexes is an isomorphism if M is a bounded complex of finitely generated projective R-modules and X = R. We refer to MacLane [14, sec. IX.1] for background on colimits. 3. Filtered colimits 3.1 Definition. Let A be a category. By a filtered colimit in A we mean the colimit of a functor F : J → A, which is denoted colimJ∈J F(J), where J is a skeletally small filtered category. We reserve the term direct limit for the colimit of a direct system, i.e. of a functor J → A where J is the filtered category associated to a directed set, i.e. a filtered preordered set. Notice that some authors, including Crawley-Boevey [6], use the term "direct limit" for any filtered colimit. For a direct system {Au → Av}u6v it is customary Au for its direct limit, i.e. its colimit, however, we shall stick to the to write lim −→ notation colim Au. Let A and B be categories that have (all) filtered colimits. Recall that a functor T : A → B is said to preserve (filtered) colimits if the canonical morphism in B, colim J∈J T(F(J)) −→ T(colim J∈J F(J)) , is an isomorphism for every (filtered) colimit colimJ∈J F(J) in A. We need a couple of facts about filtered colimits of complexes; the arguments are given in [19, lem. 2.6.14 and thm. 2.6.15]. 3.2. The following assertions hold. (a) Every homogeneous element in a filtered colimit, colimJ∈J F(J), in C(R) is in the image of the canonical morphism F(J) → colimJ∈J F(J) for some J ∈ J . (b) Filtered colimits in C(R) are exact (colimits are always right exact). THE DIRECT LIMIT CLOSURE OF PERFECT COMPLEXES 5 3.3 Lemma. Let A be a category with filtered colimits and let T′, T, T′′ : A → C(R) be functors. The following assertions hold. (a) If 0 → T′ → T → T′′ is an exact sequence and if T and T′′ preserve filtered colimits, then T′ preserves filtered colimits. (b) If T′ → T → T′′ → 0 is an exact sequence and if T′ and T preserve filtered colimits, then T′′ preserves filtered colimits. (c) If 0 → T′ → T → T′′ → 0 is an exact sequence and if T′ and T′′ preserve filtered colimits, then T preserves filtered colimits. Proof. Let J be a skeletally small filtered category and let F : J → A be a functor. (a): Exactness of the sequence 0 → T′ → T → T′′ and left exactness of filtered colimits in C(R) yield the following commutative diagram, 0 0 / colim J∈J T′(F(J)) colim J∈J T(F(J)) colim J∈J T′′(F(J)) µ′ µ µ′′ / T′(colim J∈J F(J)) / T(colim J∈J F(J)) / T′′(colim J∈J F(J)) , where µ′, µ, and µ′′ are the canonical morphisms. If µ and µ′′ are isomorphisms, then so is µ′ by the Five Lemma. Parts (b) and (c) have similar proofs. (cid:3) 3.4 Proposition. The functors Z, C, B, H : C(R) → C(R) preserve filtered colimits. Proof. A graded R-module is considered as an R-complex with zero differential; let GM(R) be the full subcategory of C(R) whose objects are all graded R-modules. Since the inclusion functor i : GM(R) → C(R) preserves filtered colimits, it suffices to argue that C preserves filtered colimits when viewed as a functor from C(R) to GM(R). However, this functor C has a right adjoint, namely the inclusion functor i, so it follows from (the dual of) [14, V§5 thm. 1] that C preserves colimits. Denote by I the identity functor on C(R). Since I and C preserve filtered colimits, Lemma 3.3 applies to the short exact sequence 0 → B → I → C → 0 to show that B preserves filtered colimits. From the short exact sequences, 0 −→ H −→ C −→ ΣB −→ 0 and 0 −→ B −→ Z −→ H −→ 0 we now conclude that H and Z preserve filtered colimits as well. (cid:3) 3.5 Proposition. For every R-complex N the functor − ⊗R N : C(R◦) → C(Z) pre- serves colimits. For every bounded R-complex P of finitely generated projective modules the functor HomR(P, −) : C(R) → C(Z) preserves colimits. Proof. The functor − ⊗R N has a right adjoint, namely HomZ(N, −), so it follows from (the dual of) [14, V§5 thm. 1] that − ⊗R N preserves colimits. If P is a bounded complex of finitely generated projective R-modules, then HomR(P, R) is a bounded complex of finitely generated projective R◦-modules. By 2.4 and 2.6 there are natural isomorphisms of functors from C(R) to C(Z), HomR(P, −) ∼= HomR(HomR◦ (HomR(P, R), R), −) ∼= HomR(R, −) ⊗R◦ HomR(P, R) ∼= − ⊗R◦ HomR(P, R) , and the desired conclusion follows from the first assertion. (cid:3) / / / / / / / / 6 L. W. CHRISTENSEN AND H. HOLM 4. Finitely presented objects in the category of complexes Let A be an additive category. Following Crawley-Boevey [6], an object A in A is called finitely presented if the functor A(A, −) preserves filtered colimits. The category A is called locally finitely presented if the category of finitely presented objects in A is skeletally small and if every object in A is a filtered colimit of finitely presented objects; see [6]. 4.1 Definition. For an R-module F and v ∈ Z denote by Dv(F ) the R-complex 0 −→ F =−→ F −→ 0 concentrated in degrees v and v − 1. 4.2 Construction. Let M be an R-complex. For a homomorphism π : F → Mv of R-modules, there is a morphism of R-complexes, eπ : Dv(F ) → M , given by 0 · · · / Mv+1 ∂M v+1 / F π / Mv 1F / F / 0 ∂M v π ∂M v / Mv−1 ∂M v−1 / Mv−2 / · · · . 4.3 Proposition. An R-complex M is bounded and degreewise finitely presented if and only if there exists an exact sequence of R-complexes L1 → L0 → M → 0 where L0 and L1 are bounded complexes of finitely generated free modules. Proof. The "if" part is trivial. To show "only if", assume that M is a bounded and degreewise finitely presented complex. Since M is, in particular, degreewise finitely generated, we can for every v ∈ Z choose a surjective homomorphism πv : F v → Mv where F v is finitely generated free, and such that F v is zero if Mv is zero. Set L0 = `v∈Z Dv(F v) and let π : L0 → M be the unique morphism whose composite, πεv, with the embedding εv : Dv(F v) ֒→ L0 equals the morphism πv from 4.2. Evidently, π is surjective and L0 is a bounded complex of finitely generated free modules. Consider the kernel M ′ = Ker π. Since L0 is bounded, so is M ′. Furthermore, as M is degreewise finitely presented, M ′ is degreewise finitely generated. Hence the argument above shows that there exists a surjective morphism L1 → M ′ where L1 is a bounded complex of finitely generated free modules. The composite L1 ։ M ′ ֒→ L0 now yields the left-hand morphism in an exact sequence L1 → L0 → M → 0. (cid:3) It is a well-known fact that every module is isomorphic to a direct limit of finitely presented modules. The following generalization to complexes can be found in Garc´ıa Rozas's [8, lems. 4.1.1(ii) and 5.1.1]. 4.4. Every R-complex is isomorphic to a direct limit of bounded and degreewise finitely presented R-complexes. (cid:3) The next theorem identifies the finitely presented objects in the category C(R), and combined with 4.4 it shows that this category is locally finitely presented; see Corollary 4.6 and Remark 4.7. 4.5 Theorem. For an R-complex M the following conditions are equivalent. (i) M bounded and degreewise finitely presented. (ii) The functor HomR(M, −) preserves filtered colimits. (iii) The functor HomR(M, −) preserves direct limits. / / / / / / / / THE DIRECT LIMIT CLOSURE OF PERFECT COMPLEXES 7 (iv ) The functor C(R)(M, −) preserves filtered colimits. (v ) The functor C(R)(M, −) preserves direct limits. Proof. The implications (ii) =⇒ (iii) and (iv ) =⇒ (v ) are trivial. By 2.2 there is for every R-complex M an identity of functors from C(R) to Z-modules, C(R)(M, −) = Z0(HomR(M, −)) . By Proposition 3.4 the functor Z0 preserves filtered colimits, and hence the implica- tions (ii) =⇒ (iv ) and (iii) =⇒ (v ) follow. It remains to show that the implications (i) =⇒ (ii) and (v ) =⇒ (i) hold. (i) =⇒ (ii): By Proposition 4.3 there is an exact sequence L1 → L0 → M → 0 where L0 and L1 are bounded complexes of finitely generated free R-modules. Thus there is an exact sequence, 0 −→ HomR(M, −) −→ HomR(L0, −) −→ HomR(L1, −) , of functors from C(R) to C(Z). By Proposition 3.5 the functors HomR(L0, −) and HomR(L1, −) preserve filtered colimits, and the conclusion follows from Lemma 3.3. (v ) =⇒ (i): By 4.4 there is a direct system {µvu : M u → M v}u6v of bounded and degreewise finitely presented R-complexes with colim M u ∼= M . By assumption, the canonical morphism α : colim C(R)(M, M u) −→ C(R)(M, colim M u) ∼= C(R)(M, M ) is an isomorphism. Write µu : M u −→ colim M u ∼= M and λu : C(R)(M, M u) −→ colim C(R)(M, M u) for the canonical morphisms, and note that αλu = C(R)(M, µu) holds for all u. Surjectivity of α yields an element χ ∈ colim C(R)(M, M u) with α(χ) = 1M . By 3.2 one has χ = λu(ψu) for some ψu ∈ C(R)(M, M u). Hence, there are equalities µuψu = C(R)(M, µu)(ψu) = αλu(ψu) = α(χ) = 1M . Thus, M is a direct summand of M u, and since M u is bounded and degreewise finitely presented, so is M . (cid:3) The equivalences above of (ii) and (iii) and of (iv ) and (v ) also follow from general principles; see [1, cor. to thm. 1.5]. 4.6 Corollary. The category C(R) is locally finitely presented, and the finitely presented objects in C(R) are exactly the bounded and degreewise finitely presented R-complexes. Proof. By the equivalence of (i) and (iv ) in Theorem 4.5, the finitely presented objects in the category C(R) are exactly the bounded and degreewise finitely pre- sented R-complexes. Evidently, the category of such complexes is skeletally small. By 4.4 every object in C(R) is a filtered colimit (even a direct limit) of finitely presented objects. (cid:3) 4.7 Remark. The fact that C(R) is locally finitely presented also follows from [5, Satz 1.5] and [6, (2.4)]; indeed, C(R) is a Grothendieck category and {Du(R) u ∈ Z} is a generating set of finitely presented objects. 8 L. W. CHRISTENSEN AND H. HOLM 5. Proof of the main theorem The notion of semi-flatness, and the related notions of semi-projectivity and semi- freeness, originate in the treatise [3] by Avramov, Foxby, and Halperin. A graded R-module L is called graded-free if it has a graded basis, that is, a basis consisting of homogeneous elements. It is easily seen that L is graded-free if and only if every component Lv is a free R-module. 5.1. An R-complex L is called semi-free if the underlying graded R-module has a graded basis E that can be written as a disjoint union E = Un>0 En such that one has E0 ⊆ Z(L) and ∂L(En) ⊆ RhSn−1 i=0 Eii for every n > 1. Such a basis is called a semi-basis of L. 5.2 Example. A bounded below complex of free modules is semi-free. 5.3. Every R-complex M has a semi-free resolution, that is, a quasi-isomorphism of R-complexes π : L → M where L is semi-free. Moreover, π can be chosen surjective and with Lv = 0 for all v < inf{n ∈ Z Mn 6= 0}. See [3, thm. 2.2]. 5.4. For an R-complex P the following conditions are equivalent. (i) The functor HomR(P, −) is exact and preserves quasi-isomorphisms. (ii) For every morphism α : P → N and for every surjective quasi-isomorphism β : M → N there exists a morphism γ : P → M such that α = βγ holds. (iii) P is a complex of projective R-modules, and the functor HomR(P, −) preserves acyclicity. A complex that satisfies these equivalent conditions is called semi-projective; see [3, thm. 3.5]. 5.5 Example. A bounded below complex of projective modules is semi-projective. By [3, thm. 3.5] a semi-free complex is semi-projective. 5.6. For an R-complex F the following conditions are equivalent. (i) The functor − ⊗R F is exact and preserves quasi-isomorphisms. (ii) F is a complex of flat R-modules and the functor − ⊗R F preserves acyclicity. A complex that satisfies these equivalent conditions is called semi-flat ; see [3, thm. 6.5]. 5.7 Example. A bounded below complex of flat modules is semi-flat. By [3, lem. 7.1] a semi-projective complex is semi-flat. As noted in 2.4, the biduality morphism δP is an isomorphism for every complex P of finitely generated projective modules. For the proof of Theorem 1.1 we need an explicit description of the inverse. 5.8 Lemma. For a complex P of finitely generated projective R-modules, the in- verse of the isomorphism δHomR(P,R) is HomR(δP , R). THE DIRECT LIMIT CLOSURE OF PERFECT COMPLEXES 9 Proof. As P is a complex of finitely generated projective R-modules, δP and hence HomR(δP , R) are isomorphisms by 2.4. For ϕ in HomR(P, R) and x in P one has (HomR(δP , R)δHomR(P,R))(ψ)(x) = (δHomR(P,R)(ψ)δP )(x) = δHomR(P,R)(ψ)(δP (x)) = δP (x)(ψ) = ψ(x) , so HomR(δP , R)δHomR(P,R) is the identity on HomR(P, R). (cid:3) Condition (iii) in Theorem 1.1 asserts the existence of a certain pure epimor- phism in C(R). In the proof below, we use that the equivalence of conditions (ii) -- (v ) has been established elsewhere and we do not directly address (iii). However, in the next section we study the relationship between purity and semi-flatness; in particular, we recall the definition of a pure epimorphism from [6, §3] in the first paragraph of Section 6. Proof of Theorem 1.1. By Corollary 4.6 the category C(R) is locally finitely pre- sented and its finitely presented objects are exactly the bounded and degreewise finitely presented complexes. It now follows from [6, (4.1)] that (ii), (iii), and (iv ) are equivalent. Furthermore, [1, thm. 1.5] shows that (iv ) and (v ) are equivalent. The remaining implications (i) =⇒ (ii) and (v ) =⇒ (i) are proved below. (i) =⇒ (ii): Let ϕ : N → F be a morphism of R-complexes where N is bounded and degreewise finitely presented. By Proposition 4.3 there is an exact sequence, L1 ψ1 −−→ L0 ψ0 −−→ N −→ 0 , of R-complexes, where L0 and L1 are bounded complexes of finitely generated free modules. Consider the exact sequence of R◦-complexes, (1) 0 −→ K ι−→ HomR(L0, R) Hom(ψ1,R) −−−−−−−−→ HomR(L1, R) , where K is the kernel of HomR(ψ1, R) and ι is the embedding. The functor Z0(−) is left exact, and the functor − ⊗R F is exact by definition, so it follows that the functor Z0(− ⊗R F ) leaves the sequence (1) exact. As L0 is bounded, so is K; set u = inf{n ∈ Z Kn 6= 0}. By 5.3 there is an exact sequence, (2) P π−−→ K −→ 0 , where π is a quasi-isomorphism and P is a semi-free R◦-complex with Pv = 0 for all v < u. As F is semi-flat, π ⊗R F is a surjective quasi-isomorphism by 5.6. A simple diagram chase shows that every surjective quasi-isomorphism is surjective on cycles, so the functor Z0(− ⊗R F ) leaves the sequence (2) exact. Consequently, there is an exact sequence, Z0(P ⊗R F ) (ιπ)⊗F −−−−→ Z0(HomR(L0, R) ⊗R F ) Hom(ψ1,R)⊗F −−−−−−−−−→ Z0(HomR(L1, R) ⊗R F ) . For every R-complex M , denote by ξM the composite morphism HomR(M, R) ⊗R F ωM RF −−−−→ HomR(M, R ⊗R F ) ∼=−−−→ HomR(M, F ) , where ωM RF is the tensor evaluation morphism 2.5 and the isomorphism is induced by the canonical one R ⊗R F ∼= F . The morphism ξM is natural in M , and by 2.5 10 L. W. CHRISTENSEN AND H. HOLM it is an isomorphism if M is a bounded complex of finitely generated projective modules. The exact sequence above now yields another exact sequence, (3) Z0(P ⊗R F ) ξL0 −−−−−−−−−→ Z0(HomR(L0, F )) ◦ ((ιπ)⊗F ) Hom(ψ1,F ) −−−−−−−−→ Z0(HomR(L1, F )) . As ϕψ0 : L0 → F is a morphism, it is an element in Z0(HomR(L0, F )); see 2.2. Since one has HomR(ψ1, F )(ϕψ0) = ϕψ0ψ1 = 0, exactness of (3) yields an element x in Z0(P ⊗R F ) with (4) (ξL0 ◦ ((ιπ) ⊗R F ))(x) = ϕψ0 . v is finitely generated and free. For v /∈ {u, . . . , w} set P ′ The graded module underlying P has a graded basis E, and x has the form x = Pn i=1 ei ⊗ fi with ei ∈ E and fi ∈ F . Set w = max{e1, . . . , en}; as one has Pv = 0 for all v < u, each basis element ei satisfies u 6 ei 6 w. For v ∈ Z set Ev = {e ∈ E e = v}. Next we define a bounded subcomplex P ′ of P such that each module P ′ v = 0; for v ∈ {u, . . . , w} the modules P ′ w be the finitely generated free submodule of Pw generated by the set E′ w = {e1, . . . , en} ∩ Ew. For v 6 w assume that a finitely generated free submodule P ′ v of Pv with finite basis E′ v } of Pv−1 is finite, there is a finite subset G′ v−1i. Now let P ′ v−1 be the submodule of Pv−1 generated by the following finite set of basis elements, v has been constructed. As the subset B′ v are constructed inductively. Let P ′ v−1 of Ev−1 with B′ v−1 ⊆ R◦hG′ v−1 = {∂P (e) e ∈ E′ E′ v−1 = G′ v−1 ∪ ({e1, . . . , en} ∩ Ev−1) . v−1 for all v ∈ Z, so P ′ is a subcomplex of By construction, one has ∂P (P ′ v) ⊆ P ′ P . The construction shows that x = Pn i=1 ei ⊗ fi belongs to P ′ ⊗R F . As F is a complex of flat R-modules, P ′ ⊗R F is a subcomplex of P ⊗R F , and as the element x is in Z0(P ⊗R F ) it also belongs to Z0(P ′ ⊗R F ). Set L = HomR◦ (P ′, R). As P ′ is a bounded complex of finitely generated free R◦-modules, L is a bounded complex of finitely generated free R-modules. Let ε : P ′ ֒→ P be the embedding and let κ′ : L0 → L be the composite morphism L0 δL0 −−−→ HomR◦ (HomR(L0, R), R) Hom(ιπε,R) −−−−−−−−→ HomR◦ (P ′, R) = L . In the commutative diagram P ′ ∼= ′ δP ιπε / HomR(L0, R) ∼= δHom (L0 ,R) HomR(HomR◦ (P ′, R), R) Hom(Hom(ιπε,R),R) / HomR(HomR◦ (HomR(L0, R), R), R) the vertical morphisms are isomorphisms by 2.4, and δHom(L0,R) is by Lemma 5.8 the inverse of HomR(δL0 , R). One now has (5) HomR(κ′, R)δP ′ = ιπε . It follows that there are equalities, HomR(κ′ψ1, R)δP ′ = HomR(ψ1, R)ιπε = 0πε = 0 , / / THE DIRECT LIMIT CLOSURE OF PERFECT COMPLEXES 11 and since δP ′ ular, HomR◦ (HomR(κ′ψ1, R), R) is zero, and hence the commutative diagram is an isomorphism, the morphism HomR(κ′ψ1, R) is zero. In partic- L1 ∼= δL1 κ′ψ1 / L δL∼= HomR◦ (HomR(L1, R), R) Hom(Hom(κ′ψ1,R),R) / HomR◦ (HomR(L, R), R) shows that κ′ψ1 = 0 holds. Again the vertical morphisms are isomorphisms by 2.4. Since κ′ vanishes on Im ψ1 = Ker ψ0 there is a unique morphism κ : N → L with κψ0 = κ′. Finally, consider the diagram, P ′ ⊗R F ε⊗F (6) ′ δP ⊗F / HomR(L, R) ⊗R F Hom(κ′,R)⊗F P ⊗R F (ιπ)⊗F / HomR(L0, R) ⊗R F ξL ξL0 / HomR(L, F ) HomR(κ′,F ) / HomR(L0, F ) , where the left-hand square is commutative by (5) and the right-hand square is commutative by naturality of ξ. Set λ = (ξL ◦ (δP ′ ⊗ F ))(x) ; it is an element in HomR(L, F ), and as x belongs to Z0(P ′ ⊗R F ), also λ is a cycle; i.e. λ : L → F is a morphism. From (6), from the definition of λ, and from (4) one gets λκ′ = ϕψ0. The identity κ′ = κψ0 and surjectivity of ψ0 now yield λκ = ϕ. (v ) =⇒ (i): Every bounded complex of finitely generated free R-modules is semi- flat, see Example 5.7, and as mentioned in the introduction a direct limit of semi- flat complexes is semi-flat. A proof of this fact can be found in [3, prop. 6.9]; for completeness we include the argument. Let {F u → F v}u6v be a direct system of semi-flat R-complexes and set F = colim F u. By Proposition 3.5 there is a natural isomorphism of functors, − ⊗R F ∼= colim(− ⊗R F u). By assumption, each functor − ⊗R F u is exact and preserves acyclicity. Since direct limits in C(Z) are exact, see 3.2, and since the homology functor preserves direct limits, see Proposition 3.4, it follows that the functor − ⊗R F is exact and preserves acyclicity; that is, F is semi-flat. (cid:3) 6. Purity Let A be a locally finitely presented category. Following [6, §3] a short exact sequence 0 → M ′ → M → M ′′ → 0 in A is called a pure if 0 −→ A(A, M ′) −→ A(A, M ) −→ A(A, M ′′) −→ 0 is exact for every finitely presented object A in A. In this case, the morphism M ′ → M is called a pure monomorphism and M → M ′′ is called pure epimorphism. In view of Corollary 4.6, a morphism α : X → Y in C(R) is a pure epimorphism if and only if for every morphism ϕ : N → Y with N bounded and degreewise finitely presented there exists a morphism β : N → X with ϕ = αβ. Semi-flat complexes have the following two-out-of-three property; see the proof of [3, prop. 6.7]. / / / / / / 12 L. W. CHRISTENSEN AND H. HOLM 6.1. Let 0 → F ′ → F → F ′′ → 0 be an exact sequence of R-complexes. If F ′′ is semi-flat, then F ′ is semi-flat if and only if F is semi-flat. The next result supplements 6.1; it shows that the class of semi-flat complexes is closed under pure subcomplexes and pure quotient complexes. 6.2 Proposition. Let 0 → F ′ → F → F ′′ → 0 be a pure exact sequence of R- complexes. If the complex F is semi-flat, then F ′ and F ′′ are semi-flat. Proof. Assume that F is semi-flat and denote the given morphism from F to F ′′ by α. Let ϕ : N → F ′′ be a morphism where N is a bounded and degreewise finitely presented R-complex. Since α is a pure epimorphism one has ϕ = αβ for some morphism β : N → F . Since F is semi-flat, the morphism β, and hence also ϕ, factors through a bounded complex of finitely generated free R-modules. Thus F ′′ is semi-flat by Theorem 1.1. It now follows from 6.1 that F ′ is semi-flat as well. (cid:3) Every surjective homomorphism M → F of R-modules with F flat is a pure epimorphism; see [13, cor. 1.3]. The next example shows that a surjective morphism M → F of R-complexes with F semi-flat need not be a pure epimorphism. 6.3 Example. Consider the Z-complexes, D0(Z) and Z. As a Z-complex, Z is semi-flat by Example 5.7. The surjective morphism π : D0(Z) → Z, given by the diagram 0 0 / Z 1Z / Z / 0 1Z / Z / 0 / 0 , is not a pure epimorphism. Indeed, the complex Z is finitely presented but the identity morphism Z → Z does not factor through π; in other words, π is not a split surjection. What can be salvaged is captured in the next proposition. 6.4 Proposition. For an R-complex F the following conditions are equivalent. (i) F is semi-flat. (ii) Every surjective quasi-isomorphism M → F a is pure epimorphism. (iii) There exists a semi-free complex L and a quasi-isomorphism L → F which is also a pure epimorphism. Proof. (i) =⇒ (ii): Let α : M → F be a surjective quasi-isomorphism and ϕ : N → F be a morphism with N bounded and degreewise finitely presented. Since F is semi- flat there exists by Theorem 1.1 a bounded complex L of finitely generated free R-modules and morphisms κ : N → L and λ : L → F with ϕ = λκ. As L is semi- projective, see Examples 5.2 and 5.5, there exists by 5.4 a morphism γ : L → M with λ = αγ, so with β = γκ one has ϕ = αβ. (ii) =⇒ (iii): Immediate from 5.3. (iii) =⇒ (i): Follows from Proposition 6.2 as a semi-free complex is semi-flat. (cid:3) / / / / / / THE DIRECT LIMIT CLOSURE OF PERFECT COMPLEXES 13 7. Semi-flat complexes of projective modules A semi-free complex is semi-projective, see Example 5.5, but a semi-projective complex of free modules need not be semi-free. Indeed, the Z/6Z-complex, · · · −→ Z/6Z 2 −→ Z/6Z 3 −→ Z/6Z 2 −→ Z/6Z 3 −→ Z/6Z −→ · · · serves as a counterexample; see [3, ex. 7.10]. It turns out that a semi-flat complex of projective modules is, in fact, semi-projective. As Murfet notes in his thesis [15, cor. 5.14], this follows from work of Neeman [17] on the homotopy category of flat modules. The purpose of this section is to provide an alternative proof of this fact. 7.1. For an R-complex C, the following conditions are equivalent. (i) The identity 1C is null-homotopic; that is, it is a boundary in HomR(C, C). (ii) The exists a degree 1 homomorphism σ : C → C with ∂C = ∂C σ∂C . (iii) There exists a graded R-module B with Cone 1B ∼= C. A complex that satisfies these conditions is called contractible; see [19, sec. 1.4]. If C is contractible, then so are all complexes HomR(C, X), HomR(X, C), and Y ⊗R C. Every contractible complex is acyclic. 7.2 Lemma. Let N be a bounded and degreewise finitely generated R-complex, and let C be a contractible complex of projective R-modules. Every morphism N → C factors as N → L → C, where L is a bounded and contractible complex of finitely generated free R-modules. Proof. There is a graded R-module P with C ∼= Cone 1P = `v∈Z Dv+1(Pv). In particular, C is a coproduct of bounded contractible complexes of projective R- modules. For each module Pv there is a complementary module Qv and a set Ev such that there is an isomorphism Pv ⊕ Qv ∼= R(Ev). Set L′ = C ⊕ (cid:0) a Dv+1(Pv)(cid:1) ∼= a v∈Z v∈Z (Dv+1(R))(Ev ) . A morphism α : N → C factors through L′, and since N is bounded and degree- wise finitely generated, it factors through a finite coproduct L = Ln i=1 Dvi+1(R). Evidently, this is a bounded and contractible complex of finitely generated free R-modules. (cid:3) The next result shows that the complexes characterized in [7, thm. 2.4], in [8, thm. 4.1.3], and in [17, fact 2.14] are precisely the acyclic semi-flat complexes. In [3] such complexes are called categorically flat, in [8] they are called flat, and in [16] they are called pure acyclic. 7.3 Theorem. For an R-complex F the following conditions are equivalent. (i) F is semi-flat and acyclic. (ii) F is a filtered colimit of bounded and contractible complexes of finitely gen- erated free R-modules. (iii) F is a direct limit of bounded and contractible complexes of finitely generated free R-modules. (iv ) F is acyclic and B(F ) is a complex of flat R-modules. 14 L. W. CHRISTENSEN AND H. HOLM Proof. (i) =⇒ (ii): By Theorem 4.5 and [6, (4.1)] it is sufficient to prove that every morphism ϕ : N → F with N bounded and degreewise finitely presented factors through a bounded and contractible complex of finitely generated free R-modules. Fix such a morphism ϕ. Let π : P ≃−−−→ F be a surjective semi-free resolution; cf. 5.3. As P is acyclic and semi-projective, the complex HomR(P, P ) is acyclic; in particular the morphism 1P is null-homotopic so P is contractible. The morphism π is by Proposition 6.4 a pure epimorphism, so ϕ factors through P and hence, by Lemma 7.2, through a bounded and contractible complex L of finitely generated free R-modules. (ii) =⇒ (iii): This follows from [1, thm. 1.5]. (iii) =⇒ (iv ): A direct limit of contractible (acyclic) complexes is acyclic, so F is acyclic. In a contractible complex L of free R-modules, the subcomplex B(L) consists of projective R-modules. The functor B(−) preserves direct limits by Propo- sition 3.4, and a direct limit of projective modules is a flat module, so B(F ) is a complex of flat R-modules. (iv ) =⇒ (i): Each sequence 0 → Bv(F ) → Fv → Bv−1(F ) → 0 is exact, so each module Fv is flat; that is, F is an acyclic complex of flat R-modules. For an acyclic R◦-complex M (actually for any R◦-complex) it follows from the Kunneth formula [19, thm. 3.6.3] that M ⊗R F is acyclic. Thus, F is semi-flat by 5.6. (cid:3) In the terminology of [3] the equivalence of (i) and (iii) above says that a complex is categorically flat if and only if it is a direct limit of categorically projective complexes of finitely generated free modules. 7.4 Corollary. Let α : F → F ′ be a quasi-isomorphism between semi-flat R-com- plexes. For every bounded and degreewise finitely presented R-complex N the morphism HomR(N, α) : HomR(N, F ) → HomR(N, F ′) is a quasi-isomorphism. Proof. By 2.1 and 6.1 the complex Cone α is acyclic and semi-flat. The functor HomR(N, −) preserves filtered colimits by Theorem 4.5 and maps contractible com- plexes to contractible complexes. Now it follows from Theorem 7.3 and Proposi- tion 3.4 that HomR(N, Cone α) is acyclic. Since the functor HomR(N, −) commutes with mapping cone, see 2.2, it follows that the complex Cone HomR(N, α) is acyclic, and thus HomR(N, α) is a quasi-isomorphism by 2.1. (cid:3) The next corollary can be proved similarly; a different proof is given in [3, 6.4]. 7.5 Corollary. Let α : F → F ′ be a quasi-isomorphism between semi-flat R-com- plexes. For every R◦-complex M the morphism M ⊗R α : M ⊗R F → M ⊗R F ′ is a quasi-isomorphism. (cid:3) The next result was proved by Neeman [17, rmk. 2.15 and thm. 8.6] in 2008. He notes, "I do not know an elementary proof, a proof which avoids homotopy categories". We show that it follows from a theorem of Benson and Goodearl [4, thm. 2.5]∗ from 2000, which asserts that if 0 → F → P → F → 0 is a short exact sequence of R-modules with F flat and P projective, then F is projective as well. Notice that if R has finite finitistic projective dimension, then this assertion follows from Jensen's [12, prop. 6], and if R has cardinality 6 ℵn for some n ∈ N, then it follows from a theorem of Gruson and Jensen [10, thm. 7.10]. ∗ Benson and Goodearl's proof uses only classical results from homological algebra. THE DIRECT LIMIT CLOSURE OF PERFECT COMPLEXES 15 7.6 Proposition. If P is an acyclic complex of projective R-modules such that the subcomplex B(P ) consists of flat R-modules, then P is contractible. Proof. For each v ∈ Z the sequence 0 → Bv(P ) → Pv → Bv−1(P ) → 0 is exact. The coproduct of all these exact sequences yields the exact sequence 0 −→ ` Bv(P ) −→ ` v∈Z Pv −→ ` v∈Z v∈Z Bv(P ) −→ 0 . By assumption, the module `v∈Z Bv(P ) is flat and `v∈Z Pv is projective, so it fol- lows from [4, thm. 2.5] that `v∈Z Bv(P ) is projective. Consequently, every module Bv(P ) is projective, and therefore P is contractible. (cid:3) Semi-projective complexes, just like semi-flat complexes, have a two-out-of-three property; see the proof of [3, prop. 3.7]. 7.7. Let 0 → P ′ → P → P ′′ → 0 be an exact sequence of R-complexes. If P ′′ is semi-projective, then P ′ is semi-projective if and only if P is semi-projective. 7.8 Theorem. A semi-flat complex of projective R-modules is semi-projective. Proof. Let π : L ≃−−→ F be a semi-free resolution, see 5.3, and consider the mapping cone sequence 0 → F → Cone π → ΣL → 0; see 2.1. Since ΣL is semi-projective, see Example 5.5, it suffices by 7.7 to argue that Cone π is semi-projective. As the complexes F and ΣL are semi-flat and consist of projective modules, it follows from 6.1 and 2.1 that Cone π is an acyclic semi-flat complex of projective modules. Thus Theorem 7.3 and Proposition 7.6 apply to show that Cone π is contractible. It remains to note that every contractible complex of projective modules is semi- projective; this is immediate from 5.4. (cid:3) Acknowledgments We thank Jan Stov´ıcek for pointing us to Ad´amek and Rosick´y's book. We also thank the anonymous referee for helpful comments on both content and exposition. References 1. Jir´ı Ad´amek and Jir´ı Rosick´y, Locally presentable and accessible categories, London Mathe- matical Society Lecture Note Series, vol. 189, Cambridge University Press, Cambridge, 1994. MR1294136 2. Luchezar L. Avramov and Hans-Bjørn Foxby, Homological dimensions of unbounded com- plexes, J. Pure Appl. Algebra 71 (1991), no. 2-3, 129 -- 155. MR1117631 3. Luchezar L. Avramov, Hans-Bjørn Foxby, and Stephen Halperin, Differential graded homolo- gical algebra, preprint (1994 -- 2013). 4. David J. Benson and Kenneth R. Goodearl, Periodic flat modules, and flat modules for finite groups, Pacific J. Math. 196 (2000), no. 1, 45 -- 67. MR1797235 5. Siegfried Breitsprecher, Lokal endlich prasentierbare Grothendieck-Kategorien, Mitt. Math. Sem. Giessen Heft 85 (1970), 1 -- 25. MR0262330 6. William Crawley-Boevey, Locally finitely presented additive categories, Comm. Algebra 22 (1994), no. 5, 1641 -- 1674. MR1264733 7. Edgar E. Enochs and J. R. Garc´ıa Rozas, Flat covers of complexes, J. Algebra 210 (1998), no. 1, 86 -- 102. MR1656416 8. Juan Ramon Garc´ıa Rozas, Covers and envelopes in the category of complexes of modules, Chapman & Hall/CRC Research Notes in Mathematics, vol. 407, Chapman & Hall/CRC, Boca Raton, FL, 1999. MR1693036 9. V. E. Govorov, On flat modules, Sibirsk. Mat. Z. 6 (1965), 300 -- 304. MR0174598 16 L. W. CHRISTENSEN AND H. HOLM 10. Laurent Gruson and Christian U. Jensen, Dimensions cohomologiques reli´ees aux foncteurs (i), Paul Dubreil and Marie-Paule Malliavin Algebra Seminar, 33rd Year (Paris, 1980), lim ←− Lecture Notes in Math., vol. 867, Springer, Berlin, 1981, pp. 234 -- 294. MR633523 11. Mark Hovey, Model categories, Mathematical Surveys and Monographs, vol. 63, American Mathematical Society, Providence, RI, 1999. MR1650134 12. Christian U. Jensen, On the vanishing of lim (i), J. Algebra 15 (1970), 151 -- 166. MR0260839 13. Daniel Lazard, Autour de la platitude, Bull. Soc. Math. France 97 (1969), 81 -- 128. MR0254100 14. Saunders MacLane, Categories for the working mathematician, Springer-Verlag, New York, ←− 1971, Graduate Texts in Mathematics, Vol. 5. MR0354798 15. Daniel Murfet, The mock homotopy category of projectives and Grothendieck duality, Ph.D. thesis, Australian National University, 2007, x+145 pp. 16. Daniel Murfet and Shokrollah Salarian, Totally acyclic complexes over Noetherian schemes, Adv. Math. 226 (2011), no. 2, 1096 -- 1133. MR2737778 17. Amnon Neeman, The homotopy category of flat modules, and Grothendieck duality, Invent. Math. 174 (2008), no. 2, 255 -- 308. MR2439608 18. Nicolas Spaltenstein, Resolutions of unbounded complexes, Compositio Math. 65 (1988), no. 2, 121 -- 154. MR0932640 19. Charles A. Weibel, An introduction to homological algebra, Cambridge Studies in Advanced Mathematics, vol. 38, Cambridge University Press, Cambridge, 1994. MR1269324 Texas Tech University, Lubbock, TX 79409, U.S.A. E-mail address: [email protected] URL: http://www.math.ttu.edu/~ lchriste University of Copenhagen, 2100 Copenhagen Ø, Denmark E-mail address: [email protected] URL: http://www.math.ku.dk/~ holm/
1805.02273	1	1805	2018-05-06T20:05:21	Extensions of integral domains and quasi-valuations	[ "math.RA" ]	Let $S$ be an integral domain with field of fractions $F$ and let $A$ be an $F$-algebra having an $S$-stable basis. We prove the existence of an $S$-subalgebra $R$ of $A$ lying over $S$ whose localization with respect to $S$ is $A$ (we call such $R$ an $S$-nice subalgebra of $A$). We also show that there is no such minimal $S$-nice subalgebra of $A$. Given a valuation $v$ on $F$ with a corresponding valuation domain $O_v$, and an $O_v$-stable basis of $A$ over $F$, we prove the existence of a quasi-valuation on $A$ extending $v$ on $F$. Moreover, we prove the existence of an infinite decreasing chain of quasi-valuations on $A$, all of which extend $v$. Finally, we present applications for the above existence theorems; for example, we show that if $A$ is commutative and $\mathcal C$ is any chain of prime ideals of $S$, then there exists an $S$-nice subalgebra of $A$, having a chain of prime ideals covering $\mathcal C$.	math.RA	math	Extensions of integral domains and quasi-valuations Shai Sarussi Abstract Let S be an integral domain with field of fractions F and let A be an F -algebra having an S-stable basis. We prove the existence of an S- subalgebra R of A lying over S whose localization with respect to S is A (we call such R an S-nice subalgebra of A). We also show that there is no such minimal S-nice subalgebra of A. Given a valuation v on F with a corresponding valuation domain Ov, and an Ov-stable basis of A over F , we prove the existence of a quasi-valuation on A extending v on F . Moreover, we prove the existence of an infinite decreasing chain of quasi-valuations on A, all of which extend v. Finally, we present applications for the above existence theorems; for example, we show that if A is commutative and C is any chain of prime ideals of S, then there exists an S-nice subalgebra of A, having a chain of prime ideals covering C. 1 Introduction Valuation theory has long been a key tool in commutative algebra, with applications in number theory and algebraic geometry. It has become a useful tool in the study of finite dimensional division algebras, particulary in the construction of examples, such as Amitsur's construction of noncrossed products division algebras. See [Wad] for a comprehensive survey. Although valuations provide a powerful tool in studying arithmetic of fields, it has been difficult to use them in noncommutative settings. For example, division algebras do not have many valuations and rings with zero divisors do not have valuations at all. This has motivated researchers to generalize the notion of valuation. Attempts to generalize the notion of valu- ation were made throughout the last few decades. Morandi's value functions (cf. [Mor]) and Tignol and Wadsworth's gauges [TW] are examples of such generalizations. Also see [KZ] for Manis valuations and PM-valuations, and [Co], [Hu], and [MH] for pseudo-valuations. See [Sa1] for a brief discussion on these other related theories. Another approach was initiated recently by the author in developing the notion of a quasi-valuation. A quasi-valuation on a ring R is a function w : R → M ∪ {∞}, where M is a totally ordered abelian monoid, to which we adjoin an element ∞ greater than all elements of M , and w satisfies the following properties: 1 (B1) w(0) = ∞; (B2) w(xy) ≥ w(x) + w(y) for all x, y ∈ R; (B3) w(x + y) ≥ min{w(x), w(y)} for all x, y ∈ R. In [Sa1] we mainly developed the theory of quasi-valuations on finite dimensional field extensions and we were able to answer questions regarding the structure of rings using quasi-valuation theory. More precisely, for a given valuation v on a field F , a corresponding valuation domain Ov, and a finite field extension E, we studied quasi-valuations on E extending v on F . We showed that every such quasi-valuation is dominated by some valuation extending v, and presented several generalizations of results from valuation theory (see [Sa1, sections 6 and 4]). We also studied the quasi-valuation rings; namely, the set of all elements of E with values greater or equal to zero. We proved that the prime spectra of Ov and its quasi-valuation ring are intimately connected. In addition, a one-to-one correspondence was obtained between exponential quasi-valuations and integrally closed quasi- valuation rings. Most importantly, we constructed the filter quasi-valuation, for any algebra over a valuation domain, and showed that if A is an F - algebra and R is an Ov-subalgebra of A lying over Ov then there exists a quasi-valuation on R ⊗Ov F (called the filter quasi-valuation) extending v on F such that the quasi-valuation ring is equal to R (under the identification of R with R ⊗Ov 1). In particular, if R is an Ov-subalgebra of A lying over Ov such that RF = A then there exists a quasi-valuation on A extending v on F . However, the existence of such subalgebras was not clear and was not proven. Existence theorems of such algebras and others of greater generality will be presented in this paper. In [Sa2] we studied the structures of algebras over valuation domains using quasi-valuation theory. We generalized some of the results of [Sa1] and presented additional connections between a valuation domain Ov and an Ov- algebra. We related the prime spectrum of a valuation domain to the prime spectrum of a (not necessarily commutative) algebra over it. We studied the classical lifting conditions of "lying over" (LO), "incomparability" (INC), "going down" (GD) and "going up" (GU) in ring extensions, as well as a subtler condition called "strong going between" (SGB), and saw how quasi- valuations play a key role in the situation under discussion (see [Sa2, section 1.2] for the definitions of LO, INC, GD, GU and SGB). Specifically, we presented a necessary and sufficient condition for an Ov-algebra to satisfy LO over Ov. We proved that if R is a torsion-free algebra over Ov such that [R ⊗Ov F : F ] < ∞, then R satisfies INC over Ov; as a result, we obtained an upper and a lower bound on the size of the prime spectrum of R. Then, we showed that if R is a torsion-free algebra over Ov then R satisfies GD over Ov; and concluded that any algebra over a commutative valuation ring satisfies SGB over it. We deduced that a torsion-free algebra over Ov satisfies GGD (generalized going down) over Ov. Moreover, a sufficient condition for a quasi-valuation ring to satisfy GU over Ov was given. 2 So, in [Sa1] and in some parts of [Sa2] we assumed that a quasi-valuation extending the valuation v exists (or equivalently by [Sa1, Theorem 9.19], we assumed the existence of an appropriate Ov-algebra). A natural and central question would be then: when does such a quasi-valuation exist? In this paper we show that quasi-valuations extending a valuation v on F exists on any finite dimensional F -algebras, and even more generally, on any F - algebra having an Ov-stable basis (see Definition 3.3 for the definition of an Ov-stable basis). In fact, we prove a more general theorem and apply it to quasi-valuation theory. Let us quote our first existence theorem: let S be an integral domain which is not a field, let F be its field of fractions, and let A be an F -algebra containing an S-stable basis; then there exists an S-subalgebra of A which lies over S and whose localization with respect to S is A. This existence theorem is then followed by several other existence theorems; for example, we prove that any such S-subalgebra of A is not minimal with respect to inclusion, and we also prove the existence of such S-subalgebras of A containing a given ideal of A. Returning to the assumption that v is a valuation on F , using the construction of the filter quasi-valuation (see section 2), we deduce the existence of an infinite descending chain of quasi- valuations on A, all of which extend v on F . Finally, as applications to the above mentioned existence theorems, in Proposition 3.23 and Theorem 3.24 we use results from valuation theory and results from [Sa2] about quasi- valuations to prove two interesting propositions regarding the prime spectra of integral domains and certain algebras over them. In this paper the symbol ⊂ means proper inclusion and the symbol ⊆ means inclusion or equality. 2 Previous results - construction of the filter quasi- valuation In this section, for the reader's convenience, we recall from [Sa1] the main steps in constructing the filter quasi-valuation. For further details and proofs, see [Sa1, section 9]. The first step is to construct a value monoid, constructed from the value group of the valuation. We call this value monoid the cut monoid. We start by reviewing some of the basic notions of cuts of ordered sets. For further information on cuts see, for example, [FKK] or [Weh]. Definition 2.1. Let T be a totally ordered set. A subset U of T is called initial if for every γ ∈ U and α ∈ T , if α ≤ γ then α ∈ U . A cut A = (AL, AR) of T is a partition of T into two subsets AL and AR, such that, for every α ∈ AL and β ∈ AR, α < β. The set of all cuts A = (AL, AR) of the ordered set T contains the two cuts (∅, T ) and (T, ∅); these are commonly denoted by −∞ and ∞, 3 respectively. However, we do not use the symbols −∞ and ∞ to denote the above cuts since we define a "different" ∞. Given α ∈ T , we denote and (−∞, α] = {γ ∈ T γ ≤ α} (α, ∞) = {γ ∈ T γ > α}. One defines similarly the sets (−∞, α) and [α, ∞). To define a cut we often write AL = U , meaning the A is defined as (U, T \ U ) when U is an initial subset of T . The ordering on the set of all cuts of T is defined by A ≤ B iff AL ⊆ BL (or equivalently AR ⊇ BR). For a group Γ, subsets U, U ′ ⊆ Γ and n ∈ N, we define U + U ′ = {α + β α ∈ U, β ∈ U ′}; nU = {s1 + s2 + ... + sn s1, s2, ..., sn ∈ U }. Now, for Γ a totally ordered abelian group, M(Γ) is called the cut monoid of Γ; M(Γ) is a totally ordered abelian monoid. For A, B ∈ M(Γ), their (left) sum is the cut defined by (A + B)L = AL + BL. The zero in M(Γ) is the cut ((−∞, 0], (0, ∞)). For A ∈ M(Γ) and n ∈ N, the cut nA is defined by (nA)L = nAL. Note that there is a natural monomorphism of monoids ϕ : Γ → M(Γ) defined in the following way: for every α ∈ Γ, ϕ(α) = ((−∞, α], (α, ∞)) For α ∈ Γ and B ∈ M(Γ), we denote B − α for the cut B + (−α) (viewing −α as an element of M(Γ)). Definition 2.2. Let v be a valuation on a field F with value group Γv. Let Ov be the valuation domain of v and let R be an algebra over Ov. For every x ∈ R, the Ov-support of x in R is the set SR/Ov x = {a ∈ OvxR ⊆ aR}. We suppress R/Ov when it is understood. 4 For every A ⊆ Ov we denote (v(A))≥0 = {v(a) a ∈ A}; in particular, (v(Sx))≥0 = {v(a)a ∈ Sx}; the reason for this notion is the fact that (v(Sx))≥0 is an initial subset of (Γv)≥0. We define v(Sx) = (v(Sx))≥0 ∪ (Γv)<0; and note that v(Sx) is an initial subset of Γv. Note that if A and B are subsets of Ov such that A ⊆ B then v(A) ⊆ v(B). Recall that we do not denote the cut (Γv, ∅) ∈ M(Γv) as ∞. So, as usual, we adjoin to M(Γv) an element ∞ greater than all elements of M(Γv); for every A ∈ M(Γv) and α ∈ Γv we define ∞ + A = A + ∞ = ∞ and ∞ − α = ∞. The following theorem holds for arbitrary algebras R. Theorem 2.3. Let v be a valuation on a field F with value group Γv. Let Ov be the valuation domain of v and let R be an algebra over Ov. Let M(Γv) denote the cut monoid of Γv. Then there exists a quasi-valuation w : R → M(Γv) ∪ {∞}. It is shown in the proof of Theorem 2.3 that w can be defined by: w(0) = ∞, and w(x) = (v(Sx), Γv \ v(Sx)) for every 0 6= x ∈ R; i.e., w(x)L = v(Sx). We call w the filter quasi-valuation induced by (R, v). Let us present several basic properties of the filter quasi-valuation. Lemma 2.4. Notation as in Theorem 2.3, assume in addition that R is torsion free over Ov; then w(cx) = v(c) + w(x) for every c ∈ Ov, x ∈ R. We note that even in the case where R is a torsion free algebra over Ov, one does not necessarily have w(c · 1R) = v(c) for c ∈ Ov, despite the fact that w(c · 1R) = v(c) + w(1R) by the previous Lemma. The reason is that w(1R) is not necessarily 0 (see, for example, [Sa1, Ex. 9.28]). Remark 2.5. Note that if R is a torsion free algebra over Ov, then there is an embedding R ֒→ R ⊗Ov F ; in this case the quasi-valuation on R ⊗Ov F extends the quasi-valuation on R. 5 Lemma 2.6. Let v, F, Γv and Ov be as in Theorem 2.3. Let R be a torsion free algebra over Ov, S a multiplicative closed subset of Ov, 0 /∈ S, and let w : R → M ∪ {∞} be any quasi-valuation where M is any totally ordered abelian monoid containing Γv and w(cx) = v(c) + w(x) for every c ∈ Ov, x ∈ R. Then there exists a quasi-valuation W on R ⊗Ov OvS−1, extending w on R (under the identification of R with R ⊗Ov 1), with value monoid M ∪ {∞}. Theorem 2.7. Let v, F, Γv, Ov and M(Γv) be as in Theorem 2.3. Let R be a torsion free algebra over Ov and let w denote the filter quasi-valuation in- duced by (R, v); then there exists a quasi-valuation W on R⊗Ov F , extending w on R, with value monoid M(Γv) ∪ {∞} and OW = R ⊗Ov 1. It is not difficult to see that for W as in the previous Theorem, (∅, Γv) /∈ im(W ). This W is also called the filter quasi-valuation induced by (R, v). We conclude the following important theorem (see [Sa1, Theorem 9.34]), Theorem 2.8. Let v, F, Γv, Ov and M(Γv) be as in Theorem 2.3 and let A be an F -algebra. Let R be a subring of A such that R ∩ F = Ov. Then there exists a quasi-valuation W on RF with value monoid M(Γv) ∪ {∞} such that R = OW and W extends v (on F ). 3 Existence Theorems In this section S denotes an integral domain which is not a field, F denotes its field of fractions, and A 6= F is an F -algebra, usually taken to contain an S-stable basis. Assuming A contains an S-stable basis, we prove the existence of S- subalgebras of A which lie over S and whose localizations with respect to S is A. we also show that any such S-subalgebra of A is not minimal with respect to inclusion. Moreover, we prove the existence of such S-subalgebras of A containing a given ideal of A. We conclude that for any such F -algebra there exists a quasi-valuation on A extending a given valuation v on F . In fact, there exists an infinite descending chain of quasi-valuations on A, all of which extend v on F . At the end, we use these existence theorems to show that any chain of prime ideals of S can be covered by a chain of prime ideals of some S-subalgebra of A with the properties mentioned above. Let v be a valuation on F with corresponding valuation domain Ov. By Theorem 2.8, one can construct a quasi-valuation on A when one is given an Ov-subalgebra of A lying over Ov and whose localization with respect to Ov is equal to A. Our preliminary goal is to prove the existence of such subalgebras of A; but in fact, we prove a more general theorem, replacing the valuation domain with an integral domain. Thus, we focus on the existence of such S-subalgebras of A. We begin, hence, with an obvious definition. 6 Definition 3.1. Let R be an S-subalgebra of A. We say that R is an S-nice subalgebra of A if R ∩ F = S and RF = A. In other words, R is an S-nice subalgebra of A if R is lying over S and the localization of R with respect to S is equal to A. The following lemma is easy to prove and we shall not prove it here. Lemma 3.2. Let R be an S-subalgebra of A. Then, RF = A iff R contains a basis of A over F . To be able to study more efficiently infinite dimensional algebras over F , we introduce the following definition. Definition 3.3. Let B be a basis of A over F . We say that B is S-stable if there exists a basis C of A over F such that for all c ∈ C and b ∈ B, one has cb ∈ Py∈B Sy. We also say that C stabilizes B (whenever S is understood). Remark 3.4. Note that if B is closed under multiplication then B is S- stable. Thus, for example, every free (noncommutative) F -algebra with an arbitrary set of generators has an S-stable basis; in particular, every polynomial algebra with an arbitrary set of indeterminates over F has an S-stable basis. The following two remarks are easy to prove. Remark 3.5. Let S1 ⊆ S2 be integral domains with field of fractions F . If B is an S1-stable basis of A over F , then B is an S2-stable basis of A over F . Remark 3.6. Let {Ri}n i=1 be a finite set of S-subalgebras of A such that RiF = A for all 1 ≤ i ≤ n. Then R = Tn i=1 Ri is an S-subalgebra of A satisfying RF = A. In particular, an intersection of a finitely many S-nice subalgebras of A is an S-nice subalgebra of A. Although we do not know whether an S-stable basis always exists, one can modify a given S-stable basis, as shown in the following lemma. Lemma 3.7. Let B be an S-stable basis of A over F and let 0 6= x0 ∈ A\B. Then there exists an S-stable basis containing x0. More precisely, there exists an S-stable basis of the form {x0} ∪ B \ {b0} for some b0 ∈ B. Proof. Let C be a basis that stabilizes B. Let B′ be a basis such that x0 ∈ B′ and B′ \ {x0} = B \ {b0} for an appropriate b0 ∈ B (of course, one can write x0 as a linear combination of elements of B and take any element of B appearing in this linear combination). Take s0 ∈ S such that s0b0 ∈ Py∈B ′ Sy. It is now easy to check that for all c ∈ C and b ∈ B′ \ {x0}, we have s0c · b ∈ Py∈B ′ Sy; we note by passing that the basis B′′ = {s0b0} ∪ B \ {b0} is S-stable, since the basis C ′ = {s0 · c}c∈C stabilizes it. As for x0, for every s0 · c ∈ C ′ there exists sc ∈ S such that scs0c · x0 ∈ Py∈B ′ Sy. Thus, the basis {scs0 · c}c∈C stabilizes B′. 7 By Lemma 3.7 and induction we conclude: Lemma 3.8. Let B be an S-stable basis of A over F and let C ⊆ A be a finite linearly independent set. Then there exists an S-stable basis containing C. More precisely, there exists an S-stable basis of the form C ∪ B \ C1 for some finite set C1 ⊆ B. In the following proposition we prove the existence of an S-subalgebra of A lying over S. Proposition 3.9. Let B be a basis of A over F . Let M = Pb∈B Sb and R = {x ∈ A xM ⊆ M }. Then R is an S-subalgebra of A satisfying R ∩ F = S. Proof. First note that M is an S-submodule of A and thus S ⊆ R and R is an S-subalgebra of A. Now, let α ∈ R ∩ F and take b ∈ B. Then, αb = Pk i=1 ⊆ B is linearly independent, α ∈ S. i=1 sibi for some si ∈ S and bi ∈ B. Since the set {b} ∪ {bi}k Note that in Proposition 3.9, B is merely a basis of A over F and we do not assume that B is S-stable. Remark 3.10. If one takes in Proposition 3.9 a basis B containing 1 (or any invertible element u ∈ S), then R ⊆ M and M ∩ F = S. Proof. We prove the remark for the case in which 1 ∈ B. For the case in which B contains an invertible element u ∈ S the proof is similar. Now, R ⊆ M and S ⊆ M because 1 ∈ M . Thus, it is enough to show that M ∩ F ⊆ S. So, let α ∈ M ∩ F . Since F is the quotient field of S, there exists t ∈ S such that tα ∈ S; clearly, tα ∈ M . Also, since α ∈ M , one can write α = Pk i=1 tsixi; but tα ∈ S ∩ M and thus it can be uniquely written as a linear combination of elements of B in the form tα = tα · 1. Therefore, the presentation of tα as Pk i=1 tsixi must be equal to tα · 1; i.e., k = 1 and x1 = 1. Consequently, α ∈ S. i=1 sixi with si ∈ S and xi ∈ B. Hence, tα = Pk Remark 3.11. In view of the previous remark, M does not necessarily lie over S. Indeed, take any basis B containing an element α ∈ F \ S. As mentioned above, the existence of an S-stable basis is not known in the general case, for any F -algebra. However, when A is finite dimensional over F , the following proposition shows not only the existence of such a basis, but even more so that every basis is S-stable. Proposition 3.12. If A is finite dimensional over F , then every basis of A over F is S-stable. 8 Proof. Let B = {x1, x2, ..., xn} be a basis of A over F and let M = Pn For all 1 ≤ i, j ≤ n one can write xixj = Pn Let γij = Qn {δ1x1, δ2x2, ..., δnxn} stabilizes B. k=1 βijk; then γijxixj ∈ M . Let δi = Qn i=1 Sxi. xk where αijk, βijk ∈ S. j=1 γij; then C = αijk βijk k=1 We note that the most restrictive assumption that we make in this paper is that A contains an S-stable basis. Therefore, in light of Proposition 3.12, all of the results presented in this paper apply to any finite dimensional F -algebra. In the following theorem we prove the existence of an S-nice subalgebra of A. Theorem 3.13. Let B be an S-stable basis of A over F . Let M = Pb∈B Sb and R = {x ∈ A xM ⊆ M }. Then R is an S-nice subalgebra of A. Proof. Let C be a basis that stabilizes B. Then R contains C and thus by Lemma 3.2, RF = A. By Proposition 3.9, R is lying over S. So, R is an S-nice subalgebra of A. In a less detailed form, Theorem 3.13 can be restated as follows: Theorem 3.14. If there exists an S-stable basis of A over F , then there exists an S-nice subalgebra of A. Note that in view of Theorem 3.14 and Proposition 3.12, if A is finite dimensional over F then an S-nice subalgebra of A always exists. The following is a generalization of Theorem 3.14. Theorem 3.15. Let I be a proper ideal of A. Assume that there exists a basis of I over F that is contained in some S-stable basis of A over F . Then there exists an S-nice subalgebra R of A such that I ⊳ R. Proof. Let B1 be a basis of I over F and let B1 ⊂ B be an S-stable basis of A over F . Let N = I + Pb∈B\B1 Sb; N is clearly an S-submodule of A and thus R = {x ∈ A xN ⊆ N } is an S-subalgebra of A. We prove that R ∩ F = S. It is clear that S ⊆ R. On the other hand, let α ∈ R ∩ F and take b ∈ B \ B1. Then, αb ∈ N ; since b /∈ I and B is linearly independent, α ∈ S. We prove now RF = A. Let M = Pb∈B Sb (note that M ⊆ N ) and let R′ = {x ∈ A xM ⊆ M }. We prove R′ ⊆ R. Indeed, Let x ∈ R′, then xN = x(I + X b∈B\B1 Sb) = xI + x X b∈B\B1 Sb ⊆ I + M = N. Now, by Theorem 3.13 (or Lemma 3.2), R′F = A. Therefore, RF = A. Finally, let x ∈ I; then xN ⊆ xA ⊆ I ⊆ N . Therefore x ∈ R and the theorem is proved. 9 In view of Theorem 3.15, we have the following two remarks. Remark 3.16. Let I be a proper ideal of A. Assume that A contains an S-stable basis. If I has a finite basis over F then, by Proposition 3.8 and Theorem 3.15, there exists an S-nice subalgebra R of A such that I ⊳ R. If [A : F ] < ∞ then there Remark 3.17. Let I be a proper ideal of A. exists an S-nice subalgebra R of A such that I ⊳ R, by Proposition 3.12 and Theorem 3.15. Let E be a field with valuation v and corresponding valuation domain Ov. It is well known (see, for example, [En, p. 62, Corollary 9.7]) that for any field K ⊇ F , there exists at least one and often many different valuation domains of K lying over Ov. Let us mention now some other well-known existence theorems regarding extensions of valuation domains. Three main classes of rings were suggested throughout the years as the noncommutative version of a valuation ring. These three types are invariant valuation rings, total valuation rings, and Dubrovin valuation rings. They are interconnected by the following diagram: {invariant valuation rings} ⊂ {total valuation rings} ⊂ {Dubrovin valuation rings}. We shall now define these classes of rings and present their existence theo- rems. An invariant valuation ring V of a division ring D is a subring V of D such that for every a ∈ D∗ we have a ∈ V or a−1 ∈ V , and also aV a−1 ⊆ V . An invariant valuation ring corresponds to a valuation on D, in the usual sense. Let D be a division ring finite dimensional over its center, E. An invari- ant valuation ring lying over Ov exists if and only if v extends uniquely to each field L with E ⊆ L ⊆ D; in particular, there exists at most one (but perhaps none) invariant valuation ring lying over Ov. See [Wa, Theorem 2.1]. A subring V of a division ring D is called a total valuation ring of D if for every d ∈ D∗, we have d ∈ V or d−1 ∈ V . Then, there exists a total valuation ring lying over Ov iff the set T = {d ∈ D d is integral over Ov} is a ring. When this occurs, there are only finitely many different total valuation rings lying over Ov. See [Wa, Theorem 9.2]. Mathiak has shown (see [Mat, p. 5]) that they have a valuation-like function whose image is a totally ordered set which is not a group. Let C be a simple Artinian ring, a subring B of C is called a Dubrovin valuation ring of C if B has an ideal J such that B/J is a simple Artinian ring, and for each c ∈ C \ B there are b, b′ ∈ B, such that cb ∈ B \ J and b′c ∈ B \ J. Let C be a central simple algebra over E; i.e., C is a simple E-algebra finite dimensional over its center E. Then there exists a Dubrovin valuation 10 ring B of C lying over Ov. Furthermore, if B′ is another Dubrovin valuation ring of C lying over Ov, then there is c ∈ C ∗ with B = cBc−1. See [Wa, Theorem 10.3]. Morandi (cf. [Mor]) defines a value function which is a quasi-valuation satisfying a few more conditions. Given an integral Dubrovin valuation ring B of a central simple algebra C, Morandi shows that there is a value function w on C with B as its value ring (the value ring of w is defined as the set of all x ∈ C such that w(x) ≥ 0). Morandi also proves the converse, that if w is a value function on C, then the value ring is an integral Dubrovin valuation ring. Now, we apply the above results to prove the existence of quasi-valuations on A extending v on F . Theorem 3.18. Let v be a valuation on F and let Ov be the valuation domain corresponding to v. Let A be an F -algebra. If there exists an Ov- stable basis of A over F , then there exists a quasi-valuation w on A extending v on F . Proof. By Theorem 3.14 there exists an Ov-nice subalgebra R of A. By Theorem 2.8, there exists a quasi-valuation w on A extending v on F whose corresponding quasi-valuation ring is R. Corollary 3.19. Let v be a valuation on F and let Ov be the valuation domain corresponding to v. Let A be an F -algebra and let I be proper ideal of A. If there exists a basis of I over F that is contained in some Ov-stable basis of A over F , then there exists a quasi-valuation w on A extending v on F such that w(I) ≥ 0. Proof. By Theorem 3.15 there exists an Ov-nice subalgebra R of A such that I ⊳ R. The result now follows from Theorem 2.8. Let I be a proper ideal of A. In view of the remarks after Theorem 3.15, if either A is finite dimensional over F or A contains an Ov-stable basis and I is finite dimensional over F , then the conclusion of the previous corollary is valid. We continue our study of S-nice subalgebras of A. The following lemma may be thought of as a going-down lemma for S-nice subalgebras. Lemma 3.20. Let S1 ⊆ S2 be integral domains with field of fractions F . Assume that there exists an S1-stable basis of A over F . Let R be an S2- nice subalgebra of A. Then there exists an S1-nice subalgebra of A, which is contained in R. Proof. Let B be an S1-stable basis of A over F . Let N = Pb∈B S1b and let R1 = {x ∈ A xN ⊆ N }. Then, by Theorem 3.13, R1 is an S1-nice 11 subalgebra of A. So, R1 ∩ R is an S1-nice subalgebra of A which is contained in R. Assuming there exists an S-stable basis of A over F , by Theorem 3.14 and Zorn's lemma it is clear that there exists a maximal (with respect to containment) S-nice subalgebra of A. In the following proposition we prove that a minimal S-nice subalgebra of A does not exist. Proposition 3.21. Assume that there exists an S-stable basis of A over F . Let R be an S-nice subalgebra of A. Then there exists an infinite decreasing chain of S-nice subalgebras of A starting from R. In particular, a minimal S-nice subalgebra of A does not exist. Proof. Let B be an S-stable basis of A over F . Let y ∈ R \ S and let s0 6= 0 be a non-invertible element of S. By Lemma 3.8 there exists an S-stable basis, say B1, containing (the linearly independent set) {1, s0y}. Let M = Pb∈B1 Sb and let R1 = {x ∈ A xM ⊆ M }. By Theorem 3.13, R1 is an S-nice subalgebra of A. Now, R1 ⊆ M since 1 ∈ M , and y /∈ R1 since y /∈ M . Thus, R * R1 and R1 ∩ R is an S-nice subalgebra of A which is strictly contained in R. Assume that there exists an S-stable basis of A over F . Let U be a maximal chain of S-nice subalgebras of A; i.e., there is no chain of S-nice subalgebras of A that strictly contains U . Note that by Zorn's lemma there exists such a chain; in fact, by Theorem 3.14 every such chain is nonempty and by Proposition 3.21 every such chain is infinite. Consider the ring T = TR∈U R, which is clearly an S-subalgebra of A lying over S. Then, T is not an S-nice subalgebra of A. Indeed, otherwise it would be a minimal S-nice subalgebra of A, in contradiction to Proposition 3.21. Thus, the localization of T with respect to S is an S-subalgebra strictly contained in A. It is easy to see that in the following two special cases T is equal to S: (a) whenever [A : F ] = 2; (b) when A is a field and [A : F ] is prime. Note that in particular, denoting U ′ = {R R is an S-nice subalgebra of A}, the ring T ′ = ∩R∈U ′R is not an S-nice subalgebra of A. We shall now discuss prime ideals of S; in particular, arbitrary chains of prime ideals of S. We will show the existence of S-subalgebras of A having a "sufficiently rich" prime spectrum with respect to S. It is well known that prime ideals can be viewed in terms of algebraic geometry, where the prime spectrum of a ring is endowed with the usual Zarisky topology, even when the ring is not commutative; see, for example, [OV, p. 36]. For a brief discussion on examples of rings having infinite chains of prime ideals we refer the reader to [Sa3]. 12 Definition 3.22. Let C be a chain of prime ideals of S and let R be a faithful S-algebra. Let D be a chain of prime ideals of R. We say that D covers C if for every P ∈ C there exists Q ∈ D lying over P ; namely, Q ∩ S = P . In Proposition 3.23 and Theorem 3.24 we apply results from valuation theory and from quasi-valuation theory. Given any chain of prime ideals C of S we prove the existence of S-subalgebras of A having a chain of prime ideals covering C. Proposition 3.23. Assume that there exists an S-stable basis of A over F . Let C be a chain of prime ideals of S. Then there exists an S-subalgebra R of A whose localization with respect to S is A and such that there exists a chain of prime ideals E of R covering C. In fact, there exists an infinite descending chain of such S-subalgebras of A. Proof. By [KO, main theorem] there exists a valuation domain S ⊆ Ov ⊂ F and a chain of prime ideals D of Ov covering C. By assumption and Remark 3.5, there exists an Ov-stable basis of A over F . Thus, by Theorem 3.14, there exists an Ov-nice subalgebra R of A. By [Sa2, Corollary 3.9], there exists a chain of prime ideals E of R covering D; it is clear that R is an S-subalgebra of A satisfying RF = A and E lies over C. The final assertion follows from Proposition 3.21 and [Sa2, Corollary 3.9]. In Proposition 3.21 we proved that the set of S-nice subalgebras of A is quite rich; we shall now prove that the prime spectra of certain S-nice subalgebras of A are rich enough with respect to the prime spectrum of S. This will be done in the case where A is commutative; in this case we can sharpen Proposition 3.23. Theorem 3.24. Assume that there exists an S-stable basis of A over F . Let C be a chain of prime ideals of S. If A is commutative then there exists an S-nice subalgebra R of A such that there exists a chain of prime ideals F of R covering C. In fact, there exists an infinite descending chain of such S-nice subalgebras of A. Proof. By Proposition 3.23 there exists an S-subalgebra R′ of A such that R′F = A and a chain of prime ideals E of R′ covering C. Clearly S ⊆ R′ ∩ F . Thus, by Lemma 3.20, there exists an S-nice subalgebra R of A which is contained in R′. Since A is commutative, every prime ideal of R′ intersects to a prime ideal of R. Thus, {Q ∩ R}Q∈E is a chain of prime ideals of R covering C. The final assertion follows from Proposition 3.21. 13 Of course, not every S-nice subalgebra of A satisfies the property of the previous theorem. Indeed, consider any S-nice subalgebra of A that does not satisfy LO over S. Let v be a valuation on F with corresponding valuation domain Ov, and let w1 and w2 be two quasi-valuations on A extending v on F . We assume that w1 and w2 are comparable; namely, there exists a totally ordered set M such that w1(A) ∪ w2(A) ⊆ M ∪ {∞}. We write w1 ≤′ w2 if for every x ∈ A, w1(x) ≤ w2(x) in M ∪ {∞}. It is clear that if w1 and w2 are both filter quasi-valuations then they are comparable. Thus, the set of all filter quasi-valuations on A extending v on F is partially ordered by ≤′. Let R1 ⊆ R2 be Ov-nice subalgebras of A. Let wi (i = 1, 2) be the filter quasi-valuation on A induced by (Ri, v). From the construction of the filter quasi-valuation one can check that w1 ≤′ w2. Now, Let U = {Ri}i∈I be a nonempty chain of S-nice subalgebras of A and let T = Ti∈I Ri. Let wi (i ∈ I) be the filter quasi-valuation on A induced by (Ri, v). Let w denote the filter quasi-valuation induced by (T, v). It is not difficult to see that for every x ∈ T F we have w(x) = Ti∈I wi(x). Moreover, if U is a maximal chain of S-nice subalgebras of A then, as shown above, T F is a proper T - In this case, for all x ∈ A, we have Ti∈I wi(x) = ∅ iff subalgebra of A. x ∈ A \ T F . By Proposition 3.21, Theorem 3.18, and the discussion above we have, Corollary 3.25. Let F be a field with valuation v and let Ov be the cor- responding valuation domain. Let A 6= F be an F -algebra having an Ov- stable basis over F . Then there exists an infinite decreasing chain of quasi- valuations on A extending v on F . Moreover, for any Ov 6= R ⊆ A, an Ov-subalgebra of A lying over Ov there exists an infinite decreasing chain of quasi-valuations on RF extending v on F starting from wR; where wR denotes the filter quasi-valuation induced by (R, v). In Theorem 3.21 we showed that there is no minimal (with respect to inclusion) S-nice subalgebra of A; in particular there exists an infinite de- scending chain of S-nice subalgebras of A. As noted above, by Zorn's Lemma there exists a maximal S-nice subalgebra of A. In case S is a valuation do- main of F and A is a field then the maximal S-nice subalgebras of A are precisely the valuation domains (whose valuations extend v) of A. We shall now show that even in the case of a central simple F -algebra, one can have an infinite ascending chain of S-nice subalgebras of A (even when S is a valuation domain). Example 3.26. Let C be a non-Noetherian integral domain with field of fractions F . Let {0} 6= I1 ⊂ I2 ⊂ I3 ⊂ ... be an infinite ascending chain of ideals of C and let A = Mn(F ). Then 14   C C ... C I1 C C ... C I1 . . . . . . C C ... C I1 C C ... C C ... ... ... . . . . . .   ⊂   C C ... C I2 C C ... C I2 . . . . . . C C ... C I2 C C ... C C ... ... ... . . . . . .   ... is an infinite accending chain of C-nice subalgebras of A. References [Bo] N. Bourbaki, Commutative Algebra, Chapter 6, Valuations, Hermann, Paris, 1961. [Co] P. M. Cohn, An Invariant Characterization of Pseudo-Valuations, Proc. Camp. Phil. Soc. 50 (1954), 159-177. [End] O. Endler, Valuation Theory. Springer-Verlag, New York, 1972. [FKK] A. Fornasiero, F.V. Kuhlmann and S. Kuhlmann, Towers of com- plements to valuation rings and truncation closed embeddings of valued fields, J. Algebra 323 (2010), no. 3, 574-600. [Hu] J. A. Huckaba, Extensions of Pseudo-Valuations. Pacific J. Math. 29, Number 2 (1969), 295 -- 302. [KO] B. G. Kang and D. Y. Oh, Lifting up an infinite chain of prime ideals to a valuation ring, Proc. Amer. Math. Soc. 126 (1998), no. 3, 645646. [KZ] M. Knebusch and D. Zhang, Manis Valuations and Prufer Extensions. Springer-Verlag, Berlin, 2002. [Mat] K. Mathiak Valuations of Skew Fields and Projective Hjelmslev Spaces, Lecture Notes in Math., Springer, Berlin, vol. 1175 (1986). [MH] M. Mahadavi-Hezavehi, Matrix Pseudo-Valuations on Rings and their Associated Skew Fields. Int. Math. J. 2 (2002), no. 1, 7 -- 30. [Mor] P. J. Morandi, Value functions on central simple algebras, Trans. Amer. Math. Soc., 315 (1989), 605-622. [Wa] A.R. Wadsworth, Valuation theory on finite dimensional division al- gebras. In Valuation Theory and its Applications, Vol. 1, eds. F.-V. Kuhlmann et al., Fields Inst. Commun. 32, American Mathematical Society, Providence, RI, 2002. 15
1803.10186	4	1803	2019-05-21T05:20:15	Representations and properties of the W-weighted core-EP inverse	[ "math.RA" ]	In this paper, we investigate the weighted core-EP inverse introduced by Ferreyra, Levis and Thome. Several computational representations of the weighted core-EP inverse are obtained in terms of singular-value decomposition, full-rank decomposition and QR decomposition. These representations are expressed in terms of various matrix powers as well as matrix product involving the core-EP inverse, Moore-Penrose inverse and usual matrix inverse. Finally, those representations involving only Moore-Penrose inverse are compared and analyzed via computational complexity and numerical examples.	math.RA	math	Representations and properties of the W -weighted core-EP inverse Yuefeng Gao1,2∗, Jianlong Chen1†, Pedro Patr´ıcio2 ‡ 1School of Mathematics, Southeast University, Nanjing 210096, China; 2CMAT-Centro de Matem´atica, Universidade do Minho, Braga 4710-057, Portugal In this paper, we investigate the weighted core-EP inverse introduced by Ferreyra, Levis and Thome. Several computational representations of the weighted core-EP inverse are obtained in terms of singular-value decomposition, full-rank decomposition and QR de- composition. These representations are expressed in terms of various matrix powers as well as matrix product involving the core-EP inverse, Moore-Penrose inverse and usual matrix inverse. Finally, those representations involving only Moore-Penrose inverse are compared and analyzed via computational complexity and numerical examples. Keywords: weighted core-EP inverse, core-EP inverse, pseudo core inverse, outer inverse, complexity AMS Subject Classifications: 15A09; 65F20; 68Q25 1 Introduction Let Cm×n be the set of all m × n complex matrices and let Cm×n be the set of all m × n complex matrices of rank r. For each complex matrix A ∈ Cm×n, A∗, Rs(A), R(A) and N (A) denote the conjugate transpose, row space, range (column space) and null space of A, respectively. The index of A ∈ Cn×n, denoted by ind(A), is the smallest non-negative integer k for which rank(Ak) =rank(Ak+1). The Moore-Penrose inverse (also known as the pseudoinverse) of A ∈ Cm×n, Drazin inverse of A ∈ Cn×n are denoted as usual by A†, AD respectively. r The Drazin inverse was extended to a rectangular matrix by Cline and Greville [1]. Let A ∈ Cm×n, W ∈ Cn×m and k =max{ind(AW ), ind(W A)}. The W -weighted Drazin inverse of A, denoted by AD,W , is the unique solution to (AW )k = (AW )k+1XW, X = XW AW X and AW X = XW A. ∗E-mail: [email protected] †Corresponding author. E-mail: [email protected] ‡E-mail: [email protected] 1 Many authors have been focusing on the weighted Drazin inverse and have achieved much in the aspect of representations (see for example, [2 -- 4]). Baksalary and Trenkler [5] introduced the notion of core inverse for a square matrix of index one. Then, Manjunatha Prasad and Mohana [6] proposed the core-EP inverse for a square matrix of arbitrary index, as an extension of the core inverse. Later, Gao and Chen [7] gave a characterization for the core-EP inverse in terms of three equations. The core-EP inverse of A ∈ Cn×n, denoted by A †(cid:13), is the unique solution to XAk+1 = Ak, AX 2 = X and (AX)∗ = AX, (.) where k =ind(A). The core-EP inverse is an outer inverse (resp. {2}-inverse), i.e., A †(cid:13)AA †(cid:13) = A †(cid:13). The core-EP inverse has the following properties: (1) R(A †(cid:13)) = R(Ak), N (A †(cid:13)) = N ((Ak)∗), (2) R(A †(cid:13)) ⊕ N (A †(cid:13)) = Cn×n, (3) AA †(cid:13) is an orthogonal projector onto R(Ak) and A †(cid:13)A is an oblique projector on to R(Ak) along N ((Ak)†A). The core inverse and core-EP inverse have applications in partial order theory (see for example, [8 -- 10]). Recently, an extension of the core-EP inverse from a square matrix to a rectangular matrix [11]. Let A ∈ Cm×n, W ∈ Cn×m and k =max{ind(AW ), was made by Ferreyra et al. ind(W A)}. The W -weighted core-EP inverse of A, denoted by A †(cid:13),W , is the unique solution to the system W AW X = (W A)k[(W A)k]† and R(X) ⊆ R((AW )k). (.) Meanwhile, the authors proved that the W -weighted core-EP inverse of A can be written as a product of matrix powers involving two Moore-Penrose inverses: A †(cid:13),W = [W (AW )l+1[(AW )l]†]† (l ≥ k). (.) Then, Mosi´c [12] studied the weighted core-EP inverse of an operator between two Hilbert spaces as a generalization of the weighted core-EP inverse of a rectangular matrix. In this paper, our main goal is to further study the weighted core-EP inverse for a rect- angular matrix and compile its new, computable representations. The paper is carried out as follows. In Section 2, first of all, the weighted core-EP inverse is characterized in terms of three equations. This could be very useful in testing the accuracy of a given numerical method (to compute the weighted core-EP inverse) via residual norms. Then, we derive the canonical form for the W -weighted core-EP inverse of A by using the singular value decompositions of A and W . Later, representations of the weighted core-EP inverse are obtained via full-rank decomposition, general algebraic structure (GAS) and QR decomposition in conjunction with the fact that the weighted core-EP inverse is a particular outer inverse. These representations are expressed eventually through various matrix powers as well as matrix product involving the core-EP inverse, Moore-Penrose inverse and usual matrix inverse. In Section 3, some properties of the weighted core-EP inverse are exhibited naturally as outcomes of given rep- resentations. As mentioned earlier, the weighted core-EP inverse is a particular outer inverse. 2 It is known that the inverse along an element [13] and (B, C)-inverse [14] are outer inverses as well. Thus, in Section 4, we wish to reveal the relations among the weighted core-EP inverse, weighted Drazin inverse, the inverse along an element, and (B, C)-inverse. In Section 5, the computational complexities of proposed representations involving pseudoinverse are estimated. In the last Section 6, corresponding numerical examples are implemented by using Matlab R2017b. 2 Representations of the weighted core-EP inverse In this section, we compile some new expressions of the weighted core-EP inverse for a rectan- gular complex matrix. First, the weighted core-EP inverse is characterized in terms of three equations.This plays a key role in examining the accuracy of a numerical method. Lemma 2.1. [7, Theorem 2.3] Let A ∈ Cn×n and let l be a non-negative integer such that l ≥ k = ind(A). Then A †(cid:13) = ADAl(Al)†. In this case, AA †(cid:13) = Al(Al)†. Theorem 2.2. Let A ∈ Cm×n, W ∈ Cn×m and k = max{ind(AW ), ind(W A)}. Then there exists a unique X ∈ Cm×n such that XW (AW )k+1 = (AW )k, AW XW X = X and (W AW X)∗ = W AW X. (.) The unique X which satisfies the above equations is X = A[(W A) †(cid:13)]2. Proof. First of all, we can check that X = A[(W A) †(cid:13)]2 satisfies the equations in (2.1). In fact, in view of Lemma 2.1, A[(W A) †(cid:13)]2W (AW )k+1 = A(W A) †(cid:13)[(W A) †(cid:13)(W A)k+1]W = A(W A) †(cid:13)(W A)kW = A(W A)D(W A)k[(W A)k]†(W A)kW = A(W A)D(W A)kW, which implies that A[(W A) †(cid:13)]2W (AW )k+1 = (AW )D(AW )k+1 = (AW )k, since A(W A)D = (AW )DA; AW A[(W A) †(cid:13)]2W A[(W A) †(cid:13)]2 = A(W A) †(cid:13)W A[(W A) †(cid:13)]2 = A[(W A) †(cid:13)]2; (W AW A[(W A) †(cid:13)]2)∗ =W A(W A) †(cid:13). Next, we would give a proof of the uniqueness of X. If XW (AW )k+1 = (AW )k, AW XW X = X and (W AW X)∗ = W AW X and Y W (AW )k+1 = (AW )k, AW Y W Y = Y and (W AW Y )∗ = W AW Y, 3 then X = AW XW X = (AW )2(XW )2X = (AW )k(XW )kX = Y W (AW )k+1(XW )kX = Y (W A)k+1(W X)k+1 = Y (W A)k+2(W X)k+2 = Y [(W A)k+2(W Y )k+2(W A)k+2](W X)k+2 = Y [(W A)k+2(W Y )k+2]∗[(W A)k+2(W X)k+2]∗ = Y [(W Y )k+2]∗[(W A)k+2(W X)k+2(W A)k+2]∗ = Y [(W Y )k+2]∗[(W A)k+2]∗ = Y (W AW Y )∗ = Y W AW Y = Y (W A)k+1(W Y )k+1 = Y W (AW )k+1(Y W )kY = (AW )k(Y W )kY = AW Y W Y = Y. This completes the proof. Theorem 2.3. Let A, X ∈ Cm×n, W ∈ Cn×m and k = max{ind(AW ), ind(W A)}. Then the following are equivalent: (1) A †(cid:13),W = X; (2) XW (AW )k+1 = (AW )k, AW XW X = X and (W AW X)∗ = W AW X. Proof. It suffices to show that X = A[(W A) †(cid:13)]2 satisfies condition (1.2). Indeed, W AW A[(W A) †(cid:13)]2 = W A(W A) †(cid:13) = W A(W A)D(W A)k[(W A)k]† = (W A)k[(W A)k]†, A[(W A) †(cid:13)]2 = AW A[(W A) †(cid:13)]3 = A(W A)k[(W A) †(cid:13)]k+2 = (AW )kA[(W A) †(cid:13)]k+2, i.e., R(A[(W A) †(cid:13)]2) ⊆ R((AW )k). This completes the proof. We now give the canonical form for the W -weighted core-EP inverse of A by using the be of the following singular value decompositions of A and W . Let A ∈ Cm×n singular value decompositions: , W ∈ Cn×m s r A = U (cid:18)Σ1 0 0 0(cid:19) V ∗ and W = S(cid:18)Σ2 0 0(cid:19) T ∗, 0 (.) where U, V, S, T are unitary matrices, Σ1 =diag(σ1, · · · , σr), σ1 ≥ · · · ≥ σr > 0, entries σi are known as the singular values of A, and Σ2 =diag(τ1, · · · , τs), τ1 ≥ · · · ≥ τs > 0, entries τi are singular values of W . Theorem 2.4. Let A ∈ Cm×n (2.2). Then r , W ∈ Cn×m s be of the singular value decompositions as in where T ∗U = (cid:20)R1 R2 R3 R4(cid:21) , R1 ∈ Cs×r, V ∗S = (cid:20)H1 H2 A †(cid:13),W = U (cid:20)Σ1H1[(Σ2R1Σ1H1) †(cid:13)]2 0 0 0(cid:21) S∗, H3 H4(cid:21) , H1 ∈ Cr×s. (.) 4 Proof. Observe that W A = S(cid:20)Σ2 0 that ind(W A) = k and X = S(cid:20)X1 X2 0(cid:21) V ∗. Now suppose X3 X4(cid:21) S∗ (X1 ∈ Cs×s) is the core-EP inverse of W A, then 0(cid:21) V ∗ = S(cid:20)Σ2R1Σ1 0 0(cid:21) T ∗U (cid:20)Σ1 0 0 0 0 X would satisfy condition (1.1). Thus, by computation, (Σ2R1Σ1H1X1)∗ = Σ2R1Σ1H1X1, Σ2R1Σ1H1X2 = 0, Σ2R1Σ1H1X 2 X1Σ2R1Σ1(H1Σ2R1Σ1)k = Σ2R1Σ1(H1Σ2R1Σ1)k−1, which implies that X1(Σ2R1Σ1H1)k+1 = (Σ2R1Σ1H1)k. 1 = X1, X3 = X4 = 0, These equalities above show that X1 = (Σ2R1Σ1H1) †(cid:13). As the core-EP inverse is an outer inverse, i.e., XW AX = X, then X2 = X1Σ2R1Σ1H1X2 = 0. Hence, (W A) †(cid:13) = S(cid:20)(Σ2R1Σ1H1) †(cid:13) 0 0(cid:21) S∗. 0 In light of Theorems 2.2 and 2.3, A †(cid:13),W = A[(W A) †(cid:13)]2 = U (cid:20)Σ1H1[(Σ2R1Σ1H1) †(cid:13)]2 0 0(cid:21) S∗. 0 This completes the proof. Additional representations of the weighted core-EP inverse can be obtained through the In 1974, Ben-Israel and full-rank decomposition. First, let us recall a concerned notion. Greville [15] introduced the notion of generalized inverse with prescribed range and null space. Let A ∈ Cm×n , T be a subspace of Cn of dimension s ≤ r and let S be a subspace of Cm of dimension m − s. If A has a {2}-inverse X such that R(X) = T and N (X) = S, then X is unique and denoted by A(2) T,S. Further, Sheng and Chen [16] gave a full-rank representation of the generalized inverse A(2) T,S, which is based on an arbitrary full-rank decomposition of G, where G is a matrix such that R(G) = T and N (G) = S. r Lemma 2.5. [16, Theorem 3.1] Let A ∈ Cm×n , T be a subspace of Cn of dimension s ≤ r and let S be a subspace of Cm of dimension m − s. Suppose that G ∈ Cn×m satisfies R(G) = T, N (G) = S. Let G be of an arbitrary full-rank decomposition, namely G = U V . If A has a {2}-inverse A(2) (1) V AU is invertible; (2) A(2) T,S, then T,S = U (V AU )−1V. r The following result shows that the weighted core-EP inverse is a generalized inverse with prescribed range and null space. 5 Theorem 2.6. Let A ∈ Cm×n, W ∈ Cn×m with ind(W A) = k. The W-weighted core-EP inverse of A is a {2}-inverse of W AW with the range R(A(W A)k[(W A)k]†) and the null space N (A(W A)k[(W A)k]†) i.e., A †(cid:13),W = (W AW )(2) R(G),N (G), (.) where G = A(W A)k[(W A)k]†. Proof. First, we check that A[(W A) †(cid:13)]2 is a {2}-inverse of W AW . Indeed, A[(W A) †(cid:13)]2W AW A[(W A) †(cid:13)]2 = A[(W A) †(cid:13)]2W A(W A) †(cid:13) = A[(W A) †(cid:13)]2. Then, we show that R(A(W A)k[(W A)k]†) = R(A[(W A) †(cid:13)]2) and N (A(W A)k[(W A)k]†) = N (A[(W A) †(cid:13)]2). Indeed, A(W A)k[(W A)k]† = A[(W A) †(cid:13)]2(W A)k+2[(W A)k]†, i.e., R(A(W A)k[(W A)k]†) ⊆ R(A[(W A) †(cid:13)]2); A[(W A) †(cid:13)]2 = A(W A)k[(W A) †(cid:13)]k+2 = A(W A)k[(W A)k]†(W A)k[(W A) †(cid:13)]k+2, i.e., R(A[(W A) †(cid:13)]2) ⊆ R(A(W A)k[(W A)k]†). If X ∈ N (A(W A)k[(W A)k]†), i.e., A(W A)k[(W A)k]†X = 0, then A[(W A) †(cid:13)]2X = A(W A) †(cid:13)(W A)D(W A)k[(W A)k]†X = A(W A) †(cid:13)[(W A)D]2W A(W A)k[(W A)k]†X = 0, namely, N (A(W A)k[(W A)k]†) ⊆ N (A[(W A) †(cid:13)]2); if X ∈ N (A[(W A) †(cid:13)]2), i.e., A[(W A) †(cid:13)]2X = 0, then A(W A)k[(W A)k]†X = AW A(W A) †(cid:13)X = AW AW A[(W A) †(cid:13)]2X = 0, namely, N (A[(W A) †(cid:13)]2) ⊆ N (A(W A)k[(W A)k]†). This completes the proof. From Theorem 2.6, it is known that the weighted core-EP inverse is a particular outer inverse. Then by applying Lemma 2.5, we derive new representations of the weighted core-EP inverse involving the usual matrix inverse. Corollary 2.7. Let A ∈ Cm×n, W ∈ Cn×m with ind(W A) = k. If A(W A)k[(W A)k]† = U V is a full-rank decomposition of A(W A)k[(W A)k]†. Then the W -weighted core-EP inverse of A possesses the following representation: A †(cid:13),W = U (V W AW U )−1V. (.) 6 Recall that the general algebraic structures (GAS) of A and W are defined as follows (see [3]): A = P (cid:20)A11 0 A22(cid:21) Q−1, W = Q(cid:20)W11 0 W22(cid:21) P −1, 0 0 (.) where P, Q, A11, W11 are non-singular matrices and A22, W22, A22W22, W22A22 are nilpotent matrices. Corollary 2.8. Let A ∈ Cm×n, W ∈ Cn×m with ind(W A) = k and let P = (cid:2)P1 P2(cid:3) , Q = (cid:2)L1 L2(cid:3), where P1, P2, L1, L2 are appropriate blocks arising from (2.6). Then the W - weighted core-EP inverse of A possesses the following representation: Proof. Suppose that Q−1 = (cid:20)Q1 1W AW P1)−1L∗ 1. A †(cid:13),W = P1(L∗ Q2(cid:21). From the GAS representations (2.6), it follows that (.) 0 0(cid:21) Q−1 = L1(W11A11)kQ1, 1L1)−1L∗ 1, 1)−1[(W11A11)k]−1(L∗ (W A)k = Q(cid:20)(W11A11)k 0 [(W A)k]† = Q∗ A(W A)k = P1A11(W11A11)kQ1 and A(W A)k[(W A)k]† = P1A11(L∗ 1(Q1Q∗ 1L1)−1L∗ 1. Therefore, it is possible to use the full-rank decomposition A(W A)k[(W A)k]† = U V , where U = P1A11 and V = (L∗ 1L1)−1L∗ 1. Then by Corollary 2.7, we obtain A †(cid:13),W = P1A11[(L∗ P1(L∗ 1. This completes the proof. 1W AW P1)−1L∗ 1L1)−1L∗ 1W AW P1A11]−1(L∗ 1L1)−1L∗ 1 = The following representation of the weighted core-EP inverse is based on the QR decom- position defined as in [2, 17, 18]. Corollary 2.9. Let A ∈ Cm×n, W ∈ Cn×m with ind(W A) = k, rank(W AW ) = r, rank[A(W A)k[(W A)k]†] = s, s ≤ r. Suppose that the QR decomposition of A(W A)k[(W A)k]† is of the form A(W A)k[(W A)k]†P = QR, where P is an n × n permutation matrix, Q ∈ Cm×m, Q∗Q = Im and R ∈ Cm×n trapezoidal matrix. Assume that P is chosen so that Q and R can be partitioned as s is an upper Q = (cid:2)Q1 Q2(cid:3) , R = (cid:20)R11 R12 0 0 (cid:21) = (cid:20)R1 0 (cid:21) , 7 R(G),N (G) = A †(cid:13),W , where G = A(W A)k[(W A)k]†, then where Q1 consists of the first s columns of Q and R11 ∈ Cs×s is non-singular. If W AW has a {2}-inverse (W AW )(2) (1) R1P ∗W AW Q1 is an invertible matrix; (2) A †(cid:13),W = Q1(R1P ∗W AW Q1)−1R1P ∗; (3) A †(cid:13),W = (W AW )(2) (4) A †(cid:13),W = Q1(Q∗ R(Q1),N (R1P ∗); 1A(W A)k[(W A)k]†W AW Q1)−1Q∗ 1A(W A)k[(W A)k]†. Various generalized inverses of complex matrices can be finally expressed in terms of the matrix product as well as matrix powers involving only Moore-Penrose inverse, so can the weighted core-EP inverse. It is crucial since in that case the operation could be implemented easily by Matlab. The main disadvantage of the representation (1.3) arises from the necessity to calculate Moore-Penrose inverses of two different matrices. The following result derives a representation of A †(cid:13),W , which involves only one Moore-Penrose inverse. Theorem 2.10. Let A ∈ Cm×n, W ∈ Cn×m and let l be a non-negative integer such that l ≥ k = max{ind(AW ), ind(W A)}. Then A †(cid:13),W can be written as follows: A †(cid:13),W = (AW )l[W (AW )l+1]†; A †(cid:13),W = A(W A)l[(W A)l+2]†. (.) (.) Proof. From Theorems 2.2 and 2.3, it follows that A †(cid:13),W = A[(W A) †(cid:13)]2. As (W A) †(cid:13) = (W A)D(W A)l[(W A)l]† = (W A)D(W A)l+2[(W A)l+2]† by Lemma 2.1, we derive that A †(cid:13),W = A[(W A) †(cid:13)]2 = A[(W A)D(W A)l+2[(W A)l+2]†]2 = A[(W A)D]2(W A)l+2[(W A)l+2]† = A(W A)l[(W A)l+2]†. One can verify (2.9) by checking three equations in Theorem 2.3. Here we omit the details. An expression of the core-EP inverse can be derived as a particular case W = I of Theo- rem 2.10. Corollary 2.11. Let A ∈ Cn×n and let l be a positive integer such that l ≥ k = ind(A). Then A †(cid:13) = Al(Al+1)†. 3 Properties of the weighted core-EP inverse In this section, we study the properties of the weighted core-EP inverse. 8 Proposition 3.1. Let A ∈ Cm×n, W ∈ Cn×m with k = max{ind(AW ), ind(W A)}. Then we have the following facts: (1) R(A †(cid:13),W ) = R((AW )k); (2) N (A †(cid:13),W ) = N ([(W A)k]∗). (1) In view of Theorems 2.2 and 2.3, A †(cid:13),W = A[(W A) †(cid:13)]2 = A(W A)k[(W A) †(cid:13)]k+2 = Proof. (AW )kA[(W A) †(cid:13)]k+2, i.e., R(A †(cid:13),W ) ⊆ R((AW )k), together with (AW )k = A[(W A) †(cid:13)]2(W A)k+2W (AW )D = A †(cid:13),W (W A)k+1W, i.e., R((AW )k) ⊆ R(A †(cid:13),W ). Thus, R(A †(cid:13),W ) = R((AW )k). (2) Suppose Y ∈ N (A †(cid:13),W ), i.e., A[(W A) †(cid:13)]2Y = 0, then [(W A)k]∗(W A)2[(W A) †(cid:13)]2Y = 0. Thus, [(W A)k]∗Y = 0, i.e., N (A †(cid:13),W ) ⊆ N ([(W A)k]∗). Conversely, suppose Z ∈ N ([(W A)k]∗), i.e., [(W A)k]∗Z = 0, then A[(W A) †(cid:13)]2[(W A) †(cid:13)]k∗[(W A)k]∗Z = 0. Therefore, A †(cid:13),W Z = A[(W A) †(cid:13)]2Z = 0, i.e., N ([(W A)k]∗) ⊆ N (A †(cid:13),W ). Hence N ([(W A)k]∗) = N (A †(cid:13),W ). Proposition 3.2. Let A ∈ Cm×n, W ∈ Cn×m with ind(W A) = k. Then we have the following facts: (1) R(A †(cid:13),W W ) ⊕ N (A †(cid:13),W W ) = Cm; (2) R(W A †(cid:13),W ) ⊕ N (W A †(cid:13),W ) = Cn. (1) Observe that A †(cid:13),W W = A[(W A) †(cid:13)]2W . For any X ∈ Cm, X = A(W A) †(cid:13)W X + Proof. [I − A(W A) †(cid:13)W ]X, where A(W A) †(cid:13)W X = A(W A) †(cid:13)W A(W A) †(cid:13)W X = A(W A) †(cid:13)(W A)k[(W A) †(cid:13)]kW X = A[(W A) †(cid:13)]2(W A)k+1[(W A) †(cid:13)]kW X = A †(cid:13),W (W A)k+1[(W A) †(cid:13)]kW X ∈ R(A †(cid:13),W W ), A †(cid:13),W W [I − A(W A) †(cid:13)W ]X = A[(W A) †(cid:13)]2W [I − A(W A) †(cid:13)W ]X = A[(W A) †(cid:13)]2W X − A[(W A) †(cid:13)]2W A(W A) †(cid:13)W X = A[(W A) †(cid:13)]2W X − A[(W A) †(cid:13)]2W X = 0, which implies that [I − A(W A) †(cid:13)W ]X ∈ N (A †(cid:13),W W ). Therefore, R(A †(cid:13),W W ) + N (A †(cid:13),W W ) = Cm. Further, suppose Y ∈ R(A[(W A) †(cid:13)]2W ) ∩ N (A[(W A) †(cid:13)]2W ), that is to say, Y = A[(W A) †(cid:13)]2W Z for some Z ∈ Cm and A[(W A) †(cid:13)]2W Y = 0. Thus, A[(W A) †(cid:13)]2W A[(W A) †(cid:13)]2W Z = 0, i.e., A[(W A) †(cid:13)]3W Z = 0. Pre-multiply this equality by W AW , then (W A) †(cid:13)W Z = 0, which deduces that Y = 0. Hence R(A †(cid:13),W W )⊕N (A †(cid:13),W W ) = Cm. (2) Note that W A †(cid:13),W = (W A) †(cid:13). From R((W A) †(cid:13)) = R(W A(W A) †(cid:13)) and N ((W A) †(cid:13)) = N (W A(W A) †(cid:13)) as well as [W A(W A) †(cid:13)]2 = W A(W A) †(cid:13) = [W A(W A) †(cid:13)]∗, it follows clearly that R(W A †(cid:13),W ) ⊕ N (W A †(cid:13),W ) = Cn. 9 Proposition 3.3. Let A ∈ Cm×n, W ∈ Cn×m with ind(W A) = k. Then we have the following facts: (1) W AW A †(cid:13),W is an orthogonal projector onto R((W A)k); (2) W A †(cid:13),W W A is an oblique projector onto R((W A)k) along N ([(W A)k]†W A). Proof. (1) Since A †(cid:13),W = A[(W A) †(cid:13)]2 by applying Theorems 2.2 and 2.3, then W AW A †(cid:13),W = W A(W A) †(cid:13) = (W A)k[(W A)k]†. Therefore, W AW A †(cid:13),W is a orthogonal projector onto R((W A)k). (2) Observe that W A †(cid:13),W W A = (W A) †(cid:13)W A. Since (W A) †(cid:13) is an outer inverse of (W A), then [(W A) †(cid:13)W A]2 = (W A) †(cid:13)W A, together with R((W A) †(cid:13)W A) = R((W A)k) and N ((W A) †(cid:13)W A) = N ([(W A)k]†W A), which implies that W A †(cid:13),W W A is a projector onto R((W A)k) along N ([(W A)k]†W A). 4 Relations among the weighted core-EP inverse and other generalized inverses In this section, we wish to reveal the relations among the weighted core-EP inverse, weighted Drazin inverse, the inverse along an element, and (B, C)-inverse. The first result states that the W -weighted core-EP inverse of A (i.e., A †(cid:13),W ) and the W -weighted Drazin inverse of A (i.e., AD,W ) can be mutually expressed by post-multiplying an oblique ( orthogonal ) projector. Theorem 4.1. Let A ∈ Cm×n, W ∈ Cn×m with ind(W A) = k. Then (1) A †(cid:13),W = AD,W P(W A)k ; (2) AD,W = A †(cid:13),W PR((W A)k),N ((W A)k). (1) It is known that A †(cid:13),W = A[(W A) †(cid:13)]2, (W A) †(cid:13) = (W A)D(W A)k[(W A)k]† and Proof. AD,W = A[(W A)D]2. Thus, A †(cid:13),W = A[(W A)D]2(W A)k[(W A)k]† = AD,W (W A)k[(W A)k]† = AD,W P(W A)k . (2) Observe that AD,W = A[(W A)D]2 = A[(W A)D(W A)k[(W A)k]†]2(W A)k[(W A)D]k = A[(W A) †(cid:13)]2W A(W A)D = A †(cid:13),W W A(W A)D = A †(cid:13),W PR((W A)k ),N ((W A)k). In what follows, we investigate the relations between the weighted core-EP inverse and the inverse along an element, (B, C)-inverse respectively. Let us recall two known notions. Definition 4.2. [13] Let A ∈ Cn×m and D, X ∈ Cm×n. Then X is the inverse of A along D if XAD = D = DAX and Rs(X) ⊆ Rs(D), R(X) ⊆ R(D). 10 Definition 4.3. [14] Let A ∈ Cn×m, B ∈ Cm×m, C ∈ Cn×n, X ∈ Cm×n. Then X is the (B, C)-inverse of A if X ∈ BCm×mX ∩ X Cn×nC and XAB = B, CAX = C. Theorem 4.4. Let A ∈ Cm×n, W ∈ Cn×m with ind(W A) = k. Then the W -weighted core-EP inverse of A ( i.e., A †(cid:13),W ) is the inverse of W AW along A(W A)k[(W A)k]∗. Proof. From Lemma 2.1 and Theorem 2.3, it is possible to verify that A †(cid:13),W W AW A(W A)k[(W A)k]∗ = A[(W A) †(cid:13)]2(W A)k+2[(W A)k]∗ = A(W A) †(cid:13)(W A)k+1[(W A)k]∗ = A(W A)k[(W A)k]∗, A(W A)k[(W A)k]∗W AW A †(cid:13),W = A(W A)k[(W A)k]∗W A(W A) †(cid:13) = A(W A)k[(W A)k]∗[W A(W A) †(cid:13)]∗ = A(W A)k[W A(W A) †(cid:13)(W A)k]∗ = A(W A)k[(W A)k]∗, A †(cid:13),W = A[(W A) †(cid:13)]2 = A(W A)k[(W A)k]†(W A)k[(W A) †(cid:13)]k+2 = A(W A)k[(W A)k]∗[(W A)k]†∗[(W A) †(cid:13)]k+2, i.e., R(A †(cid:13),W ) ⊆ R(A(W A)k[(W A)k]∗), as well as, A †(cid:13),W = A[(W A) †(cid:13)]2 = A[(W A)D]2(W A)k[(W A)k]† = A[(W A)D]2[(W A)k]†∗[(W A)k]∗ = A[(W A)D]2([(W A)k]†(W A)k[(W A)k]†)∗[(W A)k]∗ = A[(W A)D]2[(W A)k]†∗[(W A)k]†(W A)k[(W A)k]∗. Since [(W A)k]†(W A)k = [(W A)k+1]†(W A)k+1(see the dual form of Lemma 2.1), then A †(cid:13),W = A[(W A)D]2[(W A)k]†∗[(W A)k+1]†(W A)k+1[(W A)k]∗ = A[(W A)D]2[(W A)k]†∗[(W A)k+1]†W A(W A)k[(W A)k]∗, i.e., Rs(A †(cid:13),W ) ⊆ Rs(A(W A)k[(W A)k]∗). Hence A †(cid:13),W is the inverse of W AW along A(W A)k[(W A)k]∗, in view of Definition 4.2. Theorem 4.5. Let A ∈ Cm×n, W ∈ Cn×m with k = max{ind(AW ), ind(W A)}. Then the W -weighted core-EP inverse of A ( i.e., A †(cid:13),W ) is the ((AW )k, [(W A)k]∗)-inverse of W AW . 11 Proof. Clearly, we can verify that A †(cid:13),W = A[(W A) †(cid:13)]2 = A(W A)k[(W A) †(cid:13)]k+2 = (AW )kA[(W A) †(cid:13)]k+2 = (AW )kA[(W A) †(cid:13)]k+1W A †(cid:13),W ∈ (AW )kCm×mA †(cid:13),W , A †(cid:13),W = A[(W A) †(cid:13)]2 = A[(W A) †(cid:13)]2W A(W A) †(cid:13) = A[(W A) †(cid:13)]2(W A)k[W A)k]† = A[(W A) †(cid:13)]2[W A)k]†∗[(W A)k]∗ = A †(cid:13),W [W A)k]†∗[(W A)k]∗ ∈ A †(cid:13),W Cn×n[(W A)k]∗, as well as, A †(cid:13),W W AW (AW )k = A[(W A) †(cid:13)]2(W A)k+1W = A(W A) †(cid:13)(W A)kW = A(W A)D(W A)kW = (AW )D(AW )k+1 = (AW )k, [(W A)k]∗W AW A †(cid:13),W = [(W A)k]∗W A(W A) †(cid:13) = [(W A)k]∗[W A(W A) †(cid:13)]∗ = [W A(W A) †(cid:13)(W A)k]∗ = [(W A)k]∗. The above equalities show that A †(cid:13),W is the ((AW )k, [(W A)k]∗)-inverse of W AW , in light of Definition 4.3. 5 Computational complexities of representations Following from [2, 17], the computational complexity of the pseudoinverse of a singular m × n (resp. n×n) matrix is denoted by pinv(m, n) (resp. pinv(n)); the complexity of multiplying an m × n matrix by an n × k matrix is denoted by M (m, n, k), abbreviated to m · n · k; the notation M (n) is used instead of M (n, n, n) and is abbreviated to n3. Let A ∈ Cm×n, W ∈ Cn×m with k =max{ind(W A),ind(AW )} and let l be a non-negative integer such that l ≥ In general, an o(log l) algorithm for matrix exponentiation Al (see [19]) would give an k. algorithm for computing (AW )l in O(m3 log l) time, so that O((AW )l) = O(m3 log l) (see [2]). Similarly, O((W A)l) = O(n3 log l). Table 1: Computational complexity of (2.8) Expression AW Λ1 = (AW )l Λ2 = (AW )l+1 = Λ1(AW ) Λ3 = W (AW )l+1 = W Λ2 Λ4 = Λ† 3 X = (AW )l[W (AW )l+1]† = Λ1Λ4 Additional complexity m · n · m m3 log l m3 n · m · m pinv(n, m) m · m · n The computational complexity of (2.8) can be estimated from the analysis of Table 1: O(2.8) = 3m2n + m3 + m3 log l + pinv(n, m). 12 Table 2: Computational complexity of (2.9) Expression W A (W A)2 Λ1 = (W A)l Λ2 = (W A)l+2 = Λ1(W A)2 Λ3 = Λ† 2 X = A(W A)l[(W A)l+2]† = AΛ1Λ3 Additional complexity n · m · n n3 n3 log(l − 1) n3 pinv(n) 2 m · n · n Likewise, the estimation for the computational complexity of (2.9) comes from Table 2: O(2.9) = 3mn2 + 2n3 + n3 log(l − 1) + pinv(n). Obviously from O(2.8) and O(2.9), it is more appropriate to use representations involving AW while m < n, and use representations involving W A while m ≥ n. In the following, we consider the case: (0 <)m < n. Table 3: Computational complexity of (1.3) Expression AW Λ1 = (AW )l Λ2 = (AW )l+1 = Λ1(AW ) Λ3 = Λ† 1 Λ4 = W Λ2Λ3 X = [W (AW )l+1[(AW )l]†]† = Λ† 4 Additional complexity m · n · m m3 log l m3 pinv(m) 2 n · m · m pinv(n, m) The computational complexity of (1.3) is estimated from the analysis of Table 3: O(1.3) = 3m2n + m3 + m3 log l + pinv(m) + +pinv(n, m). In view of [2] and [20], the complexity pinv(m) ≥ M (m) = m3 > 0. From pinv(m) > 0, it follows that O(1.3) > O(2.8). Hence from this perspective, representation (2.8) is better than representation (1.3). 6 Numerical examples Our aim in this section is to test the time efficiency as well as the accuracy of given represen- tations involving only pseudoinverse, namely, Equalities (1.3) and (2.8). For which, randomly generated singular matrices of different sizes are employed. Time efficiency is evaluated by the CPU time and the accuracy is measured by the residual norms. All the numerical tasks have been performed by using Matlab R2017b. 13 Table 4: Comparison of representations (1.3) and (2.8). Entries of A, W are uniformly distributed random numbers from 0 to 1 Equation (1.3) Size m, n l ≥ k CPU Time r1 r2 r3 0.0300 8.7292e+10 1.6417e-25 3.4789e-16 100, 200 l = k = 4 100, 200 l = k + 5 100, 200 l = k + 15 100, 200 l = k + 25 500, 1000 l = k = 3 500, 1000 l = k + 5 500, 1000 l = k + 15 500, 1000 l = k + 25 (2.8) (1.3) (2.8) (1.3) (2.8) (1.3) (2.8) (1.3) (2.8) (1.3) (2.8) (1.3) (2.8) (1.3) (2.8) 0.0200 0.0300 0.0200 0.0400 0.0300 0.0300 0.0300 0.2600 0.2400 0.2600 0.2400 0.3200 0.2400 0.4700 4.7142e+10 5.7009e+10 8.9982e-26 1.0516e-25 3.7588e-16 2.1932e-16 1.7365e+10 4.4537e+10 3.2428e-26 8.1622e-26 6.6545e-16 2.1898e-16 2.5859e+10 6.1824e+10 4.6722e-26 1.1411e-25 2.3790e-16 2.1261e-16 1.8199e+10 1.2955e+12 3.5081e-26 1.4910e-29 2.6287e-16 4.6126e-16 5.5178e+11 1.2955e+12 1.9805e-30 1.4910e-29 5.0956e-16 4.6126e-16 5.5178e+11 1.2142e+12 1.9805e-30 1.3975e-29 5.0956e-16 4.8350e-16 5.5196e+11 7.9785e+11 3.4573e-30 9.1222e-30 5.7581e-16 8.3847e-16 0.2700 5.5190e+11 1.1077e-30 5.8500e-16 14 Let A ∈ Cm×n and W ∈ Cn×m with ind(AW ) = k. We assume that m < n. Approxima- tion derived from a numerical method for computing A †(cid:13),W will be denoted by X, and the residual norms in all numerical experiments are denoted by r1 = XW (AW )k+1−(AW )k2, r2 = AW XW X −X2 and r3 = (W AW X)∗−W AW X2. From Table 4, the following overall conclusions can be emphasized: (1) The representation (2.8) gives a better result in the aspect of the computational speed. (2) Representation (2.8) is better in accuracy with respect to the residual norms r1 and r2. (3) Contrary to the previous conclusion, the representation (1.3) is a better expression in accuracy with respect to norm r3. (4) Both (1.3) and (2.8) produce bad results with respect to the norm r1. This reason is the numerical instability caused by various matrix powers. 7 Conclusion This paper introduces several computational representations for the W -weighted core-EP inverse by using three different matrix decompositions: • singular-value decomposition; • full-rank decomposition; • QR decomposition. Based on these representations, some properties of the weighted core-EP inverse are derived. Complexity of introduced representations are estimated and numerical examples are presented. In addition, the weighted core-EP inverse is considered as a particular (B, C)-inverse, and a particular generalized inverse A(2) T,S. Acknowledgements The authors are highly grateful to the responsible editor and the anonymous referees for their valuable and helpful comments and suggestions. The first author is grateful to China Schol- arship Council for supporting her further study in University of Minho, Portugal. Funding This research is supported by the National Natural Science Foundation of China (No.11771076), the Scientific Innovation Research of College Graduates in Jiangsu Province (No.KYZZ16 0112), Partially supported by FCT- 'Funda¸cao para a Ciencia e a Tecnologia', within the project UID-MAT-00013/2013. 15 References [1] Cline RE, Greville TNE. A Drazin inverse for rectangular matrices. Linear Algebra Appl. 1980; 29: 53-62. [2] Stanimirovi´c PS, Katsikis VN, Ma H. Representations and properties of the W -weighted Drazin inverse. Linear Multilinear Algebra. 2017; 65(6): 1080-1096. [3] Wei Y. Integral representation of the W -weighted Drazin inverse. Appl. Math. Comput. 2003; 144: 3-10. [4] Zhour ZAAA, Kili¸cman A, Hassa MHA. New representations for weighted Drazin inverse of matrices. Int. J. Math. Anal. 2007; 1(15): 697-708. [5] Baksalary OM, Trenkler G. Core inverse of matrices. Linear Multilinear Algebra. 2010; 58: 681-697. [6] Manjunatha Prasad K, Mohana KS. Core-EP inverse. Linear Multilinear Algebra. 2014; 62: 792-802. [7] Gao Y, Chen J. Pseudo core inverses in rings with involution. Comm. Algebra. 2018; 46(1): 38-50. [8] Gao Y, Chen J, Ke Y. *-DMP elements in ∗-semigroups and ∗-rings. Filomat. Accept. [9] Mosi´c D. Core-EP pre-order of Hilbert space operators. Quaest. Math. http://dx.doi.org/10.2989/16073606.2017.1393021. [10] Wang H. Core-EP decomposition and its applications. Linear Algebra Appl. 2016; 508: 289-300. [11] Ferreyra DE, Levis FE, Thome N. Revisiting the core EP inverse and its extension to rectangular matrices. Quaest. Math. 2018; 41(2): 265-281. [12] Mosi´c D. Weighted core-EP inverse of an operator between Hilbert spaces. Linear Mul- tilinear Algebra. https://doi.org/10.1080/03081087.2017.1418824. [13] Mary X. On generalized inverses and Greens relations. Linear Algebra Appl. 2011; 434: 1836-1844. [14] Drazin MP. A class of outer generalized inverses. Linear Algebra Appl. 2012; 436: 1909- 1923. [15] Ben-Israel A, Greville TNE. Generalized Inverses: Theory and Applications. New York: Wiley-Interscience; 1974. [16] Sheng X, Chen G. Full-rank representation of generalized inverse A(2) T,S and its applica- tion. Comput. Math. Appl. 2007; 54: 1422-1430. 16 [17] Stanimirovi´c PS, Pappas D, Katsikis VN, Stanimirovi´c IP. Full-rank representations of outer inverses based on the QR decomposition. Appl. Math. Comput. 2012; 218: 10321- 10333. [18] Watkins D. Fundamentals of Matrix Computations. New York: Wiley-Interscience; 2002. [19] Cormen TH, Leiserson CE, Rivest RL. Introduction to Algorithms. Cambridge: The MIT Press; 2001. [20] Petkovi´c MD, Stanimirovi´c PS. Generalized matrix inversion is not harder than matrix multiplication. J. Comput. Appl. Math. 2009; 230: 270-282. 17
1503.04666	1	1503	2015-03-16T14:26:52	The isomorphism problem for graded algebras and its application to mod-p cohomology rings of small p-groups	[ "math.RA" ]	The mod-p cohomology ring of a non-trivial finite p-group is an infinite dimensional, finitely presented graded unital algebra over the field with p elements, with generators in positive degrees. We describe an effective algorithm to test if two such algebras are graded isomorphic. As application, we determine all graded isomorphisms between the mod-p cohomology rings of all p-groups of order at most 100.	math.RA	math	The isomorphism problem for graded algebras and its application to mod-p cohomology rings of small p-groups Bettina Eick and Simon King July 26, 2018 Abstract The mod-p cohomology ring of a non-trivial finite p-group is an infinite dimensional, finitely presented graded unital algebra over the field with p elements, with generators in positive degrees. We describe an effective algorithm to test if two such algebras are graded isomorphic. As application, we determine all graded isomorphisms between the mod-p cohomology rings of all p-groups of order at most 100. 1 Introduction The mod-p cohomology ring H ∗(G, F) of a non-trivial finite p-group G and the field F with p elements is an infinite dimensional graded F-algebra. It is an interesting and wide open question how good this algebra is as an isomorphism invariant for the underlying group G. More precisely: given two non-isomorphic p-groups G and H, under which circumstances are H ∗(G, F) and H ∗(H, F) isomorphic as graded F-algebras? Our aims in this paper are two-fold. First, we consider finitely presented graded unital F-algebras with generators in positive degrees over a finite field F; we call such algebras 'finitary'. We describe an effective algorithm to test if two finitary algebras are graded isomorphic. We also consider the special case of graded commutative finitary algebras and describe an improved algorithm for this case. Secondly, we apply our algorithm to the mod-p cohomology rings of the p-groups with order at most 100. The p-groups of order at most 100 are well-known, see [2] for a history on their classification. Finite presentations for their mod-p cohomology rings are also available, see [4] and also [6] for a recent account. The following theorem exhibits a brief summary of our results. A complete list of the groups with graded isomorphic cohomology ring is included in Section 7 below. 1 Theorem: The following tables list numbers of isomorphism types of groups of order pn and numbers of graded isomorphism types of the associated mod-p cohomology rings. order # groups # rings 21 1 1 22 2 2 23 5 5 24 14 14 25 51 48 26 267 239 31 1 1 32 2 2 33 5 5 34 15 15 1 There are significantly more graded isomorphisms between groups of different orders than between groups of a fixed order. In the following table we list numbers of isomorphism types of groups of order dividing pn and numbers of graded isomorphism types of the associated mod-p cohomology rings. order # groups # rings 21 1 1 22 3 3 23 8 7 24 22 18 25 73 55 26 340 260 31 1 1 32 3 2 33 8 5 34 23 14 The phenomena that there are several graded isomorphisms between mod-p cohomology rings for groups of different orders is known in various examples in the literature. A well- known example is given by the infinite families of cyclic groups with graded isomorphic cohomology rings. Further, there are infinite families of metacyclic groups with graded isomorphic mod-p cohomology rings, see [10]. Moreover, the result in [3] implies that there are infinite families of 2-groups of fixed coclass with graded isomorphic mod-2 cohomology rings. 2 Preliminaries In this preliminary section we recall some basic facts from the theory of graded algebras and we establish our notation. Let N = {0, 1, 2, . . .} and let F be a field. First, recall that an F-algebra A is graded if it can be written as a direct sum of F-vectorspaces A = M n∈N An and AiAj ⊆ Ai+j holds for each i, j ∈ N. The vectorspaces A0, A1, . . . are the graded components of A. An element a ∈ A is homogeneous, if it is contained in a graded component An for some n ∈ N; in this case we denote its degree by a = n. Let F be a free graded unital F-algebra, let ϕ : F → A be a surjective morphism of graded F-algebras, let A be a free generating set of F and let R be a generating set of ker(ϕ). Then hA Ri is called a graded presentation of A. The presentation is finite if both A and R are finite, and A is called finitely presented, if a finite presentation is given. Note that by slight abuse of notation we identify A with {ϕ(x) x ∈ A} and say that hA Ri is a presentation on the generating set A of A. Definition. Let F be a finite field and let A be a finitely presented graded unital F-algebra with generators in positive degrees. Then A is called a finitary F-algebra. 2 Lemma: Let A be a finitary F-algebra with graded components A0, A1 . . .. (1) A has a finite generating set consisting of homogeneous elements A = (a1, . . . , am). (2) A has a finite presentation on the homogeneous generating set A. (3) Let n ∈ N. The set Mn(A) = {ai1 · · · aij ai1 + . . . + aij = n} generates An as vectorspace and thus An is finite dimensional. 2 Proof: (1) Each element of A can be written as a finite sum of homogenous elements. Thus each arbitrary finite generating set of A gives rise to a finite homogeneous generating set by decomposing each generator into homogeneous summands. (2) Let A = hb1, . . . , bk R1, . . . , Rli be an arbitrary finite presentation for A and let a1, . . . , am be an arbitrary finite generating set for A. Then each bi can be written as a word in a1, . . . , am, say bi = wi(a1, . . . , am). Similarly, each aj can be written It now follows that A ∼= ha1, . . . , am as word in b1, . . . , bk, say aj = vj(b1, . . . , bk). Ri(w1, . . . , wk) for 1 ≤ i ≤ l and aj = vj(w1, . . . , wk) for 1 ≤ j ≤ mi. (3) Is elementary. • Let A be a graded F-algebra with graded components A0, A1, . . .. We define I(A) = M n≥1 An and Ij(A) = M n≥j An for j ≥ 1. Then I(A) is called the augmentation ideal of A; it is a non-unital F-algebra having the series of ideals I(A) = I1(A) ≥ I2(A) ≥ . . .. Note that A = I(A) ⋊ A0. Let I(A) ≥ I(A)2 ≥ . . . denote the series of power ideals in I(A). Then I(A)/I(A)c is a nilpotent F-algebra of class c − 1 for each c ≥ 1 by construction. 3 Lemma: Let A be a finitary F-algebra. (1) Let c ∈ N. Then I(A)c ≤ Ic(A). (2) I(A) is finitely generated and residually nilpotent. (3) Let c ∈ N. Then I(A)/I(A)c is finite dimensional. (4) There exists d ∈ N with Id+1(A) ≤ I(A)2. Proof: (1) We use induction on c. For c = 1 we note that I(A)1 = I(A) ≤ I(A) = I1(A). If I(A)c ≤ Ic(A), then I(A)c+1 = I(A)I(A)c ≤ I(A)Ic(A) ≤ Ic+1(A). (2) Let a1, . . . , am be a set of homogeneous generators in positive degrees for the unital algebra A. Then I(A) is generated by a1, . . . , am as non-unital algebra. Thus I(A) is finitely generated. Further, ∩c≥1I(A)c = {0} by (1) and thus I(A) is residually nilpotent. (3) A nilpotent quotient of a finitely generated algebra is finite dimensional. (4) By (2) the algebra I(A) is finitely generated and thus it has a finite generating set A of homogeneous elements. Let d be the maximal degree of a generator and let l > d. Then each monomial in Ml(A) is a product of at least 2 elements by the definition of d. Hence Ml(A) ⊆ I(A)2. Thus Id+1(A) = hMl(A) l > di ⊆ I(A)2. • 3 Computation with finitary algebras In this section we describe some elementary algorithms for finitary F-algebras. We assume that a finitary F-algebra A is given by a finite presentation ha1, . . . , am R1, . . . , Rli on 3 homogeneous generators A = (a1, . . . , am) with positive degrees. We denote the graded components of A by A0, A1, . . . and we assume that the Hilbert -- Poincar´e series PA(t) = Pn∈N dim(An)tn is given as rational function. It is well-known that computations with finitely presented algebraic objects is difficult in general. For example, in the case of finitely presented groups it is in general not algorithmically possible to solve the word problem (let alone the isomorphism problem). In this section we show how this and related problems can be solved in our considered case. For n ≥ 1 let ǫn : I(A) → I(A)/I(A)n : a 7→ a + I(A)n the natural epimorphism on the class-n − 1 nilpotent quotient of I(A). Then the image of ǫn is finite dimensional by Lemma 3. 4 Remark: Let n ∈ N. Then a basis and its structure constants table for I(A)/I(A)n = Im(ǫn) can be computed with the methods of [5] together with the images of a1, . . . , am in the finite-dimensional image. This computation requires an arbitrary finite presentation for I(A). Note that the given presentation ha1, . . . , am R1, . . . , Rli for A defines I(A) as non-unital algebra and hence a finite presentation for I(A) is given by our setup. 3.1 The word problem Suppose that a word w in the generators A is given; that is, w = c + Pn i=1 cili with ci ∈ F and li ∈ Mi(A). Our aim is to decide if w = 0 in A. The following lemma translates this to an easy calculation in the finite dimensional quotient I(A)/I(A)n+1. 5 Lemma: w = 0 in A if and only if c0 = 0 and ǫn+1(Pn i=1 cili) = 0. Proof: This follows from Lemma 3 (1). • 3.2 Bases for the graded components Let n ≥ 1 and recall that Mn(A) generates the graded component An of A. The following lemma shows how to reduce this generating set to a basis via a computation in the finite dimensional quotient I(A)/I(A)n+1. 6 Lemma: Bn is a basis for An if and only if ǫn+1(Bn) is a basis for hǫn+1(Mn(A))i. Proof: This follows from Lemma 3 (1). • We note that the dimensions of the graded components can be read off readily from the Hilbert -- Poincar´e series PA(t). Define P (0) (t) − dim(An−1))/t for n > 0. Then dim(An) = P (n) A (t) := PA(t) and P (n) A (0) holds. A (t) := (P (n−1) A 4 3.3 Detecting generating sets Suppose that elements b1, . . . , bk of I(A) are given. Our aim is to decide if these ele- ments generate A as unital algebra. The following lemma reduces this to an elementary computation in the finite dimensional quotient I(A)/I(A)2. 7 Lemma: b1, . . . , bk generate A (as unital algebra) if and only if hǫ2(b1), . . . , ǫ2(bk)i = I(A)/I(A)2. Proof: This follows from Lemma 3 (2). • 4 Graded isomorphisms between finitary algebras In this section we exhibit our solution to the graded isomorphism problem for finitary algebras. Recall that two graded F-algebras A and B are graded isomorphic if there exists an F-algebra isomorphism ν : A → B that respects the grading, that is, it satisfies ν(An) = Bn for each n ∈ N. We write A ∼= B if A is isomorphic to B as F-algebra and A ∼=g B if A is graded isomorphic to B. For our algorithm we assume that both finitary algebras A and B are given by finite presentations on homogeneous generators of positive degrees and we assume that their Hilbert -- Poincar´e series PA and PB are available as well. We denote the graded components of A and B by An and Bn, respectively. 8 Lemma: Let A and B be two finitary F-algebras. If there exists a graded isomorphism ϕ : A → B, then (a) PA = PB, and (b) If A is a finite homogenous generating set for A, then ϕ(a) ∈ Ba for each a ∈ A. Proof: (a) and (b) both follow from the fact that ϕ(An) = Bn for each n ∈ N, where An and Bn denotes the vectorspaces in the gradings of A and B, respectively. • Lemma 8 (b) shows that there are only finitely many possible options for graded isomor- phisms A → B, since a finite homogenous generating set for A is given and Bn is finite for each n ∈ N. 9 Theorem: Let A and B be two finitary F-algebras and suppose that PA = PB. Let A = ha1, . . . , am R1, . . . , Rli a finite homogenous presentation for A on generators of positive degree and let b1, . . . , bm ∈ B with bi ∈ Bai for 1 ≤ i ≤ m. The map ϕ : A → B : ai 7→ bi extends to a graded isomorphism if and only if (a) Rj(b1, . . . , bm) = 0 for 1 ≤ j ≤ l, and (b) b1, . . . , bm generate B. 5 Proof: First suppose that ϕ extends to a graded isomorphism. Then 0 = ϕ(0) = ϕ(Rj(a1, . . . , am)) = Rj(ϕ(a1), . . . , ϕ(an)) = Rj(b1, . . . , bn) and thus (a) holds. (b) is obvious. Now suppose that (a) and (b) hold. Then (a) yields that ϕ is an algebra homomorphism. As bi ∈ Bai for 1 ≤ i ≤ m, it follows that ϕ respects the grading and ϕ(An) ⊆ Bn for n ∈ N. (b) asserts that ϕ is surjective. Hence ϕ(An) = Bn for each n ∈ N. Finally, as PA = PB, we obtain that ϕ is also injective and hence a graded isomorphism. • Lemma 8 and Theorem 9 induce the following method to determine a graded isomorphism A → B if it exists. Let A = ha1, . . . , am R1, . . . , Rli be a finite homogenous presentation on generators of positive degrees and let di = ai for 1 ≤ i ≤ m. GradedIsomorphism(A, B) (1) Test if PA = PB; if not, then return false. (2) Determine bases for Bd1, . . . , Bdm . (3) For each (b1, . . . , bm) ∈ Bd1 × . . . × Bdm do (a) Check that Rj(b1, . . . , bm) = 0 for 1 ≤ j ≤ l. (b) Check that b1, . . . , bm generate B. (c) If (a) and (b) are satisfied, then return (b1, . . . , bm). (4) Return false; Note that bases for Bd1, . . . , Bdm can be determined as in Section 3.2. Each of these spaces is finite and thus the for-loop in Step (3) is a finite loop. Step (3a) can be implemented by the method in Section 3.1. Step (3b) can be performed as in Section 3.3. 10 Remark: Let w1 = max{di 1 ≤ i ≤ m} and let w2 denote the maximal degree of a monomial in R1, . . . , Rl. Further, let w = max{1, w1, w2}. Then the algorithm GradedIsomorphism requires the computation of ǫw+1. If I(A)/I(A)w+1 and I(B)/I(B)w+1 are both available, then this allows further reductions in the algorithm GradedIsomorphism. For example, if A and B are graded isomorphic, then dim(I(A)/I(A)c) = dim(I(B)/I(B)c) for each c ≥ 1 and this induces an additional condition that may be checked in Step (1) of the algorithm for all available nilpotent quotients. Further, if ai ∈ I(A)ci for some ci ∈ N, then ϕ(bi) ∈ I(B)ci ∩ Bdi. This can be used to obtain a reduction in Step (3) of the algorithm. 5 The graded commutative case A graded F-algebra A is called graded commutative, if for all homogeneous elements x, y ∈ A the equation x · y = (−1)x·yy · x holds. If char(F) = 2, then a graded commutative algebra is commutative and the free graded commutative F-algebra on m generators is isomorphic to the polynomial ring on m generators. If char(F) > 2, then a free graded commutative F-algebra is isomorphic to a tensor product of a polynomial ring and an exterior algebra. 6 We present graded commutative F-algebras not as quotients of free graded unital F- algebras (as in Section 2), but as quotients of free graded commutative F -algebras. Hence, if F is a free graded commutative F-algebra, and ϕ : F → A is a surjective morphism of graded F-algebras, and A is a free generating set of F and R is a generating set of K = ker(ϕ), then we call hA Ri a graded commutative presentation of A. Note that one can choose R so that its elements are homogeneous. Finitely presented graded commutative algebras are noetherian and are either commu- tative or are non-commutative G-algebras, for which a Grobner basis theory is available much similar to the commutative case [9, Chapter 1.9]. With Grobner bases, one has an alternative way to solve computational problems than by using nilpotent quotients as in Sections 3 and 4. That approach can be more effective; in particular, this is the case if the parameter w as determined in Remark 10 is large. Let ha1, . . . , am R1, . . . , Rli be a finite graded commutative presentation of A corre- sponding to ϕ : F → A with K = ker(ϕ), as above. We consider a Grobner basis B = (B1, . . . , Bk) for K. • The word problem in A can be solved by polynomial reduction with respect to B, and a basis of An is given by those elements of Mn(A) that are not divisible by any of the leading monomials of B1, ..., Bk. • By [11, Proposition 3.6.6 d)], the computation of Grobner bases also allows for an effective test whether a subset of A forms a generating set of A. • The Hilbert -- Poincar´e series PA(t) can be computed as in [9, Chapter 5.2] or [11, Chapter 5], and is a rational function. More generally, if I ≤ A is an ideal generated by homogeneous elements, then the quotient ring A/I is finitary graded commutative, and its Hilbert -- Poincar´e series PA/I (t) (to which we also refer to as the "Hilbert -- Poincar´e series of I") can be computed, too. • The nilradical of A can in principle be computed as in [11, Chapter 4.5]. There is a more efficient alternative approach is available for cohomology rings. If G is a finite group, then the nilradical of H ∗(G, Fp) is formed by the elements that have nilpo- tent restriction to all the maximal p-elementary abelian subgroups of G, by a result of Quillen (see also [3, Theorem 8.4.3]). Based on this, the nilradicals of modular cohomology rings of finite groups can be computed by intersecting the preimages of certain explicitly given ideals under morphisms (namely restrictions) of finitely pre- sented graded commutative F-algebras. The preimages can be computed as in [9, Section 1.8.10, Remark 1.8.17], and their intersection as in [9, Section 1.7.7]. • If I ≤ A is an ideal generated by homogeneous elements, then its annihilator Ann(I) = {x ∈ A ∀y ∈ I : y · x = 0} ≤ A can be computed [9, Section 2.8.4]. When we test in Step (3)(b) whether elements b1, . . . , bn ∈ B generate B according to [11, Proposition 3.6.6 d)], then the computation of a Grobner basis in elimination order for an ideal defined in terms of (b1, . . . , bn) is needed. This is potentially a very expensive operation. It is thus crucial to reduce the possible choices of (b1, . . . , bn) in Step (3) by other methods, as described in the rest of this section. Here, elimination is used as well, but it turns out that this is feasible and reduces the computation time drastically. 7 5.1 Early detection of non-isomorphic algebras Comparing PA(t) and PB(t) as in Step (1) of Algorithm GradedIsomorphism allows to disprove the existence of a graded isomorphism between A and B in many cases. In addition to that, we compute the nilradicals nilrad(A) and nilrad(B) of A and B, and test if PA/ nilrad(A)(t) = PB/ nilrad(B)(t). This may detect that A 6∼=g B even in cases where PA(t) = PB(t). 5.2 Reducing the list of potential generator images We now focus on possible reductions of the images (b1, . . . , bn) of (a1, . . . , an) to be con- sidered in Step (3) of Algorithm GradedIsomorphism. Let (AA, RA) and (AB, RB) be finite graded commutative presentations of A and B. Let AA = (ai1, . . . , aik ) be a subset of AA, and let bi1, ..., bik ∈ B. Let I = h AAi ≤ A be the ideal generated by AA, and J = hbi1, ..., bik i ≤ B. We discuss here three tests that often allow to conclude that there is no graded homomorphism mapping aij to bij for j = 1, ..., k. Firstly, if ϕ : A → B is a graded isomorphism, then PA/I (t) = PB/ϕ(I)(t). Hence, if the ideals I ≤ A and J ≤ B have different Hilbert -- Poincar´e series, then the map aij 7→ bij can not be extended to a graded isomorphism. Secondly, by elimination of the variables AA\ AA from the relation ideal RA as in [9, Section 1.8.2], one obtains relations RA,1, ..., RA,l that only involve elements of AA. If ϕ : A → B is a graded homomorphism, then RA,c (ϕ(ai1 ), ..., ϕ(aik )) = 0 for all c = 1, ..., l, which can be effectively tested using a Grobner basis of hRBi. Hence, if RA,c(bi1, ..., bik ) 6= 0 for some c = 1, ..., l, then the map aij 7→ bij can not be extended to a graded homomorphism. Thirdly, one can compute the annihilators Ann (I) ≤ A and Ann (J) ≤ B. If they have different Hilbert -- Poincar´e series, then the map aij 7→ bij can not be extended to a graded isomorphism. In principle, it would also be possible to compare the radicals of the two ideals, but we found that this does not help to improve efficiency. How do the above tests to help simplify in Step (3) of Algorithm GradedIsomorphism? It suffices to restrict Step (3) to those tuples (b1, . . . , bn) ∈ Bd1 × ... × Bdn that pass the above three tests for all subsets of (a1, . . . , an). In a practical implementation, one would start by using the three tests on one-element subsets, i.e., one would compute all possible elements of Bdi that may occur as images of ai under any graded algebra isomorphism, for i = 1, ..., n. This will normally leave very few possibilities, say, Bdi ⊆ Bdi. Next, one would consider all possible pairs (bi, bj) ∈ Bdi × Bdj , and use the three tests to determine all possible images of (ai, aj) under any graded isomorphism, for i, j = 1, ..., n. And so on, with larger subsets of AA. If A 6∼=g B, the three tests will often leave no or only very few possible choices for (b1, ..., bn) in Step (3) of Algorithm GradedIsomorphism that need to be tested in Step (3)(b). And if A ∼=g B, then often the first possible choice of (b1, ..., bn) will turn out to yield a graded isomorphism by the final test in Step (3)(b). 11 Remark: One should be aware that computing the nilradicals of A and B, a graded commutative presentation of hh Aii, or annihilators, can generally be computationally very 8 expensive. However, in all the examples that we considered, the gain of using the additional reductions in Algorithm GradedIsomorphism outweighs these additional costs by far. 6 Examples We summarise here some examples of cohomology rings of finite groups. A minimal graded commutative presentation for each ring has been computed with the optional pGroupCohomology package [7] for Sage [12]. The package uses Singular [8] for the com- putation of Grobner bases, annihilators and elimination in graded commutative rings. Nilradicals are computed as described in Section 7. 6.1 Early detection of non-isomorphy Let A be the mod-3 cohomology ring of the extraspecial 3-group of order 27 and exponent 3, which is group number 3 of order 27 in the small groups library [1]. Let B be the mod-3 cohomology ring of the Sylow 3-subgroup of U3(8), which is group number 9 of order 81 in the small groups library. Each of these algebras has a minimal graded commutative presentations with generators in degrees 1, 1, 2, 2, 2, 2, 3, 3, and 6. The Hilbert-Poincar´e series of the two algebras, respectively of their nilradicals, coincide. The power series expansion of the Hilbert-Poincar´e series is PA(t) = PB(t) = 1 + 2t + 4t2 + 6t3 + 7t4 + 8t5 + 9t6 + 10t7 + 12t8 + · · · Thus, B1, B2, B3 and B6 contain 32 − 1 = 8, 34 − 1 = 80, 36 − 1 = 728 and 19682 non-zero elements, respectively. Hence, without the reductions from Section 5.2, one would need to consider 82 · 804 · 7282 · 19682 > 1019 possible images for the generators of A. However, it turns out that there is one degree-2 generator a ∈ A so that PA/hai(t) is different from PB/hbi(t), for each of the 80 non-zero elements of B2. Hence we can readily detect that A and B are not graded isomorphic. 6.2 A more difficult to detect pair of non-isomorphic algebras Let A be the mod-2 cohomology ring of group number 27 of order 32 in the small groups library, and let B be the mod-2 cohomology ring of group number 128 of order 64. They both have minimal graded commutative presentations formed by three generators in degree 1 and three generators in degree 2, and four relations. The Hilbert-Poincar´e series of the algebras, respectively of their nilradicals, coincide. The power series expansion of the Hilbert-Poincar´e series is PA(t) = 1 + 3t + 7t2 + 13t3 + 22t4 + 34t5 + 50t6 + · · · Thus, B1 contains 23 − 1 = 7 and B2 contains 27 − 1 = 127 non-zero elements. Hence, without the reductions from Section 5.2, one would need to consider 73 · 1273 > 7 · 108 possible images for the generators of A. In contrast to the previous example, the methods from Section 5.2 applied to one-element subsets of the generating set of A are not strong enough to prove A 6∼=g B. However, when 9 applied to the triple of degree-1 generators, the tests only leave 6 candidates for the images of the triple under isomorphism. Applied to the three degree-1 and two of the degree-2 generators, still as many as 4608 different isomorphic images seem possible. And thus one needs to combine each of them with the 111 potential isomorphic images of the remaining degree-2 generator. In all but 176 cases, the mapping of generators does not extend to a homomorphism, and in the remaining 176 cases the homomorphism is not surjective. Hence, the two algebras are not graded isomorphic. 7 Application to cohomology rings Let p be a prime, let G be a finite p-group and let F be the field with p elements. Then the mod-p cohomology ring H ∗(G, F) is a graded F-algebra defined by H ∗(G, F) = M n∈N H n(G, F). By the theorem of Evens -- Venkov (see also [3, Theorem 6.5.1]), modular cohomology rings of finite groups are finitely presentable graded-commutative algebras. Each graded com- ponent H n(G, F) is a finite dimensional vectorspace over F and H 0(G, F) ∼= F. Hence H ∗(G, F) is a finitary F-algebra. The methods of [6] determine a minimal presentation of H ∗(G, F) and these allow to apply the methods described in the first part of this paper. In the following sections we exhibit the graded isomorphisms among the mod-p cohomology rings of the p-groups of order at most 100. As a preliminary step we observe that the rank of the underlying p-group is an isomorphism invariant for the cohomology ring. Recall that the rank of a finite p-group G is the rank of the finite elementary abelian quotient G/[G, G]Gp = G/φ(G) of G or, equivalently, the minimal generator number of G. In the following we consider the p-groups of order at most 100 by their generator number. The groups with 1 generator are the cyclic groups; it is well-known that the cyclic of order pn have graded isomorphic mod-p-cohomology rings (with the exception of the cyclic group of order 2). We thus omit this case in our list below. It then remains to consider the groups of order dividing 26, 34, 52 and 72. The cases 52 and 72 are again well-understood and hence we focus on 26 and 34 in the following exposition. 7.1 2-groups We give here a complete and irredundant list of all graded isomorphic mod-2 cohomology rings H ∗(G, F) for the groups G of order dividing 64. We identify a group G by its id [order, number] in the SmallGroups library, see [1]. Each of the following lists of groups statisfies that the mod-2 cohomology rings of the considered groups are pairwise graded isomorphic, and mod-2 cohomology rings of groups from different lists are not graded isomorphic. If a group of order dividing 64 does not appear in any of the lists, then the graded isomorphism type of its mod-2 cohomology ring is unique among all groups of order dividing 64. Additionally to the ids of the groups in the list, we include the rank of the groups and in many cases also a structure description. For the latter, we denote with Ck, Dk, Qk, SDk 10 the cyclic, dihedral, quaternion and semidihedral groups of order k, respectively. The symbols ×, : and . describe a direct product, a split extension and an arbitrary extension, respectively. If one of the groups in one of the following lists is metacyclic, then all groups are metacyclic and we include this information as well. We note that our result differ in one case from the theoretical description of the mod-p cohomology rings of metacyclic groups in [10]: our results imply that the mod-2 cohomology rings of the metacyclic groups [32, 15] and [64, 49] are graded isomorphic to each other, but they are not graded isomorphic to [64, 45] as Theorem E(2) of [10] suggests. This is based on the fact that the presentations of the mod-p cohomology rings of [32, 15] and [64, 49] as given in [6] and also in [4, Appendix] are not compatible with that in [10]; for example, the presentations obtained in [6] and in [4, Appendix] imply that the underlying cohomology rings have a non-nilpotent element in degree 3 in contradiction to Theorem E(2) of [10]. • groups [4, 1], [8, 1], [16, 1], [32, 1], [64, 1] rank 1 and cyclic • groups [8, 2], [16, 5], [32, 16], [64, 50] rank 2, metacyclic, and structure C2n × C2 (n > 1) • groups [8, 3], [16, 7], [32, 18], [64, 52] rank 2, metacyclic, and structure D2n (n > 2) • groups [16, 2], [32, 3], [32, 4], [64, 2], [64, 3], [64, 26], [64, 27] rank 2, metacyclic, and structure C2n : C2m (n, m > 1) • groups [16, 3], [32, 9], [64, 38] rank 2, metacyclic, and structure (C2n × C2) : C2 (n > 1) • groups [16, 4], [32, 12], [32, 13], [32, 14], [64, 15], [64, 16], [64, 44], [64, 47], [64, 48] rank 2, metacyclic, and structure C2m : C2n (m, n > 1) • groups [16, 6], [32, 17], [64, 51] rank 2, metacyclic, and structure C2n : C2 (n > 2) • groups [16, 8], [32, 19], [64, 53] rank 2, metacyclic, and structure SD2n (n > 3) • groups [16, 9], [32, 20], [64, 54] rank 2, metacyclic, and structure Q2n (n > 3) • groups [16, 10], [32, 36], [64, 183] rank 3, and structure C2n × C2 × C2 (n > 1) • groups [16, 11], [32, 39], [64, 186] rank 3, and structure C2 × D2n (n > 2) • groups [32, 2], [64, 17], [64, 21] rank 2, and structure (C2n × C2) : C4 (n > 1) • groups [32, 5], [64, 6], [64, 29] rank 2, and structure (C2n × C2m) : C2 (n > m) • groups [32, 10], [64, 39] rank 2, and structure Q2n : C4 (n > 2) • groups [32, 15], [64, 49] 11 rank 2, metacyclic, and structure C4.D2n (n > 2) • groups [32, 21], [64, 83], [64, 84] rank 3, and structure (C2n : C4) × C2 (n > 1) • groups [32, 22], [64, 95] rank 3, and structure ((C2n : C2) : C2) × C2 (n > 1) • groups [32, 23], [64, 103], [64, 106], [64, 107] rank 3, and structure (C2m : C2n) × C2 (m, n > 1) • groups [32, 25], [64, 115], [64, 118], [64, 123] rank 3 • groups [32, 26], [64, 126] rank 3, and structure C2n : Q8 (n > 1) • groups [32, 28], [64, 140], [64, 147] rank 3 • groups [32, 29], [64, 155], [64, 157] rank 3 • groups [32, 31], [64, 167] rank 3, and structure (C2n × C4) : C2 (n > 1) • groups [32, 34], [64, 174] rank 3, and structure (C2n × C4) : C2 (n > 1) • groups [32, 35], [64, 181] rank 3, and structure C2n : Q8 (n > 1) • groups [32, 37], [64, 184] rank 3, and structure (C2n : C2) × C2 (n > 2) • groups [32, 38], [64, 185] rank 3, and structure (C2n × C2) : C2 (n > 2) • groups [32, 40], [64, 187] rank 3, and structure C2 × SD2n (n > 3) • groups [32, 41], [64, 188] rank 3, and structure C2 × Q2n (n > 3) • groups [32, 42], [64, 189] rank 3, and structure (C2n : C2) × C2 (n > 2) • groups [32, 43], [64, 190] rank 3, and structure (C2 × D2n) : C2 (n > 2) • groups [32, 44], [64, 191] rank 3, and structure (C2 × Q2n) : C2 (n > 2) • groups [32, 45], [64, 246] rank 4, and structure C2n × C2 × C2 × C2 (n > 1) • groups [32, 46], [64, 250] rank 4, and structure C2 × C2 × D2n (n > 2) • groups [64, 13], [64, 14] rank 2 • groups [64, 31], [64, 40] 12 rank 2, and structure (C16 × C2) : C2 • groups [64, 112], [64, 113] rank 3, and structure (C4 : C8) : C2 • groups [64, 119], [64, 121] rank 3 • groups [64, 120], [64, 122] rank 3, and structure Q16 : C4 • groups [64, 124], [64, 125] rank 3 • groups [64, 142], [64, 148] rank 3 • groups [64, 144], [64, 146] rank 3 • groups [64, 156], [64, 158] rank 3, and structure Q8 : Q8 • groups [64, 161], [64, 162] rank 3, and structure (C2 × (C4 : C4)) : C2 • groups [64, 164], [64, 165] rank 3, and structure (Q8 : C4) : C2 • groups [64, 173], [64, 176] rank 3, and structure (C8 × C4) : C2 7.2 3-groups • groups [3, 1], [9, 1], [27, 1], [81, 1] rank 1 and cyclic • groups [9, 2], [27, 2], [81, 2], [81, 4], [81, 5] rank 2, metacyclic and structure C3n : C3m • groups [27, 4], [81, 6] rank 2, and structure C3n : C3 • groups [27, 5], [81, 11] rank 3, and structure C3n : C 2 3 References [1] H. U. Besche, B. Eick, and E. O'Brien. SmallGroups - a library of groups of small order, 2005. A refereed GAP 4 package, see [13]. [2] H. U. Besche, B. Eick, and E. A. O'Brien. A millenium project: constructing small groups. Internat. J. Algebra Comput., 12:623 -- 644, 2002. [3] J. F. Carlson. Coclass and cohomology. J. Pure Appl. Algebra, 200(3):251 -- 266, 2005. 13 [4] J. F. Carlson, L. Townsley, L. Valeri-Elizondo, and M. Zhang. Cohomology rings of finite groups, volume 3 of Algebras and Applications. Kluwer Academic Publishers, Dordrecht, 2003. With an appendix: Calculations of cohomology rings of groups of order dividing 64 by Carlson, Valeri-Elizondo and Zhang. [5] B. Eick. Computing nilpotent quotients of associative algebras and algebras satisfying a polynomial identity. Internat. J. Algebra Comput., 21(8):1339 -- 1355, 2011. [6] D. J. Green and S. A. King. The computation of the cohomology rings of all groups of order 128. J. Algebra, 325:352 -- 363, 2011. [7] D. J. Green and S. A. King. p-Group Cohomology Package (Version 2.1.4). Peer reviewed optional package for Sage [12]. Available from http://sage.math.washington.edu/home/SimonKing/Cohomology. [8] G.-M. Greuel, G. Pfister, and H. Schonemann. puter algebra system for polynomial computations, http://www.singular.uni-kl.de. Singular -- A com- 2009. from Available [9] G. M. Greuel and G. Pfister. A Singular introduction to commutative algebra. Ex- tended edition. Springer, 2008. [10] J. Huebschmann. The mod-p cohomology rings of metacyclic groups. J. Pure Appl. Algebra, 60(1):53 -- 103, 1989. [11] M. Kreuzer and L. Robbiano. Computational commutative algebra. 2. Springer-Verlag, Berlin, 2005. [12] W. Stein et al. Sage Mathematics Software. The Sage Development Team, 2009. See http://www.sagemath.org. [13] The GAP Group. GAP -- Groups, Algorithms and Programming, Version 4.4. Available from http://www.gap-system.org, 2005. 14
1609.06601	5	1609	2019-12-19T13:37:20	Positive cones on algebras with involution	[ "math.RA" ]	We introduce positive cones on algebras with involution. These allow us to prove analogues of Artin's solution to Hilbert's 17th problem, the Artin-Schreier theorem characterizing formally real fields, and to define signatures with respect to positive cones. We consider the space of positive cones of an algebra with involution and investigate its topological properties, showing in particular that it is a spectral space. As an application we solve the problem of the existence of positive involutions.	math.RA	math	Positive cones on algebras with involution∗ Vincent Astier Thomas Unger Dedicated to Manfred Knebusch on the occasion of his eightieth birthday Abstract We introduce positive cones on algebras with involution. These allow us to prove analogues of Artin's solution to Hilbert's 17th problem, the Artin- Schreier theorem characterizing formally real fields, and to define signatures with respect to positive cones. We consider the space of positive cones of an algebra with involution and investigate its topological properties, showing in particular that it is a spectral space. As an application we solve the problem of the existence of positive involutions. Keywords. Real algebra, algebras with involution, orderings, hermitian forms Contents 1 Introduction 2 Preliminaries 3 Prepositive cones and positive cones 4 Prepositive cones under Morita equivalence . . 4.1 Going up . . 4.2 Going down . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 3 7 13 15 16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . ∗Accepted manuscript c(cid:13)2019, made available under the CC-BY-NC-ND 4.0 license http://creativecommons.org/licenses/by-nc-nd/4.0/. Published article: Advances in Mathematics 361 (2020) 106954. DOI: 10.1016/j.aim.2019.106954 V. Astier, T. Unger (Corresponding author): School of Mathematics and Statistics, University College Dublin, Belfield, Dublin 4, Ireland; e-mail: [email protected], [email protected] Mathematics Subject Classification (2010): 13J30, 16W10, 06F25, 16K20, 11E39 1 Positivecones on algebras with involution 5 Prepositive cones under field extension 5.1 Basic results on convex cones over ordered fields 5.2 Prepositive cones under field extensions . . . . . . . 6 Existence of positive involutions 6.1 Prepositive cones in the almost split case . . 6.2 Prepositive cones and positive involutions . 7 Positive cones 7.1 Description of positive cones . . 7.2 Formally real algebras with involution . . 7.3 Intersections of positive cones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Signatures at positive cones 8.1 Real splitting and maximal symmetric subfields . . . . 8.2 The signature at a positive cone . . . . . . . . . 9 The topology of X(A,σ) 1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 19 19 20 23 24 25 27 27 29 31 33 33 37 42 In a series of papers [2], [3], [4], [5] we initiated an investigation of central sim- ple algebras with involution from a real algebraic point of view, inspired by the classical correspondences between signatures of quadratic forms over a field F, morphisms from the Witt ring W(F) into Z, prime ideals of W(F), and orderings on F, which form one of the foundations of real algebra. More precisely, in [2] we defined signatures of hermitian forms over algebras with involution (A, σ), and in [3] we showed that these provide the desired natural correspondences with morphisms from the Witt group W(A, σ) into Z and with "prime ideals" of W(A, σ). In the present paper we show that these correspondences can be extended to include a notion of partial ordering on algebras with involution, which we call (pre-)positive cone (Definition 3.1). Prepositive cones are inspired by both posi- tive semidefinite matrices and one-dimensional hermitian forms of maximal sig- nature, as well as by Prestel's notion of quadratic semiorderings [20], and can be interpreted as those orderings on the base field that extend to the algebra (Propo- sition 3.9). In addition to the above correspondences, positive cones allow us to obtain analogues of the Artin-Schreier theorem on the characterization of for- mally real fields (Theorem 7.9) and Artin's solution of Hilbert's 17th problem (Theorem 7.14, Corollary 7.15). We also define signatures with respect to posi- tive cones (Section 8.2) and show that the space of all positive cones is a spectral Positivecones on algebras with involution 3 space (Theorem 9.12), whose topology is linked to the topology of the space of orderings of the base field (Proposition 9.9 and Corollary 9.10). Finally, as an application, we answer the question of the existence of positive involutions (Theorem 6.8), a question that does not seem to have been treated before in full generality. 2 Preliminaries We present the notation and main tools used in this paper and refer to the standard references [12], [13], [14] and [24] as well as to [2], [3] and [4] for the details. Let F be a field of characteristic different from 2. We denote by W(F) the Witt ring of F, by XF the space of orderings of F, and by FP a real closure of F at an ordering P ∈ XF. We allow for the possibility that F is not formally real, i.e. that XF = ∅. By an F-algebra with involution we mean a pair (A, σ) where A is a finite-dimensional simple F-algebra with centre a field K = Z(A), equipped with an involution σ : A → A, such that F = K ∩ Sym(A, σ), where Sym(A, σ) := {a ∈ A σ(a) = a}. We also define Skew(A, σ) := {a ∈ A σ(a) = −a}. Observe that dimF K 6 2. We say that σ is of the first kind if K = F and of the second kind (or of unitary type) otherwise. Involutions of the first kind can be further subdivided into those of orthogonal type and those of symplectic type, depending on the dimension of Sym(A, σ). We let ι = σK and note that ι = idF if σ is of the first kind. The class of F-algebras with involution is often enlarged to include algebras with unitary involution with centre F × F, cf. [13, 2.B] since it then becomes stable under scalar extension. We are not interested in such algebras in this paper, since the objects we are interested in are trivial in this situation, cf. Remark 3.4. If A is a division algebra, we call (A, σ) an F-division algebra with involution. We denote Brauer equivalence by ∼ and isomorphism by (cid:27). Quadratic and her- mitian forms are often just called forms. The notation ϕ 6 ψ indicates that ϕ is a subform of ψ and ϕ ≃ ψ indicates that ϕ and ψ are isometric. Let (A, σ) be an F-algebra with involution. We denote by W(A, σ) the Witt group of Witt equivalence classes of nonsingular hermitian forms over (A, σ), de- fined on finitely generated right A-modules. Note that W(A, σ) is a W(F)-module. For a1, . . . , ak ∈ F the notation ha1, . . . , aki stands for the quadratic form i ∈ F, as usual, whereas for a1, . . . , ak in Sym(A, σ) (x1, . . . , xk) ∈ F k 7→ Pk the notation ha1, . . . , akiσ stands for the diagonal hermitian form i=1 ai x2 (cid:0)(x1, . . . , xk), (y1, . . . , yk)(cid:1) ∈ Ak × Ak 7→ In each case, we call k the dimension of the form. kX i=1 σ(xi)aiyi ∈ A. Positivecones on algebras with involution 4 Let h : M × M → A be a hermitian form over (A, σ). We sometimes write (M, h) instead of h. The rank of h, rank(h), is the rank of the A-module M. The set of elements represented by h is denoted by D(A,σ)(h) := {u ∈ Sym(A, σ) ∃x ∈ M such that h(x, x) = u}. We say that h is weakly isotropic if there exists ℓ ∈ N such that ℓ × h is isotropic, and strongly anisotropic otherwise. Similarly, we say that h is weakly hyperbolic if there exists ℓ ∈ N such that ℓ × h is hyperbolic. We denote by Int(u) the inner automorphism determined by u ∈ A×, where Int(u)(x) := uxu−1 for x ∈ A. It follows from the structure theory of F-algebras with involution that A is isomorphic to a full matrix algebra Mℓ(D) for a unique ℓ ∈ N (called the matrix size of A) and an F-division algebra D (unique up to isomorphism) that is equipped with an involution ϑ of the same kind as σ, cf. [13, Theorem 3.1]. For B = (bi j) ∈ Mℓ(D) we let ϑt(B) = (ϑ(b ji)). By [13, 4.A], there exists ε ∈ {−1, 1} and an invertible matrix Φ ∈ Mℓ(D) such that ϑ(Φ)t = εΦ and (A, σ) (cid:27) (Mℓ(D), adΦ), where adΦ = Int(Φ) ◦ ϑt. (In fact, Φ is the Gram matrix of an ε-hermitian form over (D, ϑ).) Note that Φ is only defined up to a factor in F×, with adΦ = adλΦ for all λ ∈ F× and that ε = 1 when σ and ϑ are of the same type, cf. [13, Theorem 4.2]. We fix an isomorphism of F-algebras with involution f : (A, σ) → (Mℓ(D), adΦ). Given an F-algebra with involution (B, τ) we denote by Hermε(B, τ) the cate- gory of (possibly singular) ε-hermitian forms over (B, τ), cf. [12, p. 12]. We drop the subscript ε when ε = 1. The isomorphism f trivially induces an equivalence of categories f∗ : Herm(A, σ) −→ Herm(Mℓ(D), adΦ). Furthermore, the F-algebras with involution (A, σ) and (D, ϑ) are Morita equivalent, cf. [12, Chapter I, Theo- rem 9.3.5]. In this paper we make repeated use of a particular Morita equivalence between (A, σ) and (D, ϑ), following the approach in [17] (see also [2, §2.4] for the case of nonsingular forms and [2, Proposition 3.4] for a justification of why using this equivalence is as good as using any other equivalence for the purpose of computing signatures), namely: s / Hermε(Mℓ(D), ϑt) g f∗ Herm(A, σ) / Herm(Mℓ(D), adΦ) / Hermε(D, ϑ), (2.1) where s is the scaling by Φ−1 Morita equivalence, given by (M, h) 7→ (M, Φ−1h) and g is the collapsing Morita equivalence, given by (M, h) 7→ (Dk, b), where k is the rank of M as Mℓ(D)-module and b is defined as follows: fixing an isomorphism M (cid:27) (Dℓ)k, h can be identified with the form (Mk,ℓ(D), hBiϑt ) for some matrix B ∈ Mk(D) that satisfies ϑt(B) = εB and we take for b the ε-hermitian form whose Gram matrix is B. In particular the ε-hermitian form hdiag(d1, . . . , dℓ)iϑt is mapped to hd1, . . . , dℓiϑ by g. Note that hBiϑt(X, Y) := ϑ(X)t BY for all X, Y ∈ Mk,ℓ(D). / / / Positivecones on algebras with involution 5 Given an ordering P ∈ XF we defined a signature map signη P : W(A, σ) → Z via scalar extension to FP and Morita theory in [2]. This map has many properties in common with the usual Sylvester signature signP of quadratic forms (cf. [2] and [3, §2]) and reduces to ± signP when (A, σ) = (F, idF). See also [4] for a concise presentation as well as for the notation that we will use in this paper. In partic- ular, recall that η denotes a tuple of reference forms for (A, σ) and that a Morita equivalence between F-algebras with involution of the same type sends a tuple of reference forms to a tuple of reference forms, cf. the proof of [3, Theorem 4.2]. Furthermore Nil[A, σ] := {P ∈ XF signη P = 0} denotes the set of nil orderings for (A, σ), and depends only on the Brauer class of A and the type of σ. Finally, let eXF := XF \ Nil[A, σ], which does not indicate the dependence on (A, σ) in order to avoid cumbersome and is therefore compact. We denote the restriction of the Harrison topology to notation. By [2, Corollary 6.5], eXF is clopen in XF for the Harrison topology TH, eXF also by TH. Definition 2.1. Let h ∈ Herm(A, σ). We say that h is universal if D(A,σ)h = Sym(A, σ). We denote the set of invertible elements in Sym(A, σ) by Sym(A, σ)×. Lemma 2.2. Let h be a hermitian form of rank k over (A, σ). There are a1, . . . , ak ∈ Sym(A, σ)× ∪ {0} such that ℓ × h ≃ ha1, . . . , akiσ. If (A, σ) = (Mℓ(D), ϑt) we can take a1, . . . , ak ∈ Sym(D, ϑ) · Iℓ. Proof. Since f∗ and s preserve diagonal forms and isometries, we may assume that (A, σ) = (Mℓ(D), ϑt). Let d1, . . . , dk ∈ Sym(D, ϑ) be such that g(h) = hd1, . . . , dkiϑ. Then g(ℓ × h) ≃ (ℓ × hd1iϑ) ⊥ · · · ⊥ (ℓ × hdkiϑ). Therefore ℓ × h ≃ g−1(ℓ × hd1iϑ) ⊥ · · · ⊥ g−1(ℓ × hdkiϑ) ≃ hdiag(d1, . . . , d1)iϑt ⊥ · · · ⊥ hdiag(dk, . . . , dk)iϑt ≃ hdiag(d1, . . . , d1), · · · , diag(dk, . . . , dk)iϑt . (cid:3) Lemma 2.3. Let h be a nonsingular isotropic hermitian form over (A, σ). Then ℓ × h is universal. Positivecones on algebras with involution 6 Proof. By Morita theory g(s( f∗(h))) is a nonsingular isotropic hermitian form over (D, ϑ), and thus there are d3, . . . , dk ∈ Sym(D, ϑ) such that g(s( f∗(h))) ≃ h−1, 1, d3, . . . , dkiϑ. Using the same method as in the proof of Lemma 2.2 we obtain ℓ × s( f∗(h)) ≃ h−Iℓ, Iℓ, d3 · Iℓ, . . . , dk · Iℓiϑt , and so ℓ × h ≃ h−a, aiσ ⊥ h′, for some a ∈ Sym(A, σ)× and some form h′ over (A, σ). Since a is invertible, we have h−a, aiσ ≃ h−1, 1iσ and a standard argument shows that the hermitian form h−1, 1iσ over (A, σ) is universal. The result follows. Lemma 2.4. Let a ∈ Sym(A, σ) \ {0}. Then ℓ × haiσ represents an element in Sym(A, σ)× and ℓ × ha, −aiσ is universal. Proof. By Lemma 2.2, ℓ×haiσ ≃ ha1, . . . , aℓiσ for some a1, . . . , aℓ ∈ Sym(A, σ)×∪ {0}. At least one ai, say a1, is nonzero since D(A,σ)(ℓ × haiσ) contains a , 0. Also, ha1, −a1iσ 6 ℓ × ha, −aiσ and we conclude as in the proof of Lemma 2.3 since a1 is invertible. (cid:3) (cid:3) P h0 = k0 for every P ∈ eXF. Lemma 2.5. There exist a nonsingular diagonal hermitian form h0 over (A, σ) and k0 ∈ N such that signη Proof. Let ϕ0 be a nonsingular diagonal hermitian form over (A, σ) such that P ϕ0 , 0 for every P ∈ eXF, cf. [3, Proposition 3.2 and the remark follow- signη ing it]. Let k1, . . . , ks be the different values that signη P ϕ0 takes when P varies in eXF, and let Ui be the clopen set {P ∈ eXF signη P ϕ0 = ki} for i = 1, . . . , s. (Ui is clopen since signη ϕ0 is continuous by [2, Theorem 7.2] and eXF is clopen.) The Fact: There is a diagonal hermitian form ϕ′ over (A, σ) such that signη ϕ′ has result is obtained by induction on s, using the following fact. constant non-zero values on U1 ∪ U2, U3, . . . , Us. Proof of the fact: Let q be a quadratic form over F and let r ∈ N be such that sign q is equal to 2rk2 on U1, to 2rk1 on U2 and to 2r on U3∪· · ·∪Us, cf. [14, Chapter VIII, Lemma 6.10]. Then q ⊗ ϕ0 is a hermitian form over (A, σ) and signη(q ⊗ ϕ0) is equal to 2rk2k1 on U1, to 2rk1k2 on U2 and to 2rki on Ui for i = 3, . . . , s. (cid:3) Let T0 be the coarsest topology on eXF that makes signη h continuous, for all nonsingular hermitian forms h over (A, σ), and T1 the coarsest topology on eXF that makes all signηhaiσ continuous, for a ∈ Sym(A, σ)×. Proposition 2.6. The topologies TH, T0 and T1 on eXF are equal. Positivecones on algebras with involution 7 Proof. Clearly T1 ⊆ T0. We also know that the maps signη h are all continuous from eXF with the topology TH to Z by [2, Theorem 7.2], so T0 ⊆ TH. The result will follow if we check that TH ⊆ T1. Let h0 = ha1, . . . , aniσ be the hermitian form from Lemma 2.5. Observe that since h0 is nonsingular we have a1, . . . , an ∈ Sym(A, σ)×. We consider the open set H(u) ∩ eXF in TH. Let q be a nonsingular diagonal quadratic form over F such that sign q is equal to 2r on H(u) ∩ eXF and 0 on eXF \ H(u) (cf. [14, Chapter VIII, Lemma 6.10] and since eXF is clopen in XF). The hermitian form q ⊗ h is a diagonal form with invertible coefficients, so the map signη q ⊗ h is continuous on eXF equipped with T1. By choice of h0 and q, signη q ⊗ h0 is equal to 2rk0 on H(u) ∩ eXF and to 0 on eXF \ H(u). Then H(u) ∩ eXF = (signη q ⊗ h0)−1(2rk0) ∈ T1. 3 Prepositive cones and positive cones (cid:3) The objective of this section is to introduce a notion of ordering on algebras with involution that corresponds to non-zero signatures of hermitian forms. The precise statement of this correspondence will appear later in the paper, in Corollary 7.6. For the remainder of the paper we fix some field F of characteristic not 2 and some F-algebra with involution (A, σ). Let (D, ϑ), adΦ, ℓ and ε be as in Section 2. We make the convention that orderings in XF always contain 0. Definition 3.1. A prepositive cone P on (A, σ) is a subset P of Sym(A, σ) such that (P1) P , ∅; (P2) P + P ⊆ P; (P3) σ(a) · P · a ⊆ P for every a ∈ A; (P4) PF := {u ∈ F uP ⊆ P} is an ordering on F. (P5) P ∩ −P = {0} (we say that P is proper). A prepositive cone P is over P ∈ XF if PF = P. A positive cone is a prepositive cone that is maximal with respect to inclusion. Observe that {0} is never a prepositive cone on (A, σ) by (P4). Definition 3.2. We denote by Y(A,σ) the set of all prepositive cones on (A, σ), and by X(A,σ) the set of all positive cones on (A, σ). We say that (A, σ) is formally real if it has at least one prepositive cone. Positivecones on algebras with involution 8 Remark 3.3. (1) We will show later in the paper that positive cones only exist over non-nil orderings, and that for any such ordering there are exactly two positive cones over it, cf. Proposition 6.6 and Theorem 7.5. (2) There are prepositive cones that are not positive cones. A natural example can be constructed using the theory of Tignol-Wadsworth gauges, cf. [25], as well as their links with positive cones, cf. [6], and will be presented in a forthcoming paper. Remark 3.4. The reason we do not consider the case where Z(A) = F × F in this paper is as follows: in this case assume that P is a prepositive cone on (A, σ) and take a ∈ P \ {0}. Consider the hermitian form haiσ. Then D(A,σ)(k × haiσ) ⊆ P for every k ∈ N by (P2) and (P3) and it follows from [4, Propositions A.3 and B.8] that P = Sym(A, σ), contradicting (P5). The following proposition justifies why we use the terminology "proper" for (P5). Proposition 3.5. Let P ⊆ Sym(A, σ) satisfy properties (P1) up to (P3) and as- sume that P does not satisfy (P5). Then P = Sym(A, σ). Proof. By hypothesis there is a ∈ P \ {0} such that a, −a ∈ P. Therefore D(A,σ)(k × ha, −aiσ) ⊆ P for every k ∈ N by (P2) and (P3). Since a , 0, Lemma 2.4 tells us that ℓ × h−a, aiσ is universal and the conclusion follows. Lemma 3.6. Let P be a prepositive cone on (A, σ). Then P contains an invert- ible element. (cid:3) Proof. Consider the hermitian form haiσ, where a ∈ P \ {0}. By Lemma 2.4, ℓ × haiσ represents an element b ∈ Sym(A, σ)×. Then b ∈ P since by (P2) and (P3), D(A,σ)(ℓ × haiσ) ⊆ P. Proposition 3.7. If (A, σ) is formally real, then W(A, σ) is not torsion. (cid:3) Proof. Let P be a prepositive cone on (A, σ). Suppose W(A, σ) is torsion and consider the hermitian form haiσ, where a ∈ P is invertible (cf. Lemma 3.6). Then there exists k ∈ N such that k × haiσ is hyperbolic. By Lemma 2.3 there exists r ∈ N such that r×haiσ is universal. By (P2) and (P3), D(A,σ)(r×haiσ) ⊆ P, contradicting (P5). (cid:3) The converse to the previous proposition also holds, as we will show in Propo- sition 7.11. Positivecones on algebras with involution 9 Corollary 3.8. If (A, σ) is formally real, then the involution ϑ on D can be chosen such that ε = 1. Proof. If it is not possible to choose ϑ as indicated, then (D, ϑ, ε) = (F, idF, −1) by [4, Lemma 2.3]. Diagram (2.1) then induces an isomorphism of Witt groups be- tween W(A, σ) and W−1(F, idF), the Witt group of skew-symmetric bilinear forms over F, which is well-known to be zero, contradicting Proposition 3.7. (cid:3) Assumption for the remainder of the paper: In view of this corollary we may and will assume without loss of generality that the involution ϑ on D is chosen to be of the same type as σ (i.e. such that ε = 1) whenever (A, σ) is formally real. We collect some simple, but useful, properties of prepositive cones in the fol- lowing proposition. Proposition 3.9. Let P be a prepositive cone on (A, σ). (1) Assume that 1 ∈ P. Then PF = P ∩ F. (2) Let α ∈ PF \ {0}. Then αP = P. Proof. (1) If α ∈ PF, then α = α·1 ∈ P∩F. Conversely, assume that α ∈ P∩F. If α < PF, then −α ∈ PF \ {0}. Therefore −α = −α · 1 ∈ P, contradicting (P5) since α ∈ P. (2) We know that αP ⊆ P. It follows from (P3) that α−1Pα−1 ⊆ P and so α−1P ⊆ Pα = αP ⊆ P. The result follows. (cid:3) Note that a characterization of the condition 1 ∈ P is given in Corollary 7.7. Example 3.10. Let (A, σ) be an F-algebra with involution and let P ∈ XF. Let h be a hermitian form over (A, σ) such that h¯ui ⊗ h is anisotropic for every finite tuple ¯u of elements of P. Then P := [ ¯u∈Pk, k∈N D(A,σ)(h¯ui ⊗ h) is a prepositive cone on (A, σ). Also, if P is a prepositive cone on (A, σ) over P ∈ XF, then for any diagonal hermitian form h over (A, σ) with coefficients in P we have D(A,σ)(h¯ui ⊗ h) ⊆ P for every finite tuple ¯u of elements of P. In particular the hyperbolic plane is never a subform of h¯ui ⊗ h. If h is in addition nonsingular, then h is strongly anisotropic by Lemma 2.3. Positivecones on algebras with involution 10 Most of the motivation behind the definition of prepositive cones comes from Examples 3.11 and 3.13 below. Example 3.11. Let (A, σ) = (Mn(F), t), where t denotes the transpose involution. Let P ∈ XF and let P consist of those symmetric matrices in Mn(F) that are positive semidefinite with respect to P. Then P is a prepositive cone on (Mn(F), t) over P. Note that if M ∈ P, then there exists T ∈ GLn(F) such that T t MT is a diagonal matrix whose elements all belong to P. In particular, if M is invertible, M is positive definite, and therefore signhInit hMit = n. (Note that hInit is a reference form for (Mn(F), t)). We will observe later in Example 4.11 that P and −P are the only prepositive cones on (Mn(F), t) over P. P Definition 3.12. Let (A, σ) be an F-algebra with involution and let P ∈ XF. We define mP(A, σ) := max{signη Phaiσ a ∈ Sym(A, σ)×} (note that mP(A, σ) does not depend on the choice of η) and M η P (A, σ) := {a ∈ Sym(A, σ)× signη Phaiσ = mP(A, σ)} ∪ {0}. Example 3.13. For every F-division algebra with involution (D, ϑ), every tuple of reference forms η for (D, ϑ) and every P ∈ XF \ Nil[D, ϑ], the set M η P (D, ϑ) is a prepositive cone on (D, ϑ) over P. Properties (P1) and (P4) are clear (for (P4) note that signη P h, cf. [3, Theorem 2.6]). We justify the others. P(q ⊗ h) = signP q · signη (P2): Let a, b ∈ M η Phaiσ = − signη P (D, ϑ) \ {0}. Then a + b is represented by ha, biϑ. Observe that since P ∈ XF \ Nil[D, ϑ], mP(D, ϑ) > 0 and thus a , −b (because a = −b would imply mP(D, ϑ) = signη Phbiσ = −mP(D, ϑ), impossible). Since D is a division algebra, a + b is invertible and ha, biϑ ≃ ha + b, ciϑ for some c ∈ Sym(D, ϑ)×. Comparing signatures and using the fact that a, b ∈ M η P (D, ϑ) we obtain a + b ∈ M η forms hbiϑ and hϑ(a)baiϑ are isometric. (P3) follows from the fact that for any a ∈ D× and any b ∈ Sym(D, ϑ)× the (P5) follows from the fact that P < Nil[D, ϑ] and thus mP(D, ϑ) > 0. We will see later, in Proposition 7.1, that M η P (D, ϑ). P (D, ϑ) is in fact a positive cone on (D, ϑ) over P. Remark 3.14. Observe that P is a prepositive cone on (A, σ) if and only if −P is a prepositive cone on (A, σ). This property of prepositive cones corresponds to the fact that if signη P . We will return to this observation later, cf. Proposition 9.9. P : W(A, σ) → Z is a signature, then so is − signη P = sign−η Remark 3.15. The reason for requiring PF to be an ordering on F in (P4) in- stead of just a preordering is as follows: in case (A, σ) = (Mn(F), t), we want the Positivecones on algebras with involution 11 prepositive cone P on (Mn(F), t) to be a subset of the set of positive definite (or negative definite) matrices (see Proposition 6.5 where we prove this result). Hav- ing this requirement while only asking that PF is a preordering forces F to be a SAP field as we explain in the remainder of this remark. We first recall that a field F has the Strong Approximation Property (SAP) if for any two disjoint closed subsets (in the Harrison topology) A and B of XF, there exists a ∈ F such that A ⊆ H(a) and B ⊆ H(−a), cf. [21, p. 66]. This is equivalent to F satisfying the Weak Hasse Principle, which asserts that a quadratic form over F is weakly isotropic whenever it is indefinite with respect to all orderings of F, cf. [21, Section 9]. We thus assume for the remainder of this remark that axiom (P4) only re- quires PF to be a preordering. Let q = ha1, . . . , ani be a quadratic form over F that is strongly anisotropic. Let M = diag(a1, . . . , an) ∈ Mn(F). Then M ∈ Sym(Mn(F), t). Let P be the smallest subset of Mn(F) containing M, that is closed under (P2) and (P3). In other words, P consists of all the elements in Mn(F) that are weakly represented by the hermitian form hMit. Since hMit is Morita equivalent to q (via the map g), hMit is strongly anisotropic. It follows that P is a prepositive cone on (Mn(F), t) with PF a preordering on F. If P is to only contain positive definite (or negative definite) matrices with respect to some P ∈ XF, all the ai must have the same sign at P and thus q must be definite at P. This shows that F must satisfy the Weak Hasse Principle, i.e., that F is SAP, as recalled above. Lemma 3.16. Let P ⊆ Q be prepositive cones in (A, σ). Then PF = QF. Proof. Let α ∈ QF. Assume that α < PF. Then −α ∈ PF and thus αP = −P, cf. Proposition 3.9. Using that P ⊆ Q we also obtain αP ⊆ αQ ⊆ Q. So P, −P ⊆ Q, contradicting that Q is proper since P , {0} (as observed after Definition 3.2). Therefore, QF ⊆ PF and the equality follows since they are both orderings on F. (cid:3) Definition 3.17. Let (A, σ) be an F-algebra with involution and let P ∈ XF. For a subset S of Sym(A, σ), we define CP(S ) := n kX i=1 uiσ(xi)si xi(cid:12)(cid:12)(cid:12)(cid:12) k ∈ N, ui ∈ P, xi ∈ A, si ∈ So. For a prepositive cone P on (A, σ) over P and a ∈ Sym(A, σ), we define P[a] := np + kX i=1 uiσ(xi)axi(cid:12)(cid:12)(cid:12)(cid:12) p ∈ P, k ∈ N, ui ∈ P, xi ∈ Ao = P + CP({a}). Positivecones on algebras with involution 12 It is easy to see that both CP(S ) and P[a] are prepositive cones on (A, σ) over P if and only if they are proper. Prepositive cones on (A, σ) only give rise to partial orderings on Sym(A, σ). The following result gives an approximation, for a positive cone P, to the prop- erty F = P ∪ −P for any ordering P ∈ XF: Lemma 3.18. Let P be a positive cone on (A, σ) over P ∈ XF and a ∈ Sym(A, σ)\ P. Then there are k ∈ N, u1, . . . , uk ∈ P \ {0} and x1, . . . , xk ∈ A \ {0} such that kX i=1 uiσ(xi)axi ∈ −P. Proof. Assume for the sake of contradiction that the conclusion of the lemma does not hold, and consider P[a]. It is easy to check that P[a] satisfies axioms (P1) to (P4), and obviously it properly contains P since it contains a. We now check that P[a] is proper, thus reaching a contradiction. If P[a] were not proper, we would have kX i=1 p + uiσ(xi)axi = −(q + rX j=1 v jσ(y j)ay j), for some k, r ∈ N ∪ {0}, ui, v j ∈ P \ {0} and xi, y j ∈ A \ {0} (and where neither side is 0). Therefore kX i=1 uiσ(xi)axi + rX j=1 viσ(y j)ay j ∈ −P, a contradiction. (cid:3) Remark 3.19. Several notions of orderings have previously been considered in the special case of division algebras with involution, most notably Baer orderings (see the survey [8]). The essential difference between our notion of (pre-)positive cone and (for instance) Baer orderings, is that our definition is designed to cor- respond to a pre-existing algebraic object: the notion of signature of hermitian forms (see in particular axiom (P4) which reflects the fact that the signature is a morphism of modules, cf. [3, Theorem 2.6(iii)], and Example 3.13). In order to achieve this, it was necessary to accept that the ordering defined by a positive cone on the set of symmetric elements is in general only a partial order- ing. It gives positive cones a behaviour similar to the set of positive semidefinite matrices in the algebra with involution (Mn(F), t) (which provides, as seen above, one of the main examples of (pre-)positive cones, cf. Example 3.11). Positivecones on algebras with involution 13 4 Prepositive cones under Morita equivalence For a subset S of Sym(A, σ), we denote by Diag(S ) the set of all diagonal hermi- tian forms with coefficients in S . We also recall: Proposition 4.1 ([4, Proposition A.3]). Let h be a hermitian form over (A, σ). Then where hns is a nonsingular hermitian form (uniquely determined up to isometry) and 0 is the zero form of suitable rank. h ≃ hns ⊥ 0, Theorem 4.2. Let (A, σ) and (B, τ) be two Morita equivalent F-algebras with involution with σ and τ of the same type, and fix a Morita equivalence m : Herm(A, σ) → Herm(B, τ). The map m∗ : Y(A,σ) → Y(B,τ), defined by m∗(P) := [{D(B,τ)(m(h)) h ∈ Diag(P)} is an inclusion-preserving bijection from Y(A,σ) to Y(B,τ) that thus restricts to a bijection from X(A,σ) to X(B,τ). Furthermore, if P is over P ∈ XF, then m∗(P) is also over P. Proof. We first show that m∗(P) is a prepositive cone over (B, τ). By definition, m∗(P) satisfies properties (P1), (P2) and (P3). We now show (P5): Assume there is b ∈ m∗(P)∩−m∗(P) such that b , 0. Then there are diagonal hermitian forms h1 and h2 in Diag(P) such that b ∈ D(B,τ)(m(h1)) and −b ∈ D(B,τ)(m(h2)). By Lemma 2.4 there is k ∈ N and b1 ∈ Sym(B, τ)× such that b1 ∈ D(B,τ)(m(k× h1)) and −b1 ∈ D(B,τ)(m(k × h2)). Since b1 is invertible we deduce (as for quadratic forms) that hb1iτ 6 m(k × h1) and h−b1iτ 6 m(k × h2). Therefore hb1, −b1iτ 6 m(k × (h1 ⊥ h2)) and hb1, −b1iτ is nonsingular and clearly isotropic. So m−1(hb1, −b1iτ) 6 k × (h1 ⊥ h2) and m−1(hb1, −b1iτ) is also nonsingular and isotropic by Morita theory. By Lemma 2.3 there is k′ ∈ N such that kk′ × (h1 ⊥ h2) is universal, a contradiction to (P5) since D(A,σ)(kk′ × (h1 ⊥ h2)) ⊆ P. We prove (P4) by showing that (m∗(P))F = PF. Let u ∈ PF and let b ∈ m∗(P). Then there is a diagonal hermitian form h over (A, σ) with coefficients in P such that b ∈ D(B,τ)(m(h)). Since the map W(A, σ) → W(B, τ) induced by m is a morphism of W(F)-modules, the forms hui ⊗ m(h) and m(hui ⊗ h) are isometric and thus ub ∈ D(B,τ)(hui⊗ m(h)) = D(B,τ)(m(hui⊗ h)). This shows that ub ∈ m∗(P) since hui ⊗ h ∈ Diag(P) and thus PF ⊆ (m∗(P))F. Assume now that PF $ (m∗(P))F, so that there is u ∈ −PF \ {0} such that u ∈ (m∗(P))F. Let b ∈ m∗(P) \ {0}. Then by choice of u, ub ∈ m∗(P), and since Positivecones on algebras with involution 14 −u ∈ PF ⊆ (m∗(P))F we also obtain −ub ∈ m∗(P). Since u , 0 we have ub , 0 and so obtain a contradiction to property (P5). Finally, we show that m∗ is a bijection by showing that (m−1)∗ is a left inverse of m∗. The invertibility of m∗ then follows by swapping m and m−1. Let P ∈ Y(A,σ) and let c ∈ (m−1)∗(m∗(P)). Then there are elements b1, . . . , br ∈ m∗(P) such that c ∈ D(A,σ)m−1(hb1, . . . , briτ) = D(A,σ)m−1(hb1, . . . , brins τ ), and there are a1, . . . , as ∈ P such that b1, . . . , br ∈ D(B,τ)m(ha1, . . . , asiσ). Observe that by [4, Proposition 2.10] each form hbiins is a subform of sufficiently many copies of m(ha1, . . . , asiσ). Hence there exists r′ ∈ N such that hb1, . . . , brins 6 r′ × m(ha1, . . . , asiσ). It follows that τ τ m−1(hb1, . . . , brins τ ) = (cid:0)m−1(hb1, . . . , briτ)(cid:1)ns 6 m−1[r′ × m(ha1, . . . , asiσ)] = r′ × ha1, . . . , asiσ. Therefore, and since D(A,σ)m−1(hb1, . . . , brins τ ) = D(A,σ)m−1(hb1, . . . , briτ), we ob- tain c ∈ D(A,σ)(r′ × ha1, . . . , asiσ) ⊆ P. Let now a ∈ P and let h = m(haiσ). By Lemma 2.2 there are t, ℓ′ ∈ N and b1, . . . , bt ∈ D(B,τ)(ℓ′ × h) such that ℓ′ × h ≃ hb1, . . . , btiτ. In particular b1, . . . , bt ∈ m∗(P) and h 6 hb1, . . . , btiτ. Then haiσ = m−1(h) 6 m−1(hb1, . . . , btiτ), which implies a ∈ (m−1)∗(m∗(P)). Therefore P = (m−1)∗(m∗(P)). (cid:3) We can refine the description of the map m∗: Proposition 4.3. Let P ∈ Y(A,σ). Then, with the same hypotheses and notation as in Theorem 4.2, m∗(P) := [{D(B,τ)m(h) h ∈ Diag(P ∩ A×)}. Proof. The inclusion from right to left is obvious. Let b ∈ m∗(P). Then there exist a1, . . . , ar ∈ P such that b ∈ D(B,τ)m(ha1, . . . , ariσ). By Lemma 2.2 there are c1, . . . , cs ∈ Sym(A, σ)× such that ℓ × ha1, . . . , ariσ ≃ hc1, . . . , cs, 0, . . . , 0iσ. It follows from this isometry that c1, . . . , cs ∈ P, and thus b ∈ D(B,τ)m(hc1, . . . , cs, 0, . . . , 0iσ) = D(B,τ)m(hc1, . . . , csiσ), since m preserves forms with constant value zero. This proves the other inclusion. (cid:3) In addition to the general result, Theorem 4.2, we now give explicit descrip- tions of transferring prepositive cones between (A, σ) (cid:27) (Mℓ(D), Int(Φ) ◦ ϑt) and (Mℓ(D), ϑt) (scaling) and between (Mℓ(D), ϑt) and (D, ϑ) (going up and going down). Positivecones on algebras with involution 15 Proposition 4.4. Let P be a prepositive cone on (A, σ) over P. Let a ∈ Sym(A, σ)×. Then aP is a prepositive cone on (A, Int(a) ◦ σ) over P. This defines a natural inclusion-preserving bijection between Y(A,σ) and Y(A,Int(a)◦σ) and between X(A,σ) and X(A,Int(a)◦σ). Proof. Properties (P1), (P2), (P4) and (P5) are clear. Let b ∈ P and let x ∈ A. Then (Int(a) ◦ σ)(x)abx = aσ(x)a−1abx = aσ(x)bx ∈ aP, which proves (P3). The fact that the map P 7→ aP is a bijection, and preserves being proper as well as inclusions is clear. (cid:3) In the remainder of this section we describe the going up and going down correspondences, which are reminiscent of the behaviour of positive semidefinite matrices. 4.1 Going up Let P be a prepositive cone on (D, ϑ) over P ∈ XF. We define PSDℓ(P) := {B ∈ Sym(Mℓ(D), ϑt) ∀X ∈ Dℓ ϑ(X)t BX ∈ P}. The following result is straightforward, since any matrix in Sym(Mℓ(D), ϑt) is the matrix of a hermitian form over (D, ϑ) and thus can be diagonalized by congru- ences. Lemma 4.5. For B ∈ Sym(Mℓ(D), ϑt) the following are equivalent: (1) B ∈ PSDℓ(P). (2) There is G ∈ GLℓ(D) such that ϑ(G)t BG is diagonal with diagonal elements in P. (3) For every G ∈ GLℓ(D) such that ϑ(G)tBG is diagonal, the diagonal elements are in P. Lemma 4.6. PSDℓ(P) is the closure of P · Iℓ under sums and the operation Z 7→ ϑ(X)tZX, for X ∈ Mℓ(D). Proof. That PSDℓ(P) is closed under these operations is clear. Let B ∈ PSDℓ(P). By Lemma 4.5 there is G ∈ GLℓ(D) such that ϑ(G)tBG = diag(a1, . . . , aℓ) with a1, . . . , aℓ ∈ P. Observe that, with C = diag(0, . . . , 0, 1, 0, . . . , 0), diag(0, . . . , 0, ai, 0, . . . , 0) = ϑ(C)t(ai · Iℓ)C, and the result follows using sums of matrices. (cid:3) Positivecones on algebras with involution 16 Proposition 4.7. Let P be a prepositive cone on (D, ϑ) over P ∈ XF. (1) PSDℓ(P) is a prepositive cone on (Mℓ(D), ϑt) over P. (2) If P is a positive cone, then so is PSDℓ(P). Proof. (1) The properties (P1) to (P4) are easily verified. For (P5), we assume there is B ∈ Sym(Mℓ(D), ϑt) \ {0} such that that PSDℓ(P) is not proper, i.e. B ∈ PSDℓ(P) ∩ − PSDℓ(P). Let X ∈ Dℓ be such that ϑ(X)t BX , 0 (such an X exists since we may assume that B is diagonal). Then ϑ(X)t BX ∈ P ∩ −P, contradicting that P is proper. For (2), assume that there is B ∈ Sym(Mℓ(D), ϑt) such that PSDℓ(P) $ PSDℓ(P)[B] and PSDℓ(P)[B] is proper (cf. Definition 3.17 for the notation). Since B < PSDℓ(P) there is X0 ∈ Dℓ such that b0 := ϑ(X0)tBX0 < P. Let Z ∈ Mℓ(D) be the matrix whose columns are all X0. Then B0 := ϑ(Z)t BZ ∈ Sym(Mℓ(D), ϑt) is the matrix with b0 everywhere. We have B0 ∈ PSDℓ(P)[B], so PSDℓ(P)[B0] is proper (since PSDℓ(P)[B] is proper by assumption). Let Ei := diag(0, . . . , 0, 1, 0, . . . , 0) (with 1 in position i). Then B1 := Pℓ i=1 ϑ(Ei)tB0Ei = diag(b0, . . . , b0) ∈ PSDℓ(P)[B0], so PSD(P)[B1] is proper. We claim that P[b0] is proper, contradicting that P is maximal: otherwise we would have p + Pk j=1 v jϑ(y j)b0y j) , 0 for some p, q ∈ P, k, r ∈ N, ui, v j ∈ P and xi, y j ∈ D. Then if Xi = diag(xi, . . . , xi) and Y j = diag(y j, . . . , y j), we have i=1 uiϑ(xi)b0 xi = −(q + Pr pIn + kX i=1 uiϑ(Xi)B1Xi = −(qIn + rX j=1 v jϑ(Y j)B1Y j) , 0, contradicting that PSDℓ(P)[B1] is proper. (cid:3) 4.2 Going down Let P be a prepositive cone over P on (Mℓ(D), ϑt). Denoting the matrix trace by tr, we define Trℓ(P) := {tr(B) B ∈ P}. Lemma 4.8. For d ∈ Sym(D, ϑ)× the following are equivalent: (1) d ∈ Trℓ(P). (2) d ∈ {ϑ(X)t BX B ∈ P, X ∈ Dℓ}. (3) diag(d, d2, . . . , dℓ) ∈ P for some d2, . . . , dℓ ∈ Sym(D, ϑ). (4) diag(d, 0, . . . , 0) ∈ P. Positivecones on algebras with involution 17 Proof. (1) ⇒ (2): Let d = tr(B) for some B = (bi j) ∈ P and consider the matrix Ei := diag(0, . . . , 0, 1, 0, . . . , 0) (where the element 1 is in position i). Then Bi := ϑ(Ei)tBEi = diag(0, . . . , 0, bii, 0, . . . , 0) ∈ P and so diag(b11, . . . , bℓℓ) = B1 + · · · + Bℓ ∈ P. Thus d = ϑ(X)t(B1 + · · · + Bℓ)X, where X = (1 · · · 1)t ∈ Dℓ. (2) ⇒ (3): Let d = ϑ(X)t BX for some B ∈ P and X ∈ Dℓ. The matrix B is the matrix of a hermitian form h over (D, ϑ) and by hypothesis d is represented by h. Since D is a division algebra, there is a diagonalization of h that has d as its first entry, i.e. there exist G ∈ GLℓ(D) and d2, . . . , dℓ ∈ Sym(D, ϑ) such that ϑ(G)t BG = diag(d, d2, . . . , dℓ). (3) ⇒ (4): Follows from diag(d, 0, . . . , 0) = ϑ(X)t diag(d, d2, . . . , dℓ)X, where X = diag(1, 0, . . . , 0). (4) ⇒ (1): Clear. For future use we note the equality (cid:3) Trℓ(P) = {ϑ(X)t BX B ∈ P, X ∈ Dℓ}, (4.1) trivially given by (1) ⇔ (2) in Lemma 4.8. Proposition 4.9. Let P be a prepositive cone on (Mℓ(D), ϑt) over P. (1) Trℓ(P) is a prepositive cone on (D, ϑ) over P. (2) If P is a positive cone, then so is Trℓ(P). Proof. (1) Axioms (P1) and (P2) are straightforward, while (P3) follows from (4.1). We check axiom (P5) and assume that Trℓ(P) is not proper, i.e. there is a ∈ D \ {0} such that a ∈ Trℓ(P) ∩ − Trℓ(P). By Lemma 4.8, diag(a, 0, . . . , 0) and diag(−a, 0, . . . , 0) are in P, contradicting that P is proper. Axiom (P4) now follows using (P5). (2) Assume that Trℓ(P) is not maximal. Then there is d ∈ Sym(D, ϑ) such that d < Trℓ(P) and Trℓ(P) $ Trℓ(P)[d], where Trℓ(P)[d] is proper. Since d < Trℓ(P), the matrix C := diag(d, 0, . . . , 0) is not in P by Lemma 4.8. We show that P[C] is proper, which will contradict the hypothesis that P is maximal, finishing the proof. If P[C] is not proper, then P[C] ∩ −P[C] , {0}, so there are B, B′ ∈ P, k, r ∈ N, ui, v j ∈ P and Xi, Y j ∈ Mℓ(D) \ {0} such that B + kX i=1 uiϑ(Xi)tCXi = −(B′ + rX j=1 v jϑ(Y j)tCY j) , 0. Up to multiplying all terms in the above equality on the left by ϑ(J)t and on the right by J for some well-chosen invertible matrix J, we can assume that the matrix B0 := B +Pk i=1 uiϑ(Xi)tCXi is diagonal. Positivecones on algebras with involution 18 Let k0 ∈ {1, . . . , ℓ} be such that the (k0, k0)-th coordinate of B0 is non-zero, and let E ∈ Dℓ be the column vector with all coordinates 0 except for a 1 at coordinate k0. Then ϑ(E)tBE + kX i=1 uiϑ(XiE)tCXiE = −(ϑ(E)t B′E + rX j=1 v jϑ(Y jE)tCY jE), where the left-hand side is non-zero by choice of E. Since both sides belong to Trℓ(P)[d] which is proper, we get a contradiction. (cid:3) Proposition 4.10. The maps Y(D,ϑ) → Y(Mℓ (D),ϑt), P 7→ PSDℓ(P) and are inverses of each other, and restrict to maps Y(Mℓ (D),ϑt) → Y(D,ϑ), P 7→ Trℓ(P) and X(D,ϑ) → X(Mℓ (D),ϑt), P 7→ PSDℓ(P) X(Mℓ(D),ϑt) → X(D,ϑ), P 7→ Trℓ(P). Proof. By Propositions 4.7 and 4.9 we only need to show that these maps are in- verses of each other. The equality Trℓ(PSDℓ(P)) = P for P ∈ Y(D,ϑ) is straight- forward using (4.1). We now show that, given P ∈ Y(Mℓ (D),ϑt), we have PSDℓ(Trℓ(P)) = P. Let B ∈ PSDℓ(Trℓ(P)). Then by Lemma 4.5 there exists G ∈ GLℓ(D) such that ϑ(G)t BG = diag(d1, . . . , dℓ) with d1, . . . , dℓ ∈ Trℓ(P). By Lemma 4.8, diag(di, 0, . . . , 0) ∈ P for i = 1, . . . , ℓ. Using (P3) with permutation matri- ces, followed by (P2) we obtain that diag(d1, . . . , dℓ) ∈ P and therefore B = ϑ(G−1)t diag(d1, . . . , dℓ)G−1 ∈ P. Conversely, let B ∈ P. Since B ∈ Sym(Mℓ(D), ϑt) there exists G ∈ GLℓ(D) such that ϑ(G)tBG = diag(d1, . . . , dℓ) ∈ P. It follows from (4.1) that d1, . . . , dℓ are in Trℓ(P). Therefore, by Lemma 4.5, diag(d1, . . . , dℓ) and thus B are in PSDℓ(Trℓ(P)). (cid:3) Example 4.11. Let P ∈ XF. The only two prepositive cones on (F, id) over P are P and −P. Therefore, by Proposition 4.10, the only two prepositive cones on (Mn(F), t) over P are the set of symmetric positive semidefinite matrices and the set of symmetric negative semidefinite matrices with respect to P. Positivecones on algebras with involution 19 5 Prepositive cones under field extension 5.1 Basic results on convex cones over ordered fields In this section we fix some P ∈ XF and write >P for the order relation defined by P. We recall some basic concepts from convex geometry. We consider the usual euclidean inner product on F n, hx; yi := x1y1 + · · · + xnyn, and the topology TP on F n that comes from the order topology on F. We say that a nonempty subset C of F n is a cone over P if it satisfies (C1) below and a convex cone over P if it satisfies (C1) and (C2). (C1) For every a ∈ C and r ∈ P, ra ∈ C; (C2) For every a, b ∈ C, a + b ∈ C. We say that a cone C over P is pointed if C ∩ −C = {0}, that it is closed if it is closed in the topology TP, and that it is full-dimensional if Span(C) = F n. We define the dual cone of C by C∗ := {v ∈ F n hv; Ci >P 0}. Observe that C∗ is always a closed convex cone over P. We say that a convex cone C over P is finitely generated if there are a1, . . . , ak ∈ F n such that C = {λ1a1 + · · · + λkak λ1, . . . , λk ∈ P}. We recall the following well-known result, cf. [7], as well as some immediate consequences: Theorem 5.1 (Farkas' Lemma). A finitely generated convex cone over P is the in- tersection of a finite number of closed half-spaces (i.e. is polyhedral) with respect to >P. Corollary 5.2. A finitely generated convex cone over P is closed in TP. Corollary 5.3. If C is a finitely generated convex cone over P, then (C∗)∗ = C. Lemma 5.4. Let C be a cone over P. The following are equivalent: (1) C is full-dimensional. (2) C∗ is pointed. Positivecones on algebras with involution 20 Proof. (1) ⇒ (2): Assume that C∗ is not pointed. Then there is u ∈ F n \ {0} such that u, −u ∈ C∗. By definition of C∗ we have hu; xi >P 0 and h−u; xi >P 0 for every x ∈ C, so hu; xi = 0 for every x ∈ C, i.e. u ∈ C⊥. But C⊥ = {0} since C is full-dimensional, contradiction. (2) ⇒ (1): Assume that C is not full-dimensional, so F n = Span(C) ⊥ W for some non-zero subspace W. Then W ⊆ C∗ and −W ⊆ C∗, contradicting that C∗ is pointed. (cid:3) 5.2 Prepositive cones under field extensions Let P ∈ XF and let dimF A = m. We identify A with F m as F-vector space, so that we can use coordinates in F. Let t ∈ N and let b1, . . . , bt : A × A → A be F-bilinear maps. We write ¯b = (b1, . . . , bt) and define C ¯b := n sX tX j=1 i=1 ai, jbi(xi, j, xi, j) (cid:12)(cid:12)(cid:12)(cid:12) s ∈ N, ai, j ∈ P, xi, j ∈ Ao. In other words, C ¯b is the convex cone in A over P generated by the elements bi(x, x) for i = 1, . . . , t and x ∈ A. Lemma 5.5. C ¯b is a finitely generated convex cone. ukb(cid:0) mX Proof. It suffices to prove the statement for t = 1, so for a single bilinear map b. Let {e1, . . . , em} be a basis of F m. Every element of F m can be written as Pm i=1 εiaiei with εi ∈ {−1, 1} and ai ∈ P. It follows that the convex cone generated by the elements b(x, x) for x ∈ F m is n qX δ jc je j(cid:1)(cid:12)(cid:12)(cid:12)(cid:12) q ∈ N, uk, ai, c j ∈ P, εi, δ j ∈ {−1, 1}o mX εiδ jukaic jb(ei, e j)(cid:12)(cid:12)(cid:12)(cid:12) q ∈ N, uk, ai, c j ∈ P, εi, δ j ∈ {−1, 1}o mX = n qX vrbr(cid:12)(cid:12)(cid:12)(cid:12) s ∈ N, vr ∈ P, br ∈ {±b(ei, e j) i, j = 1, . . . , n}o, = n sX εiaiei, k=1 i, j=1 k=1 i=1 j=1 r=1 which is finitely generated. Lemma 5.6. Let a1, . . . , as ∈ Sym(A, σ). Then CP(a1, . . . , as) is a finitely gener- ated convex cone over P. (cid:3) Proof. We have CP(a1, . . . , as) = n kX i=1 uiσ(xi)ci xi (cid:12)(cid:12)(cid:12)(cid:12) k ∈ N, ui ∈ P, xi ∈ A, ci ∈ {a1, . . . , as}o. Positivecones on algebras with involution 21 Define bi : A × A → A by bi(x, y) = σ(x)aiy. The map bi is F-bilinear and CP(a1, . . . , as) = n rX sX j=1 i=1 = C ¯b. ui, jbi(xi, j, xi, j) (cid:12)(cid:12)(cid:12)(cid:12) r ∈ N, ui, j ∈ P, xi, j ∈ Ao So by Lemma 5.5, the cone CP(a1, . . . , as) = C ¯b is finitely generated. (cid:3) Lemma 5.7. Let (L, Q) be an ordered field extension of (F, P). Let F be a set of F-linear forms on A such that T f∈F f −1(P) is closed under x 7→ σ(y)xy for every y ∈ A. Let a ∈ T f∈F f −1(P). Then (σ ⊗ id)(z) · (a ⊗ 1) · z ∈ \ ( f ⊗ id)−1(Q) f∈F for every z ∈ A ⊗F L. Proof. Let z = Pk i=1 xi ⊗ αi ∈ A ⊗F L and f ∈ F . Then ( f ⊗ id)[(σ ⊗ id)(z) · (a ⊗ 1) · z] = kX i, j=1 f (σ(xi)ax j)αiα j. (5.1) The map q f : A → F, x 7→ f (σ(x)ax) is a quadratic form over F, so there is an orthogonal basis {e1, . . . , em} of A for q f . Therefore, writing z = Pm i=1(ei ⊗ 1)βi with βi ∈ L, we obtain in (5.1): ( f ⊗ id)[(σ ⊗ id)(z) · (a ⊗ 1) · z] = = i, j=1 mX mX i=1 f (σ(ei)ae j)βiβ j f (σ(ei)aei)β2 i . Since f (σ(ei)aei) ∈ P by choice of a, we obtain that ( f⊗id)[(σ⊗id)(z)(a⊗1)z] ∈ Q, proving the result. (cid:3) Proposition 5.8. Let (L, Q) be an ordered field extension of (F, P) and let P be a prepositive cone on (A, σ) over P. Then P ⊗ 1 := {a ⊗ 1 a ∈ P} is contained in a prepositive cone on (A ⊗F L, σ ⊗ id) over Q. Proof. It suffices to show that CQ(P ⊗ 1) is a prepositive cone on (A⊗F L, σ⊗ id) over Q, i.e. that it is proper. This is equivalent to showing that it is pointed as a convex cone, and for this it suffices to show that CQ({a1⊗ 1, . . . , ar⊗ 1}) is pointed, for every a1, . . . , ar ∈ P and r ∈ N. Positivecones on algebras with involution 22 Since P is a prepositive cone over P, we know that C := CP({a1, . . . , ar}) is pointed, therefore by Corollary 5.3 and Lemma 5.4, C∗ is full-dimensional. It follows that C∗Q, the convex cone generated by {b ⊗ 1 b ∈ C∗} over Q is full- dimensional in A ⊗F L, and thus, by Lemma 5.4, that its dual (C∗Q)∗ is pointed. Claim: (1) (C∗Q)∗ contains ai ⊗ 1 for i = 1, . . . , r; (2) (C∗Q)∗ contains (σ ⊗ id)(z) · (ai ⊗ 1) · z for i = 1, . . . , r and z ∈ A ⊗F L; (3) (C∗Q)∗ contains CQ({a1⊗ 1, . . . , ar⊗ 1}), the prepositive cone on (A⊗F L, σ⊗ id) over Q generated by a1 ⊗ 1, . . . , ar ⊗ 1. Proof of the claim: We first observe that (C∗Q)∗ = \ b∈C∗ {w ∈ A ⊗F L hb ⊗ 1; wi ∈ Q}. (1) By definition of C∗ we have, for every b ∈ C∗, hb; aii ∈ P, i.e. α := Pm j=1 b j· j=1(b j ⊗ 1)((ai) j ⊗ 1) ∈ Q, i.e. hb ⊗ 1; ai ⊗ 1i ∈ Q, (2) By Lemma 5.6 we know that C is a finitely generated convex cone over P, (ai) j ∈ P. Therefore α ⊗ 1 = Pm proving that ai ⊗ 1 ∈ (C∗Q)∗. so by Corollary 5.3, (C∗)∗ = C, i.e. C = \ b∈C∗ ( fb)−1(P), where fb(x) := hb; xi. By Lemma 5.7 it follows that \ b∈C∗ ( fb ⊗ 1)−1(Q) contains (σ ⊗ id)(z) · (ai ⊗ 1) · z for every i = 1, . . . , r and z ∈ A ⊗F L. The result follows since a direct verification shows that ( fb ⊗ id)(z) = hb ⊗ 1; zi for every z ∈ A ⊗F L. (3) By construction (C∗Q)∗ is a convex cone over Q, so is closed under sum and multiplication by elements of Q. Using the second item in the claim, it follows at once that it contains CQ({a1 ⊗ 1, . . . , ar ⊗ 1}). This finishes the proof of the Claim. It follows from (3) that CQ({a1 ⊗ 1, . . . , ar ⊗ 1}) is pointed, which proves the result. (cid:3) Positivecones on algebras with involution 23 6 Existence of positive involutions The notion of positive involution seems to go back to Albert, see for instance [1], and Weil [26] and plays a central role in the paper [23] by Procesi and Schacher. Denote the reduced trace of A by TrdA and let P ∈ XF. Recall from [23, Definition 1.1] that the involution σ is called positive at P whenever the form A × A → K, (x, y) 7→ TrdA(σ(x)y) is positive semidefinite at P (hence positive definite at P since it is nonsingular). For more details we refer to [4, Section 4]. In this section we will show that for any given P ∈ eXF there exists an involution Let P ∈ XF. If σ is an involution of the first kind, Z(A) = F and A ⊗F FP is isomorphic to a matrix algebra over FP or (−1, −1)FP . If σ is of the second kind, Z(A) is a quadratic extension of F and A ⊗F FP is isomorphic to a matrix algebra over FP( √ −1), FP × FP or (−1, −1)FP × (−1, −1)FP , where the last two cases occur whenever Z(A) ⊗F FP (cid:27) FP × FP, in which case A ⊗F FP is semisimple. Hence there exists a unique integer nP such that on A that is positive at P, cf. Theorem 6.8. A ⊗F FP (cid:27) MnP(DP), where DP ∈ {FP, (−1, −1)FP , FP( √ −1), FP × FP, (−1, −1)FP × (−1, −1)FP}. In light of this discussion we define the following subsets of XF: Xrcf := {P ∈ XF DP = FP} Xquat := {P ∈ XF DP = (−1, −1)FP} Xacf := {P ∈ XF DP = FP( √ −1)} Xd-rcf := {P ∈ XF DP = FP × FP} Xd-quat := {P ∈ XF DP = (−1, −1)FP × (−1, −1)FP}. (6.1) (6.2) (6.3) Note that the value of nP is constant on each of these sets, since it only depends on dimFP DP by (6.1). Lemma 6.1. Each of the five sets in (6.3) is a clopen subset of XF. Proof. Since XF is the disjoint union of these five sets, it suffices to show that they are all open. We only do this for Xrcf, the other cases are similar. Let P ∈ Xrcf. Then there is a finite field extension L of F such that L ⊆ FP and A⊗F L (cid:27) MnP(L). Since an ordering Q ∈ XF extends to L if and only if signQ((TrL/F)∗h1i) > 0 (see [24, Chapter 3, Theorem 4.4]), the set U := {Q ∈ XF Q extends to L} is clopen in XF and contains P. Then for Q ∈ U we have L ⊆ FQ and thus A ⊗F FQ (cid:27) (A ⊗F L) ⊗L FQ (cid:27) MnP(L) ⊗L FQ (cid:27) MnP(FQ), so U ⊆ Xrcf. (cid:3) Positivecones on algebras with involution 24 Remark 6.2. The algebra DP carries an involution ϑP of the same kind as σ, and σ ⊗ idFP is adjoint to an εP-hermitian form over (DP, ϑP) with εP ∈ {−1, 1}. Note that if σ and ϑP are of the first kind, they have the same type if εP = 1. Recall from [2, Section 3.2] that P ∈ Nil[A, σ] if we can choose εP and ϑP such that εP = −1 and (DP, ϑP) ∈ {(FP, id), ((−1, −1)FP , ), (FP × FP,b), ((−1, −1)FP × (−1, −1)FP ,b)} and that P ∈ eXF = XF \ Nil[A, σ] if we can choose εP and ϑP such that εP = 1 and (6.4) (DP, ϑP) ∈ {(FP, id), ((−1, −1)FP , ), (FP( √ −1), )}. Here id denotes the (orthogonal) identity involution on FP, denotes the (unitary) conjugation involution or the (symplectic) quaternion conjugation involution and b denotes the (unitary) exchange involution. Note that the list in [2, Section 3.2] is missing the double quaternion case, an omission corrected in [3, Section 2]. 6.1 Prepositive cones in the almost split case For convenience we record the following trivial fact: Fact 6.3. Let B = diag(b1, . . . , bn) be a diagonal matrix in Mn(F). (1) Let π ∈ S n and denote the associated permutation matrix by Pπ. Then PπBPπ diag(bπ(1), . . . , bπ(n)). t = (2) Let i, j ∈ {1, . . . , n}. Then there is E ∈ Mn(F) such that EBEt is the matrix with zeroes everywhere, except for bi at coordinates ( j, j). Lemma 6.4. Let P ∈ XF and let S ∈ Mn(F) be a diagonal matrix with at least two nonzero entries of different sign with respect to P. Let ¯ε ∈ {−1, 0, 1}n. Then there is a diagonal matrix S ′ ∈ Mn(F) such that (1) S ′ is weakly represented by hSit in (Mn(F), t); (2) signP S ′ii = εi for i = 1, . . . , n. Proof. Without loss of generality (by Fact 6.3(1)) we may reorder the elements in ¯ε such that its non-zero entries are ε1, . . . , εk. For i = 1, . . . , k, by Fact 6.3(2) there are matrices E1, . . . , Ek such that S i := EiS Et i has zeroes everywhere, except for an element of sign εi at coordinates (i, i). Now take S ′ = S 1 + · · · + S k. (cid:3) The next result is a step towards the proof of Proposition 6.7 and shows that our definition of prepositive cone corresponds to positive semidefinite matrices (or negative semidefinite matrices) in all cases that are relevant. Positivecones on algebras with involution 25 Proposition 6.5. Let P ∈ XF. Let P be a prepositive cone on (Mℓ(D), ϑt) over P, where (D, ϑ) is one of (F, id), (F( √ −d), ), ((−a, −b)F, ) with a, b, d ∈ P \ {0}. Then there is ε ∈ {−1, 1} such that P ⊆ {C ∈ Sym(Mℓ(D), ϑt) ∃G ∈ GLℓ(D) such that ϑ(G)tCG ∈ Diag(εP)}. Proof. Let C ∈ P. Since C ∈ Sym(Mℓ(D), ϑt), it is the matrix of some hermitian form over (D, ϑ), so there is G ∈ GLℓ(D) such that ϑ(G)tCG is diagonal, and since P is a prepositive cone, this new matrix is also in P. Therefore, we can assume that C is diagonal, and thus has diagonal coefficients in F. Suppose for the sake of contradiction that C has two non-zero diagonal elements of different signs with respect to P. By Lemma 6.4, there are u, v ∈ P \ {0} such that the matrices diag(u, 0, . . . , 0) and diag(−v, 0, . . . , 0) belong to P. Since P is closed under products by elements of P, it follows that the matrices diag(1, 0, . . . , 0) and diag(−1, 0, . . . , 0) are both in P, contradicting that P is proper. Assume now that we have two matrices B and C in P such that ϑ(G)t BG ∈ Diag(P) and ϑ(H)tCH ∈ Diag(−P) for some G, H ∈ GLℓ(D). By the properties of P and Fact 6.3(1), we may assume that B and C are diagonal with non-zero first diagonal element. As above, it follows that there are u, v ∈ P \ {0} such that the matrices diag(u, 0, . . . , 0) and diag(−v, 0, . . . , 0) both belong to P, and thus that diag(1, 0, . . . , 0) and diag(−1, 0, . . . , 0) belong to P, contradiction. (cid:3) 6.2 Prepositive cones and positive involutions Proposition 6.6. Let P ∈ XF. The following statements are equivalent: (1) There is a prepositive cone on (A, σ) over P; (2) P ∈ eXF. Proof. (1) ⇒ (2): Assume that P ∈ Nil[A, σ]. Then the unique ordering on FP is in Nil[A ⊗F FP, σ ⊗ id] by Remark 6.2. Thus W(A ⊗F FP, σ ⊗ id) is torsion by Pfister's local-global principle [16, Theorem 4.1]. But (A⊗F FP, σ⊗id) is formally real by Proposition 5.8, which contradicts Proposition 3.7. (2) ⇒ (1): By [4, Remark 2.7] we may assume that ε = 1, i.e. that σ and ϑ are of the same type. It follows that Nil[D, ϑ] = Nil[A, σ] and so there exists a prepositive cone on (D, ϑ) over P by Example 3.13. By Theorem 4.2 there exists a prepositive cone on (A, σ) over P. (cid:3) Proposition 6.7. Let P ∈ XF and let P be a prepositive cone on (A, σ) over P. Then there exists ε ∈ {−1, 1} such that P ∩ A× ⊆ {a ∈ Sym(A, σ)× signη Phaiσ = εnP}. In particular, mP(A, σ) = nP for every P ∈ eXF. Positivecones on algebras with involution 26 Proof. Observe that P ∈ eXF by Proposition 6.6. By Proposition 5.8 there exists a prepositive cone Q on (A ⊗F FP, σ ⊗ id) such that P ⊗ 1 ⊆ Q. Recall from (6.1) that there is an isomorphism fP : A ⊗F FP → MnP(DP). We now apply part of diagram (2.1) to (A ⊗F FP, σ ⊗ id), adjusting the diagram mutatis mutandis: ( fP)∗ / sP / HermεP(MnP(DP), ϑt / Herm(MnP (DP), adΦP) Herm(A ⊗F FP, σ ⊗ id) Since Q is a prepositive cone on (A⊗F FP, σ⊗ id), we may assume that εP = 1 by Corollary 3.8, i.e. that ϑt P(ΦP) = ΦP. Thus the prepositive cone Q is transported to P fP(Q) by Proposition 4.4. Since P ∈ eXF, (DP, ϑP) is in the the prepositive cone Φ−1 list (6.4). Thus, by Proposition 6.5, there exists ε′ ∈ {−1, 1} such that ε′Φ−1 P fP(Q) only contains positive semidefinite matrices with respect to the ordering on FP. In particular, every invertible element in Φ−1 P fP(Q) will have Sylvester signature ε′nP. P) Let a ∈ P be invertible (cf. Lemma 3.6). Then there exists δ ∈ {−1, 1}, depending only on η, such that signη Phaiσ = signη⊗1ha ⊗ 1iσ⊗id = signsP◦( fP)∗(η)hΦ−1 where the final equality follows from the fact that Φ−1 element in Φ−1 P ( fP(a ⊗ 1))iϑt P ( fP(a ⊗ 1)) is an invertible Finally, if P ∈ eXF, there exists a prepositive cone on (A, σ) over P by Propo- sition 6.6 and thus, by the argument above, there is at least one element with signature nP. (cid:3) P fP(Q). = δε′nP, P The following result solves the question of the existence of positive involutions at a given ordering, a problem that seems not to have been treated as yet despite the appearance of positive involutions in the literature. Theorem 6.8. Let P ∈ XF. The following statements are equivalent: (1) There is an involution τ on A which is positive at P and of the same type as σ; (2) P ∈ eXF = XF \ Nil[A, σ]. (Recall that for τ of the same type as σ, Nil[A, σ] = Nil[A, τ].) Proof. Let τ be an involution on A of the same type as σ and let η be a tuple of reference forms for (A, τ). Thus τ = Int(b) ◦ σ for some b ∈ Sym(A, σ)× and signη Ph1iτ = signb−1η P hb−1iσ, (6.5) cf. [3, Theorem 4.2]. / Positivecones on algebras with involution 27 (1) ⇒ (2): If τ is positive at P, then hb−1iσ has nonzero signature at P by [4, Corollary 4.6] and thus P ∈ eXF. (2) ⇒ (1): If P ∈ eXF, there exists b ∈ Sym(A, σ)× such that signη′ Phb−1iσ = nP by Proposition 6.7. Therefore, τ = Int(b) ◦ σ is positive at P by (6.5) and [4, Corollary 4.6], and of the same type as σ, cf. [13, Propositions 2.7 and 2.18]. (cid:3) 7 Positive cones We remind the reader of the notation of diagram (2.1), which will be used through- out this section. 7.1 Description of positive cones In this subsection, we show that there are exactly two positive cones for a given algebra with involution over any given non-nil ordering of the base field, cf. The- orem 7.5. The following result is a reformulation of Proposition 6.7 for F-division al- gebras (D, ϑ) with involution of any kind and completely describes their positive cones. In particular M η P (D, ϑ) (cf. Definition 3.12). P (D, ϑ) and −M η P (D, ϑ) or P ⊆ −M η P (D, ϑ), M η Proposition 7.1. Let P be a prepositive cone on (D, ϑ) over P ∈ XF. Then P ⊆ M η P (D, ϑ) are the only positive cones over P, i.e. X(D,ϑ) = {−M η Lemma 7.2. CP(M g−1(η) Proof. Let B ∈ PSDℓ(M η that ϑ(G)t BG = diag(a1, . . . , aℓ) with a1, . . . , aℓ ∈ M η P (D, ϑ)). Then by Lemma 4.5 there is G ∈ GLℓ(D) such P (D, ϑ), and we may assume P (D, ϑ) P ∈ XF \ Nil[D, ϑ]}. P (D, ϑ)). (Mℓ(D), ϑt)) = PSDℓ(M η P that with a1, . . . , ar ∈ M η easy to represent ϑ(G)t BG, and thus B, as an element of CP(M g−1(η) proving that PSDℓ(M η PSDℓ(M η ϑ(G)tBG = diag(a1, . . . , ar, 0, . . . , 0) P (D, ϑ) \ {0}. Using that aiIℓ ∈ M g−1(η) P (D, ϑ)) ⊆ CP(M g−1(η) (Mℓ(D), ϑt), it is now (Mℓ(D), ϑt)), (Mℓ(D), ϑt)). The result follows since (cid:3) P (D, ϑ)) is a positive cone by Proposition 7.1 and Proposition 4.7. P P P Using the notation from Proposition 4.1 we recall [4, Definition 3.1]: Definition 7.3. Let P ∈ XF and let η be a tuple of reference forms for (A, σ). An element u ∈ Sym(A, σ) is called η-maximal at P if for every nonsingular hermitian Positivecones on algebras with involution 28 P h, where huins If Y ⊆ XF, then we say that u is η-maximal on Y if u is η-maximal at P for all form h of rank equal to the rank of huins denotes the nonsingular part of the form huiσ. P ∈ Y. σ , we have signη Phuins > signη σ σ Observe that if u is invertible, huins only if signη we have the following equivalence: Phuiσ = mP(A, σ) if and only if u ∈ M η σ = huiσ and thus u is η-maximal at P if and P (A, σ). If u is not invertible, only if u is η-maximal at P. Lemma 7.4. Let u ∈ Sym(A, σ) and let P ∈ eXF. Then u ∈ CP(M η Proof. Let k = rankhuins u ∈ CP(M η P (A, σ)) σ . We have the following sequence of equivalences: P (A, σ)) if and P (Mℓ(D), Int(Φ) ◦ ϑt)) ⇔ f (u) ∈ CP(M f∗(η) ⇔ Φ−1 f (u) ∈ CP(M s◦ f∗(η) ⇔ Φ−1 f (u) ∈ PSDℓ(M g◦s◦ f∗(η) ⇔ ∃G ∈ GLℓ(D) ∃d1, . . . , dk ∈ M g◦s◦ f∗(η) (Mℓ(D), ϑt)) (D, ϑ)) P P ϑ(G)tΦ−1 f (u)G = diag(d1, . . . , kk, 0, . . . , 0) [by Lemma 7.2] (D, ϑ) \ {0} P (D, ϑ) [by Lemma 4.5] g ◦ s ◦ f∗(huiσ) ≃ hd1, . . . , dkiϑ ⊥ 0 ⇔ ∃d1, . . . , dk ∈ M g◦s◦ f∗(η) ⇔ signη ⇔ u is η-maximal at P. Phuins P σ is maximal among all forms of rank k (cid:3) Theorem 7.5. Let η be a tuple of reference forms for (A, σ) and let P be a prepos- itive cone on (A, σ) over P ∈ XF. Then P ⊆ CP(M η P (A, σ)) or P ⊆ −CP(M η P (A, σ)). In particular X(A,σ) = {−CP(M η P (A, σ)), CP(M η P (A, σ)) P ∈ eXF} and for each P ∈ X(A,σ), there exists ε ∈ {−1, 1} such that P ∩ A× = εM η {0}. Proof. We only prove the second part of the theorem, since the first part follows from it. We start with the description of X(A,σ). Let P be a positive cone on P (A, σ)\ (A, σ) over P. By Proposition 6.6 we know that P ∈ eXF. By Proposition 7.1, there are exactly two positive cones in (D, ϑ) over P and therefore, by Theorem 4.2, there are exactly two positive cones in (A, σ) over P, necessarily P and −P. Positivecones on algebras with involution 29 Thus we only need to show that CP(M η P (A, σ)) is a positive cone in (A, σ) over P. Upon identifying (A, σ) with (Mℓ(D), adΦ) = (Mℓ(D), Int(Φ) ◦ ϑt) (without loss of generality) and using the scaling from Proposition 4.4, it suffices to show that ΦCP(M η P (A, σ)) is a positive cone on (Mℓ(D), ϑt) over P. A direct com- putation shows that ΦCP(M η P (Mℓ(D), ϑt)), which is equal to PSDℓ(M g(Φη) (D, ϑ)) by Lemma 7.2, which in turn is maximal by Proposition 7.1 and Proposition 4.10. P (A, σ)) = CP(M Φη P Finally, for P ∈ X(A,σ), we show that P ∩ A× = εM η P (A, σ) \ {0} for some ε ∈ {−1, 1}. Only the left to right inclusion is not obvious. Let u ∈ P ∩ A×. Since P = ±CP(M η P (A, σ)). By Lemma 7.4, huins σ = huiσ is η-maximal, i.e. u ∈ M η P (A, σ). P (A, σ)) we assume for instance that u ∈ CP(M η σ is η-maximal. But u is invertible, so huins (cid:3) An immediate consequence of this theorem is the following result, correspond- ing to the classical bijection between orderings on F and signatures of quadratic forms with coefficients in F: Corollary 7.6. The map εCP(M η X(A,σ) → {signµ P (A, σ)) 7→ ε signη Q Q ∈ eXF, µ a tuple of reference forms for (A, σ)} P = signεη P is a bijection (where ε ∈ {−1, 1} and P ∈ eXF). As another consequence of Theorem 7.5, we can now clarify the hypothesis used in Proposition 3.9(1): Corollary 7.7. Let P ∈ XF. The following statements are equivalent: (1) σ is positive at P; Ph1iσ = nP; (2) signη (3) there exists a prepositive cone P on (A, σ) over P such that 1 ∈ P. Proof. The first two statements are equivalent by [4, Corollary 4.6] and the last two statements are equivalent by Theorem 7.5. (cid:3) 7.2 Formally real algebras with involution In this subsection we prove the positive cone analogue of the classical Artin- Schreier theorem for fields, which states that F is formally real if and only if 0 is not a nontrivial sum of squares (see Theorem 7.9 (1) ⇔ (3)). Positivecones on algebras with involution 30 Lemma 7.8. The following statements are equivalent: (1) There is d ∈ Sym(D, ϑ)× such that hdiϑ is strongly anisotropic; (2) There is a ∈ Sym(Mℓ(D), ϑt)× such that haiϑt is strongly anisotropic. Proof. (1) ⇒ (2): Take a = diag(d, . . . , d). The result follows since g(haiϑt ) = ℓ × hdiϑ and Morita equivalence preserves (an)isotropy. (2) ⇒ (1): Let d1, . . . , dℓ ∈ Sym(D, ϑ)× such that g(haiϑt ) = hd1, . . . , dℓiϑ. By hypothesis and Morita theory, hd1, . . . , dℓiϑ is strongly anisotropic. It follows that the form hd1iϑ is also strongly anisotropic. Theorem 7.9. The following statements are equivalent: (cid:3) (1) (A, σ) is formally real; By definition (A, σ) is formally real if and only if X(A,σ) , ∅, which is equiva- (1) ⇒ (3): Let P ∈ eXF, and let d ∈ Sym(D, ϑ)× such that signη (2) There is a ∈ Sym(A, σ)× and P ∈ XF such that CP(a) ∩ −CP(a) = {0}; (3) There is b ∈ Sym(A, σ)× such that hbiσ is strongly anisotropic. Proof. (1) ⇔ (2) is clear, so we prove the equivalence of (1) and (3). lent to eXF , ∅, by Proposition 6.6. Phdiϑ = mP(D, ϑ). Assume that hdiϑ is weakly isotropic, i.e. 0 = Pk i=1 σ(xi)dxi, for some k ∈ N, xi ∈ D×. Since D is a division ring, we have σ(x1)dx1 , 0 and thus k > 2. Then −d = Pk 1 , and in particular h−d, b2, . . . , bk−1iϑ ≃ hd, . . . , diϑ for some b2, . . . , bk ∈ Sym(D, ϑ)×. We obtain a contradiction by taking signatures on both sides, since d has maximal signature. Therefore there is d ∈ Sym(D, ϑ)× such that hdiϑ is strongly anisotropic, and (3) ⇒ (1): Assume that (A, σ) is not formally real. Then eXF = ∅ and by Pfis- ter's local-global principle [16, Theorem 4.1], every hermitian form over (A, σ) is weakly hyperbolic, and in particular weakly isotropic, a contradiction. (cid:3) i=2 σ(xi x−1 1 )dxi x−1 the result follows by Lemma 7.8 and diagram (2.1). Remark 7.10. In Theorem 7.9, the element a in statement (2) obviously belongs to a prepositive cone on (A, σ). However, the element b from statement (3) may be chosen so that it does not belong to any prepositive cone, as the following example shows. Let (A, σ) = (Mn(F), t) and let b ∈ Sym(Mn(F), t)× be such that hbit is strongly anisotropic. Let q be the quadratic form over F with Gram matrix b. Assume that b belongs to some positive cone P over P ∈ XF. Thus P = εCP(M η Phbit = εmP(Mn(F), t) by Theorem 7.5. Since signη Ph1it = n, it follows that mP(Mn(F), t) = n. By definition of signη Phbit = δn, P there exists δ ∈ {−1, 1} such that signP q = signη P (Mn(F), t)) for some ε ∈ {−1, 1}, and signη Positivecones on algebras with involution 31 i.e. q is definite at P. If this can be done for every b ∈ Sym(A, σ)×, i.e. for every nonsingular quadratic form over F, we obtain that every strongly anisotropic nonsingular quadratic form over F is definite, i.e. that F is a SAP field (cf. [10] and [20]), but not every formally real field is SAP. The analogy between the field and algebra with involution cases presented in Theorem 7.9 can be completed by the following result: Proposition 7.11. The following statements are equivalent: (1) (A, σ) is formally real; (2) W(A, σ) is not torsion; (3) XF , Nil[A, σ]. Proof. (1) ⇒ (2): This is Proposition 3.7. (2) ⇒ (3): By Pfister's local-global principle [16, Theorem 4.1]. (3) ⇒ (1): By Proposition 6.6. (cid:3) 7.3 Intersections of positive cones Recall that Hilbert's 17th problem asks if a polynomial that takes only non-negative values over the reals can be represented as a sum of squares of rational functions. This question was answered in the affirmative by E. Artin and a key step in his proof consists of showing that the totally positive elements in a field are precisely the sums of squares. In [23] Procesi and Schacher considered noncommutative generalizations of Artin's result, and in [4] we proved a general sums-of-hermitian squares version of one of the main results in their paper: Theorem 7.12 ([4, Theorem 3.6]). Let b1, . . . , bt ∈ F× and consider the Harrison set Y = H(b1, . . . , bt). Let a ∈ Sym(A, σ)× be η-maximal on Y. Then the following two statements are equivalent for u ∈ Sym(A, σ): (1) u is η-maximal on Y; (2) There is s ∈ N such that u ∈ D(A,σ)(s × hhb1, . . . , btii ⊗ haiσ). Our proof uses signatures of hermitian forms and is in essence the same as the one in the field case based on Pfister's local-global principle and is straightforward when A is an F-division algebra. Lemma 7.13. Let Y = H(b1, . . . , bt) and let a ∈ Sym(A, σ)× be such that a ∈ CP(M η P (A, σ)) for every P ∈ Y (i.e. a is η-maximal on Y by Lemma 7.4). Let u ∈ Sym(A, σ). The following two statements are equivalent: Positivecones on algebras with involution 32 (1) u is η-maximal on Y; (2) For every Q ∈ X(A,σ) with QF ∈ Y, u ∈ Q if and only if a ∈ Q. Proof. (1) ⇒ (2): By Theorem 7.5, the only two prepositive cones over any Q ∈ eXF are are −CQ(M η Q(A, σ)). Since a is η-maximal on Y, the following are equivalent, for Q ∈ X(A,σ) with QF ∈ Y: Q(A, σ)) and CQ(M η a ∈ Q ⇔ Q = CQ(M η Q(A, σ)) ⇔ u ∈ Q (since u is η-maximal on Y). (2) ⇒ (1): Let P ∈ Y. By hypothesis a ∈ CP(M η P (A, σ)), i.e. u is η-maximal at P by Lemma 7.4. P (A, σ)) and by (2) we have (cid:3) u ∈ CP(M η We can now reformulate Theorem 7.12 in terms of intersections of positive cones. Observe that the element a in the next theorem plays the role of the element 1 in the field case. Theorem 7.14. Let b1, . . . , bt ∈ F×, let Y = H(b1, . . . , bt) and let a ∈ Sym(A, σ)× be such that, for every P ∈ X(A,σ) with PF ∈ Y, a ∈ P ∪ −P. Then \{P ∈ X(A,σ) PF ∈ Y and a ∈ P} = [ s∈N Proof. Let u ∈ Sym(A, σ) \ {0}. Observe that D(A,σ)(s × hhb1, . . . , btii ⊗ haiσ). u ∈ \{P ∈ X(A,σ) PF ∈ Y and a ∈ P} (7.1) if and only if for every Q ∈ X(A,σ) such that QF ∈ Y, (u ∈ Q ⇔ a ∈ Q). Indeed, assume that (7.1) holds and let Q ∈ X(A,σ) be such that QF ∈ Y. If u ∈ Q but a < Q, then a ∈ −Q and thus u ∈ −Q, contradicting that Q is proper. If a ∈ Q then u ∈ Q by choice of u. The converse is immediate. Using this observation and Lemma 7.13, we obtain that u ∈ T{P ∈ X(A,σ) PF ∈ Y and a ∈ P} if and only if u is η-maximal on Y, and the result follows by Theorem 7.12. (cid:3) Under stronger conditions, we obtain the following characterization of sums of hermitian squares in (A, σ), which reduces to Artin's theorem in the case of fields: Corollary 7.15. Assume that for every P ∈ X(A,σ), 1 ∈ P ∪ −P. Then \{P ∈ X(A,σ) 1 ∈ P} = n sX i=1 σ(xi)xi(cid:12)(cid:12)(cid:12)(cid:12) s ∈ N, xi ∈ Ao. Positivecones on algebras with involution 33 Remark 7.16. The hypothesis of Corollary 7.15 is exactly Xσ = eXF in the ter- minology of [4]. More precisely, this property characterizes the algebras with involution for which there is a positive answer to the (sums of hermitian squares version of the) question formulated by Procesi and Schacher in [23, p. 404], cf. [4, Section 4.2]. Remark 7.17. In the field case, results such as Theorem 7.14 are a direct con- sequence of the fact that if c ∈ F does not belong to some preordering T , then T [−c] is a proper preordering. In our setting of prepositive cones on F-algebras with involution such a direct approach does not seem to be obvious. 8 Signatures at positive cones Recall from [2] that to define a signature map from W(A, σ) to Z at an ordering of XF, we had to introduce a reference form η in order to resolve a sign choice at each P ∈ XF. This sign choice is reflected in the structure of X(A,σ) in the fact that there are exactly two positive cones over P ∈ eXF, P and −P (if P denotes one of them). In this section we define, directly out of P ∈ X(A,σ) over P ∈ eXF, the signature at P of hermitian forms, which we denote by signP , and show that it coincides with signη P, the definition of signP does not require a sign choice since P and −P are different points in X(A,σ) and thus both sign choices occur, one for each of the two positive cones over P. P or − signη P, cf. Proposition 8.16. In contrast to signη 8.1 Real splitting and maximal symmetric subfields Let E be a division algebra. It is well-known that any maximal subfield of E is a splitting field of E. We are interested in particular in the situation where E is equipped with a positive involution τ. In this case, any maximal symmetric sub- field of (E, τ) is a real splitting field of (E, τ), cf. Theorem 8.5 and Definition 8.6. Lemma 8.1. Let E be a division algebra with centre K and let K ⊆ L ⊆ M be subfields of E, where M is maximal. Then E ⊗K L is Brauer equivalent to a quaternion division algebra over L if and only if [M : L] = 2. Proof. Let k = [L : K] and ℓ = [M : L], then by [19, Chapter 1, §2.9, Prop., p. 139], E ⊗K L ≃ CE(L) ⊗L Mk(L), where CE(L) denotes the centralizer of L in E. Thus, dimK E = dimL CE(L) · k2. Also, E ⊗K M ≃ Mkℓ(M) since M is a splitting field of E, cf. [19, Theorem 2, p. 139]. Thus, dimK E = (ℓk)2. It follows that dimL CE(L) = ℓ2 and thus CE(L) is Positivecones on algebras with involution 34 a quaternion algebra over L if and only if ℓ = 2. Since E ⊗K L is Brauer equivalent to CE(L), the result follows. (cid:3) Lemma 8.2. Let (E, τ) be an F-division algebra with involution and let L be a subfield of Sym(E, τ), maximal for inclusion and containing F. Then either L is a maximal subfield of E or there is u ∈ Skew(E, τ) such that u2 ∈ L and L(u) is a maximal subfield of E. Proof. We assume that L is not a maximal subfield of E. Then L $ CE(L). Let v ∈ CE(L) \ L, w = (v + τ(v))/2 and u = (v − τ(v))/2. Since L ⊆ Sym(E, τ) we have τ(CE(L)) ⊆ CE(L) and so w, u ∈ CE(L). Since w + u = v < L we have w < L or u < L. If w < L, we get a contradiction since L(w) ⊆ Sym(E, τ). So u < L, but u2 ∈ L (because u2 ∈ CE(L) and is symmetric, so L(u2) is a subfield containing L and included in Sym(E, τ), so equal to L). We now check that L(u) is a maximal subfield of E. Assume that this is not the case. As above there is x ∈ CE(L(u)) \ L(u), and since τ(L(u)) = L(u) we have τ(x) ∈ CE(L(u)) and thus x can be written as y + z with τ(y) = y, τ(z) = −z and y, z ∈ CE(L(u)). Since x < L(u) we have either y < L(u) or z < L(u). If y < L(u) then L $ L(y) ⊆ Sym(E, τ), impossible. So z < L(u). Then uz = zu (since z ∈ CE(L(u))), uz < L (since z < L(u)) and uz is symmetric. Therefore L $ L(uz) ⊆ Sym(E, τ), a contradiction. Lemma 8.3. Let P ∈ XF and let (B, τ) be an F-algebra with involution such that τ is positive at P. Let L be a subfield of Sym(B, τ), maximal for inclusion and containing F. Then P extends to an ordering on L. (cid:3) Proof. Since τ is positive at P, by Corollary 7.7 there is P ∈ X(B,τ) over P such that 1 ∈ P and it follows that P ⊆ P by Proposition 3.9. Since, for p ∈ P and x ∈ L, px2 = τ(x)px, the preordering generated by P in L is contained in P, so is proper, and is therefore included in an ordering on L. (cid:3) Proposition 8.4. Let (E, τ) be an F-division algebra with involution of the second kind. Let d ∈ F× be such that Z(E) = F( √ −d). Then Nil[E, τ] = H(−d) = {P ∈ XF P extends to Z(E)}, where H(−d) denotes the usual Harrison set. Proof. Observe that for P ∈ XF, E ⊗F FP (cid:27) E ⊗Z(E) (Z(E) ⊗F FP) (cid:27)  E ⊗Z(E) FP( √ E ⊗Z(E) (FP × FP) −d) ∼ FP( √ −1) if d ∈ P if d ∈ −P (8.1) . Positivecones on algebras with involution 35 Assume first that P extends to Z(E), i.e. that −d ∈ P. By (8.1), E ⊗F FP is an FP- algebra which is not simple. By [2, Lemma 2.1(iv)] W(E ⊗F FP, τP) = 0, implying that P ∈ Nil[E, τ]. Conversely, assume that P ∈ Nil[E, τ], but that d ∈ P. Then E ⊗F FP ∼ FP( √ −1). Using [5, (2.1)] this contradicts [5, Proposition 2.2(1)]. (cid:3) Theorem 8.5. Let (E, τ) be an F-division algebra with involution, and let L be a subfield of Sym(E, τ), maximal for inclusion and containing F. Then one of the following holds: (1) L is a maximal subfield of E and so E ⊗F L ∼ L. This can only occur when τ is of the first kind. (2) There exists u ∈ Skew(E, τ)× such that u2 ∈ L and L(u) is a maximal subfield of E. (a) If τ is of the first kind, E⊗F L is Brauer equivalent to a quaternion division (b) If τ is of the second kind and Z = F( √ −d), we may take u = √ algebra (u2, c)L for some c ∈ L. E ⊗F L ∼ L( √ −d). −d and Furthermore, if τ is positive at P ∈ XF, then there is an ordering Q on L that extends P, τ is orthogonal in (1) and τ is symplectic in (2a). In addition, for every ordering Q on L extending P, we have u2, c ∈ −Q in (2a), and d ∈ Q in (2b). Proof. If L is a maximal subfield of E, we must have Z(E) = F since L ⊆ Sym(E, τ), and so τ is of the first kind and E ⊗F L ∼ L. If L is not a maximal subfield of E, there exists u ∈ Skew(E, τ)× such that u2 ∈ L and L(u) is a maximal subfield of E by Lemma 8.2. Suppose that τ is of the first kind, so that Z(E) = F. Since [L(u) : L] = 2, E⊗F L is Brauer equivalent to a quaternion division algebra B over L by Lemma 8.1. Since L(u) is a maximal subfield of E, we have (E⊗F L)⊗L L(u) (cid:27) E⊗F L(u) ∼ L(u) and so B ⊗L L(u) ∼ L(u). By [14, Chapter III, Theorem 4.1], there exists c ∈ L such that B (cid:27) (u2, c)L. −d) for some d ∈ F× and τ( √ √ −d. Since √ −d ∈ L(u) \ L and [L(u) : L] = 2, we have −d) = − L(u) = L( √ −d. Since L( √ −d) and we take u = √ −d) is a maximal subfield of E, it follows that Suppose that τ is of the second kind. Then Z(E) = F( √ E⊗F L (cid:27) E⊗Z(E)(Z(E)⊗F L) (cid:27) E⊗Z(E)(F( √ Finally, assume that τ is positive at P. −d). In particular, P < Nil[E, τ]. By Lemma 8.3 there exists an ordering Q on L that extends P. In particular, Q < Nil[E ⊗F L, τ⊗ idL]. By Remark 6.2 it follows that τ must be orthogonal in (1) and −d)⊗F L) (cid:27) E⊗Z(E) L( √ −d) ∼ L( √ Positivecones on algebras with involution 36 symplectic in (2a). Furthermore, for any ordering Q on L that extends P we have: in (2a), (u2, c)L ⊗L LQ must be division and so u2, c ∈ −Q; in (2b), if −d ∈ Q, then P extends to Z(E) and so P ∈ Nil[E, τ] by Proposition 8.4, contradiction. Definition 8.6. Let (B, τ) be an F-algebra with involution and let P ∈ XF. An ordered extension (L, Q) of (F, P) is called a real splitting field of (B, τ) and we say that (L, Q) real splits (B, τ) if B ⊗F L is Brauer equivalent to one of the L- algebras L, L( √ −d) or (−a, −b)L, where a, b, d ∈ Q and the involution τ ⊗ id is positive at Q. (cid:3) Remark 8.7. The condition that τ ⊗ id be positive at Q is equivalent to τ being positive at P and is also equivalent to τ⊗id being adjoint to a quadratic or hermitian form that is positive definite at Q. Proposition 8.8. Let (E, τ) be an F-division algebra with involution and let P ∈ XF. Assume that τ is positive at P. Let L be a subfield of Sym(E, τ), maximal for inclusion and containing F. Then there is a Galois extension N of F such that L ⊆ N, P extends to N and for every extension Q of P to N, we have that (N, Q) is a real splitting field of (E, τ). In addition, [N : F] 6 (deg E)!. Proof. Observe that, under the hypothesis for L and τ and by Theorem 8.5, E ⊗F L (cid:27) Mt(E0) where E0 is one of L, L( √ −d), (−a, −b)L, and for every ordering P′ on L extending P, d, a, b ∈ P′. Moreover P extends to L, so we fix a real closure FP of F at P such that L ⊆ FP. Write L = F(a) for some a ∈ Sym(E, τ). Since τ is positive at P, (E ⊗F FP, τ⊗ id) is isomorphic to a matrix algebra over one of FP, FP( √ −1) or (−1, −1)FP , equipped with the (conjugate) transpose involution. We denote this isomorphism by λ. Then the matrix λ(a ⊗ 1) is symmetric with respect to conjugate transposi- tion. Thus λ(a ⊗ 1) is congruent to a diagonal matrix with entries in FP, cf. [15, Theorem 9] and all the roots of the reduced characteristic polynomial of a (which has coefficients in F) are in FP. Hence all the roots of the minimal polynomial mina of a over F are in FP. Let N be the splitting field of mina in FP. Then L ⊆ N and N/F is a Galois extension. Since N ⊆ FP, the ordering P extends to N. Now let Q be any ordering on N extending P. Then P′ := Q∩ L is an ordering on L extending P and, as mentioned above, E ⊗F L (cid:27) Mt(E0) with d, a, b ∈ P′. Therefore E ⊗F N (cid:27) Mt(E0) with d, a, b ∈ Q. Concerning [N : F], observe that F(a) is a subfield of E and therefore [F(a) : F] 6 deg E by the Centralizer Theorem, cf. [19, §2.9]. The conclusion follows. (cid:3) Positivecones on algebras with involution 37 8.2 The signature at a positive cone Let P ∈ X(A,σ) be a positive cone over P ∈ XF. Recall that P ∈ eXF by Theo- rem 7.5. Our first objective is to show the following: Theorem 8.9. Let h be a nonsingular hermitian form over (A, σ). Then there are c ∈ P ∩ A× and β1, . . . , βt ∈ P× such that n2 P × hβ1, . . . , βti ⊗ h ≃ (ha1, . . . , ari ⊥ hb1, . . . , bsi) ⊗ hciσ with a1, . . . , ar ∈ P×, b1, . . . , bs ∈ −P×, t 6 2nP(deg D)! and where nP is the matrix size of A ⊗F FP. Observe that the hypothesis that h is nonsingular in Theorem 8.9 is not restric- tive, cf. Proposition 4.1. In order to prove Theorem 8.9 we need the following three lemmas. Lemma 8.10. Assume (D, ϑ) ∈ {(F, id), (F( √ P× and let h be a nonsingular hermitian form over (Mℓ(D), ϑt). Then −d), −), ((−a, −b)F, −)} where a, b, d ∈ ℓ2 × h ≃ (ha1, . . . , ari ⊥ hb1, . . . , bsi) ⊗ hIℓiϑt for some a1, . . . , ar ∈ P× and b1, . . . , bs ∈ −P×. Proof. Recall that by Lemma 2.2, the hermitian form ℓ × h is diagonal with co- efficients in Sym(Mℓ(D), ϑt)×, so we can assume without loss of generality that h = hαiϑt with α ∈ Sym(Mℓ(D), ϑt)×. Observe that X(D,ϑ) = {R, −R R ∈ XF} since Sym(D, ϑ) = F and thus, by Proposition 4.10, X(Mℓ (D),ϑt) = {PSDℓ(R), − PSDℓ(R) R ∈ XF}. Without loss of generality we can assume that P = PSDℓ(P). Diagonalizing the matrix α by congruences, we obtain h ≃ hα′iϑt where α′ = diag(u1, . . . , uℓ) with u1, . . . , uℓ ∈ F×. Let g be the collapsing Morita equivalence from (2.1), then g(h) = hu1, . . . , uℓiϑ and ℓ × g(h) ≃ hu1, . . . , u1iϑ ⊥ · · · ⊥ huℓ, . . . , uℓiϑ. Applying g−1 we obtain ℓ × h ≃ hu1Iℓ, . . . , uℓIℓiϑt = hu1, . . . , uℓi ⊗ hIℓiϑt . (cid:3) Positivecones on algebras with involution 38 Let L/F be a finite field extension. The trace map TrL/F induces an A-linear homomorphism TrA⊗F L = idA ⊗ TrL/F : A ⊗F L → A. Let (M, h) be a hermitian form over (A ⊗F L, σ ⊗ id). The Scharlau transfer Tr∗(M, h) is defined to be the hermitian form (M, TrA⊗F L ◦h) over (A, σ). Let q be a quadratic form over L, then Frobenius reciprocity, i.e. Tr∗(q ⊗ (h ⊗ L)) ≃ Tr∗(q) ⊗ h (where Tr∗(q) is the usual Scharlau transfer of q with respect to TrL/F) can be proved just as for quadratic forms, cf. [14, VII, Theorem 1.3]. For L/F as above and P ∈ XF, we denote the set of orderings on L that extend P by XL/P. Lemma 8.11. Assume that σ is positive at P and that D ∈ {F, F( √ −d), (−a, −b)F} where a, b, d ∈ P×. Let h be a nonsingular hermitian form over (A, σ). Then there are a1, . . . , ar ∈ P×, b1, . . . , bs ∈ −P× and α1, . . . , α2k ∈ F× such that ℓ2 × hα1, . . . , α2ki ⊗ h ≃ (ha1, . . . , ari ⊥ hb1, . . . , bsi) ⊗ h1iσ, where k 6 ℓ. Furthermore, for every ordering Q ∈ XF such that σ is positive at Q and either d ∈ Q in case D = F( √ −d), or a, b ∈ Q in case D = (−a, −b)F, we have α1, . . . , α2k ∈ Q. Proof. Let σ = adϕ for some form ϕ over (D, ϑ), where ϑ is as in Lemma 8.10. Since σ is positive at P, we may assume that ϕ = hd1, . . . , dℓi with d1, . . . , dℓ ∈ P×. Up to renumbering, let {d1, . . . , dk} be a minimal subset of {d1, . . . , dℓ} such that F( √d1, . . . , √dk) = F( √d1, . . . , √dℓ) =: L. Since the extension L/F is obtained by successive proper quadratic extensions by elements of P we have XL/P = 2k. After scalar extension to L we obtain (A⊗F L, σ⊗id) (cid:27) (Mℓ(D⊗F L), adϕ⊗L) (cid:27) (Mℓ(D⊗F L), adh1,...,1iϑ) = (Mℓ(D⊗F L), ϑt). Observe that D ⊗F L is an element of {L, L( √ −d), (−a, −b)L} and is an L-division algebra since a, b, d ∈ P×. Thus, by Lemma 8.10 (with F replaced by L) there are u1, . . . , ut ∈ L× such that ℓ2 × (h ⊗ L) ≃ hu1, . . . , uti ⊗ h1iϑt . Applying the Scharlau transfer induced by the trace map TrL/F gives ℓ2 × Tr∗(h ⊗ L) ≃ Tr∗(hu1, . . . , uti ⊗ h1iϑt ). Thus, by Frobenius reciprocity and since σ ⊗ id (cid:27) ϑt, we obtain ℓ2 × Tr∗(h1i) ⊗ h ≃ Tr∗(hu1, . . . , uti) ⊗ h1iσ. Positivecones on algebras with involution 39 Let hα1, . . . , α2ki be a diagonalization of the form Tr∗(h1i). Since L/F is separable, TrL/F is nonzero and it follows from [14, VII, Proposition 1.1] that Tr∗(hu1, . . . , uti) is a nonsingular quadratic form over F. Therefore we can separate the coefficients of a diagonalization of Tr∗(hu1, . . . , uti) into elements of P× and elements of −P×. Finally, let Q ∈ XF be such that σ is positive at Q and either d ∈ Q in case D = F( √ −d), or a, b ∈ Q in case D = (−a, −b)F. We have d1, . . . , dℓ ∈ Q and, as above, XL/Q = 2k. By the quadratic Knebusch trace formula [24, Chap- ter 3, Theorem 4.5], signQ(Tr∗(h1i)) = 2k = [L : F] = dim Tr∗(h1i), proving that α1, . . . , α2k ∈ Q. Lemma 8.12. Let h be a nonsingular hermitian form over (A, σ) where σ is a positive involution at P. Then there are β1, . . . , βt ∈ P× such that (cid:3) n2 P × hβ1, . . . , βti ⊗ h ≃ (ha1, . . . , ari ⊥ hb1, . . . , bsi) ⊗ h1iσ with a1, . . . , ar ∈ P×, b1, . . . , bs ∈ −P× and t 6 2nP(deg D)!. Proof. By Theorem 6.8 there exists a positive involution on D. Thus, by Propo- sition 8.8 there exists a finite Galois extension N of F such that P extends to an ordering Q of N and such that (N, Q) real splits (A, σ). In particular we can take N ⊆ FP. Therefore we can apply Lemma 8.11 to (A ⊗F N, σ ⊗ id) and we obtain M2 × hα1, . . . , α2ki ⊗ (h ⊗ N) ≃ (hu1, . . . , uri ⊥ hv1, . . . , vsi) ⊗ h1iσ⊗id, where M is the matrix size of A⊗F N, u1, . . . , ur ∈ Q×, v1, . . . , vs ∈ −Q×, α1, . . . , α2k are as given in Lemma 8.11 and k 6 M. Observe that in fact M = nP (as defined in (6.1)) since (N, Q) real splits (A, σ), so that the passage from A⊗F N to A⊗F FP does not change the degree of the underlying division algebra. Furthermore, since σ is positive at P, σ is positive at every R ∈ XN/P and by Proposition 8.8, (N, R) is a real splitting field of (A, σ). Therefore, by Lemma 8.11, α1, . . . , α2k ∈ R for every R ∈ XN/P. Applying the Scharlau transfer induced by the trace map TrN/F and Frobenius reciprocity give n2 P × Tr∗(hα1, . . . , α2ki) ⊗ h ≃ Tr∗(hu1, . . . , ur, v1, . . . , vsi) ⊗ h1iσ. Observe now that dim Tr∗(hα1, . . . , α2ki) = 2k[N : F] and that, since N is an ordered Galois extension of (F, P), the number XN/P of extensions of P to N is [N : F] (this is a direct consequence of [22, Corollary 1.3.19] and the fact that the extension is Galois). In particular, and using that α1, . . . , α2k belong to every R ∈ XN/P, we obtain, using the quadratic Knebusch trace formula [24, Chapter 3, Theorem 4.5], that signP Tr∗(hα1, . . . , α2ki) = X R∈XN/P signPhα1, . . . , α2ki = 2k[N : F]. Positivecones on algebras with involution 40 It follows that if we write Tr∗(hα1, . . . , α2ki) ≃ hβ1, . . . , βti with β1, . . . , βt ∈ F×, we must have β1, . . . , βt ∈ P× since t = 2k[N : F]. Therefore n2 P × hβ1, . . . , βti ⊗ h ≃ Tr∗(hu1, . . . , ur, v1, . . . , vsi) ⊗ h1iσ. The result follows and the bound on t is obtained from Proposition 8.8. (cid:3) Proof of Theorem 8.9. Let γ be an involution on A, positive at P, and let c ∈ Sym(A, σ)× be such that γ = Int(c−1) ◦ σ. It follows from [4, Remark 4.3 and Proposition 4.4] and Theorem 7.5 that c ∈ P ∩ A× or c ∈ −(P ∩ A×). By Proposition 4.4, c−1P is a positive cone on (A, γ) over P. Furthermore, c−1h is a hermitian form over (A, γ) by scaling. So by Lemma 8.12, there are β1, . . . , βt ∈ P× such that P × hβ1, . . . , βti ⊗ c−1h ≃ (ha1, . . . , ari ⊥ hb1, . . . , bsi) ⊗ h1iγ n2 with a1, . . . , ar ∈ P× and b1, . . . , bs ∈ −P×. Scaling by c gives n2 P × hβ1, . . . , βti ⊗ h ≃ (ha1, . . . , ari ⊥ hb1, . . . , bsi) ⊗ hciσ and the result follows since c ∈ P ∩ A× or c ∈ −(P ∩ A×). (cid:3) The following corollary is an immediate weaker version of Theorem 8.9. Corollary 8.13. Let h be a nonsingular hermitian form over (A, σ). Then there exist β1, . . . , βt ∈ P× (with t 6 2nP(deg D)!) , a1, . . . , ar ∈ P ∩ A× and b1, . . . , bs ∈ −(P ∩ A×) such that n2 P × hβ1, . . . , βti ⊗ h ≃ ha1, . . . , ariσ ⊥ hb1, . . . , bsiσ. A result in the style of Sylvester inertia is immediate: Lemma 8.14. Let h be a hermitian form over (A, σ) and suppose h ≃ ha1, . . . , ariσ ⊥ hb1, . . . , bsiσ ≃ ha′1, . . . , a′piσ ⊥ hb′1, . . . , b′qiσ with a1, . . . , ar, a′1, . . . , a′p ∈ P ∩ A× and b1, . . . , bs, b′1, . . . , b′q ∈ −(P ∩ A×). Then r = p and s = q. Proof. For dimension reasons we have r + s = p + q. The result will follow if we show r − s = p − q, i.e. r + q = p + s. Assume for instance that r + q > p + s. We have the following equality in W(A, σ): ha1, . . . , ar, −b′1, . . . , −b′qiσ = ha′1, . . . , a′p, −b1, . . . , −bsiσ. Positivecones on algebras with involution 41 Relabelling the entries gives hα1, . . . , αr+qiσ = hβ1, . . . , βp+siσ, with αi, β j ∈ P ∩ A×. Since r + q > p + s there is i ∈ {1, . . . , r + q} such that hαi+1, . . . , αr+qiσ ≃ h−α1, . . . , −αi, β1, . . . , βp+siσ. If follows that −α1 ∈ D(A,σ)(hαi+1, . . . , αr+qiσ) ⊆ P, contradicting that P is proper. (cid:3) We use Corollary 8.13 to define the signature at P. Definition 8.15. Let h be a nonsingular hermitian form over (A, σ), P ∈ X(A,σ) and let t, βi, a j and bk be as in Corollary 8.13. We define the signature of h at P by signP h := r − s nPt . Corollary 8.13 suggests that signP h should be a multiple of r− s by some con- stant. The reason for the particular normalization applied to r − s in the definition above is given by: Proposition 8.16. Let η be a reference form for (A, σ). Then there is εP ∈ {−1, 1} such that signP = εP signη P. Proof. Let h be a nonsingular hermitian form over (A, σ) and consider n2 P × hβ1, . . . , βti ⊗ h ≃ ha1, . . . , ariσ ⊥ hb1, . . . , bsiσ as in Corollary 8.13. By Theorem 7.5 there exists εP ∈ {−1, 1} such that P ∩ A× = εPM η Phaiiσ = εPnP, signη P h = εPnP(r − s), which proves the result. P (A, σ)\{0}. Applying the map signη Phb jiσ = −εPnP we obtain n2 P to this isometry, and since signη Pt signη (cid:3) Remark 8.17. Observe that for P ∈ X(A,σ), sign −P = − signP . Furthermore, the following are immediate consequences of Proposition 8.16 and [3, Theorem 2.6]: let h1, h2 be hermitian forms over (A, σ), let q be a quadratic form over F and let P ∈ X(A,σ) be over P ∈ XF, then signP(h1) = 0 if h1 is hyperbolic, signP (h1 ⊥ h2) = signP h1 + signP h2, signP (q ⊗ h1) = (signP q)(signP h1). In particular, the map signP is well-defined on W(A, σ). After introducing the topology Tσ on X(A,σ) we will show in the final section that the total signature map sign h : X(A,σ) → Z, P 7→ signP h is continuous, cf. Theorem 9.6. Positivecones on algebras with involution 42 9 The topology of X(A,σ) One of the main objectives of this section is to show that the natural topology on X(A,σ) is spectral, cf. Theorem 9.12. Spectral topologies were studied in great detail by Hochster [11] in order to completely describe the topology on Spec(A) for any commutative ring A. We also refer to [18, Section 6.3] and the recent monograph on spectral spaces [9]. Let (X, T ) be a topological space. We denote by B(T ) the set of all subsets of X that are compact open in T . (Note that for us compact means quasicompact.) The space (X, T ) is spectral if it is T0 and compact, B(T ) is closed under finite intersections and forms an open basis of T , and every nonempty irreducible closed subset has a generic point, cf. [11, Section 0]. We define the patch topology on X, denoted Tpatch, as the topology with sub- basis {U, X \ V U, V ∈ B(T )}, cf. [11, Sections 2 and 8]. Observe that if X is compact, this subbasis can be replaced by the subbasis {U \ V U, V ∈ B(T )}, which is actually a basis of Tpatch if B(T ) is in addition closed under finite unions and finite intersections. The interesting aspects of a spectral topology derive from the properties of Tpatch and its links with T . We define, for a1, . . . , ak ∈ Sym(A, σ), Hσ(a1, . . . , ak) := {P ∈ X(A,σ) a1, . . . , ak ∈ P}. We denote by Tσ the topology on X(A,σ) generated by the sets Hσ(a1, . . . , ak), for a1, . . . , ak ∈ Sym(A, σ), and by T ×σ the topology on X(A,σ) generated by the sets Hσ(a1, . . . , ak), for a1, . . . , ak ∈ Sym(A, σ)×. Recall that we denote the usual Har- rison topology on XF or eXF by TH. We first show that the topologies Tσ and T ×σ are equal. Lemma 9.1. (1) Let a ∈ Sym(A, σ)×. The scaling map X(A,σ) → X(A,Int(a)◦σ), P 7→ aP is a homeomorphism in the following two cases, where (a) X(A,σ) is equipped with Tσ and X(A,Int(a)◦σ) with TInt(a)◦σ; (b) X(A,σ) is equipped with T ×σ and X(A,Int(a)◦σ) with T ×Int(a)◦σ. (2) The topologies Tϑt and T ×ϑt on X(Mℓ(D),ϑt) are equal. Positivecones on algebras with involution 43 Proof. (1) Since the inverse of this map is of the same type, it suffices to prove that it is continuous. The result is clear in both cases since the inverse image of HInt(a)◦σ(a1, . . . , ak) is Hσ(a−1a1, . . . , a−1ak), and a−1ai is invertible if and only if ai is. (2) We consider the set Hϑt (a) for some a ∈ Sym(Mℓ(D), ϑt). By (P3) we may assume that a is diagonal. Therefore, let r ∈ N, let d1, . . . , dr ∈ D× and let P ∈ X(Mℓ (D),ϑt). Then, using (P2) and (P3), we obtain diag(d1, . . . , dr, 0, . . . , 0) ∈ P ⇔ d1Iℓ, . . . , drIℓ ∈ P. Thus, Hϑt (diag(d1, . . . , dr, 0, . . . , 0)) = Hϑt (d1Iℓ) ∩ · · · ∩ Hϑt (drIℓ) and the result follows. (cid:3) Proposition 9.2. The topologies Tσ and T ×σ on X(A,σ) are equal. Proof. Use the homeomorphisms from Lemma 9.1(1) to bring both topologies to X(Mℓ (D),ϑt) on which they are equal by Lemma 9.1(2). (cid:3) As a consequence of this proposition, we may use the sets Hσ(a1, . . . ak) as a basis of open sets for Tσ = T ×σ with a1, . . . , ak in Sym(A, σ) or in Sym(A, σ)×, whichever is more convenient for the problem at hand. From now on, and unless specified otherwise, we assume that X(A,σ) is equipped with the topology Tσ and that eXF is equipped with the topology TH. Lemma 9.3. The map XF → Z, P 7→ mP(A, σ) (cf. Definition 3.12) is continuous. Proof. Since the five clopen sets defined in (6.3) cover XF, it suffices to show that the map P 7→ mP(A, σ) is continuous on each of them. Let U be one of these clopen sets. We know from Proposition 6.7 that mP(A, σ) = nP, and from the observation after (6.3) that the value of nP is constant on U. The map is then constant on U and therefore continuous. (cid:3) Lemma 9.4. There are c1, . . . , ct ∈ Sym(A, σ)× such that X(A,σ) = Hσ(c1) ∪ · · · ∪ Hσ(ct). Let P ∈ eXF. By definition there exists cP ∈ Sym(A, σ)× such that signη Proof. For c ∈ Sym(A, σ)× we define Hmax(c) := {Q ∈ eXF signη By Lemma 9.3 and [2, Theorem 7.2] the set Hmax(c) is clopen in eXF. PhcPiσ = mP(A, σ), i.e., P ∈ Hmax(cP). Therefore eXF = Sc∈Sym(A,σ)× Hmax(c) and by com- pactness of eXF we get eXF = Hmax(c1) ∪ · · · ∪ Hmax(cr) for some c1, . . . , cr ∈ Sym(A, σ)×. It follows from Theorem 7.5 that X(A,σ) = Hσ(c1) ∪ · · · ∪ Hσ(cr) ∪ Hσ(−c1) ∪ · · · ∪ Hσ(−cr). Qhciσ = mQ(A, σ)}. (cid:3) Positivecones on algebras with involution 44 Remark 9.5. It will be convenient in several proofs in this section to express the property u ∈ PF in terms of some element belonging to P. Observe that if c ∈ P \ {0} and u ∈ F \ {0}, then u ∈ PF implies uc ∈ P, and u ∈ −PF implies uc ∈ −P. Therefore u ∈ PF is equivalent to uc ∈ P. Moreover, the choice of c is essentially independent of P. Indeed, by Lemma 9.4 we have X(A,σ) = Hσ(c1)∪· · ·∪ Hσ(ct) for some c1, . . . , ct ∈ Sym(A, σ)×. Thus, for P ∈ Hσ(c j) and u ∈ F \ {0} we have u ∈ PF if and only if uc j ∈ P. Theorem 9.6. Let h be a nonsingular hermitian form over (A, σ). Then the to- tal signature map sign h : X(A,σ) → Z, P 7→ signP h (cf. Definition 8.15) is continuous, where Z is equipped with the discrete topology. Proof. By Lemma 2.2, ℓ × h ≃ ha1, . . . , akiσ for some a1, . . . , ak ∈ Sym(A, σ)×. Therefore (sign h)−1(i) = (signha1iσ + · · · + signhakiσ)−1(ℓi) for any i ∈ Z and it suffices to show that signhaiσ is continuous, for a ∈ Sym(A, σ)×. By Lemma 9.4 we have X(A,σ) = Hσ(c1) ∪ · · · ∪ Hσ(ct) for some c1, . . . , ct ∈ Sym(A, σ)× and so it is sufficient to show that the map signhaiσ is continuous on each Hσ(c j). By Definition 8.15 we have signPhaiσ = i if and only if there exist r, s, t ∈ N, β1, . . . , βt ∈ PF \ {0} and a1, . . . , ar, b1, . . . , bs ∈ P ∩ A× such that (r − s)/(nPF t) = i and (nPF )2 × hβ1, . . . , βti ⊗ haiσ ≃ ha1, . . . , ariσ ⊥ h−b1, . . . , −bsiσ. Therefore (signhaiσ)−1(i) ∩ Hσ(c j) = {P ∈ Hσ(c j) ∃r, s, t ∈ N, a1, . . . , ar, b1, . . . , bs ∈ P ∩ A×, β1, . . . , βt ∈ PF \ {0} such that (r − s)/nPF t = i, and (nPF )2 × hβ1, . . . , βti ⊗ haiσ ≃ ha1, . . . , ariσ ⊥ h−b1, . . . , −bsiσ} = [{Hσ(c j, β1c j, . . . , βtc j, a1, . . . , ar, b1, . . . , bs) r, s, t ∈ N, a1, . . . , ar, b1, . . . , bs ∈ Sym(A, σ)×, β1, . . . , βt ∈ F× such that (r − s)/nPF t = i, and (nPF )2 × hβ1, . . . , βti ⊗ haiσ ≃ ha1, . . . , ariσ ⊥ h−b1, . . . , −bsiσ} is open for Tσ (note that the second equality follows from Remark 9.5, since for P ∈ Hσ(c j) and u ∈ F \ {0} we have u ∈ PF if and only if uc j ∈ P). (cid:3) We now introduce the maps π : X(A,σ) → eXF, P 7→ PF, ξ : eXF → X(A,σ), P 7→ CP(M η P (A, σ)), with reference to Theorem 7.5 for the definition of ξ. Positivecones on algebras with involution 45 Proposition 9.7. (1) The map π is open. (2) The map π is continuous. Proof. (1) Let a1, . . . , ak ∈ Sym(A, σ)×. Then, using Theorem 7.5, π(Hσ(a1, . . . , ak)) = {P ∈ eXF signη Phaiiσ = mP(A, σ), i = 1, . . . , k}, which is open since P 7→ mP(A, σ) and P 7→ signη and [2, Theorem 7.2]. Phaiσ are continuous by Lemma 9.3 (2) Let u ∈ F. We show that π−1(H(u)) is open. Observe that π−1(H(u)) = {P ∈ X(A,σ) u ∈ PF}. Using Remark 9.5 we have for P ∈ Hσ(c j) that u ∈ PF if and only if P ∈ Hσ(uc j). Therefore, π−1(H(u)) = t[ j=1 π−1(H(u)) ∩ Hσ(c j) = t[ j=1 Hσ(c j, uc j), which is in Tσ. Proposition 9.8. (cid:3) (1) The maps ξ and −ξ are open. (2) Let b ∈ Sym(A, σ)×. Then ξ−1(Hσ(b)) is clopen in TH. In particular ξ is continuous. Proof. (1) Let u1, . . . , uk ∈ F and let ε ∈ {−1, 1}. We show that εξ(H(u1, . . . , uk)) is open for Tσ. By definition εξ(H(u1, . . . , uk)) = {εCP(M η P (A, σ)) P ∈ H(u1, . . . , uk)}. By Lemma 9.4 it suffices to show that εξ(H(u1, . . . , uk)) ∩ Hσ(ci) is open for i = 1, . . . , t. We have εξ(H(u1, . . . , uk)) ∩ Hσ(ci) = {P ∈ X(A,σ) P = εCP(M η = {P ∈ X(A,σ) signη P (A, σ)), P ∈ H(u1, . . . , uk), ci ∈ P} PFhciiσ = εmPF (A, σ), P ∈ Hσ(u1ci, . . . , ukci), P ∈ Hσ(ci)} [by Remark 9.5 and Theorem 7.5] = {P ∈ X(A,σ) signη PFhciiσ = εmPF (A, σ)} ∩ Hσ(ci, u1ci, . . . , ukci), Positivecones on algebras with involution 46 the first set is open since the maps π and P 7→ signη Proposition 9.7(2) and [2, Theorem 7.2]. which is open. Indeed, the second set in the intersection is open by definition and Phciiσ are both continous, by Phbiσ = mP(A, σ)}, which is clopen since the map XF → Z, P 7→ mP(A, σ) is continuous by Lemma 9.3 and the signature map is also continuous by [2, Theorem 7.2]. (2) Let b ∈ Sym(A, σ)×. Then ξ−1(Hσ(b)) = {P ∈ eXF signη (cid:3) Proposition 9.9. For every open set U of eXF, π−1(U) = (π−1(U) ∩ Im ξ) ∪ (π−1(U) ∩ − Im ξ), map. and ππ−1(U)∩ε Im ξ is a homeomorphism from π−1(U) ∩ ε Im ξ to U, for ε ∈ {−1, 1}. In particular, πIm ξ and π− Im ξ are homeomorphisms onto eXF and π is a covering Proof. The first statement is obvious since X(A,σ) = Im ξ∪− Im ξ. That ππ−1(U)∩ε Im ξ is a homeomorphism follows from the fact that it is continous by Proposition 9.7(2) and has inverse εξ, which is also continuous by Proposition 9.8(2). The remaining statements follow immediately. (cid:3) Corollary 9.10. X(A,σ) is homeomorphic to a disjoint union of two copies of eXF. Proof. The statement follows from Propositions 9.8(1) and 9.9. (cid:3) Proposition 9.11. (1) The sets Hσ(a1, . . . , ak) are compact for any k ∈ N and a1, . . . , ak ∈ Sym(A, σ). (2) The compact open subsets of Tσ are precisely the finite unions of sets of the form Hσ(a1, . . . , ak), for k ∈ N and a1, . . . , ak ∈ Sym(A, σ)× (or a1, . . . , ak ∈ Sym(A, σ)). (3) X(A,σ) is compact. Proof. (1) Assume that Hσ(a1, . . . , ak) = Si∈I Hσ(¯bi) for some finite tuples ¯bi of elements of Sym(A, σ). Then Hσ(a1, . . . , ak) ∩ Im ξ = Si∈I Hσ(¯bi) ∩ Im ξ, and applying π we get π(Hσ(a1, . . . , ak) ∩ Im ξ) = [ i∈I π(Hσ(¯bi) ∩ Im ξ), where π(Hσ(a1, . . . , ak) ∩ Im ξ) = ξ−1(Hσ(a1, . . . , ak)) and π(Hσ(¯bi) ∩ Im ξ) = ξ−1(Hσ(¯bi)). Furthermore, these two sets are clopen by Proposition 9.8(2), and since TH is compact there is a finite subset I1 of I such that π(Hσ(a1, . . . , ak) ∩ Im ξ) = [ π(Hσ(¯bi) ∩ Im ξ). i∈I1 Positivecones on algebras with involution 47 Similarly, with −ξ instead of ξ, there is a finite subset I2 of I such that π(Hσ(a1, . . . , ak) ∩ − Im ξ) = [ π(Hσ(¯bi) ∩ − Im ξ). i∈I2 It follows that proving the result. Hσ(a1, . . . , ak) = [ Hσ(¯bi), i∈I1∪I2 (2) If U is open in Tσ, then U is a union of open sets of the form Hσ(a1, . . . , ak), and if U is compact, this union can be taken finite (with a1, . . . , ak in Sym(A, σ)× or Sym(A, σ), cf. Proposition 9.2). The other part of the statement is immediate since the sets Hσ(a1, . . . , ak) are compact in Tσ by (1). (3) The statement follows immediately from (1) and Lemma 9.4. (cid:3) Theorem 9.12. Tσ is spectral. Proof. By [11, Proposition 4 (i)⇔(v)], and in view of Proposition 9.11, to show that Tσ is spectral it suffices to show that Tσ is T0 and that (Tσ)patch is compact. Tσ is T0: Let P1 , P2 ∈ X(A,σ). Without loss of generality we may assume that there exists a ∈ P1 \ P2. Then P1 ∈ Hσ(a) and P2 < Hσ(a). (Tσ)patch is compact: Since Tσ is compact, the sets of the form U \ V with U, V compact open in Tσ, form a subbasis of open sets of (Tσ)patch. Assume X(A,σ) = [ (\ i∈I j∈Ji Ui, j \ Vi, j), where each set Ji is finite, and the sets Ui, j, Vi, j are compact open in Tσ. Then Im ξ = [ (\ i∈I j∈Ji (Ui, j ∩ Im ξ) \ (Vi, j ∩ Im ξ)), where each Ui, j ∩ Im ξ, Vi, j ∩ Im ξ is compact open in Im ξ since both Im ξ and − Im ξ are clopen by Proposition 9.8(1). Applying π, we get eXF = [ i∈I (\ j∈Ji π(Ui, j ∩ Im ξ) \ π(Vi, j ∩ Im ξ)), with each π(Ui, j ∩ Im ξ), π(Vi, j ∩ Im ξ) compact open in eXF by Proposition 9.9, so in particular clopen, and thus each T j∈Ji π(Ui, j ∩ Im ξ) \ π(Vi, j ∩ Im ξ)) is open in eXF. By compactness there is a finite subset I1 of I such that eXF = [ i∈I1 (\ j∈Ji π(Ui, j ∩ Im ξ) \ π(Vi, j ∩ Im ξ)). Positivecones on algebras with involution 48 Reasoning similarly with −ξ, we obtain a finite subset I2 of I such that eXF = [ i∈I2 (\ j∈Ji π(Ui, j ∩ − Im ξ) \ π(Vi, j ∩ − Im ξ)). Therefore X(A,σ) = [ (\ i∈I1∪I2 j∈Ji Ui, j \ Vi, j), (cid:3) proving that (Tσ)patch is compact. Proposition 9.13. Let (A, σ) and (B, τ) be two Morita equivalent F-algebras with involution. With notation as in Theorem 4.2, the map m∗ is a homeomorphism from (X(A,σ), Tσ) to (X(B,τ), Tτ). Proof. Let b1, . . . , bk ∈ Sym(B, τ)×. We have m∗(P) ∈ Hτ(b1, . . . , bk) if and only if b1, . . . , bk ∈ m∗(P) if and only if there are a1, . . . , ar ∈ P (see Theorem 4.2) such that b1, . . . , bk ∈ D(B,τ)m(ha1, . . . , ariσ). Therefore (m∗)−1(Hτ(b1, . . . , bk)) = {P ∈ X(A,σ) ∃r ∈ N ∃a1, . . . , ar ∈ P such that b1, . . . , bk ∈ D(B,τ)m(ha1, . . . , ariσ)} = {P ∈ X(A,σ) ∃r ∈ N ∃a1, . . . , ar ∈ Sym(A, σ) such that = [{Hσ(a1, . . . , ar) r ∈ N, a1, . . . , ar ∈ Sym(A, σ), P ∈ Hσ(a1, . . . , ar) and b1, . . . , bk ∈ D(B,τ)m(ha1, . . . , ariσ)} b1, . . . , bk ∈ D(B,τ)m(ha1, . . . , ariσ)}, which is a union of open sets in the topology Tσ. The fact that (m∗)−1 is also continuous is obtained as above, using that (m∗)−1 = (m−1)∗ as observed in the proof of Theorem 4.2. (cid:3) Acknowledgement We are very grateful to the referee, in particular for some suggestions that led to a significantly improved final section. References [1] Albert, A.A.: On involutorial algebras. Proc. Nat. Acad. Sci. U.S.A., 41:480 -- 482, 1955. Positivecones on algebras with involution 49 [2] Astier, V., Unger, T.: Signatures of hermitian forms and the Knebusch trace formula. Math. Ann., 358(3-4):925 -- 947, 2014. [3] Astier, V., Unger, T.: Signatures of hermitian forms and "prime ideals" of Witt groups. Adv. Math., 285:497 -- 514, 2015. [4] Astier, V., Unger, T.: Signatures of hermitian forms, positivity, and an an- swer to a question of procesi and schacher. J. Algebra, 508:339 -- 363, 2018. [5] Astier, V., Unger, T.: Stability index for algebras with involution, Contem- porary Mathematics 697:41 -- 50, 2017. [6] Astier, V., Unger, T.: Positive cones and gauges on algebras with involution, http://arxiv.org/abs/1806.05489, 2019. [7] Charnes, A., Cooper, W.W.: The strong Minkowski-Farkas-Weyl theorem for vector spaces over ordered fields. Proc. Nat. Acad. Sci. U.S.A., 44:914 -- 916, 1958. [8] Craven, T.C.: Orderings, valuations, and Hermitian forms over ∗-fields. In K-theory and algebraic geometry: connections with quadratic forms and division algebras (Santa Barbara, CA, 1992), volume 58 of Proc. Sympos. Pure Math., pages 149 -- 160. Amer. Math. Soc., Providence, RI, 1995. [9] Dickmann, M., Schwartz, N., Tressl, M.: Spectral spaces, New Mathemati- cal Monographs, 35. Cambridge University Press, Cambridge, 2019. [10] Elman, R., Lam, T.Y., Prestel, A.: On some Hasse principles over formally real fields. Math. Z., 134:291 -- 301, 1973. [11] Hochster, M.: Prime ideal structure in commutative rings. Trans. Amer. Math. Soc., 142:43 -- 60, 1969. [12] Knus, M.-A.: Quadratic and Hermitian forms over rings, volume 294 of Grundlehren der Mathematischen Wissenschaften. Springer-Verlag, Berlin, 1991. [13] Knus, M.-A., Merkurjev, A., Rost, M., Tignol, J.-P.: The book of involu- tions, volume 44 of American Mathematical Society Colloquium Publica- tions. American Mathematical Society, Providence, RI, 1998. [14] Lam, T.Y.: Introduction to quadratic forms over fields, volume 67 of Grad- uate Studies in Mathematics. American Mathematical Society, Providence, RI, 2005. Positivecones on algebras with involution 50 [15] Lee, H.C.: Eigenvalues and canonical forms of matrices with quaternion coefficients. Proc. Roy. Irish Acad. Sect. A., 52:253 -- 260, 1949. [16] Lewis, D.W., Unger, T.: A local-global principle for algebras with involution and Hermitian forms. Math. Z., 244(3):469 -- 477, 2003. [17] Lewis, D.W., Unger, T.: Hermitian Morita theory: a matrix approach. Irish Math. Soc. Bull., (62):37 -- 41, 2008. [18] Marshall, M.A.: Spaces of orderings and abstract real spectra, volume 1636 of Lecture Notes in Mathematics. Springer-Verlag, Berlin, 1996. [19] Platonov, V.P., Yanchevskii, V.I.: Finite-dimensional division algebras. In Algebra, IX, volume 77 of Encyclopaedia Math. Sci., pages 121 -- 239. Springer, Berlin, 1995. Translated from the Russian by P. M. Cohn. [20] Prestel, A.: Quadratische Semi-Ordnungen und quadratische Formen. Math. Z., 133:319 -- 342, 1973. [21] Prestel, A.: Lectures on formally real fields, volume 1093 of Lecture Notes in Mathematics. Springer-Verlag, Berlin, 1984. [22] Prestel, A., Delzell, C.N.: Positive polynomials. Springer Monographs in Mathematics. Springer-Verlag, Berlin, 2001. [23] Procesi, C., Schacher, M.: A non-commutative real Nullstellensatz and Hilbert's 17th problem. Ann. of Math. (2), 104(3):395 -- 406, 1976. [24] Scharlau, W.: Quadratic and Hermitian forms, volume 270 of Grundlehren der Mathematischen Wissenschaften. Springer-Verlag, Berlin, 1985. [25] Tignol, J.-P., Wadsworth, A.R.: Valuations on algebras with involution. Math. Ann., 351(1):109 -- 148, 2011. [26] Weil, A.: Algebras with involutions and the classical groups. J. Indian Math. Soc. (N.S.), 24:589 -- 623 (1961), 1960.
1609.02605	1	1609	2016-09-08T21:45:45	Cube term blockers without finiteness	[ "math.RA" ]	We show that an idempotent variety has a $d$-dimensional cube term if and only if its free algebra on two generators has no $d$-ary compatible cross. We employ Hall's Marriage Theorem to show that a variety of finite signature whose fundamental operations have arities $n_1, \ldots, n_k$ has a $d$-dimensional cube term if and only if it has one of dimension $d=1+\sum_{i=1}^k (n_i-1)$. This lower bound on dimension is shown to be sharp. We show that a pure cyclic term variety has a cube term if and only if it contains no $2$-element semilattice. We prove that the Maltsev condition "existence of a cube term" is join prime in the lattice of idempotent Maltsev conditions.	math.RA	math	CUBE TERM BLOCKERS WITHOUT FINITENESS KEITH A. KEARNES AND ´AGNES SZENDREI Abstract. We show that an idempotent variety has a d-dimensional cube term if and only if its free algebra on two generators has no d-ary compatible cross. We employ Hall's Marriage Theorem to show that a variety of finite signature whose fundamental operations have arities n1, . . . , nk has a d-dimensional cube term if i=1(ni − 1). This lower bound on dimension is shown to be sharp. We show that a pure cyclic term variety has a cube term if and only if it contains no 2-element semilattice. We prove that the Maltsev condition "existence of a cube term" is join prime in the lattice of idempotent Maltsev conditions. and only if it has one of dimension d = 1 +Pk 1. Introduction This note concerns a recently identified Maltsev condition, which promises to be significant. It is called "existence of a cube term". We begin by discussing relations. A binary cross on a set A is a subset of A2 of the form (U0 × A) ∪ (A × U1), where U0 and U1 are nonempty proper subsets of A. A U1 U0 A If U0 = U1 = 1, the cross is called If U0 = U1, the cross is called symmetric. thin. The sequence (U0, U1) is called the base (sequence) for the cross. If the cross is symmetric, i.e., if the base has the form (U, U), then we also refer to U as the base. 2010 Mathematics Subject Classification. Primary: 08B05; Secondary: 08A30, 08B20. Key words and phrases. cube term, cyclic term, Maltsev condition, idempotent variety, compat- ible cross. This material is based upon work supported by the National Science Foundation grant no. DMS 1500254 and the Hungarian National Foundation for Scientific Research (OTKA) grant no. K104251 and K115518. 1 2 KEITH A. KEARNES AND ´AGNES SZENDREI The definition of a cross makes sense for higher arity relations, i.e. a d-ary cross is a subset of Ad of the form (U0 × A × · · · × A) ∪ (A × U1 × · · · × A) ∪ · · · ∪ (A × A × · · · × Ud−1) where U0, . . . , Ud−1 are nonempty proper subsets of A. For d = 1 this means that a 1-ary cross is a nonempty proper subset of A. The definitions of symmetric cross, thin cross, and base for a d-ary cross are the expected ones. The arity of a cross is also called its dimension. Our use in this paper of the notion of a cross follows the earlier use of crosses in the 1987 paper [15], which concerns the description of the maximal, locally closed subclones of the clone of all idempotent operations on a given set. In [15], symmetric and asymmetric thin crosses play a central role, although they are just called 'crosses'. In the current paper, we need to consider arbitrary ('thick') crosses as well. Now we turn to cube terms. Let V be a variety and let F = FV (x, y) be the V-free algebra generated by the set {x, y}. Since F has an automorphism that switches x and y, it follows that exactly one of the following two conditions holds: (i) the set {x, y}d−{y} generates y, where y = (y, y, . . . , y) is the constant tuple with range {y}, or (ii) different subsets of {x, y}d generate different subalgebras of Fd. For condition (i) to hold, V must have a term c which applied to elements of {x, y}d − {y} yields y, i.e., (1.1) V = c(z1, z2, . . .) = y with all zi in {x, y}d − {y}. This is a vector identity. By considering this single vector identity coordinatewise, this means that V satisfies d identities of the form (1.2) c(. . . , x, . . .) = y where the only variables that appear in the identity are x and y, and for each place of c there is an identity that has x in that place. For example, if V has a term c satisfying c(cid:18)(cid:20)x y(cid:21) ,(cid:20)x x(cid:21) ,(cid:20)y x(cid:21)(cid:19) =(cid:20)y y(cid:21) , or equivalently, both of (cid:26)c(x, x, y) = y c(y, x, x) = y, then c is a term of the desired type for d = 2, which is called a Maltsev term. A term c satisfying the condition described in (1.1) is called a d-dimensional cube term or just d-cube term for V. Equivalently, c is a d-cube term if d identities of the type in (1.2) suffice to establish the condition in (1.2) for each place of c. Clearly, a d-cube term for a variety V is automatically a d′-cube term for all d′ ≥ d. Cube terms were introduced in [4] as part of an investigation of finite algebras with few subalgebras of powers. Terms of equal strength, called parallelogram terms, were discovered independently and at the same time in the study of finitely related clones, [9]. Cube terms and their equivalents have played roles in [8] in the study of constraint satisfaction problems, in [1, 2, 9] in the study of finitely related clones, CUBE TERM BLOCKERS 3 in [10, 13] in natural duality theory, and in [5] concerning the subpower membership problem. Theorem 2.1 of [12] gives a method for recognizing if a finite idempotent algebra has no cube term. Namely, a finite idempotent algebra fails to have a d-cube term for any d if and only if it has a cube term blocker. A cube term blocker of a finite idempotent algebra A is defined to be a pair (U, B) of nonempty subuniverses of A, with U ( B, such that U serves as a base for a compatible, symmetric, d-ary cross of B for every d. It follows that a finite idempotent algebra A fails to have a d-cube term for any d if and only if some subalgebra B ≤ A has compatible symmetric crosses of every arity. The result of [12] does not help if one wants to show that A has no d-cube term for a fixed d. The result also does not help if A is infinite. But Lemma 2.8 of [11] shows that an idempotent variety V fails to have a Maltsev term (i.e. a 2-cube term) if and only if the free V-algebra on 2 generators has a compatible 2-ary cross. Here V need not satisfy any finiteness hypothesis, but the cross involved turns out to be asymmetric, while the notion of a cube term blocker involves symmetric crosses only. Furthermore, Lemma 2.8 of [11] is a result about 2-cube terms only. The current paper may be viewed as establishing a generalization of both Theo- rem 2.1 of [12] and Lemma 2.8 of [11]. We will prove that an idempotent variety V fails to have a d-cube term if and only if the free V-algebra on 2 generators, F = FV(x, y), has a compatible d-ary cross. We will also show that an idempotent variety V fails to have a d-cube term for every d if and only if F has a nonempty proper subuniverse U that serves as a base for symmetric crosses of all arities, i.e. (U, F ) is a cube term blocker of F. The message to take away from this is that to avoid cube terms of a fixed dimension one should work with a not-necessarily-symmetric cross of that di- mension, but to avoid cube terms of all dimensions it suffices to work with symmetric crosses or cube term blockers. This note evolved in response to a question we learned from Cliff Bergman, which we state and answer in Section 5. Section 3 contains our proof that F = FV(x, y) has compatible crosses of all arities if and only if F has a subuniverse U such that (U, F ) is a cube term blocker. In Section 4 we develop a tight lower bound on the minimal dimension of a cube term for idempotent varieties of finite signature. In Section 6 we use our results to establish that the Maltsev condition "existence of a cube term" is join prime in the lattice of idempotent Maltsev conditions. 2. Cube terms and crosses A nonempty subset B of a product A0 × · · · × Ad−1 is a block if it is a product subset: B = B0 × · · · × Bd−1 with Bi ⊆ Ai for all i. A block is full in the i-th coordinate if Bi = Ai. Thus, for any sequence (U0, . . . , Ud−1) of nonempty proper 4 KEITH A. KEARNES AND ´AGNES SZENDREI subsets of A, the cross Cross(U0, . . . , Ud−1) = (U0 × A × · · · × A) ∪ · · · ∪ (A × · · · × A × Ud−1) = B0 ∪ · · · ∪ Bd−1 on A is defined to be a subset of Ad that is a union of d blocks B0, . . . , Bd−1 where Bi is full in all coordinates except the i-th. If t is an operation on a set A and U ⊆ A, then t is U-absorbing in its i-th variable if t(A, A, . . . , U , . . . , A) ⊆ U. {z}i An operation t on a set A is idempotent if t(a, a, . . . , a) = a for all a ∈ A. Lemma 2.1. Let A be a set with nonempty proper subsets U, U0, . . . , Ud−1, and let t be an n-ary idempotent operation on A. (1) If t is compatible with Cross(U0, . . . , Ud−1) and π ∈ Sd is a permutation, then t is compatible with Cross(Uπ(0), . . . , Uπ(d−1)). If (i0, . . . , ie−1) is a subsequence of (0, . . . , d − 1), then t is also compatible with Cross(Ui0, . . . , Uie−1). (2) If t is compatible with Cross(U0, . . . , Ud−1), then each Ui is a subuniverse of (A, t). (3) t is compatible with Cross(U0, . . . , Ud−1) if and only if (∗) for every function m : {0, . . . , n − 1} → {0, . . . , d − 1} there is some i ∈ im(m) such that (2.1) aj ∈ Ui for all j ∈ m−1(i) =⇒ t(a0, . . . , an−1) ∈ Ui. (4) If t is compatible with Cross(U0, . . . , Ud−1), and n ≤ d, then for all except possibly n − 1 choices of j < d it is the case that t is Uj-absorbing in one of its variables. (5) The following are equivalent for t: (i) t is compatible with the d-ary symmetric cross Cross(U, . . . , U) for some d ≥ n. (ii) t is U-absorbing in one of its variables. (iii) t is compatible with the d-ary symmetric cross Cross(U, . . . , U) for every d ≥ 1. Proof. For the first statement in (1) observe that Cross(Uπ(0), . . . , Uπ(d−1)) differs from Cross(U0, . . . , Ud−1) by a permution of coordinates. Therefore the desired conclusion follows from the fact that if we permute the coordinates of a subuniverse of (A; t)d we again get a subuniverse of (A; t)d. CUBE TERM BLOCKERS 5 For the second statement in (1), choose a0 /∈ U0. Since t is idempotent, the singleton {a0} is a subuniverse of (A; t). Hence {a0} × Cross(U1, . . . , Ud−1) = Cross(U0, . . . , Ud−1) ∩ ({a0} × Ad−1) is a subuniverse of (A; t)d. It follows that Cross(U1, . . . , Ud−1) is a subuniverse of (A; t)d−1. Similarly, Cross(U0, . . . , Ui−1, Ui+1, . . . , Ud−1) is a subuniverse of (A; t)d−1 for all i < d. Repeating this procedure for every i not occurring in (i0, . . . , ie−1) we get that Cross(Ui0, . . . , Uie−1) is a subuniverse of (A; t)e, that is, t is compatible with Cross(Ui0, . . . , Uie−1). Item (2) is the special case e = 1 of the second statement in (1). For item (3), assume first that (∗) fails. Then there is a function m : {0, . . . , n − 1} → {0, . . . , d − 1} such that for every i ∈ im(m) the implication in (2.1) fails. Choose witnessing elements: i.e. choose, for each i ∈ im(m) = {i0, . . . , ie−1}, elements ai,j ∈ A (j < n) satisfying ai,j ∈ Ui for all j ∈ m−1(i) such that t(ai,0, . . . , ai,n−1) /∈ Ui. The columns of the matrix [ak,ℓ] lie in Cross(Ui0, . . . , Uie−1), because every j < n belongs to m−1(i) for some i ∈ im(m) = {i0, . . . , ie−1}, and hence by construction, the j-th column of [ak,ℓ] has i-th entry ai,j ∈ Ui. However, by construction, the column obtained by applying t to the rows of this matrix does not lie in Cross(Ui0, . . . , Uie−1). Thus, t is not compatible with Cross(Ui0, . . . , Uie−1), so it is not compatible with Cross(U0, . . . , Ud−1) either, according to item (1). Conversely, assume that t is not compatible with Cross(U0, . . . , Ud−1). Then we can select elements from this relation and allow them to serve as columns for a d × n matrix [ai,j] where (i) for each column j, there is a row index i (=: m(j)) such that ai,j ∈ Ui and (ii) the value obtained by applyig t to the i-th row fails to belong to Ui for every i. This yields a function m witnessing that (∗) fails. For item (4), define a bipartite graph with one part X = {x0, . . . , xn−1}, other part equal to d := {0, . . . , d − 1}, and edge relation containing exactly those pairs (xi, j) where t is not Uj-absorbing in variable xi. Claim 2.2. There is no matching from X to d. (A matching from X to d is a subset of the edge set that is an injective function X → d.) Proof of Claim 2.2. Assume that there is a matching µ : X → d. By item (1) we may reorder the Uj's so that µ(xi) = i is the matching. It follows that for every i < n, t is not Ui-absorbing in its i-th variable, so there must exist ui,i ∈ Ui and elements ai,k ∈ A such that t(ai,0, ai,1, . . . , ui,i, . . . , ai,n−1) /∈ Ui. 6 KEITH A. KEARNES AND ´AGNES SZENDREI There also exist aj /∈ Uj for every j < d. These ingredients yield a situation (2.2) t an−1,0 , an−1,1 , · · · , un−1,n−1 = u0,0 a1,0 ... an ... ad−1       a0,1 u1,1 ... an ... ad−1     a0,n−1 a1,n−1 ... an ... ad−1         /∈ U0 /∈ U1 ... /∈ Un−1 an( /∈ Un) ... ad−1( /∈ Ud−1) .   The operands are in Cross(U0, . . . , Ud−1), but the value is not, a contradiction. ⋄ By the Marriage Theorem, there is a subset Y ⊆ X such that the set K ⊆ d of elements adjacent to elements of Y satisfies K < Y ≤ n. Thus, Y 6= ∅ and the set K has size at most n − 1; moreover, if j ∈ d − K then no element of Y is adjacent to j. Hence t is Uj-absorbing in its variables in Y . For item (5), the implication (iii) ⇒ (i) is a tautology, and the implcation (i) ⇒ (ii) follows from the statement in (4) we just proved. To establish the remaining implication (ii) ⇒ (iii) assume without loss of generality that t is U-absorbing in its first variable, and consider the d-ary symmetric cross Cross(U, . . . , U) for some d ≥ 1. Let [ai,j] be a d × n matrix of element of A whose columns are in Cross(U, . . . , U). In particular, the first column lies in Cross(U, . . . , U), so ai,0 ∈ U for some i < d. Since t is U-absorbing in its first variable, we get that t(ai,0, . . . , ai,n−1) ∈ U. Hence the column obtained by applying t to the rows of the matrix [ai,j] lies in Cross(U, . . . , U). This proves that t is compatible with Cross(U, . . . , U). (cid:3) Corollary 2.3. Let A be an idempotent algebra. If A has a d-cube term, then A has no compatible cross of dimension d or larger. Proof. Let c be a d-cube term for A, and assume A has an e-dimensional compatible cross Cross(U0, . . . , Ue−1) with d ≤ e. By Lemma 2.1 (4) there exists j < d such that c is Uj-absorbing in one of its variables. This contradicts the cube identities (see (1.1) or (1.2)). (cid:3) Our goal in this section is to characterize idempotent varieties which have no d-cube term (for a fixed d ≥ 2) or have no cube term (of any dimension). In Theorem 2.4 below the first property will be characterized by the existence of a d-dimensional cross, while the second one will be characterized by the existence of a specific in- finite system of crosses, which we call a 'cross sequence'. A cross sequence for an idempotent algebra A is an ω-sequence (U0, U1, . . .) of proper nonempty subsets of A such that Cross(Ui0, . . . , Uik−1) is a compatible relation of A for every finite sub- sequence (Ui0, . . . , Uik−1) of (U0, U1, . . .). A cross sequence (U0, U1, . . .) is proper if CUBE TERM BLOCKERS 7 Ti<ω Ui 6= ∅ and Si<ω Ui 6= A. It follows from Lemma 2.1(1) that any subsequence and any reordering of a (proper) cross sequence is again a (proper) cross sequence. Theorem 2.4. Let V be an idempotent variety, let F = FV (x, y) be the V-free algebra over the free generating set {x, y}, and let d be a positive integer. (1) V has no d-cube term iff F has a compatible d-dimensional cross. (2) V has no cube term of any dimension iff F has a proper cross sequence iff F has a cross sequence. We will prove the two statements of Theorem 2.4 simultaneously. In Theorem 2.5 a uniform formulation of these two statements is based on the observation that for any variety V, the condition 'V has no d-cube term' for a fixed d is equivalent to V has no e-cube term for any e < δ for δ = d + 1, while the condition 'V has no cube term of any dimension' is equivalent to the same displayed condition for δ = ω. For 0 < δ ≤ ω, let δ− :=(δ − 1 ω if δ < ω, if δ = ω. Theorem 2.5. Let V be an idempotent variety, let F = FV (x, y) be the free V- algebra generated by {x, y}, and let 2 ≤ δ ≤ ω. The variety V fails to have a d-cube for any d < δ if and only if there is a δ−-sequence, σ = (U0, U1, . . .) = (Ui)i<δ−, of subuniverses of F such that (1) x ∈ Ui and y /∈ Ui for every i < δ−, and (2) Cross(Ui0, . . . , Uik−1) is a compatible relation of F for every finite subsequence (Ui0, . . . , Uik−1) of σ. Proof. The "if" assertion follows from Corollary 2.3: if F has compatible crosses of every arity < δ, then it cannot have a d-cube term for any d < δ. For the converse, assume that V has no d-cube term for any d < δ. This implies, in particular, that V is nontrivial. Recursively define the sequence σ = (U0, U1, . . .) = (Ui)i<δ− with Ui ≤ F, according to the following rule: for i < δ−, Ui is chosen to be a subuniverse maximal for the properties that (i) x ∈ Ui, and (ii) y /∈ hCross(U0, U1, . . . , Ui, {x}, . . . , {x})iFk for any k with i < k < δ. It is possible to make these choices, for the following reasons. The fact that V does not have a d-cube term for any d < δ means exactly that y /∈ hCross({x}, {x}, . . . , {x})iFk for every k < δ. Thus the subuniverse {x} satisfies all the properties required of U0, except that it need not be maximal among the subuniverses satisfying (i) and (ii) for i = 0. But the union of a chain of subuniverses of F satisfying (i) and (ii) for 8 KEITH A. KEARNES AND ´AGNES SZENDREI a given i again satisfies these conditions for that i, so {x} can be extended to a maximal subuniverse U0 satisfying (i) and (ii) for i = 0. Similarly, if 0 < i < δ− and U0, . . . , Ui−1 have been chosen, then condition (ii) for i − 1 guarantees that {x} satisfies all the properties required of Ui except maximality. Extend {x} to a maximal Ui, etc. Observe that y /∈ Ui for any i < δ−, since otherwise {x, y} ⊆ Ui, leading to F = Ui, leading to Cross(U0, . . . , Ui) = F i+1, contradicting item (ii) above. Hence item (1) of the theorem statement holds for σ = (U0, U1, . . .) = (Ui)i<δ−. Our remaining task is to show that Cross(Ui0, . . . , Uik−1) is a compatible relation of F for every finite subsequence (Ui0, . . . , Uik−1) of σ. Every finite subsequence of σ is a subsequence of an initial segment of σ, therefore, in view of Lemma 2.1 (1), it suffices to prove that Cross(U0, . . . , Ud−1) is a compatible relation of F for every d < δ. We will do this simultaneously for every d with an indirect argument. Assume that there is some d < δ and some element (2.3) p = (p0, . . . , pd−1) ∈ hCross(U0, . . . , Ud−1)iFd − Cross(U0, . . . , Ud−1). Necessarily pi /∈ Ui for any i < d. There is no harm in assuming that, among all possible choices of d and p, we have chosen those such that pi 6= y holds in the fewest number of coordinates. That is, we assume that (2.3) holds and also that for no e < δ do we have q ∈ hCross(U0, . . . , Ue−1)iFe − Cross(U0, . . . , Ue−1) with q differing from y in strictly fewer coordinates than p. Since p ∈ hCross(U0, . . . , Ud−1)iFd, there exists a term s and there exist elements b0, . . . , bm−1 ∈ Cross(U0, . . . , Ud−1) such that s(b0, . . . , bm−1) = p. Observe that one may lengthen all tuples involved by adding some number of y's (say g with d + g < δ) to the end in order to obtain (2.4) b0 y ... y  s     , . . . ,  bm−1 y ... y     p y ... y =    . The columns appearing as arguments to s in (2.4) belong to Cross(U0, . . . , Ud−1, Vd, . . . , Vd+g−1) for any choice of (∅ 6=) Vi ⊂ F . There must exist some coordinate of p that is not y, else condition (ii) from the definition of σ is violated when k = d. Let i be the first coordinate of p where pi 6= y; hence i < d. Since pi /∈ Ui, the subuniverse U ′ i = hUi ∪ {pi}iF properly extends Ui. By the maximality of Ui, there must exist some k with i < k < δ such that y ∈ hCross(U0, . . . , U ′ i, {x}, . . . , {x})iFk. To understand how y could be generated, observe that since CUBE TERM BLOCKERS 9 Cross(U0, . . . , U ′ i , {x}, . . . , {x}) = B0(U0) ∪ · · · ∪ Bi−1(Ui−1) ∪ Bi(U ′ i ) ∪ Bi+1({x}) ∪ · · · ∪ Bk−1({x}), we get that hCross(U0, . . . , U ′ i, {x}, . . . , {x})iFk equals the subalgebra join B0(U0) ∨ · · · ∨ Bi−1(Ui−1) ∨ Bi(U ′ i ) ∨ Bi+1({x}) ∨ · · · ∨ Bk−1({x}). i ) = F i × U ′ i × F k−i−1 is generated by {x, y}i × (Ui ∪ {pi}) × {x, y}k−i−1, However Bi(U ′ and all elements of this product set already belong to Cross(U0, . . . , Ui, {x}, . . . , {x}) except the tuple (y, . . . , y, pi, y, . . . , y). Hence hCross(U0, . . . , U ′ i , {x}, . . . , {x})iFk is generated by Cross(U0, . . . , Ui, {x}, . . . , {x}) ∪ {(y, . . . , y, pi, y, . . . , y)}. Since y is generated by this set, there is a term t and elements (columns) ci ∈ Cross(U0, . . . , Ui, {x}, . . . , {x}) ⊆ F k such that (2.5) t c0, . . . , cn−1, = = y.   y ... y pi y ... y       y ... y y y ... y     Lengthen these tuples, by adding y's at the end, to some length e satisfying max{k, d} ≤ e < δ (hence the columns have length at least d). Equation (2.5) still holds. Write the extension of ci as c′ i, and note that i ∈ Cross(U0, . . . , Ui, {x}, . . . , {x}) (⊆ F e) c′ where there may be more {x}'s than before. Let ε = (ε0, . . . , εe−1) be the endomorphism of Fe defined coordinatewise as follows: (a) εj : F → F is the identity if 0 ≤ j ≤ i (≤ d) or d ≤ j < e, and (b) εj : F → F : x 7→ x, y 7→ pj if i < j < d. Observe that ε maps Cross(U0, . . . , Ui, {x}, . . . , {x}) into itself. Indeed, ε(Bj(Uj)) ⊆ Bj(Uj) for j ≤ i because εj = id, while ε(Bj({x})) ⊆ Bj({x}) for j > i because εj fixes x. 10 KEITH A. KEARNES AND ´AGNES SZENDREI Applying ε to (2.5) yields (2.6) t ε(c′ 0), . . . , ε(c′ n−1), = =: q.   y ... y pi pi+1 ... pd−1 y ... y       y ... y y pi+1 ... pd−1 y ... y     By the choice of i, the last argument of t in (2.6) is the column (p, y, . . . , y). Therefore, the expression for q in (2.6) can be rewritten as (2.7) ε(c′ 0), . . . , ε(c′ t   n−1), s   b0 y ... y   , . . . ,  bm−1 y ... y       = q using (2.4). The columns, ε(c′ but qj /∈ Uj for any j. Hence (2.7) asserts that u) and (bv, y, . . . , y) all belong to Cross(U0, . . . , Ue−1), q ∈ hCross(U0, . . . , Ue−1)iFe − Cross(U0, . . . , Ue−1) with q differing from the constant y-tuple in strictly fewer coordinates than p. This is so because q differs from p only in that it may have more y's at the end and qi = y while pi 6= y. This conclusion contradicts the choice of p. This proves that Cross(U0, . . . , Ud−1) is a compatible relation of F for every d < δ. (cid:3) Proof of Theorem 2.4. For item (1), we assume first that V has no d-cube term. Hence, V has no e-cube term for any e < d + 1. It follows from Theorem 2.5 (for δ = d + 1) that there is a sequence (U0, . . . , Ud−1) of subuniverses of F such that x ∈Ti<d Ui, y /∈Si<d Ui, and Cross(U0, . . . , Ud−1) is a compatible relation of F. Conversely, if F has a compatible d-dimensional cross, then, by Corollary 2.3, F has no d-cube term. Hence V has no d-cube term. For item (2), let us assume first that V has no cube term. Applying Theorem 2.5 (for δ = ω) we conclude that there is an ω-sequence, σ = (U0, U1, . . . ), of subuniverses of F such that σ is a proper cross sequence for F satisfying x ∈ Ti<ω Ui and y /∈ Si<ω Ui. If F has a proper cross sequence, then F has a cross sequence. Finally, if F has a cross sequence, then F has compatible crosses of arbitrarily high dimensions. CUBE TERM BLOCKERS 11 Therefore, by Corollary 2.3, F has no cube term of any dimension. Hence V has no cube term of any dimension either, completing the proof. (cid:3) 3. Producing symmetric crosses We have shown in Theorem 2.4 (2) that an idempotent variety V fails to have a cube term if and only if its 2-generated free algebra F has a proper cross sequence. In this section we will use a combinatorial argument to show that this cross sequence can be converted to a constant cross sequence (U, U, U, . . .). This produces a nonempty proper subuniverse U of F that is the base of a compatible symmetric d-ary cross for every d. In fact, our construction of 'symmetrizing' a proper cross sequence works for any idempotent algebra, as the theorem below shows. Theorem 3.1. The following are equivalent for an idempotent algebra A. (1) A has a proper cross sequence. (2) A has a nonempty proper subuniverse U such that (U, A) is a cube term blocker for A. (That is, U is a base for compatible symmetric d-ary crosses of A for all d.) Proof. The implication (2) ⇒ (1) is clear from the definitions: If (U, A) is a cube term blocker of A, then the d-ary cross Cross(U, . . . , U) is a compatible relation of A for every d, so the constant ω-sequence (U, U, . . .) is a proper cross sequence for A. To prove the reverse implication (1) ⇒ (2), assume that σ = (U0, U1, . . .) is a proper cube term blocker of A. Note that cross sequence for A. We shall prove that if U := S∞ \j=i+1 Uj ⊆ · · · ⊆ \j=1 \j=i Uj ⊆ Uj ⊆ ∞ ∞ ∞ ∞ i=0(cid:16)T∞ j=i Uj(cid:17), then (U, F ) is a [j=0 Uj, ∞ Uj ⊆ · · · ⊆ \j=0 i=0(cid:16)T∞ U is a nonempty proper subuniverse of A. j=i Uj(cid:17) is the union of an increasing ω-sequence of subuniverses of A. It follows that U is a subuniverse of A. Moreover, since the cross sequence j=0 Uj 6= A. This implies that so U := S∞ σ = (U0, U1, . . . ) is proper, we haveT∞ j=0 Uj 6= ∅ andS∞ The nontrivial part of the proof, therefore, is the argument that U is a base for symmetric crosses of all arities. To see that this is so, choose an arbitrary term operation t(x0, . . . , xn−1) of A and as in the proof of Lemma 2.1 (4), use it to define a bipartite graph as follows: the vertices of the two parts are X = {x0, . . . , xn−1}, the set of variables of t, and ω, the set indexing the terms of the cross sequence σ = (U0, U1, . . . ); the edges of the graph are the pairs (xi, j) such that t is not Uj-absorbing in variable xi. Claim 3.2. There is no matching from X to ω. 12 KEITH A. KEARNES AND ´AGNES SZENDREI Proof of Claim 3.2. Suppose that there is a matching µ : X → ω. Since a reordering of finitely many terms of σ produces a new cross sequence for A and leaves the set U unchanged, we may assume without loss of generality that µ(xi) = i for all i < n. This mean that for every i < n the term t is not Ui-absorbing in variable xi, so for every i < n there must exist ui,i ∈ Ui and elements ai,k ∈ A such that t(ai,0, ai,1, . . . , ui,i, . . . , ai,n−1) /∈ Ui. There also exist aj /∈ Uj for every j < d. Now the same calculation (2.2) as in the proof of Lemma 2.1 (4) yields that t is not compatible with Cross(U0, . . . , Ud−1). This contradicts our assumption that (U0, U1, . . . ) is a cross sequence for A, and hence ⋄ proves the claim. Since X is finite, the Marriage Theorem holds in this situation. It asserts that, since there is no matching from X to ω, there is a subset Y ⊆ X such that the set K ⊆ ω of elements adjacent to elements of Y satisfies K < Y ≤ X = n. Therefore Y 6= ∅, the set K has size at most Y − 1 ≤ n − 1, and if xj ∈ Y then for all ℓ ∈ ω − K we have that t is Uℓ-absorbing in variable xj. It follows that if xj ∈ Y then t is (T∞ finitely many i, and hence that t is (cid:0)S∞ i=0(T∞ proves that every term operation t of A has a variable xj in which t is U-absorbing. By Lemma 2.1 (5), this is equivalent to the statement that all term operations of A are compatible with all symmetric crosses with base U. (cid:3) j=i Uj)-absorbing in variable xj for all but j=i Uj)(cid:1)-absorbing in variable xj. This In Theorem 2.4 (2) we characterized the idempotent varieties that fail to have cube terms. Using Theorem 3.1 we can strengthen this characterization as follows. Theorem 3.3. Let V be an idempotent variety, and let F = FV (x, y) be the V-free algebra over the free generating set {x, y}. The following conditions are equivalent. (1) V has no cube term. (2) F has a nonempty proper subuniverse U such that (U, F ) is a cube term blocker for F. (That is, U is a base for compatible symmetric d-ary crosses of F for all d.) Proof. Combine Theorem 3.1 for A = F with Theorem 2.4. (cid:3) As we mentioned in the introduction, cube term blockers were intorduced in [12] to prove that a finite idempotent algebra A fails to have a d-cube term for any d if and only if A has a cube term blocker (U, B), or equivalently, some subalgebra B of A has compatible symmetric crosses with base U of every arity. Next we show how to derive this theorem of [12] from the results of our paper. Lemma 3.4. In any given signature, the class of idempotent algebras with no cube term blockers is closed under the formation of homomorphic images, subalgebras and finite products. CUBE TERM BLOCKERS 13 Proof. We prove that the nonexistence of cube term blockers is preserved under ho- momorphic images, subalgebras, and finite products by arguing the contrapositive. If ϕ : A → C is a surjective homomorphism and (U, B) is a cube term blocker of C, then it is easy to see that (ϕ−1(U), ϕ−1(B)) is a cube term blocker of A. This can be done by checking that the inverse image, under ϕ, of each compatible symmetric cross Cross(U, . . . , U) of B is a compatible relation of the subalgebra ϕ−1(B) of A, and is equal to the symmetric cross Cross(cid:0)ϕ−1(U), . . . , ϕ−1(U)(cid:1). If C ≤ A and (U, B) is a cube term blocker of C, then it is also a cube term blocker of A. Before turning to products, first note that if (U, B) is a cube term blocker of A, and V is a subuniverse of A, then (U ∩ V, B ∩ V ) is a cube term blocker for A unless U ∩ V = ∅ or U ∩ V = B ∩ V . Now, to prove the statement for products, we assume that (U, B) is a cube term blocker of A × C, and explain how to find a cube term blocker for either A or C. Choose (a, c) ∈ B − U, and let πA, πC be the coordinate projections of A × C. If πA(U) 6= πA(B), then (πA(U), πA(B)) is a cube blocker of A, and we are done. Otherwise πA(U) = πA(B), hence a ∈ πA(B) = πA(U), implying the existence of an element (a, d) ∈ U. Now we let V = {a} × C and apply the observation of the previous paragraph: Since ∅ 6= U ∩ V 6= B ∩ V , the pair (U ∩ V, B ∩ V ) is a cube term blocker for A × C. We further have that ∅ 6= πC(U ∩ V ) 6= πC(B ∩ V ), since d ∈ πC(U ∩ V ) and c ∈ πC(B ∩ V ) − πC(U ∩ V ). Hence (πC(U ∩ C), πC(B ∩ V )) is a cube term blocker for C. (cid:3) Corollary 3.5. If A is a finite idempotent algebra, then the following are equivalent: (1) A has no cube term. (2) V(A) has no cube term. (3) FV(A)(x, y) has a cube term blocker. (4) A has a cube term blocker. Proof. (1) ⇒ (2) is clear, because a cube term for V(A) would be a cube term for A. The implication (2) ⇒ (3) follows from Theorem 3.3. For (3) ⇒ (4) we prove the contrapositive. If A has no cube term blocker, then by Lemma 3.4 and by the fact that FV(A)(x, y) lies in HSP fin(A) when A is finite, we get that FV(A)(x, y) has no cube term blocker. Finally, (4) ⇒ (1) follows from Corollary 2.3. (cid:3) 4. The influence of finite signature By a signature we mean a pair τ = (O, arity) where O is a set of operation symbols and arity : O → ω is a function assigning arity. We consider only idempotent varieties, so we may and do consider only signatures where arity(f ) ≥ 2 for all f ∈ O. We will call such signatures "suitable for idempotent varieties". 14 KEITH A. KEARNES AND ´AGNES SZENDREI In this section we consider only finite signatures, which are those where O < ω. For such a signature τ we define τ = max f ∈O(cid:0)arity(f )(cid:1) and kτ k = 1 +Xf ∈O(cid:0)arity(f ) − 1(cid:1). It is easy to see that τ = kτ k if there is only one operation in the signature, while τ < kτ k if there is more than one. Now let us consider an idempotent variety V of finite signature τ , and let F = FV (x, y) be the V-free algebra generated by {x, y}. The main results of this section are the following: • If F has a compatible cross of arity kτ k or more, then it has compatible crosses of all arities. Equivalently, if V has a cube term, then it has a kτ k-cube term (Theorem 4.1). • For any suitable finite signature τ there exists an example for V where F has a compatible cross of arity kτ k − 1, but no compatible crosses of higher arity. Equivalently, for any suitable finite signature τ there exists an example for V such that V has a kτ k-cube term, but has no d-cube term for d < kτ k (Example 4.4). For symmetric crosses the corresponding statements are as follows: • If F has a compatible symmetric cross of arity τ or more, then it has com- patible symmetric crosses of all arities (Corollary 4.3). • For any suitable finite signature there exists an example where F has a com- patible symmetric cross of arity τ − 1, but no compatible symmetric crosses of higher arity (Example 4.8). By adding operations to a signature one can make kτ k large while τ remains small. Thus one can create varieties with cube terms where the least dimension of a cube term is much greater than the arities of the symmetric crosses of F. These results show that we can't use only symmetric crosses to characterize the existence or nonexistence of cube terms of a fixed dimension. Theorem 4.1. Let V be an idempotent variety of finite signature τ . If V has no kτ k-cube term, then it has no cube term at all. In particular, if the signature of V consists of a single binary operation symbol, then either V has a Maltsev term or it has no cube term at all. Proof. In the second statement, our assumption on the signature τ forces that kτ k= 2. Hence the claim is an easy consequence of the first statement of the theorem and the fact that a variety has a Maltsev term if and only if it has a 2-cube term. To prove the first statement, assume that V is an idempotent variety of finite signature τ , and let F = FV (x, y). We know from Theorem 2.4 (1) that for every d ≥ 1, V has no d-cube term ⇔ F has a compatible d-ary cross. CUBE TERM BLOCKERS 15 We also know from Theorem 3.3, from the definition of a cube term blocker, and from Corollary 2.3 that V has no cube term Cor 2.3 ⇑ Thm 3.3⇔ F has a cube term blocker (U, F ) where U is a nonempty proper subuniverse of F m def F has compatible crosses of all arities ⇐ U is the base for compatible symmetric crosses of F of all arities; hence all four conditions displayed here are equivalent. Therefore it suffices to prove the first statement of Theorem 4.1 in the following equivalent formulation. Claim 4.2. Let V be an idempotent variety of finite signature τ , and let F = FV(x, y) be the V-free algebra over the free generating set {x, y}. If F has a compatible cross of arity ≥ kτ k, then F has a nonempty proper subuniverse U such that (U, F ) is a cube term blocker for F. (That is, U is a base for compatible symmetric crosses of F of all arities.) (i < k). So, d ≥ kτ k = 1 +Pk−1 j ∈ d − Ki. Since d −Sk−1 Proof of Claim 4.2. Assume that F has a compatible d-ary cross Cross(U0, . . . , Ud−1) where d ≥ kτ k. Let f0, . . . , fk−1 be the operation symbols of τ , and let arity(fi) = ni i=0 (ni − 1). Applying Lemma 2.1 (4) to the basic operations f0, . . . , fk−1 of F we see that for each i < k there exists a subset Ki of d := {0, . . . , d − 1} such that Ki ≤ ni−1 and fi has a Uj-absorbing variable for every i=0 (ni − 1) ≥ 1, we obtain that there exists at least one j ∈ d such that every fi (i < k) has a Uj-absorbing variable. It follows from Lemma 2.1 (5) that every fi (i < k) is compatible with the symmetric crosses Cross(Uj, . . . , Uj) of all arities. Hence, the symmetric crosses Cross(Uj, . . . , Uj) of all arities are compatible relations of F. Equivalently, (Uj, F ) is a cube term blocker for ⋄ F. i=0 Ki ≥ d −Pk−1 This finishes the proof of Theorem 4.1. (cid:3) The analog of Claim 4.2 for symmetric crosses can be proved similarly. Corollary 4.3. Let V be an idempotent variety of finite signature τ , and let F = FV (x, y) be the V-free algebra with free generators x, y. If F has a compatible sym- metric cross Cross(U, . . . , U) of arity ≥ τ , then (U, F ) is a cube term blocker for F. (That is, U is a base for compatible symmetric crosses of F of all arities.) Proof. Assume that F has a compatible symmetric d-ary cross Cross(U, . . . , U) where d ≥ τ . As before, let f0, . . . , fk−1 be the operation symbols of τ , and let arity(fi) = ni (i < k). So, d ≥ ni for all i. It follows from Lemma 2.1 (4) that every basic operation f0, . . . , fk−1 of F has a U-absorbing variable. Hence, by Lemma 2.1 (5), f0, . . . , fk−1 are compatible with the symmetric crosses Cross(U, . . . , U) of all arities. Thus, the symmetric crosses Cross(U, . . . , U) of all arities are compatible relations of F, or equivalently, (U, F ) is a cube term blocker for F. (cid:3) 16 KEITH A. KEARNES AND ´AGNES SZENDREI Example 4.4. Our goal in this example is to show that the bound in Theorem 4.1 is sharp. That theorem shows that if an idempotent variety of finite signature τ has a d-cube term for some d, then it has one for d = kτ k. In this example we construct, for any suitable finite signature τ , an idempotent variety that has a kτ k-cube term, but no d-cube term for d < kτ k. If one revisits the definition of "d-cube term", one sees that the concept of a 1-cube term is degenerate: the only varieties with 1-cube terms are varieties of 1-element algebras, and for these varieties any term without nullary symbols is a 1-cube term. As noted earlier, a variety has a 2-cube term if and only if it has a Maltsev term. Thus the simplest nondegenerate example to be exhibited is that of a nontrivial variety with a Maltsev term in a signature τ satisfying kτ k = 2. If τ is suitable for idempotent varieties, then kτ k = 2 implies exactly that τ is a signature with one operation, which is binary. For this signature, take as an example the variety generated by hZ3; f (x, y)i where f (x, y) = 2x + 2y. This variety is nontrivial, has kτ k = 2, and has a Maltsev term m(x, y, z) := f (f (x, z), y) = x + 2y + z. The cases where kτ k > 2 will be handled by a uniform construction. Suppose that τ has m operation symbols. Set ni = arity(fi), 1 ≤ i ≤ m, and set n := kτ k − 1 = i=1(ni − 1). If Pm C1 = {1, 2, · · · , (n1 − 1)}, C2 = {(n1 − 1) + 1, (n1 − 1) + 2, · · · , (n1 − 1) + (n2 − 1)}, j−1 ... Cj =n(cid:16) Xi=1 ... Cm =n(cid:16)m−1 Xi=1 (ni − 1)(cid:17) + 1, · · · ,(cid:16) j Xi=1 (ni − 1)(cid:17)o, (ni − 1)(cid:17) + 1, · · · , no, then {C1, . . . , Cm} is a partition of [n] := {1, . . . , n} whose cells that are in 1–1 correspondence with the operation symbols: Ci ↔ fi. Moreover, Ci = arity(fi) − 1. The elements of [n] are going to be coordinates in a product algebra. To describe the construction of the algebra we will use the terminology that an element j ∈ [n] belongs to fi, (or fi belongs to j) if j ∈ Ci. The universe of the product algebra will be A = {0, 1}n. We will explain how to interpret each symbol fi on this set, by describing its behavior in each coordinate. We need some terminology to do this. Let ∨ (join) and ∧ (meet) be the lattice operations on {0, 1} for the order 0 < 1. In any given coordinate we will interpret fi as either: CUBE TERM BLOCKERS 17 (1) the ni-ary join on {0, 1}: fi(x1, . . . , xni) = _1≤j≤ni xj, or (2) the "canonical" ni-ary near-unanimity operation on {0, 1}, namely fi(x1, . . . , xni) = _1≤j<k≤ni (xj ∧ xk). (This operation is a near-unanimity operation only when ni > 2, but we shall use the terminology even when ni = 2. In this situation fi(x1, x2) = x1 ∧ x2.) We interpret fi on A = {0, 1}n by stipulating that it acts coordinatewise, and that it acts like the canonical ni-ary near-unanimity operation in the coordinates that belong to fi and like the ni-ary join operation in the coordinates that do not belong to fi. The set A equipped with the operations f1, . . . , fm just defined is the algebra we call A. Each fi is idempotent on A, since join, meet and near-unanimity are idempotent. Therefore, the set Uj ⊆ A = {0, 1}n consisting of all n-tuples with 1 in the j-th coordinate is a nonempty proper subuniverse of A for each j between 1 and n. Claim 4.5. Cross(U1, . . . , Un) is a compatible (kτ k − 1)-ary cross of A. Proof of Claim 4.5. It is only the n > 1 case of the claim that is interesting, and we are in that case since kτ k > 2 and n = kτ k − 1. Elements of A = {0, 1}n will be represented by rows of length n consisting of 0's and 1's. The elements of Cross(U1, . . . , Un) are n-tuples of elements of A, so could be represented by columns of length n, where each entry in the column is a row of length n. But instead of doing this, we drop parentheses and consider elements of Cross(U1, . . . , Un) to be n × n matrices of 0's and 1's. For such a matrix to belong to this relation we must have the first row in U1 or the second row in U2, etc. Since a row of length n belongs to Ui if and only if it has a 1 in the i-th place, it follows that an n × n matrix belongs to Cross(U1, . . . , Un) exactly if it has a 1 somewhere on the diagonal. The operations of A act coordinatewise on the columns in a relation, and, in a given coordinate, act coordinatewise on rows. Thus, the operations of A act coordinatewise on matrices. We must show that if fi is one of the operations of A and M1, . . . , Mni ∈ Cross(U1, . . . , Un), then fi(M1, . . . , Mni) ∈ Cross(U1, . . . , Un). Suppose that some Mk has a 1 on its diagonal in the j, j-th entry, where j does not belong to fi. Then, as fi acts as ni-ary join in the j, j-position, it follows that fi(M1, . . . , Mni) has a 1 in its j, j-th entry, so fi(M1, . . . , Mni) ∈ Cross(U1, . . . , Un). Now suppose that each Mk only has 1's in entries j, j where j does belong to fi. Since there are ni arguments of fi, and only ni − 1 distinct j's that belong to fi, it 18 KEITH A. KEARNES AND ´AGNES SZENDREI must be that there are two matrices Mk and Mℓ, 1 ≤ k < ℓ ≤ ni which both have 1 in the j, j-position for some j belonging to fi. Since fi acts like the canonical ni-ary near unanimity operation in position j, j, it follows that fi(M1, . . . , Mni) has a 1 in its j, j-th entry, so fi(M1, . . . , Mni) ∈ Cross(U1, . . . , Un). These arguments establish ⋄ the claim. Claim 4.6. A has a cube term. Proof of Claim 4.6. The fact that the operations of A are defined coordinatewise on {0, 1}n implies that A is a product of its 2-element coordinate factor algebras. Thus, to prove this claim, it is enough to show that each coordinate factor of A has a cube term. That this is enough follows from our Corollary 3.5, combined with Lemma 3.4, or from Corollary 2.5 of [12]. Namely, each result implies that if each algebra in a finite family has a cube term, then the product also has a cube term. But it is easy to see that each coordinate factor of A has a cube term (in fact, a near-unanimity term). To see this, consider the j-th coordinate factor algebra and suppose that fi belongs to j. If arity(fi) > 2, then fi interprets as the canonical ni-ary near-unanimity operation in the j-th coordinate and we are done. If arity(fi) = 2, then fi interprets as binary meet in the j-th coordinate and fi belongs to no coordinate other than j. Since n = kτ k − 1 > 1, there exists some coordinate different from j. Hence there must exist some fk 6= fi belonging to a coordinate other than j. In this case, fk will interpret as nk-ary join in the j-th coordinate. With join coming from fk and meet coming from fi one can construct a ternary near-unanimity operation ⋄ in coordinate j. Claim 4.6 shows that A has a d-cube term for some d. Hence, we can use Theo- rem 4.1 to conclude that A has a kτ k-cube term. On the other hand, by Corollary 2.3, Claim 4.5 prevents A from having a d-cube term for any d < kτ k. This proves all required properties of A. (cid:3) Remark 4.7. Example 4.4 was discovered with the help of UACalc, a universal algebra calculator. After including it here we learned that the preprint [6] by Camp- enella, Conley and Valeriote contains essentially the same example. We are informed that they also discovered the example with the help of UACalc. The purpose of the example in their paper is roughly the same as ours (i.e., lower bounds for dimension estimates), except our application is to cube terms and their application is to near unanimity terms. Example 4.8. In this example our goal is to show that the bound in Corollary 4.3 is sharp. Accordingly, we want to construct, for any suitable finite signature τ , an idempotent variety V such that the free algebra F = FV (x, y) has a compatible symmetric cross of arity τ − 1, but no compatible symmetric crosses of higher arity. If τ is suitable for idempotent varieties and τ = 2, then τ is a signature with binary operation symbols only, say f1, . . . , fm. In this case we can choose V to be CUBE TERM BLOCKERS 19 the variety generated by an algebra hZ3; f1, . . . , fmi where f1(x, y) = 2x + 2y and for 2 ≤ i ≤ m, fi is interpreted as a projection. Since V has a Maltsev term, F has no compatible symmetric crosses of arity ≥ 2. However, Cross({x}) is a compatible symmetric cross of F of arity 1. Let us assume from now on that τ ≥ 3, and let f1, . . . , fm denote the opera- tion symbols in τ where fi is ni-ary (1 ≤ i ≤ m). We may assume without loss of generality that n1 ≥ · · · ≥ nm, so τ = n1 (≥ 3). Now consider an algebra B = h{0, 1}; f1, . . . , fmi where f1 is interpreted on {0, 1} as the "canonical" n1-ary near-unanimity operation (see the definition in Example 4.4), and for 2 ≤ i ≤ m, fi is interpreted as a projection. Let V be the variety generated by B, and let F = FV(x, y). Since V has an n1-ary near-unanimity operation, we know from Corol- lary 2.3 that F has no compatible crosses of arity ≥ n1 = τ . On the other hand, since the (n1 − 1)-ary cross Cross({1}, . . . , {1}) = {0, 1}n1−1 \ {(0, . . . , 0)} is a com- patible relation of B, its inverse image under the homomophism F → B sending x to 1 and y to 0 is a compatible symmetric cross of F of arity n1 − 1 = τ − 1. 5. A fact about cyclic term varieties This note emerged in response to a question we learned from Cliff Bergman: Sup- pose that C2 is the variety defined with one binary operation and axiomatized by the identities (1) w(x, x) = x, and (2) w(x, y) = w(y, x). Is it true that every subvariety of C2 either contains the 2-element semilattice or is congruence permutable? Bergman's question arose out of a certain line of investigation into constraint satis- faction problems. Namely, it is of interest to understand whether the algebras in the pure cyclic term varieties have tractable CSP's. The d-ary pure cyclic term variety Cd is defined with one d-ary operation satisfying (1) w(x, x, . . . , x) = x, and (2) w(x1, x2, . . . , xd) = w(x2, x3, . . . , x1). If one could show that each finite algebra in each variety Cd has tractable associated CSP's, then one would have solved the Feder–Vardi Conjecture. Bergman and David Failing showed in [3] that if V is a subvariety of C2 that is the join of a congruence permutable variety and the variety of semilattices, then the finite algebras in V have tractable associated CSP's. So, Bergman was really asking whether this theorem applied to every subvariety of C2 that is a join of the variety of semilattices and a disjoint subvariety. When Bergman asked the question, he men- tioned that an affirmative answer was supported by extensive computer computations performed by Bergman, William DeMeo, and Jiali Li. 20 KEITH A. KEARNES AND ´AGNES SZENDREI Here we explain why the answer to Bergman's question is affirmative, even with 2 replaced by d. That is, we explain why a subvariety of the d-ary pure cyclic term variety either contains a 2-element semilattice or else has a d-cube term. For this, define a 2-element semilattice in Cd to be an algebra with 2-element universe, say {0, 1} and operation defined by (5.1) w(x1, . . . , xd) =(1 if x1 = x2 = · · · = xd = 1; 0 else. More generally, a 2-element semilattice for a variety V is one in which, for every d, each d-ary fundamental operation satisfies (5.1). Theorem 5.1. A subvariety of the pure d-ary cyclic term variety either has a d-cube term or contains a 2-element semilattice. A finite algebra in the pure d-ary cyclic term variety either has a d-cube term or has a 2-element semilattice section. One should note that a 2-element semilattice has no d-cube term for any d, so the two cases described in the theorem are complementary. Proof. Let V be a subvariety of the pure d-ary cyclic term variety. If V does not have a d-cube term, then by Theorem 2.4 (1) the free algebra F = FV (x, y) has a compatible d-ary cross. But the signature τ of V satisfies d = kτ k, so Claim 4.2 shows that F has a cube term blocker of the form (U, F ). It follows from Lemma 2.1 (5) that the cyclic term of the variety must be U-absorbing in at least one of its variables, so by cyclicity this term is U-absorbing in all of its variables. This implies that (i) the congruence θ on F generated by U ×U is the union of U ×U and the equality relation, and (ii) if t ∈ F \ U is chosen arbitrarily, then S = U ∪ {t} is a subuniverse of F. The algebra S/θS must then be a 2-element semilattice in V. For the second statement, assume that A is a finite algebra in the pure d-ary cyclic term variety and that A does not have a d-cube term. Then V(A) cannot have a d-cube term, so by Theorem 4.1, V(A) cannot have a cube term at all. Therefore Corollary 3.5 guarantees that A has a cube term blocker, say (U, B). Now construct a 2-element semilattice from this blocker in the same manner one was constructed from the blocker (U, F ) in the preceding paragraph. It will be a section of A. (cid:3) What matters to us in Theorem 5.1 is that the d in "d-ary cyclic term" agrees with the d in "d-cube term"; that is, the theorem establishes the existence of a cube term under some condition, and bounds its index. If one is not concerned with such a bound, then one can establish a result about varieties with many cyclic fundamental operations, namely: Theorem 5.2. Let V be an idempotent variety whose fundamental operations each satisfy cyclic identities. That is, for each fundamental operation w(x1, . . . , xn) it is the case that V = w(x1, x2, . . . , xn) = w(x2, x3, . . . , x1). Then V either has a cube term or contains a 2-element semilattice. CUBE TERM BLOCKERS 21 Proof. Let F = FV (x, y). If V has no cube term, then by Theorem 3.3 there is a cube term blocker (U, F ) for F. By repeating the argument in the first paragraph of the proof of Theorem 5.1 we find that V contains a 2-element semilattice. (cid:3) 6. Generic crosses In this section we focus on idempotent varieties of finite type. We show that if a nontrivial member of such a variety has a compatible d-ary cross, then some countably infinite algebra A in the variety has a 'generic' compatible d-ary cross. By a 'generic cross' we mean a cross Cross(U1, . . . , Un) where the sets U1, . . . , Un are as independent as possible. Specifically, when Cross(U1, . . . , Un) is a cross on a countably infinite set A, we call Cross(U1, . . . , Un) generic if every nonzero Boolean combination of the sets U1, . . . , Un is countably infinite. Theorem 6.1. If X is an idempotent variety of finite type and some member of X has a compatible d-ary cross, then some countably infinite member of X has a compatible d-ary generic cross. Proof. If some member of X has a compatible d-ary cross, then by Corollary 2.3, X cannot have a d-cube term. Hence Theorem 3.3 implies that the algebra F = FX (x, y) must have a compatible d-ary cross, say Cross(U1, . . . , Ud) = B1 ∪ · · · ∪ Bd where Bi = F × · · · × F × Ui × F × · · · × F is full in all coordinates except the ith. Here F need not be infinite, and this cross need not be generic, so we modify the situation as follows. Let A ∈ X be a countably infinite algebra. Now define F = Fd × A, U1 = B1 × A, ... Ud = Bd × A. It is easy to see that F is countably infinite and that each Ui is a nonempty proper subuniverse of F . Claim 6.2. Cross(U1, . . . , Ud) is a compatible generic d-ary cross of F . Proof of Claim 6.2. We first argue that Cross(U1, . . . , Ud) is compatible, i.e. a sub- universe of F d. For this we consider F = Fd × A to have d + 1 coordinates, so F d = Fd × A × Fd × A × · · · × Fd × A has d(d + 1) coordinates. Notice that all coordinate algebras in this direct represen- tation of F d are F except those whose coordinates lie in the arithmetical progression d + 1, 2(d + 1), · · · , d(d + 1), in which case the coordinate algebras are A. 22 KEITH A. KEARNES AND ´AGNES SZENDREI There is a projection homomorphism π : F d → Fd which projects onto the coordi- nates in the arithmetic progression 1, (d + 1) + 2, 2(d + 1) + 3, · · · , (d − 1)(d + 1) + d, which projects onto the first coordinate of the first block of d + 1 factors of F d, the second factor of the second block of d + 1 factors, etc. It is not hard to verify that thereby establishing that Cross(U1, . . . , Ud) is compatible. Cross(U1, . . . , Ud) = π−1(cid:0)Cross(U1, . . . , Ud)(cid:1), d To show that Cross(U1, . . . , Ud) is generic, it suffices to show that any intersection 1 ∩ · · · ∩ U εd U ε1 contains infinitely many elements, where εi = ±1 for each i and U +1 1 = U1 while U −1 1 = F \ U1. Observe that a (d + 1)-tuple (u1, . . . , ud, a) belongs to the set U ε1 d exactly when ui ∈ Ui if εi = +1, ui ∈ F \ Ui if εi = −1, and a ∈ A. Such choices are possible since Ui is a nonempty proper subuniverse of F for each i and A is nonempty. If we fix the ui's and let the last coordinate a range over the infinite set A we obtain infinitely many elements in U ε1 ⋄ 1 ∩ · · · ∩ U εd 1 ∩ · · · ∩ U εd d . This completes the proof of Theorem 6.1. (cid:3) Corollary 6.3. The class of varieties having a d-cube term represents a join-prime filter in the lattice of idempotent Maltsev conditions. Proof. If not, then there are idempotent varieties X and Y that do not have a d-cube term, but their coproduct X ⊔ Y does have a d-cube term. But if X ⊔ Y has a d-cube term, then so must X ′ ⊔ Y ′ for some finitely presentable varieties X ′ interpretable in X and Y ′ interpretable in Y. Replacing X and Y by X ′ and Y ′ we may assume that X and Y are finitely presentable, in particular of finite type. We prove the corollary by arguing that if X and Y have finite type and no d-cube term, then X ⊔ Y has no d-cube term. By Theorem 6.1 there exist countably infinite algebras A ∈ X and B ∈ Y which have generic compatible d-ary crosses, say Cross(U1, . . . , Ud) and Cross(V1, . . . , Vd). By genericity, it is possible to find a bijection α : A → B such that α(Ui) = Vi for all i. There is a unique X -structure B′ on B that makes α : A → B′ an isomorphism. Thus Cross(V1, . . . , Vd) = α(Cross(U1, . . . , Ud)) is a compatible cross of B′. Since it is also a compatible cross of B ∈ Y, the algebra on B obtained by merging B and B′ is a model of the identities of X ⊔ Y which has a compatible d-ary cross. This is enough to show that X ⊔ Y has no d-cube term. (cid:3) Remark 6.4. The results in this section were discovered during the 2016 'Alge- bra and Algorithms' workshop after hearing a talk by Matthew Moore on the join- primeness among idempotent linear Maltsev conditions of the condition expressing the existence of a cube term. Later, Jakub Oprsal pointed us to his preprint [14] where he proves our Corollary 6.3 (among other things). Oprsal told us that he learned of our characterization of cube terms in terms of crosses from a talk of Szen- drei at the AAA90 conference in Novi Sad in 2015, and then developed a similar CUBE TERM BLOCKERS 23 characterization of his own which allowed him to prove Corollary 6.3. His discovery of Corollary 6.3 predates ours. His argument depends on an analogue of Theorem 6.1, which he proves for varieties in arbitrary languages. Our proof also works for arbi- trary languages, but we decided not to change ours after learning about Oprsal's work. References [1] Aichinger, Erhard; Mayr, Peter; McKenzie, Ralph On the number of finite algebraic structures. J. Eur. Math. Soc. (JEMS) 16 (2014), no. 8, 1673–1686. [2] Barto, Libor Finitely related algebras in congruence modular varieties have few subpowers. J. Eur. Math. Soc. (JEMS), to appear. [3] Bergman, Clifford; Failing, David Commutative, idempotent groupoids and the constraint sat- isfaction problem. Algebra Universalis 73 (2015), no. 3-4, 391–417. [4] Berman, Joel; Idziak, Pawe l; Markovi´c, Petar; McKenzie, Ralph; Valeriote, Matthew; Willard, Ross Varieties with few subalgebras of powers. Trans. Amer. Math. Soc. 362 (2010), no. 3, 1445–1473. [5] Bulatov, Andrei; Mayr, Peter; Szendrei, ´Agnes The subpower membership problem for finite algebras with cube terms. Preprint. [6] Campanella, Maria; Conley, Sean; Valeriote, Matt Preserving near unanimity terms under products. Algebra Universalis, to appear. [7] Freese, Ralph; Kiss, Emil; Valeriote, Matthew Universal Algebra Calculator, Available at: www.uacalc.org, 2011. [8] Idziak, Pawe l; Markovi´c, Petar; McKenzie, Ralph; Valeriote, Matthew; Willard, Ross Tractabil- ity and learnability arising from algebras with few subpowers. SIAM J. Comput. 39 (2010), no. 7, 3023–3037. [9] Kearnes, Keith A.; Szendrei, ´Agnes Clones of algebras with parallelogram terms. Internat. J. Algebra Comput. 22 (2012), no. 1, 1250005, 30 pp. [10] Kearnes, Keith A.; Szendrei, ´Agnes Dualizable algebras with parallelogram terms. Algebra Uni- versalis, to appear. http://arxiv.org/abs/1502.02192 [11] Kearnes, Keith A.; Tschantz, Steven T. Automorphism groups of squares and of free algebras. Internat. J. Algebra Comput. 17 (2007), no. 3, 461–505. [12] Markovi´c, Petar; Mar´oti, Mikl´os; McKenzie, Ralph Finitely related clones and algebras with cube terms. Order 29 (2012), no. 2, 345–359. [13] Moore, Matthew Naturally dualisable algebras omitting types 1 and 5 have a cube term. Algebra Universalis 75 (2016), 221–230. [14] Oprsal, Jakub Taylor's modularity conjecture and related problems for idempotent varieties. http://arxiv.org/abs/1602.08639v1 [15] Szendrei, ´Agnes Idempotent algebras with restrictions on subalgebras. Acta Sci. Math. (Szeged) 51 (1987), no. 1-2, 251–268. 24 KEITH A. KEARNES AND ´AGNES SZENDREI (Keith Kearnes) Department of Mathematics, University of Colorado, Boulder, CO 80309-0395, USA E-mail address: [email protected] ( ´Agnes Szendrei) Department of Mathematics, University of Colorado, Boulder, CO 80309-0395, USA E-mail address: [email protected]
1609.00271	2	1609	2016-09-20T15:23:00	On structure and TKK algebras for Jordan superalgebras	[ "math.RA" ]	We compare a number of different definitions of structure algebras and TKK constructions for Jordan (super)algebras appearing in the literature. We demonstrate that, for unital superalgebras, all the definitions of the structure algebra and the TKK constructions fall apart into two cases. Moreover, one can be obtained as the Lie superalgebra of superderivations of the other. We also show that, for non-unital superalgebras, more definitions become non-equivalent. As an application, we obtain the corresponding Lie superalgebras for all simple finite dimensional Jordan superalgebras over an algebraically closed field of characteristic zero.	math.RA	math	ON STRUCTURE AND TKK ALGEBRAS FOR JORDAN SUPERALGEBRAS SIGISWALD BARBIER AND KEVIN COULEMBIER Abstract. We compare a number of different definitions of structure algebras and TKK con- structions for Jordan (super)algebras appearing in the literature. We demonstrate that, for unital superalgebras, all the definitions of the structure algebra and the TKK constructions fall apart into two cases. Moreover, one can be obtained as the Lie superalgebra of superderiva- tions of the other. We also show that, for non-unital superalgebras, more definitions become non-equivalent. As an application, we obtain the corresponding Lie superalgebras for all simple finite dimensional Jordan superalgebras over an algebraically closed field of characteristic zero. 1. Introduction There is an acclaimed principle that associates a 3-graded Lie algebra to a Jordan algebra, as developed by Tits, Kantor and Koecher in three variations, see [Ti, Kan, Ko]. These three constructions have natural analogues for Jordan superalgebras and some also extend to Jordan (super)pairs. The principle behind these three constructions, and further variations appearing in the literature, is loosely referred to as "the" TKK construction. A common feature of TKK constructions is that, under the appropriate conditions, they associate simple Lie superalgebras to simple Jordan superalgebras or superpairs. They were as such used to classify simple Jordan superalgebras and superpairs, see [Ka2, CK, KMZ, Kan2, Kr], but also to study representations of Jordan superalgebras, see [MZ, Sh, KS]. When the constructions of Tits, Kantor and Koecher are applied to a simple finite dimensional Jordan algebra over the field of complex numbers, they all yield the same Lie algebra, as follows a posteori from the classification. However, if one applies the TKK constructions to more general algebras, they can give different outcomes. The aim of the paper is to create more structure in this plethora of TKK constructions, by (dis)proving equivalences of some of the definitions under certain conditions and describing concrete links between the different constructions. First we consider the zero component of the 3-graded Lie (super)algebra associated to a Jordan (super)algebra, which is often referred to as the structure algebra. Then we construct the 3-graded Lie superalgebra out of the structure algebra and the Jordan superalgebra. We refer to this algebra as the TKK algebra. We consider four definitions of the structure algebra and show that, for unital Jordan superal- gebras, they lead to two non-equivalent versions of the structure algebra. For non-unital Jordan superalgebras, all four definitions are non-equivalent. For completeness, we also review two further definitions of structure algebras of unital Jordan superalgebras, with no direct link to TKK constructions, and prove that these are both equivalent to one of the above definitions. One of these definitions also applies to non-unital Jordan superalgebras, and we prove that it is non-equivalent to the previous definitions. 1 2 SIGISWALD BARBIER AND KEVIN COULEMBIER Then we consider the TKK algebras. First we introduce Kantor's construction. Koecher's construction appears in several forms in the literature, depending on the choice of structure algebra. Finally, the construction by Tits depends on the structure algebra and an auxiliary three-dimensional Lie algebra, which we assume to be sl2 for now. This yields 5 definitions of TKK superalgebras associated to a Jordan superalgebra V , corresponding to constructions of Tits, Koecher and Kantor: Ti(V, Inn(V ), sl2) Ko(V ) Kan(V ) Ti(V, Der(V ), sl2) fKo(V ) If V is a simple finite dimensional Jordan algebra over the field of complex numbers, it is known that all five Lie algebras are isomorphic. We prove that so long as V is unital, the three Lie superalgebras in the top row are isomorphic. Under the same assumption, the two algebras in the bottom row are then also isomorphic and given by the algebra of derivations of the Lie superalgebras in the top row. For arbitrary V , even when finite dimensional, we show that all five algebras can be pairwise non-isomorphic and that the link between bottom and top row through derivations generally fails. We derive these results for the super case, but they are already pertinent for ordinary Jordan algebras. However, the differences in definitions are more exposed for Jordan superalgebras, as they already appear for finite dimensional simple Jordan superalgebras over the field of complex numbers. Contrary to simple Lie algebras, simple Lie superalgebras can admit outer derivations, and contrary to Jordan algebras, there is a simple finite dimensional Jordan superalgebra which is non-unital. Therefore we apply our results to obtain a table with all versions of the TKK construction for the simple finite dimensional Jordan superalgebras over an algebraically closed field of characteristic zero. For this, we can rely on the classification of simple Jordan superalgebras in [CK, Ka2] and the calculation of derivations in [Ka1, Sc]. We organise the paper as follows. In Section 2 we introduce some concepts and terminology regarding Jordan superalgebras and superpairs. In Section 3 we investigate the different defini- tions of the structure algebra. In Section 4 we compare the constructions of Tits, Kantor and Koecher. In Section 5 we study further variations of the Koecher construction, based on the choice of structure algebra. In Section 6 we use the above to list all the versions of the TKK al- gebras for the finite dimensional simple Jordan superalgebras over an algebraically closed field of characteristic zero. Finally, in Appendix A we introduce the notation for the Lie superalgebras of type A, P , Q and H as used in Section 6. 2. Jordan superalgebras and superpairs In the following, we will consider a super vector space V = V¯0 ⊕ V¯1 over a field K. For an element x of Vi we write x = i for i ∈ {¯0, ¯1} = Z2. By hAi we will denote the K-linear span of a set A. As is customary in the theory of Jordan superalgebras and superpairs, we will always assume that the characteristic of K is different from 2 and 3. At this stage we make no other assumptions on K. We note furthermore that the main results of section 3, 4 and 5 still hold if we replace K by a ring containing 1 2 and 1 3 . 2.1. Jordan superalgebras. STRUCTURE AND TKK ALGEBRAS 3 Definition 2.1. A Jordan superalgebra is a super vector space V equipped with a bilinear product which satisfies • ViVj ⊂ Vi+j, i, j ∈ Z2 • xy = (−1)xyyx (commutativity) • (−1)xz[Lx, Lyz] + (−1)yx[Ly, Lzx] + (−1)zy[Lz, Lxy] = 0 (Jordan identity), for all homogeneous x, y, z ∈ V . Here, the operator Lx : V → V is defined by Lx(y) = xy. A Jordan superalgebra V is unital if there exists an element e ∈ V such that ex = x = xe for all x ∈ V . We stress that we do not restrict to finite dimensional algebras. A Jordan superalgebra satisfies the following relation, see [Ka2, Section 1.2], (1) [[Lx, Ly], Lz] = Lx(yz) − (−1)xyLy(xz). Define the following operators on V : Dx,y := 2Lxy + 2[Lx, Ly]. The Jordan triple product is given by (2) {x, y, z} := Dx,yz = 2(cid:16)(xy)z + x(yz) − (−1)xyy(xz)(cid:17) . This triple product satisfies the symmetry property {x, y, z} = (−1)xy+yz+xz{z, y, x}, and the 5-linear Jordan identity {x, y, {u, v, w}} − {{x, y, u}, v, w} = (−1)(x+y)(u+v)(−{u, {v, x, y}, w} + {u, v, {x, y, w}}). The 5-linear identity can be rewritten as (3) [Dx,y, Du,v] = D{x,y,u},v − (−1)(x+y)(u+v)Du,{v,x,y} = Dx,{y,u,v} − (−1)(x+y)(u+v)D{u,v,x},y. 2.2. Jordan superpairs. A Jordan superpair is a pair of super vector spaces (V +, V -) equipped with two even trilinear products, known as the Jordan triple products, {·, ·, ·}σ : V σ × V -σ × V σ → V σ, for σ ∈ {+, −}. These triple products satisfy symmetry in the outer variables {x, y, z}σ = (−1)xy+yz+zx{z, y, x}σ, and the 5-linear identity {x, y, {u, v, w}σ }σ − {{x, y, u}σ, v, w}σ = (−1)(x+y)(u+v)(−{u, {v, x, y}-σ, w}σ + {u, v, {x, y, w}σ }σ), for homogeneous x, z, u, w ∈ V σ and y, v ∈ V -σ. We will use the following operators Dx,y : V σ → V σ; z 7→ {x, y, z}σ, 4 SIGISWALD BARBIER AND KEVIN COULEMBIER for x ∈ V σ and y ∈ V -σ. Example 2.2. By the previous subsection, the doubling of a Jordan superalgebra V gives a Jordan superpair (V +, V -) := (V, V ) with products {xσ, y−σ, zσ}σ := {x, y, z} for σ ∈ {+, −}. Here we use the notation x+, resp. x-, for an element x ∈ V interpreted as in V +, resp. V -. When the context clarifies in which space we interpret x ∈ V , we will leave out the indices. In the following sections we will often omit the σ in the notation for the triple product since it can be derived from the elements it acts on. 3. Derivations and the structure algebra In this section we show that the (inner) structure algebra of a unital Jordan superalgebra is isomorphic to the algebra of (inner) derivations of the corresponding superpair. We provide counterexamples to both claims when the Jordan superalgebra is non-unital. 3.1. The structure algebra for a Jordan superalgebra. Definition 3.1. Let V be a Jordan superalgebra. An element D in End(V ) is called a deriva- tion of V if D(xy) = D(x)y + (−1)xDxD(y). We use the notation Der(V ) for the space of derivations, and Inn(V ) for the subspace of inner derivations, which is spanned by the operators [Lx, Ly] for x, y ∈ V . The condition on D ∈ End(V ) to be a derivation is equivalent with (4) [D, Lx] = LD(x) for all x ∈ V . Hence equation (1) implies that [Lx, Ly] is a derivation. One verifies easily that Der(V ) is a sub- algebra of gl(V ). The Jacobi identity on gl(V ) combined with equation (4), for any derivation D, implies that Inn(V ) is an ideal in Der(V ). We will use the following definition for the structure algebra of Jordan superalgebras, since this is the one that will be required for the Kantor functor. There exist other definitions of the structure algebra in the literature which are not immediately connected to TKK constructions. We will review them in Section 3.4 and show that for unital Jordan superalgebras they are all equivalent to our definition. Definition 3.2. The structure algebra str(V ) is a subalgebra of gl(V ), defined as Definition 3.3. The inner structure algebra istr(V ) is a subalgebra of gl(V ), defined as str(V ) = {Lx x ∈ V } + Der(V ). istr(V ) = {Lx x ∈ V } + Inn(V ) = hLx, [Lx, Ly] x, y ∈ V i. By the above, istr(V ) is an ideal in str(V ). Remark 3.4. For a unital Jordan superalgebra the sum in Definitions 3.2 and 3.3 is a direct sum of super vector spaces: str(V ) = {Lx x ∈ V } ⊕ Der(V ) and istr(V ) = {Lx x ∈ V } ⊕ Inn(V ), since D(e) = 0 for all D in Der(V ), while Lx(e) = x. For non-unital Jordan superalgebras the sums are not necessarily direct, as follows from Example 3.5 and Remark 3.18. Example 3.5. Consider the commutative three dimensional algebra V = he1, e1, e3i with product STRUCTURE AND TKK ALGEBRAS 5 e2 1 = e1, e1e2 = 1 2 e2, e2 2 = e3, and all other products of basis elements zero. This is the Jordan algebra J19 in [KM, Sec- tion 3.3.3]. Because we conclude that Le2 is an element of Inn(V ). [Le1, Le2] = − 1 2 Le2, 3.2. Derivations of Jordan superpairs. Definition 3.6. Let (V +, V -) be a Jordan superpair. An element D = (D+, D-) ∈ End(V +) ⊕ End(V -) is called a derivation of (V +, V -) if Dσ({x, y, z}) = {Dσ(x), y, z} + (−1)xD-σ{x, D-σ(y), z} + (−1)(x+y)Dσ{x, y, Dσ(z)}. We use the notation Der(V +, V -) for the space of all derivations of (V +, V -) and the notation Inn(V +, V -) for the subspace of inner derivations, which is spanned by the operators Dx,y := (Dx,y, −(−1)xyDy,x), for x ∈ V +, y ∈ V -. Observe that any derivation (D+, D-) can be written as the sum of derivations where D+ and D- have the same parity. The space Der(V +, V -) hence inherits a grading from the super vector space End(V +) ⊕ End(V -). By construction, the space Der(V +, V -) is a subalgebra of gl(V +) ⊕ gl(V -). The operator D = (D+, D-) ∈ End(V +) ⊕ End(V -) is a derivation if and only if (5) [Dσ, Dx,y] = DDσ(x),y + (−1)DxDx,D-σ(y). One can then easily verify that Inn(V +, V -) is an ideal in Der(V +, V -). 3.3. Connections. The main result of this section is the following connection between the structure algebra of a unital Jordan superalgebra and the derivations of the associated Jordan superpair in Example 2.2. Proposition 3.7. For a unital Jordan superalgebra V we have (1) str(V ) ∼= Der(V, V ), (2) istr(V ) ∼= Inn(V, V ). Remark 3.8. For a unital Jordan algebra, we have that str(V ) is the Lie algebra of the structure group (Section 3.4) and that Der(V, V ) is the Lie algebra of the automorphism group of the Jordan pair (V, V ), ([Lo, I.1.4]. Since the structure group is isomorphic to the automorphism group of the Jordan pair, [Lo, Proposition 1.8], we can immediately conclude that str(V ) ∼= Der(V, V ). Remark 3.9. Both parts of the proposition do not extend, as stated, to non-unital Jordan superalgebras. As a counterexample consider again Example 3.5. One can easily check that istr(V ) = hDx,y x, y ∈ V i = hLe1, Le2i and Inn(V, V ) = h(Le1, −Le1), (Le2 , 0), (0, Le2 )i. Define A ∈ End(V ) by A(e1) = 0, A(e2) = 2e2 and A(e3) = e3. Then we also obtain str(V ) = istr(V ) + hAi and Der(V, V ) = Inn(V, V ) + h(A, A), (A, −A)i. 6 SIGISWALD BARBIER AND KEVIN COULEMBIER Subsection 6.2 also contains a counterexample where istr(V ) ∼= Inn(V, V ) but str(V ) 6∼= Der(V, V ). Even without the existence of a multiplicative identity e, we still have a chain of inclusions if Lx is not a derivation for any x in V . Proposition 3.10. For a Jordan superalgebra V for which Lx 6∈ Der(V ), for all x in V , we have Inn(V, V ) ⊂ istr(V ) ⊂ str(V ) ⊂ Der(V, V ). Remark 3.11. Examples where the second inclusion is strict can be found in Subsection 6.1 while an example for the third inclusion to be strict can be found in Subsection 6.2. Now we start the proofs of the propositions. Lemma 3.12. Let V be a Jordan superalgebra. For x in V and D in Der(V ), we have that are elements of Der(V, V ). (Lx, −Lx) and (D, D) Proof. Using the Jordan identity and equation (1), we get for x, y, z in V [Lx, Dy,z] = 2[Lx, Lyz] + 2[Lx, [Ly, Lz]] = −2(−1)x(y+z)[Ly, Lzx] − 2(−1)z(x+y)[Lz, Lxy] + 2L(xy)z − 2(−1)xyLy(xz) = DLx(y),z − (−1)xyDy,Lx(z). Thus (Lx, −Lx) satisfies equation (5) and hence belongs to Der(V, V ). Let D ∈ Der(V ). By equation (4), the Jacobi identity and the definition of Der(V ), we find [D, Dx,y] = 2[D, Lxy] + 2[D, [Lx, Ly]] = 2LD(xy) − 2(−1)D(x+y)[Lx, [Ly, D]] − 2(−1)y(D+x)[Ly, [D, Lx]] = 2LD(x)y + 2[LD(x), Ly] + 2(−1)DxLxD(y) + 2(−1)xD[Lx, LD(y)] = DD(x),y + (−1)xDDx,D(y). Therefore, also (D, D) satisfies equation (5) and is thus an element of Der(V, V ). (cid:3) Proof of Proposition 3.10. Since Dx,y = 2Lxy + 2[Lx, Ly], the map (6) ψ : Inn(V, V ) → istr(V ); Dx,y = (Dx,y, −(−1)xyDy,x) 7→ Dx,y is well-defined and clearly a Lie superalgebra morphism. Assume Dx,y = 0. Then Lxy = −[Lx, Ly] is a derivation. So by our assumption Lxy = 0, an thus also Dy,x = 0. Therefore ψ is injective. From the definitions it follows immediately that istr(V ) ⊂ str(V ). By assumption, str(V ) is a direct sum of {Lx x ∈ V } and Der(V ). Together with Lemma 3.12 this implies that the map (7) φ : str(V ) → Der(V, V ); Lx + D 7→ (Lx + D, −Lx + D), is well-defined. This map is clearly injective. The fact that this is also a Lie superalgebra morphism follows from a direct computation, which finishes the proof. (cid:3) STRUCTURE AND TKK ALGEBRAS 7 The rest of this section is devoted to the proof of Proposition 3.7, so we consider a unital Jordan superalgebra V . From Remark 3.4 it follows then that the assumption of Proposition 3.10 is satisfied, so we can use that result. We will also use the following immediate consequences of equation (2), (8) Consider the map σ 1 2 {x, e, y} = xy = Lx(y) and 1 2 {e, x, e} = x. σ : Der(V, V ) → Der(V, V ); (D+, D-) 7→ (D-, D+). Then σ2 = id and Der(V, V ) decomposes in two subspaces h := {D ∈ Der(V, V ) σ(D) = D} and q := {D ∈ Der(V, V ) σ(D) = −D}. Lemma 3.13. We have h = {D ∈ Der(V, V ) D-(e) = 0}. Proof. Assume first that D-(e) = 0. By equation (8), we have D-(e) = 1 2 D-{e, e, e} = 1 2 {e, D+(e), e} = D+(e), and hence also D+(e) = 0. Then we also get for all x in V 2D+(x) = D+{e, x, e} = {e, D-(x), e} = 2D-(x). Hence D+ = D-. Now assume that D+ = D-. Equation (8) then implies D-(e) = 1 2 D-{e, e, e} = 1 2 {D-(e), e, e} + 1 2 {e, D+(e), e} + 1 2 {e, e, D-(e)} = 3D-(e). Hence D-(e) = 0. This concludes the proof. (cid:3) Lemma 3.14. We have a Lie superalgebra isomorphism Der(V ) ∼= h, given by φ : Der(V ) → h; D 7→ (D, D). Proof. The map φ is a restriction to Der(V ) of the morphism φ : str(V ) → Der(V, V ) defined in (7). From there we know that it is injective. The image of φ is clearly contained in h. To show that φ is surjective, we let D = (D+, D-) be an element of Der(V +, V -) with D+ = D-, i.e. D ∈ h. Then D-(e) = 0 by Lemma 3.13. Hence, using equation (8), D+(xy) = {x, D-e, y} + (−1)Dx 1 2 {x, e, D+y} 1 2 1 2 D+{x, e, y} {D+x, e, y} + (−1)Dx 1 2 = D+(x)y + (−1)DxxD+(y). = We conclude that D+ = D- is an element of Der(V ), so (D+, D-) is in the image of φ. (cid:3) Lemma 3.15. We have an isomorphism of super vector spaces {Lx x ∈ V } → q, given by Lx 7→ (Lx, −Lx). 8 SIGISWALD BARBIER AND KEVIN COULEMBIER Proof. The assignment La → (La, −La) is a restriction to {Lx x ∈ V } of the injective morphism φ : str(V ) → Der(V, V ) considered in (7). Its image is clearly contained in q. So the map is well-defined and injective. For an element D = (D, −D) in q we claim that (LD(e), −LD(e)) = D. Indeed, using equation (8), we have D(x) = 1 2 D({e, x, e}) = 1 2 {D(e), x, e} − 1 2 {e, D(x), e} + (−1)Dx 1 2 {e, x, D(e)} = 2D(e)x − D(x), which implies that D(x) = LD(e)(x). This proves surjectivity. (cid:3) Proof of Proposition 3.7. Consider the injective morphism φ : Der(V ) ⊕ {Lx x ∈ V } → Der(V, V ); Lx + D 7→ (Lx + D, −Lx + D) of (7). From Lemmata 3.14 and 3.15 it follows that φ is also surjective. This proves part (1) of the proposition. For part (2), consider the injective morphism ψ : Inn(V, V ) → istr(V ); (Dx,y, −(−1)xyDy,x) 7→ Dx,y. defined in (6). Since 1 4 the map ψ is surjective, which concludes the proof. Dx,e and [Lx, Ly] = Lx = 1 2 (Dx,y, −(−1)xyDy,x), (cid:3) 3.4. Alternative definitions for the (inner) structure algebra. We review some further definitions appearing in the literature. Set Ux,y : V → V ; z 7→ (−1)yz{x, z, y}. Then for a unital Jordan algebra we define, see [GN, Section 3.1], fstr(V ) := {X ∈ gl(V ) UX(a),b+(−1)XbUa,X(b) = XUa,b+(−1)X(a+b)Ua,bX ∗ for all a, b ∈ V }, where X ∗ = −X + 2LX(e). In the non-super case, this algebra is the Lie algebra of the structure group, see [Ja, Section 9]. In the literature we did not find an explicit definition of the structure algebra for the non-unital case using this approach. However, we will define a natural generalization which for a unital the Lie superalgebra consisting of the elements (X, Y ) ∈ gl(V ) ⊕ gl(V )op for which Jordan superalgebra will reduce to fstr(V ). So, for V a Jordan superalgebra, define strw(V ) as UX(a),b + (−1)XaUa,X(b) = XUa,b + (−1)Y (a+b)Ua,bY and UY (a),b + (−1)Y aUa,Y (b) = Y Ua,b + (−1)X(a+b)Ua,bX (9) holds for all a, b in V . If V is a Jordan algebra, one can check that strw(V ) is the Lie algebra of the group consisting of pairs of 'weakly structural transformations', as defined in [McC, II.18.2]. Using the equality Ux,y(z) = (−1)yzDx,z(y), one finds that the defining conditions of strw(V ) are equivalent with (X, −Y ) ∈ Der(V, V ). So we conclude that strw(V ) ∼= Der(V, V ) in full generality. Lemma 3.16. For a unital Jordan superalgebra V , we have strw(V ) ∼= fstr(V ) ∼= str(V ). STRUCTURE AND TKK ALGEBRAS 9 Proof. Let X be an element of fstr(V ). Note that by definition X satisfies equation (9) for Y = −X + 2LX(e). From Lemma 3.12 we know that (LX(e), −LX(e)) ∈ Der(V, V ). Combining this, one shows easily that UY (a),b + (−1)Y bUa,X(b) = Y Ua,b + (−1)X(a+b)Ua,bX for all a, b ∈ V holds for Y = −X + 2LX(e). Thus the map ϕ : fstr(V ) → strw(V ); X 7→ (X, −X + 2LX(e)) is well-defined. Let (X, Y ) ∈ strw(V ). Then setting a and b equal to the unit e in equation (9) gives us Y = −X + 2LX(e), hence ϕ is an isomorphism. Since strw(V ) ∼= Der(V, V ), Proposition 3.7, immediately implies that strw(V ) is also isomorphic to str(V ). (cid:3) The inner structure algebra is also often defined as the Lie superalgebra spanned by the operators Dx,y ∈ End(V ), see for example [Ja, Section 9], [Sp, Chapter 4] and [GN, Section 3.1]. For this algebra we will use the notation Lemma 3.17. For a Jordan superalgebra V for which Lx 6∈ Der(V ), for all x in V , we have In particular, for a unital Jordan superalgebra, we have fistr(V ) := hDx,y x, y ∈ V i. fistr(V ) ∼= Inn(V, V ). fistr(V ) ∼= istr(V ). Proof. By assumption Lxy 6∈ Der(V ), so we have that Dx,y = 0 implies Dy,x = 0. Hence algebra morphism. This proves the first part of the lemma. Since unital Jordan superalgebras satisfy Lx 6∈ Der(v), for all x in V , Proposition 3.7 immediately implies the second part of the lemma. (cid:3) the map Inn(V, V ) → fistr(V ); (Dx,y, −(−1)xyDy,x) 7→ Dx,y is bijective. It is also clearly an Remark 3.18. In the non-unital case we can both have fistr(V ) 6∼= istr(V ) and fistr(V ) 6∼= Inn(V, V ). The example in Remark 3.9 is a counterexample for the second part, while a coun- terexample for the first part is as follows. Consider V := tK[t]/(tk), the algebra of polynomials in the variable t without constant term, modulo the ideal (tk) = tkK[t] of polynomials without term in degree lower than k for some k ∈ Z>2. This is an (associative) Jordan algebra, for the standard multiplication of polynomials, which does not have multiplicative identity. In this example we have Df,g = 2Lf g, for all f, g ∈ V . We hence find that On the other hand, by definition, we have Inn(V, V ) ∼= fistr(V ) = SpanK{Lt2, Lt3 , . . . , Ltk−2 }. istr(V ) = SpanK{Lt, Lt2 , . . . , Ltk−2 }. As the dimensions of both abelian Lie algebras do not agree, we find istr(V ) 6∼= fistr(V ). Observe further that Ltk−2 is an element of Der(V ). Therefore the structure algebra str(V ) also does not have a direct sum decomposition as in Remark 3.4 4. The Tits-Kantor-Koecher construction In this section, we will study the three different TKK constructions, dating back to Tits, Kantor and Koecher, and show that, for unital Jordan superalgebras, they are equivalent. Again this claim does not extend to non-unital cases. 10 SIGISWALD BARBIER AND KEVIN COULEMBIER 4.1. TKK for Jordan superalgebras (Kantor's approach). In [Ka2], Kac uses the "Kantor functor" Kan to classify simple finite dimensional Jordan superalgebras over an algebraically closed field of characteristic zero. This functor is a generalisation to the supercase of the one considered by Kantor in [Kan]. In particular this functor provides a TKK construction, which we review for arbitrary Jordan superalgebras over arbitrary fields. We associate to a Jordan superalgebra V , the 3-graded Lie superalgebra Kan(V ) := g = g− ⊕ g0 ⊕ g+, with g0 := istr(V ) = hLa, [La, Lb]i ⊂ End(g−). Finally, g+ is defined as the subspace hP, [La, P ]i of End(g− ⊗ g−, g−), with g− := V and P (x, y) = xy and [La, P ](x, y) := a(xy) − (ax)y − (−1)xy(ay)x. Note that P = −[Le, P ] for a unital Jordan superalgebra. As the notation suggests, [La, P ] corresponds to the superbracket of La ∈ g0 and P ∈ g+. The Lie superbracket is then completely defined by • [g−, g−] = 0 = [g+, g+]. • [a, x] = a(x), for a ∈ g0, x ∈ g−. • [A, x](y) = A(x, y), for A ∈ g+, x, y ∈ g−. • [a, B](x, y) = a(B(x, y)) − (−1)aBB(a(x), y)) − (−1)aB+xyB(a(y), x), for a ∈ g0, B ∈ g+ and x, y ∈ g−. To verify that Kan(V ) is a Lie superalgebra, one can use the following relations (see Proposi- tion 5.1 in [CK]) • [P, x] = Lx • [[La, P ], x] = [La, Lx] − Lax • [La, [Lb, P ]] = −[Lab, P ] • [[La, Lb], P ] = 0 • [[La, Lb], [Lc, P ]] = (−1)bc[La(cb)−(ac)b, P ]. 4.2. TKK for Jordan superpairs (Koecher's approach). In [Ko], Koecher defined a prod- uct on a triple consisting of two vector spaces and a Lie algebra acting on these vector spaces. This product makes the triple into a 3-graded anti-commutative algebra, which is a Lie algebra if and only if the vector spaces form a Jordan pair and the Lie algebra acts by derivations on the vector spaces. Hence the Koecher construction gives rise to a TKK construction, not only for Jordan algebras, but for Jordan pairs, which is the most natural formulation. Note that, as the concept of Jordan pairs was not yet studied at the time, Koecher did not use this terminology. This TKK construction can be generalised to the supercase, which was for example used by Krutelevich to classify simple finite dimensional Jordan superpairs over an algebraically closed field in characteristic zero in [Kr]. We associate a 3-graded Lie superalgebra Ko(V +, V -) to the Jordan superpair (V +, V -) in the following way. As vector spaces we have Ko(V +, V -) = V + ⊕ Inn(V +, V -) ⊕ V -. The Lie super bracket on Ko(V +, V -) is defined by STRUCTURE AND TKK ALGEBRAS 11 [x, u] = Dx,u [Dx,u, y] = Dx,u(y) = {x, u, y} [Dx,u, v] = Dx,u(v) = −(−1)xu{u, x, v} [Dx,u, Dy,v] = DDx,u(y),v + (−1)(x+u)yD y,Dx,u(v) [x, y] = [u, v] = 0, for x, y ∈ V +, u, v ∈ V -. Recall that Dx,u = (Dx,u, −(−1)xuDu,x) ∈ Inn(V +, V −). In case V is a Jordan superalgebra, we simply write Ko(V ) for Ko(V, V ). Conversely, with each 3-graded Lie superalgebra g = g−1 ⊕ g0 ⊕ g+1 we can associate a Jordan superpair by J (g) = (g+1, g−1) with the Jordan triple product given by {xσ, y−σ, zσ}σ := [[xσ, y−σ], zσ]. Definition 4.1. A 3-graded Lie superalgebra g = g− ⊕ g0 ⊕ g+ is called Jordan graded if [g+, g−] = g0 and g0 ∩ Z(g) = 0. We have the following result by Lemmata 4 and 5 in [Kr]. Proposition 4.2. For every Jordan superpair (V +, V -), we have J (Ko(V +, V -)) ∼= (V +, V -). Let g be a Jordan graded Lie superalgebra, then Ko(J (g)) ∼= g. Note that the main results in [Kr] are only concerned with finite dimensional pairs, over alge- braically closed fields with characteristic zero. However, the mentioned lemmata still hold for arbitrary Jordan superpairs over a field with characteristic different from 2 or 3. 4.3. Connection. The main result of this section is the following proposition, which shows that Kantor's and Koecher's constructions for unital Jordan superalgebras coincide. Proposition 4.3. For a unital Jordan superalgebra V , we have Kan(V ) ∼= Ko(V ). Proof. Let Kan(V ) = g+ ⊕ g0 ⊕ g−. The relations [P, a] = La and [[La, P ], b] = [La, Lb] − Lab in Subsection 4.1 imply [g−, g+] = g0. For all x ∈ g0, it follows from the definition of the bracket that, if [x, g−] = 0, then x = 0. So g0 ∩ Z(g) = 0 and Kan(V ) is Jordan graded. Hence Proposition 4.2 implies Ko(J (Kan(V ))) ∼= Kan(V ). Set (V +, V -) := J (Kan(V )). Then V - = V and V + = hP, [La, P ] a ∈ V i. One can check that the map φ : (V +, V -) → (V, V ) defined by • φ(x) = x for x ∈ V -, • φ(P ) = − e 2 , where e is the unit of V , • φ([La, P ]) = a 2 . 12 SIGISWALD BARBIER AND KEVIN COULEMBIER is an isomorphism of Jordan pairs. From this it follows that Ko(V ) = Ko(V, V ) ∼= Ko(J (Kan(V ))) ∼= Kan(V ), which proves the proposition. (cid:3) Remark 4.4. The proposition as stated does not extend to Jordan superalgebras without multiplicative identity e. If V is finite dimensional but not unital, we will generally have dim Kan(V )+ 6= dim V = dim Ko(V )+, and hence Kan(V ) 6∼= Ko(V ). This difference in dimension can for instance be caused by the occurrence of elements of V for which the left multiplication operator is trivial , since this lowers the dimension of hP, [La, P ]i, or by P 6∈ h[La, P ]i, which raises the dimension. Another source of counterexamples comes from Jordan superalgebras V which satisfy Inn(V, V ) 6= istr(V ), see e.g. Remark 3.9. 4.4. Tits' approach. There is a third version of the TKK construction, which appeared in [Ti] and historically was the first to appear. In this section, we will give the super version of this construction by Tits. Consider an arbitrary Jordan superalgebra V . Let D be a Lie superalgebra, containing Inn(V ), with a Lie superalgebra morphism ψ : D → Der(V ); d 7→ ψd, such that ψ acts as the identity on the subalgebra Inn(V ). Finally, let Y be an arbitrary three- dimensional simple Lie algebra Y . For example, for K = C, we only have Y ∼= sl2(C) and for K = R either Y ∼= sl2(R) ∼= su(1, 1) or Y ∼= su(2). Let (y, y′) := 1 2 tr(ad(y)ad(y′)) be the Killing form on Y . Then we define a Lie superalgebra where D is a subalgebra, and the rest of the multiplication is defined by Ti(V, D, Y ) := D ⊕ (Y ⊗ V ), [d, y ⊗ v] = y ⊗ ψd(v), [y ⊗ v, y′ ⊗ v′] = (y, y′)[Lv, Lv′ ] + [y, y′] ⊗ vv′, for arbitrary d ∈ D, y, y′ ∈ Y and v, v′ ∈ V . For Y = sl2(K) we can use the 3-grading on sl2(K) to define a 3-grading on Ti(V, D, sl2(K)): Ti(V, D, sl2(K))- = sl2(K)-⊗V, Ti(V, D, Y )0 = D⊕(sl2(K)0⊗V ), Ti(V, D, sl2(K))+ = sl2(K)+⊗V. For unital Jordan superalgebras, this contains, as a special case, Koecher's and hence also Kantor's construction, as we prove in the following proposition. Proposition 4.5. For a unital Jordan superalgebra V we have Ti (V, Inn(V ), sl2(K)) ∼= Ko(V ). Proof. Consider a K-basis e, f, h of sl2(K), such that [e, f ] = h, [h, e] = 2e and [h, f ] = −2f . For a unital Jordan superalgebra V , we have Inn(V, V ) ∼= istr(V ), by Proposition 3.7(2). Then an isomorphism between Ti (V, Inn(V ), sl2(K)) and Ko(V ) = V + ⊕ istr(V ) ⊕ V - is given by e ⊗ a 7→ a+, f ⊗ a 7→ a-, h ⊗ a 7→ 2La, [La, Lb] 7→ [La, Lb]. It follows from the definitions that this is a Lie superalgebra morphism. STRUCTURE AND TKK ALGEBRAS 13 (cid:3) Remark 4.6. From the proof, it is clear that the proposition still holds for non-unital Jordan superalgebras so long as istr(V ) ∼= Inn(V, V ). Now we consider the opposite direction of the above construction. Let N be a Lie superalgebra and Y a simple Lie algebra of dimension 3. We say that Y acts on N if there is an (even) Lie superalgebra homomorphism from Y to Der(N ). For example, we can define an action of Y on Ti (V, D, Y ) as follows y · (d + y′ ⊗ v) = [y, y′] ⊗ v. Under this action, Ti (V, D, Y ) viewed as an Y -module decomposes as a trivial part given by D and dim(V ) copies of the adjoint representation. Now consider an arbitrary Lie superalgebra N with Y -action which decomposes as above, viz. as a trivial representation D and some copies of the adjoint representation, N = D ⊕ (Y ⊗ A), for some vector space A. As a direct generalisation of [Ti], we show that there is a Jordan algebra structure on A where D acts on A by derivations and Ti (V, D, Y ) is the inverse of this construction. Proposition 4.7. Let N be a Lie superalgebra and Y a 3-dimensional simple Lie algebra which acts on N such that N decomposes as N = D ⊕ (Y ⊗ A) where D is a trivial representation and Y the adjoint representation. Then A is a Jordan superalgebra and D is a superalgebra containing the inner derivations on A equipped with a morphism ψ : D → Der(A), for which the restriction to the inner derivations is the identity. Furthermore N ∼= Ti(A, D, Y ). Proof. Proposition 1 in [Ti] and its proof, which extend trivially to the super case, imply that under these conditions, D is a subalgebra of N , and there are bilinear maps α(·, ·) : D × A → A, h·, ·i : A × A → D and µ : A × A → A, such that [d, y ⊗ a] = y ⊗ α(d, a) and [y ⊗ a, y′ ⊗ a′] = (y, y′)ha, a′i + [y, y′]µ(a, a′). Furthermore (A, µ) is a Jordan superalgebra and d 7→ α(d, ·) is a Lie superalgebra morphism φ : D → Der(A). Finally, by equation (2.6) in [Ti], we have Comparison with the definition of Ti(A, D, Y ) concludes the proof. (cid:3) φ(ha, bi) = [La, Lb]. 5. Further TKK constructions In this section we consider variations of the TKK constructions for a Jordan superalgebra V , which also appear in the literature, by using str(V ) and Der(V, V ), instead of istr(V ) and Inn(V, V ). 14 SIGISWALD BARBIER AND KEVIN COULEMBIER 5.1. Definition. The Lie superalgebra Ti(V, D, sl2) had more freedom compared to the con- structions by Kantor and Koecher, due to the choice of D. Also in the Koecher construction, we can replace Inn(V +, V -) by any Lie superalgebra containing Inn(V +, V -) with a morphism to Der(V +, V -) which restricts to the identity on the inner derivations. For example, we can set g0 = Der(V +, V -) in the TKK construction of Section 4.2. This gives a 3-graded Lie superalge- bra fKo(V +, V -) = V + ⊕ Der(V +, V -) ⊕ V -, is the superalgebra of derivations of Ko(V ) for unital Jordan superalgebras. We can also relate see [GN] for more details. Remark that, by construction, Ko(V +, V -) is an ideal in fKo(V +, V -). We will again use the notation fKo(V ) for fKo(V, V ). In Subsection 5.3, we will prove that fKo(V ) fKo(V ) to the Tits' construction in Subsection 4.4 as follows. fKo(V ) ∼= Ti(V, Der(V ), sl2(K)). Lemma 5.1. For a unital Jordan superalgebra V , we have This will be proved in greater generality in Subsection 5.2. 5.2. Comparison of further TKK constructions. Let D be a Lie superalgebra containing Inn(V ) with a morphism ψ to Der(V ) such that ψInn(V ) = id. Define the Lie superalgebra where D is a subalgebra of eD, the product of Lx and Ly is given by [Lx, Ly] interpreted via the embedding of Inn(V ) in D, and [D, Lx] := Lψ(D)x for D ∈ D, x ∈ V . eD := D ⊕ {Lx x ∈ V }, Set eψ : eD → Der(V, V ); D + Lx 7→ (ψ(D) + Lx, ψ(D) − Lx). From Lemma 3.12, it follows that this map is well defined, while from the definition of the bracket on eD it follows that it is a Lie superalgebra morphism. The morphism eψ yields an action of eD on V + and V -, which allows us to define a TKK construction similar to the Koecher construction in Subsection 4.2. Concretely, the bracket on KoD(V ) := V ⊕ eD ⊕ V [d, x] = eψ(d)x, [d1, d2] = [d1, d2] eD, [d, u] = eψ(d)u, [x, y] = 0 = [u, v], is given by [x, u] = 2Lxu + 2[Lx, Lu], for x, y in V +, u, v in V -, d, d1, d2 in eD and [·, ·] eD the product in eD. Proposition 5.2. Consider a (not necessarily unital) Jordan superalgebra V and a Lie super- algebra D as above. We have an isomorphism of Lie superalgebras Ti(V, D, sl2(K)) ∼= KoD(V ). Proof. The following generalisation of the map used in Proposition 4.5 e ⊗ a 7→ a+, f ⊗ a 7→ a-, h ⊗ a 7→ 2La, D 7→ D, is an isomorphism between Ti(V, D, sl2(K)) and KoD(V ). (cid:3) STRUCTURE AND TKK ALGEBRAS 15 The case D = Inn(V ) yields Ti(V, Inn(V ), sl2(K)) ∼= KoInn(V )(V ), with ^Inn(V ) = {Lx x ∈ V } ⊕ Inn(V ). This is a generalisation of Proposition 4.5 to the non-unital case. Note that if there exists an x ∈ V , such that Lx is in Der(V ), then ^Der(V ) contains two copies of Lx, one in Der(V ) and one in {Lx x ∈ V }. The isomorphism between ^Der(V ) and Der(V, V ) maps the first to (Lx, Lx), while the second copy gets mapped to (Lx, −Lx). For unital Jordan superalgebras we have canonical isomorphisms ^Inn(V ) ∼= istr(V ) and ^Der(V ) ∼= str(V ), by Remark 3.4, and thus KoInn(V )(V ) = Ko(V ) and KoDer(V )(V ) = fKo(V ). Hence we find that Proposition 5.2 implies Lemma 5.1. Remark 5.3. Let g be an arbitrary 3-graded Lie superalgebra and set (V +, V -) = J (g). Then we have a morphism of Lie superalgebras g0 → Der(V +, V -); x 7→ (adx g+ , adx g- ), V , then one can easily check that I = 0 (and thus D ⊆ Der(V )) is equivalent with the condition that the only ideal of g contained in D is the zero ideal. and its kernel I is an ideal in g0 and by construction even in g. By definition of fKo(V +, V -), we have an embedding of g/I into fKo(V +, V -). If g = Ti (V, D, sl2) for a unital Jordan superalgebra Another "universality property" of fKo(V +, V −) will be discussed in Subsection 5.4. 5.3. Outer derivations. Definition 5.4 (See [AMR]). For a Lie superalgebra g, denote the Lie superalgebra of derivations by Der(g). The inner derivations Inn(g) = {adX X ∈ g} form an ideal isomorphic to the quotient of g by its centre. The Lie superalgebra of outer derivations is Out(g) = Der(g)/Inn(g). An extension e of a Lie superalgebra g over a Lie superalgebra h is a Lie superalgebra e such that the following is a short exact sequence: In particular h is an ideal in e. 0 → h → e → g → 0. Let h be a Lie superalgebra with trivial centre. Then we will freely use the isomorphism between the space of extensions of g over h, and the space of Lie superalgebra morphisms g → Out(h), see e.g. Corollary 8 in [AMR]. The main result of this section is the following proposition. Proposition 5.5. For a unital Jordan superalgebra V , we have and thus fKo(V ) ∼= Der(Ko(V )), fKo(V )/Ko(V ) ∼= str(V )/istr(V ) ∼= Out(Ko(V )). Remark 5.6. Again the assumption of a multiplicative identity is essential for this proposition. A counterexample of the statement for non-unital Jordan superalgebras is given in Subsec- tion 6.2. 16 SIGISWALD BARBIER AND KEVIN COULEMBIER Remark 5.7. For any Z-graded Lie superalgebra, the Lie superalgebra Der(g) ⊂ EndK(g) is Z- graded by construction. The endomorphisms in Der(g)i map elements in gj to elements in gi+j. Clearly Inn(g) is then a graded ideal in Der(g), so that Out(g) is also Z-graded. In particular, when g is 3-graded then Der(g) and Out(g) will be 5-graded. The following reformulation of Proposition 5.5 holds for arbitrary Jordan superpairs and thus a fortiori also for non-unital Jordan superalgebras. Proposition 5.8. For a Jordan superpair (V +, V -), we have fKo(V +, V -)/Ko(V +, V -) ∼= Der(V +, V -)/Inn(V +, V -) ∼= Out(Ko(V +, V -))0. In particular, for a (non-unital) Jordan superalgebra V we have that fKo(V ) is the extension over Ko(V ) of Out(Ko(V ))0 corresponding to the embedding Out(Ko(V ))0 ֒→ Out(Ko(V )). The rest of the subsection is devoted to the proofs of Propositions 5.5 and 5.8. Lemma 5.9. We have a Lie superalgebra isomorphism φ : Der(V +, V -) → Der(Ko(V +, V -))0, x 7→ adxKo(V +,V -). Proof. As Ko(V +, V -) is an ideal in fKo(V +, V -) and Der(V +, V -) ⊂ fKo(V +, V -) is the zero component of the Z-grading, the map φ is well-defined. By construction it is an injective Lie superalgebra morphism. Now let D be a Z-grading preserving derivation of Ko(V +, V -), then (DV +, DV -) is an element of Der(V +, V -) since, using the definition of the bracket on Ko(V +, V -) in Subsection 4.2, we find D({x, y, z}) = D([[x, y], z]) = [[D(x), y], z] + (−1)xD[[x, D(y)], z] + (−1)(x+y)D[[x, y], D(z)] = {D(x), y, z} + (−1)xD{x, D(y), z} + (−1)(x+y)D{x, y, D(z)}. One can check that So we have indeed Der(V +, V -) ∼= Der(Ko(V +, V -))0 as Lie superalgebras. D = ad(DV + ,DV - ) and x = (adxV +, adxV -). (cid:3) Using this lemma, we can immediately prove Proposition 5.8. Proof of Proposition 5.8. The first isomorphism follows immediately from Koecher's construc- tion in Subsection 4.2. Furthermore, since the intersection of the centre of Ko(V +, V -) with Ko(V +, V -)0 is trivial, we have Inn(Ko(V +, V -))0 ∼= Ko(V +, V -)0 = Inn(V +, V -). Hence, from Lemma 5.9 we conclude that Der(V +, V -)/Inn(V +, V -) ∼= Out(Ko(V +, V -))0. (cid:3) To prove Proposition 5.5, we will show that, for a unital Jordan superalgebra V , all outer derivations of Ko(V ) are grading preserving for the 3-grading we consider. This is not true for non-unital algebras, see Subsection 6.2. Lemma 5.10. For a unital Jordan superalgebra V , we have Der(Ko(V ))−2 = 0 = Der(Ko(V ))2. STRUCTURE AND TKK ALGEBRAS 17 Proof. Let D ∈ Der(Ko(V ))2. First remark that D acts trivially on Ko(V )0 and Ko(V )1. We will show that it must also acts trivially on Ko(V )-1. To use the definition of Ko(V ) we use the Jordan superpair (V +, V -) := (V, V ). For x ∈ V , we use the notation x+ and x-, as in Example 2.2. We find, using the definition of the bracket on Ko(V ) and the property D(Ko(V )0) = 0, that D(e-) = 1 2 D([e-, De,e]) = [D(e-), De,e] + 1 2 1 2 [e-, D(De,e)] = 1 2 [D(e-), De,e] = −D(e-). Hence D(e-) = 0, which then implies that D(x-) = 1 2 D([e-, Dx,e]) = [D(e-), Dx,e] + 1 2 1 2 [e-, D(Dx,e)] = 0. We conclude that D = 0 for all D ∈ Der(Ko(V ))2. The proof that Der(Ko(V ))−2 = 0 is completely similar. (cid:3) Lemma 5.11. For a unital Jordan superalgebra V , we have isomorphisms V → Der(Ko(V ))1; x 7→ adx+ and V → Der(Ko(V ))-1; x 7→ adx-, as super vector spaces. Proof. Let x+ be an element of Ko(V )1 = V +, then adx+ ∈ Der(Ko(V ))1. With an element D in Der(Ko(V ))1 we can associate the element − 1 2 D(De,e) ∈ V +. We will now show that x+ 7→ adx+ and D 7→ − 1 2 D(De,e) are each others inverse. This follows from and the following three calculations, for arbitrary x, y ∈ V , − 1 2 adx+ (De,e) = − 1 2 [x+, De,e] = x+ − adD(De,e)(y-) = − 1 2 1 2 − adD(De,e)(Lx) = − − 1 2 adD(De,e)([Lx, Ly]) = − 1 2 1 2 1 2 [D(De,e), y-] = − 1 D([De,e, y-]) + 2 1 [D(De,e), Lx] = − 2 1 2 D([De,e, Lx]) + [De,e, D(y-)] = D(y-) 1 2 [De,e, D(Lx)] = D(Lx) [D(De,e), [Lx, Ly]] = − 1 2 D([De,e, [Lx, Ly]]) + 1 2 [De,e, D([Lx, Ly])] = D([Lx, Ly]). We conclude that V ∼= Der(Ko(V ))1. Similarly Ko(V )-1 → Der(Ko(V ))-1; x− 7→ adx- is an isomorphism with inverse D 7→ 1 (cid:3) 2 D(De,e). Proof of Proposition 5.5. Consider the following morphism of Lie superalgebras Combining Lemmata 5.9, 5.10 and 5.11, we see that this is an isomorphism. fKo(V ) → Der(Ko(V )); x 7→ adxKo(V ). (cid:3) Proposition 5.8 fits into a more general construction. In [BDS, Section 4.1], the authors start 5.4. Alternative construction. The construction of fKo(V +, V -) starting from Ko(V +, V -) in from an arbitrary (2n + 1)-graded Lie superalgebra L = Li∈Z Li (strictly speaking only Lie algebras are considered, but the procedure carries over naturally to the super case). Then [BDS, Construction 4.1.2] constructs an extension L over L, which is again a (2n + 1)-graded Lie superalgebra which satisfies Li = Li if i 6= 0. 18 SIGISWALD BARBIER AND KEVIN COULEMBIER It is not difficult to show that in the case of a 3-graded Lie superalgebra L we have L0 = Der(L+, L-) and hence the Jordan superpair associated with L in Subsection 4.2. In other words, L = fKo(L+, L-) with (L+, L-) := J (L), fKo(V +, V -) ∼= Ko(V +, V -). This reveals a universality principle behind fKo(V +, V -), as the construction of L starting from L in [BDS] does not depend on L0. An interesting consequence of [BDS, Lemma 4.1.3] is then for arbitrary Jordan superpairs (V +, V -), so also for arbitrary (unital or non-unital) Jordan superalgebras. Out(fKo(V +, V -)) = 0, 6. Examples In this section, we use the results of the previous sections to calculate fKo(V ) for V any finite dimensional simple Jordan superalgebra over an algebraically closed field of characteristic zero. We assume these conditions on the ground field for the entire section. 6.1. Unital finite dimensional simple Jordan superalgebras. A complete list of unital finite dimensional simple Jordan superalgebras V and the corresponding Kan(V ) is given in [Ka2, CK]. This gives us Ko(V ) and Ti(V, Inn(V ), sl2), by Propositions 4.3 and 4.5. For the Jordan superalgebras we use the notation of [CK], where also the definitions can be found. We introduce our convention for the notation of Lie superalgebras in Appendix A. In [Ka1, Theorem 5.1.2] and [Sc, Chapter III, Proposition 3], Der(g) is calculated for any simple finite dimensional Lie superalgebra g. Together with Proposition 5.5 and Lemma 5.1, this gives us fKo(V ) ∼= Ti(V, Der(V ), sl2), leading to the following table. fKo(V ) Ko(V ) V pgl(2m2m) gl(m, n)+ gl(m, m)+ osp(m, 2n)+ (m − 3, 2n)+ sl(2m2n) psl(2m2m) osp(4n2m) osp(m2n) Remarks m 6= n m > 1 (n, m) 6= (1, 0) m ≥ 3, (m, 2n) 6= (4, 0) p(n)+ q(n)+ Dt E F spe(2n) psq(2n) D(2, 1, t) E(7) F (4) pe(2n) pq(2n) n > 1 n > 1 t 6∈ {0, −1} n ≥ 5 gl(1, 1)+ JP (0, n − 3) H(0, n) = H(n) KC ⋉ eH(n) When fKo(V ) is isomorphic to Ko(V ), we only wrote it once. D(2, 1, −1) psl(22) Taking the zero component of the 3-graded algebras in the above table gives us istr(V ) ∼= Inn(V, V ) and str(V ) ∼= Der(V, V ). These are listed in the following table, where the same restrictions on the indices are assumed as in the previous table. V gl(m, n)+ gl(m, m)+ osp(m, 2n)+ (m − 3, 2n)+ p(n)+ q(n)+ Dt E F JP (0, n − 3) gl(1, 1)+ STRUCTURE AND TKK ALGEBRAS 19 istr(V ) sl(mn) ⊕ sl(mn) ⊕ K str(V ) s(gl(mm) ⊕ gl(mm))/hI4mi (gl(mm) ⊕ gl(mm))/hI4mi gl(2nm) osp(m − 22n) ⊕ K sl(nn) s(q(n) ⊕ q(n))/hI4ni sl(21) ⊕ K ∼= osp(22) ⊕ K E(6) ⊕ K gl(nn) (q(n) ⊕ q(n))/hI4ni osp(24) ⊕ K eH(n − 2) ⋉ (Λ(n − 2)/hξ1 · · · ξn−2i) KC ⋉(cid:16)eH(n − 2) ⋉ Λ(n − 2)(cid:17) s(gl(11) ⊕ gl(11))/hI4i sl2 ⋉ istr(gl(1, 1)+) Again, if str(V ) is isomorphic to istr(V ), we only wrote it once and the notation is explained in Appendix A. The action of sl2 on istr(gl(1, 1)+) is the adjoint action by using the embedding of sl2 in D(2, 1; −1). The following isomorphisms exist in the list of Jordan superalgebras: (1, 2)+ ∼= D1, Dt ∼= Dt−1 . Furthermore, also the simple Jordan superalgebras JP (0, 1) and D−1 appear in the literature, but they are isomorphic to gl(1, 1)+, so they are already included in the table. 6.2. The non-unital finite dimensional simple Jordan superalgebra. The full list of finite dimensional simple Jordan superalgebras in [Ka2, CK] contains only one Jordan superalgebra which is non-unital. In [Ka2] it is denoted by K. The algebra K is defined as K = hai ⊕ hξ1, ξ2i, with multiplication satisfying a2 = a, aξi = 1 a = ¯0, ξ1 = ξ2 = ¯1, 2 ξi and ξ1ξ2 = a. A straightforward calculation implies istr(K) = str(K) = Inn(K, K) ∼= sl(12), and Der(K, K) ∼= gl(12). This gives a counterexample to the statement in Proposition 3.7(1) for non-unital Jordan su- peralgebras. For K, the sums in Definitions 3.2 and 3.3 are direct. One also finds Ko(K) ∼= psl(22). terms of Out(Ko(K)) ∼= sl2: By construction, fKo(K) is an extension over Ko(K). As istr(K) ∼= Inn(K, K), it follows easily that the same is true for Kan(K). The algebras fKo(K) and Kan(K) can hence be described in • fKo(K) ∼= pgl(22) is the extension of K over Ko(K) corresponding to the morphism K → sl2, where 1 ∈ K is mapped to a semisimple element of sl2. • Kan(K) is the extension of K over Ko(K) corresponding to the morphism K → sl2, where 1 ∈ K is mapped to a nilpotent element of sl2. In particular we find that fKo(K) 6∼= Der(Ko(K)) and Kan(K) 6∼= Ko(K). 20 SIGISWALD BARBIER AND KEVIN COULEMBIER This gives counterexamples to the statements in Propositions 5.5 and 4.3, for non-unital Jordan superalgebras. By Remark 4.6 and the above, we do have Ti(K, Inn(K), sl2) ∼= Ti(K, Der(K), sl2) ∼= Ko(K) ∼= psl(22). For the 3-grading on psl(22) corresponding to the interpretation as Ko(K), the algebra g = Out(psl(22) ∼= sl2 is 3-graded where gi has dimension one for i ∈ {−1, 0, 1}. This is in sharp contrast with Lemma 5.11 for the unital case. By Proposition 5.8, fKo(K) is the subalgebra of Der(Ko(K)) where only the degree 0 derivations are added to Ko(K). In the same way, Kan(K) is the subalgebra of Der(Ko(K)) where only the degree 1 derivations are added to Ko(K). Appendix A. The Lie superalgebras of type A, P , Q and H Consider an algebraically closed field K of characteristic zero. We quickly review the Lie super- algebras of type A, P , Q, and H, as different notations appear in literature. Our nomenclature is based on [CW]. See also [CW, Theorem 1.11] for the list of simple finite dimensional Lie superalgebras. A.1. Type A. The general linear superalgebra gl(mn) is defined as End(Kmn), with multi- plication given by the super commutator. Define the supertrace for a matrix A ∈ gl(mn) as str(A) :=Pi(−1)iAii, where i = ¯0 for i ≤ m and i = ¯1 for m + 1 ≤ i ≤ m + n. The special linear superalgebra is sl(mn) = {A ∈ gl(mn) str(A) = 0}, If m 6= n then sl(mn) is simple. If m = n then hI2ni, with I2n the identity matrix, is an ideal in sl(nn) and is simple for n > 1. Similarly, we set psl(nn) := sl(nn)/hI2ni pgl(nn) := gl(nn)/hI2ni. A.2. Type P. The periplectic Lie superalgebra is the subalgebra of gl(nn) defined as pe(n) :=(cid:26)(cid:18) a b c −at (cid:19)(cid:12)(cid:12)(cid:12) a, b, c ∈ Kn×n with bt = b, ct = −c(cid:27) . The special periplectic Lie superalgebra is defined as It is simple for n ≥ 3. spe(n) := {x ∈ p(n)tr(a) = 0}. A.3. Type Q. The queer Lie superalgebra is the subalgebra of gl(nn) defined as Remark that str(X) = 0 for all X ∈ q(n). The special queer Lie superalgebra is defined as q(n) :=(cid:26)(cid:18) a b b a (cid:19)(cid:12)(cid:12)(cid:12) a, b ∈ Kn×n(cid:27) . b a (cid:19)(cid:12)(cid:12)(cid:12) a, b ∈ Kn×n, tr(b) = 0(cid:27) . sq(n) :=(cid:26)(cid:18) a b The projective special queer Lie superalgebra is defined as psq(n) := sq(n)/hI2ni. STRUCTURE AND TKK ALGEBRAS 21 It is simple for n ≥ 3. We also define the projective queer Lie superalgebra as pq(n) := q(n)/hI2ni. A.4. Type H. Let Λ(n) be the exterior algebra generated by ξ1, . . . , ξn. The indeterminates hence satisfy This is an associative superalgebra where the generators are odd, ξi = ¯1. We also consider a compatible Z-grading, by setting deg ξi = 1. Denote by W (n) the algebra of derivations of the associative superalgebra Λ(n). The Lie superalgebra W (n) is simple for n ≥ 2. ξiξj = −ξjξi. On Λ(n), we define the following Poisson superbracket {f, g} := (−1)f n−2Xi=1 ∂ξif ∂ξig + ∂ξn−1f ∂ξng + ∂ξnf ∂ξn−1g! , for f and g in Λ(n). Then (Λ(n), {·, ·}) becomes a Lie superalgebra with ideal h1i. Consider the following Lie superalgebras W (n). eH(n) := Λ(n)/h1i and H(n) := [eH(n), eH(n)]. Note that eH(n) = H(n)⊕ Kξ1 · · · ξn as super vector spaces. The Lie superalgebra H(n) is simple for n ≥ 4. We can embed H(n) and eH(n) into W (n), using f 7→ {f, ·}. Consider C := Pn i=1 ξi∂ξi ∈ W (n), then KC ⋉ eH(n) is naturally defined as a subalgebra of We also define the semidirect product eH(n − 2) ⋉ Λ(n − 2), where the action of eH(n − 2) on We further introduce, KC ⋉(cid:16)eH(n − 2) ⋉ Λ(n − 2)(cid:17), where C acts by [C, f ] = (deg f − 2)f for f ∈ eH(n − 2) and by [C, g] = deg g for g ∈ Λ(n − 2). Λ(n − 2) is given by the Poisson superbracket on Λ(n − 2), while the bracket of Λ(n − 2) is trivial. Acknowledgment. SB is a PhD Fellow of the Research Foundation - Flanders (FWO). KC is supported by Australian Research Council Discover-Project Grant DP140103239 and a post- doctoral fellowship of the Research Foundation - Flanders (FWO). The authors thank Hendrik De Bie, Tom De Medts and Erhard Neher for helpful discussions and comments. References [AMR] D. Alekseevsky, P. Michor, W. Ruppert. Extensions of super Lie algebras. J. Lie Theory 15 (2005), no. 1, 125-134. [BDS] L. Boelaert, T. De Medts, A. Stavrova. Moufang sets and structurable division algebras. Preprint: [CK] [CW] [FK] [GN] arXiv:1603.00780. N. Cantarini, V. G. Kac. Classification of linearly compact simple Jordan and generalized Poisson super- algebras. J. Algebra 313 (2007), no. 1, 100-124. S.J. Cheng, W. Wang. Dualities and representations of Lie superalgebras. Graduate Studies in Mathe- matics, 144. American Mathematical Society, Providence, RI, 2012. J. Faraut, A. Kor´anyi. Analysis on symmetric cones. Oxford Mathematical Monographs. Oxford Science Publications. The Clarendon Press, Oxford University Press, New York, (1994) E. Garc´ıa, E. Neher. Tits-Kantor-Koecher superalgebras of Jordan superpairs covered by grids. Comm. Algebra 31 (2003), no. 7, 3335-3375. 22 [Ja] SIGISWALD BARBIER AND KEVIN COULEMBIER N. Jacobson. Structure groups and Lie algebras of Jordan algebras of symmetric elements of associative algebras with involution. Advances in Math. 20 (1976), no. 2, 106-150. [Ka1] V. G. Kac. Lie superalgebras. Advances in Math. 26 (1977), no. 1, 8-96. [Ka2] V. G. Kac. Classification of simple Z-graded Lie superalgebras and simple Jordan superalgebras. Comm. Algebra 5 (1977), no. 13, 1375-1400. [KMZ] V. G. Kac, C. Martinez, E. Zelmanov. Graded simple Jordan superalgebras of growth one. Mem. Amer. [Kan] Math. Soc. 150 (2001), no. 711. I. L. Kantor. Transitive differential groups and invariant connections in homogeneous spaces. Trudy Sem. Vektor. Tenzor. Anal. 13 (1966) 310-398. [Kan2] I. L. Kantor. Jordan and Lie superalgebras determined by a Poisson algebra. Amer. Math. Soc. Transl. [KM] [KS] Ser. 2, 151, Amer. Math. Soc., Providence, RI, 1992. I. Kashuba, M. E. Martin. The variety of three-dimensional real Jordan algebras. J. Algebra Appl. 15 (2016), no. 8, 1650158 I. Kashuba, V. Serganova. On the Tits-Kantor-Koecher construction of unital Jordan bimodules. Preprint: arXiv:1502.07407. [Ko] M. Koecher. Imbedding of Jordan algebras into Lie algebras. I. Amer. J. Math. 89 (1967) 787-816. [Kr] [Lo] [MZ] S. V. Krutelevich. Simple Jordan superpairs. Comm. Algebra 25 (1997), no. 8, 2635-2657. O. Loos. Jordan pairs. Lecture Notes in Mathematics, Vol. 460. Springer-Verlag, Berlin-New York, 1975. C. Martinez, E. Zelmanov. Representation theory of Jordan superalgebras I. Trans. AMS, 362, (2010), no.2, 815-846. [McC] K. McCrimmon. A taste of Jordan algebras. Universitext, Springer-Verlag, New York, 2004. [Sc] M. Scheunert. The theory of Lie superalgebras. An introduction. Lecture Notes in Mathematics, 716. Springer, Berlin, 1979. A. S. Shtern. Representations of finite dimensional Jordan superalgebras of Poisson bracket. Comm. Algebra, 23, (1995), no. 5, 1815 -- 1823. T. Springer. Jordan algebras and algebraic groups. Ergebnisse der Mathematik und ihrer Grenzgebiete, Band 75. Springer-Verlag, New York-Heidelberg, 1973 J. Tits. Une classe d'alg`ebres de Lie en relation avec les alg`ebres de Jordan. Nederl. Akad. Wetensch. Proc. Ser. A 65 = Indag. Math. 24 (1962) 530-535. [Sh] [Sp] [Ti] SB: Department of Mathematical Analysis, Faculty of Engineering and Architecture, Ghent University, Krijgslaan 281, 9000 Gent, Belgium; E-mail: [email protected] KC: School of Mathematics and Statistics, University of Sydney, NSW 2006, Australia; E-mail: [email protected]
1204.1721	1	1204	2012-04-08T08:55:54	A characterization of nilpotent Leibniz algebras	[ "math.RA" ]	W. A. Moens proved that a Lie algebra is nilpotent if and only if it admits an invertible Leibniz-derivation. In this paper we show that with the definition of Leibniz-derivation from W. A. Moens the similar result for non Lie Leibniz algebras is not true. Namely, we give an example of non nilpotent Leibniz algebra which admits an invertible Leibniz-derivation. In order to extend the results of paper W. A. Moens for Leibniz algebras we introduce a definition of Leibniz-derivation of Leibniz algebras which agrees with Leibniz-derivation of Lie algebras case. Further we prove that a Leibniz algebra is nilpotent if and only if it admits an invertible Leibniz-derivation. Moreover, the result that solvable radical of a Lie algebra is invariant with respect to a Leibniz-derivation was extended to the case of Leibniz algebras.	math.RA	math	A CHARACTERIZATION OF NILPOTENT LEIBNIZ ALGEBRAS ALICE FIALOWSKI, A.KH. KHUDOYBERDIYEV AND B.A. OMIROV Abstract. W. A. Moens proved that a Lie algebra is nilpotent if and only if it admits an invertible Leibniz-derivation. In this paper we show that with the definition of Leibniz-derivation from [17] the similar result for non Lie Leibniz algebras is not true. Namely, we give an example of non nilpotent Leibniz algebra which admits an invertible Leibniz-derivation. In order to extend the results of paper [17] for Leibniz algebras we introduce a definition of Leibniz-derivation of Leibniz algebras which agrees with Leibniz-derivation of Lie algebras case. Further we prove that a Leibniz algebra is nilpotent if and only if it admits an invertible Leibniz-derivation. Moreover, the result that solvable radical of a Lie algebra is invariant with respect to a Leibniz-derivation was extended to the case of Leibniz algebras. Mathematics Subject Classification 2000: 17A32, 17B30. Key Words and Phrases: Lie algebra, Leibniz algebra, derivation, Leibniz-derivation, solvability, nilpotency. 1. Introduction In 1955, Jacobson [11] proved that a Lie algebra over a field of characteristic zero admitting a non-singular (invertible) derivation is nilpotent. The problem, whether the inverse of this statement is correct, remained open until work [8], where an example of an nilpotent Lie algebra, whose derivations are nilpotent (and hence, singular), was constructed. Such types of Lie algebras are called character- istically nilpotent Lie algebras. The study of derivations of Lie algebras lead to appearance of natural generalization -- pre-derivations of Lie algebras [16]. In [2] it is proved that Jacobson's result is also true in terms of pre-derivations. Similar to the example of Dixmier and Lister [8] several examples of nilpotent Lie algebras, whose pre-derivations are nilpotent were presented in [2], [4]. Such Lie algebras are called strongly nilpotent [4]. In paper [17] a generalized notion of derivations and pre-derivation of Lie algebras is defined as In fact, a Leibniz-derivation is a derivation of a Leibniz k-algebra Leibniz-derivation of order k. constructed by Lie algebra [6]. Below we present the characterization of nilpotency for Lie algebras in terms of Leibniz-derivations. Theorem 1.1. [17] A Lie algebra over a field of characteristic zero is nilpotent if and only if it has an invertible Leibniz-derivation. Leibniz algebras were introduced by Loday in [13]-[14] as a non-antisymmetric version of Lie algebras. Many results of Lie algebras are extended to Leibniz algebras case. Since the study of derivations and automorphisms of a Lie algebra plays essential role in the structure theory, the natural question arises whether the corresponding results for Lie algebras can be extended to more general objects. In [12] it is proved that a finite dimensional complex Leibniz algebra admitting a non-singular derivation is nilpotent. Moreover, it was shown that similarly to the case of Lie algebras, the inverse of this statement does not hold and the notion of characteristically nilpotent Lie algebra can be extended for Leibniz algebras [18]. In this paper we show that if we define Leibniz-derivations for Leibniz algebra as in [17], then Theorem 1.1 does not hold. In order to avoid the confusion we need to modify the notion of Leibniz- derivation for Leibniz algebras. Recall, in the definition of Leibniz-derivation of order k for Lie algebras the k-ary bracket is defined as multiplication of k elements on the left side. For the case of Leibniz algebras we propose the definition of Leibniz-derivation of order k as k-ary bracket on the right side. Due to anti-commutativity of multiplication in Lie algebras this definition agrees with the case of Lie algebras. The research of the first author was partially supported by the grant OTKA K77757. 1 2 ALICE FIALOWSKI, A.KH. KHUDOYBERDIYEV AND B.A. OMIROV Note that a vector space equipped with right sided k-ary multiplication is not a Leibniz k-algebra defined in [6]. For Leibniz-derivation of Leibniz algebra we prove the analogue of Theorem 1.1 for finite dimensional Leibniz algebras over a field of characteristic zero. Through the paper all spaces an algebras are assumed finite dimensional. 2. Preliminaries In this section we present some known facts about Leibniz algebras and Leibniz n-algebras. Definition 2.1. A vector space L over a field F with a binary operation [−, −] is a (right) Leibniz algebra, if for any x, y, z ∈ L the so-called Leibniz identity [x, [y, z]] = [[x, y], z] − [[x, z], y] holds. Every Lie algebra is a Leibniz algebra, but the bracket in a Leibniz algebra needs not to be skew- symmetric. For a Leibniz algebra L consider the following central lower and derived sequences: L1 = L, Lk+1 = [Lk, L1], k ≥ 1, L[1] = 1, L[s+1] = [L[s], L[s]], s ≥ 1. Definition 2.2. A Leibniz algebra L is called nilpotent (respectively, solvable), if there exists p ∈ N (q ∈ N) such that Lp = 0 (respectively, L[q] = 0). Levi's theorem, which has been proved for left Leibniz algebras in [3], is also true for right Leibniz algebras. Theorem 2.3. (Levi's Theorem). Let L be a Leibniz algebra over a field of characteristic zero and R be its solvable radical. Then there exists a semisimple subalgebra Lie S of L, such that L = S +R. The following theorem from linear algebra characterizes the decomposition of a vector space into the direct sum of characteristic subspaces. Theorem 2.4. [15] Let A be a linear transformation of the vector space V. Then V decomposes into the direct sum of characteristic subspaces V = Vλ1 ⊕ Vλ2 ⊕ · · · ⊕ Vλk with respect to A, where Vλi = {x ∈ V (A − λiI)k(x) = 0 for some k ∈ N} and λi, 1 ≤ i ≤ k, are eigenvalues of A. In Leibniz algebras a derivation is defined as follows Definition 2.5. A linear transformation d of a Leibniz algebra L is a derivation if for any x, y ∈ L d([x, y]) = [d(x), y] + [x, d(y)]. Consider for an arbitrary element x ∈ L the operator of right multiplication Rx : L → L, defined by Rx(z) = [z, x]. Operators of right multiplication are derivations of the Leibniz algebra L. The set R(L) = {Rx x ∈ L} is a Lie algebra with respect to the commutator, and the following identity holds: RxRy − RyRx = R[y,x]. (2.1) A subset S of an associative algebra A over a field F is called a weakly closed subset if for every pair (a, b) ∈ S × S there is an element γ(a,b) ∈ F such that ab + γ(a,b)ba ∈ S. We will need the following result concerning weekly closed sets Theorem 2.6. [11] Let S be a weakly closed subset of the associative algebra A of linear transformations of a vector space V over F. Assume that every W ∈ S is nilpotent, that is, W k = 0 for some positive integer k. Then the enveloping associative algebra S∗ of S is nilpotent. The classical Engel's theorem for Lie algebras has the following analogue for Leibniz algebras. Theorem 2.7. [1] A Leibniz algebra L is nilpotent if and only if Rx is nilpotent for any x ∈ L. The following Theorem generalizes Jacobson's theorem to Leibniz algebras. Theorem 2.8. [12] Let L be a complex Leibniz algebra which admits a non-singular derivation. Then L is nilpotent. The next example presents n-dimensional Leibniz algebra possessing only nilpotent derivations. A CHARACTERIZATION OF NILPOTENT LEIBNIZ ALGEBRAS 3 Example 2.9. Let L be an n-dimensional Leibniz algebra and let {e1, e2, . . . , en} be a basis of L with the following table of multiplication:   (omitted products are equal to zero). [e1, e1] = e3, [ei, e1] = ei+1, 2 ≤ i ≤ n − 1, [e1, e2] = e4, [ei, e2] = ei+2, 2 ≤ i ≤ n − 2, Using derivation property it is easy to see that every derivation of L has the following matrix form: 0 0 0 0 0 0 a3 a3 0 . . . . . . . . . 0 0 0 0 0 0 a4 a4 a3 . . . 0 0 a5 a5 a4 . . . 0 0 . . . . . . . . . . . . . . . . . . an bn an−1 an−1 an−2 an−1 . . . . . . 0 0 a3 0     . Thus, every derivation of L is nilpotent, i.e., L is characteristically nilpotent. Let us give the definition of Leibniz n-algebras. Definition 2.10. [6] A vector space L with an n-ary multiplication [−, −, ..., −] : L⊗n → L is a Leibniz n-algebra if it satisfies the following identity: n [[x1, x2, . . . , xn], y2, . . . , yn] = [x1, . . . , xi−1, [xi, y2, . . . , yn], xi+1, . . . , xn]. (2.2) Xi=1 Let L be a Leibniz algebra with the product [−, −]. Then the vector space L can be equipped with a Leibniz n-algebra structure with the following product: Definition 2.11. A derivation of a Leibniz n-algebra L is a K-linear map d : L → L satisfying [x1, x2, . . . , xn] = [x1, [x2, . . . , [xn−1, xn]]]. d([x1, x2, . . . , xn]) = n Xi=1 [x1, . . . , d(xi), . . . , xn]. The notion of Leibniz-derivation of Lie algebra was introduced in [17] and it generalizes the notions of derivation and pre-derivation of Lie algebra. Definition 2.12. A Leibniz-derivation of order n for a Lie algebra G is an endomorphism P of G satisfying the identity P ([x1, [x2, . . . , [xn−1, xn]]]) = [P (x1), [x2, . . . , [xn−1, xn]]]+ +[x1, [P (x2), . . . , [xn−1, xn]]] + · · · + [x1, [x2, . . . , [xn−1, P (xn)]]] for every x1, x2, . . . , xn ∈ G. In other words, a Leibniz-derivation of order n for a Lie algebra G is a derivation of G viewed as a Leibniz n-algebra. 3. Leibniz-derivation of Leibniz algebras The following example shows that Definition 2.12 is not substantial for the case of Leibniz algebras. Example 3.1. Let R be an (n + 1)-dimensional solvable Leibniz algebra and {e1, e2, . . . , en, en+1} be a basis of R with the table of multiplication given by [e1, e1] = e3, [ei, e1] = ei+1, 2 ≤ i ≤ n − 1, [e1, en+1] = e2 + αiei, [e2, en+1] = e2 + [ei, en+1] = ei + αj−i+2ei, 3 ≤ i ≤ n, n−1 n Pi=4 Pj=i+2 n−1 Pi=4 αiei,   4 ALICE FIALOWSKI, A.KH. KHUDOYBERDIYEV AND B.A. OMIROV It is easy to see that [R, [R, R]] = 0. For the identity map d we have 0 = d([x, [y, z]]) = [d(x), [y, z]] + [x, [d(y), z]] + [x, [y, d(z)]] = 0. Therefore, the invertible map d satisfies the condition of Definition 2.12, but the Leibniz algebra R is not nilpotent, i.e. analogue of Theorem 1.1 for Leibniz algebras is not true. Remark 3.2. The Example 3.1 can be extended for any non nilpotent solvable Leibniz algebra L such that L2 lies in the right annihilator of L. Let us introduce n-ary multiplication as follows [x1, x2, . . . , xn]r = [[[x1, x2], x3] . . . , xn]. The next example shows that, in general, a vector space equipped with defined n-ary multiplication [x1, x2, . . . , xn]r is not a Leibniz n-algebra. Example 3.3. Let R be a solvable Leibniz algebra and let {e1, e2, . . . , en, x} be a basis of R such that multiplication table of R in this basis has the following form [7]: [ei, e1] = ei+1, 1 ≤ i ≤ n − 1, [x, e1] = e1, [ei, x] = −iei, 1 ≤ i ≤ n.   It is not difficult to check that the vector space R with k-ary multiplication [x1, x2, . . . , xk]r does not define Leibniz k-algebra structure. Indeed, we have [[e1, e1, . . . , e1]r, x, . . . , x]r = [. . . [[[. . . [[e1, e1], e1], . . . , e1 , ] x], x], . . . , x ] = [. . . [[ek, x], x], . . . , x ] = (−k)k−1ek. On the other hand k−times {z {z } k−1−times k−1−times {z } k Xi=1 [e1, . . . , e1, [e1, x, . . . , x]r , e1, . . . , e1]r = [e1, . . . , e1, (−1)k−1e1 , e1, . . . , e1]r = } Xi=1 k = (−1)k−1 } i−th k {z Xi=1 k−times Hence identity (2.2) does not hold for k ≥ 3. {z } [. . . [[e1, e1], e1], . . . , e1 ] = (−1)k−1 i−th {z } ek = (−1)k−1kek. k Xi=1 Now we define the notion of Leibniz-derivation for Leibniz algebras. Definition 3.4. A Leibniz-derivation of order n ∈ N for a Leibniz algebra L is a K-linear map d : L → L satisfying d([x1, x2, . . . , xn]r) = n Xi=1 [x1, . . . , d(xi), . . . , xn]r. Proposition 3.5. For Lie algebras Definition 3.4 agrees with Definition 2.12. Proof. Let L be a Lie algebra, then we have P ([x1, [x2, . . . , [xn−1, xn]]]) = (−1)nP ([[[xn, xn−1], . . . , x2], x1]) = (−1)nP ([xn, xn−1, . . . , x1]r). On the other hand, [P (x1), [x2, . . . , [xn−1, xn]]] + [x1, [P (x2), . . . , [xn−1, xn]]] + · · · + [x1, [x2, . . . , [xn−1, P (xn)]]] = = (−1)n[[[xn, xn−1], . . . , x2], P (x1)]+(−1)n[[[xn, xn−1], . . . , P (x2)], x1]+· · ·+(−1)n[[[P (xn), xn−1], . . . , x2], x1] = = (−1)n ([xn, xn−1, . . . , x2, P (x1)]r + [xn, xn−1, . . . , P (x2), x1]r + · · · + [P (xn), xn−1, . . . , x2, x1]r) = = (−1)n n Xi=1 [xn, . . . , P (xi), . . . , x1]r. This implies the equality P ([x1, [x2, . . . , [xn−1, xn]]]) = [P (x1), [x2, . . . , [xn−1, xn]]]+ +[x1, [P (x2), . . . , [xn−1, xn]]] + · · · + [x1, [x2, . . . , [xn−1, P (xn)]]], A CHARACTERIZATION OF NILPOTENT LEIBNIZ ALGEBRAS which is equivalent to P ([xn, xn−1, . . . , x1]r) = n Xi=1 [xn, . . . , P (xi), . . . , x1]r. Relabeling xi with xn+1−i for 1 ≤ i ≤ n we complete the proof of the proposition. 5 (cid:3) Let LDern(L) denote the set of all Leibniz-derivations of order n for a Leibniz algebra L and let LDer(L) be the set of all Leibniz-derivations, i.e. LDer(L) = Sn∈N LDern(L). Note that a Leibniz derivation of order 2 is a derivation. Moreover, any derivation is a Leibniz- derivation of any order n. Thus, the order of a Leibniz-derivation is not unique. Lemma 3.6. The following statements are true 1) If s, t ∈ N and st, then LDers+1(L) ⊂ LDert+1(L); 2) for any k, l ∈ N, LDerk(L) ∩ LDerl(L) ⊂ LDerk+l−1. Proof. The proof is similar to that of Lemma 2.3 [17]. (cid:3) Similarly to the case of Lie algebras we call a Leibniz-derivation of order 3 a pre-derivation of Leibniz algebra. A nilpotent Leibniz algebra is called strongly nilpotent if all its Leibniz pre-derivations are nilpotent. Note that a strongly nilpotent Leibniz algebra is characteristically nilpotent, but the inverse is not true in general. Example 3.7. Any pre-derivation of the characteristically nilpotent Leibniz algebra in Example 2.9 with n = 6 have the matrix form:   a1 0 0 0 0 0 a1 2a1 0 0 0 0 a4 a4 a3 a3 3a1 −a1 + a3 0 0 0 4a1 0 0 a5 b5 c5 2a1 + a3 a6 b6 c6 a4 5a1 0 a1 + a3 6a1   Thus, this Leibniz algebra is not strongly nilpotent. Proposition 3.8. The Leibniz algebra L in Example 2.9 is strongly nilpotent if n > 6. Proof. Let d : L → L be a pre-derivation of L. Put d(e1) = n Xi=1 aiei, d(e2) = n Xi=1 biei, d(e3) = n Xi=1 ciei. Consider the property of pre-derivation d(e4) = d([e1, e1, e1]r) = (3a1 + a2)e4 + (a3 + 2a2)e5 + n−2 Xi=4 aiei+2. On the other hand, d(e4) = d([e2, e1, e1]r) = (2a1 + b1 + b2)e4 + (b3 + 2a2)e5 + Comparing coefficients of basis elements we have The equality d([e1, e1, e3]r) = 0 implies 0 = c1e4 + c2e5, hence c1 = c2 = 0. b1 + b2 = a1 + a2, bi = ai, 3 ≤ i ≤ n − 2. The chain of equalities n−2 Xi=4 biei+2. b1e4 + (2a1 + a2 + b2)e5 + (a3 + a2)e6 + n−3 Xi=4 aiei+3 = d([e1, e2, e1]r) = d(e5) = d([e3, e1, e1]r) = (2a1 + c3)e5 + (2a2 + c4)e6 + n−2 Xi=5 ciei+2. 6 ALICE FIALOWSKI, A.KH. KHUDOYBERDIYEV AND B.A. OMIROV deduce b1 = 0, c3 = a2 + b2, c4 = a3 − a2, ci = ai−1, 4 ≤ i ≤ n − 2. From the equalities (3a1 + 3a2)e6 + n−4 Xi=3 aiei+4 = d([e1, e2, e2]r) = d(e6) = d([e4, e1, e1]r) = (5a1 + a2)e6 + (4a2 + a3)e7 + n−4 Xi=4 aiei+4 we get a1 = a2 = 0. Since b2 = a1 + a2 and c3 = a2 + b2, we have b2 = c3 = 0. Thus we obtain d(e1) = n Xi=3 aiei, d(e2) = n−2 Xi=3 aiei + bn−1en−1 + bnen, d(e3) = n−3 Xi=3 aiei+1 + cn−1en−1 + cnen. Finally, from the expression d([ei−2, e1, e1]r) we derive d(ei) = a3ei+1 + a4ei+2 + · · · + an+2−jen, i ≥ 4 which completes the proof of Proposition. (cid:3) Below we present 7- and 8-dimensional characteristically nilpotent Leibniz algebras, which are not strongly nilpotent. Example 3.9. The 7-dimensional Leibniz algebra with table of multiplication: [e1, e1] = e3, [ei, e1] = ei+1, [e1, e2] = e4 − 2e5, [ei, e2] = ei+2 − 2ei+3, 2 ≤ i ≤ 4, [e5, e2] = e7 2 ≤ i ≤ 6,   is characteristically nilpotent, but not strongly nilpotent. Example 3.10. The 8-dimensional filiform Leibniz algebra with table of multiplication: [e1, e1] = e3, [ei, e1] = ei+1, [e1, e2] = e4 − 2e5 + 5e6, [ei, e2] = ei+2 − 2ei+3 + 5ei+4, 2 ≤ i ≤ 4, [e5, e2] = e7 − 2e8 [e6, e2] = e8 2 ≤ i ≤ 6,   is characteristically nilpotent, but not strongly nilpotent. Following the proofs of Lemmas in [9] and [5] for derivations of Lie and Leibniz n-algebras respec- tively, we get the following statement for Leibniz-derivations of order n of Leibniz algebras. Lemma 3.11. For a Leibniz-derivation d : L → L of order n of a Leibniz algebra L over a field of characteristic zero, the following formula holds for any k ∈ N: dk([x1, . . . , xn]r) = Xi1+i2+···+in=k k! i1!i2! . . . in! [di1 (x1), di2 (x2), . . . , din (xn)]r (3.1). A CHARACTERIZATION OF NILPOTENT LEIBNIZ ALGEBRAS 7 4. Nilpotent Leibniz algebras Starting with a Leibniz algebra L, we denote the n-ary algebra with multiplication [−, −, . . . , −]r by Ln(L). A subalgebra I is called an n-ideal of L or an ideal of Ln(L), if it satisfies Let M be any Leibniz subalgebra of L. Consider the following sequences: n Xi=1 [L, . . . , I, . . . , L]r ⊆ I. L1 n(M ) = M, Lk+1 (M ) = [Lk n (M ) = [L[s] n(M ), M, . . . , M ]r, n (M ) . . . , L[s] n (M ), L[s] L[1] n (M ) = M, L[s+1] n k ≥ 1, n (M )]r, s ≥ 1. Definition 4.1. A Leibniz algebra L is called n-nilpotent (n-solvable) if there exists a natural number p ∈ N (q ∈ N) such that Lp n (L) = 0). n(L) = 0 (L[q] Lemma 4.2. Let M be an ideal of L. The following inclusions are true L[k] n (M ) ⊆ M [k], Lk n(M ) ⊆ M k. Proof. It is easy to check that M k and M [k] are also ideals of L for any k. We shall proof the first embedding by induction on k for any n. If k = 2, then L[2] n (M ) = [M, M, M, . . . , M ]r = [[[M, M ], M ], . . . , M ] = [[M [2], M ], . . . , M ] ⊆ M [2]. Suppose that the statement holds for some k and we will prove it for k + 1. L[k+1] n (M ) = [[[L[k] n (M ), L[k] n (M )], L[k] n (M )], . . . , L[k] n (M )] ⊆ ⊆ [[[M [k], M [k]], M [k]], . . . , M [k]] = [[M [k+1], M [k]], . . . , M [k]] ⊆ M [k+1]. The second inclusion is established in a similar way. (cid:3) Lemma 4.3. M [tk+1] ⊆ L[k+1] n (M ), where k ∈ N and t is a natural number such that 2t ≥ n. Proof. Since M [p] ⊆ M [p+q] for any p, q ∈ N, it is sufficient to prove embedding for the minimal t such that 2t ≥ n. We shall use induction. If n = 3 then t = 2. For k = 1 we have M [3] = [M [2], M [2]] = [M [2], [M [1], M [1]]] ⊆ [[M [2], M [1]], M [1]] ⊆ [M [1], M [1], M [1]]r ⊆ L[2] Suppose that the statement holds for some k and we will prove it for k + 1. 3 (M ). M [2(k+1)+1] = M [2k+1+2] = [[M [2k+1], M [2k+1]], [M [2k+1], M [2k+1]]] ⊆ ⊆ [[M [2k+1], M [2k+1]], M [2k+1]] ⊆ [L[k+1] 3 (M ), L[k+1] 3 (M ), L[k+1] 3 (M )]r = L[k+2] 3 (M ). Let us prove the statement for any n. Since 2t ≥ n for k = 1 we get M [t+1] ⊆ M 2t = [[M [1], M [1]], . . . , M [1]] ⊆ [[M [1], M [1]], . . . , M [1]] The following chain equalities and inclusions 2t−times {z } n−times {z } = L[2] n (M ). M [t(k+1)+1] = M [tk+1+t] = (M [tk+1])[t+1] ⊆ (M [tk+1])2t = [[[M [tk+1], M [tk+1]], . . . , M [tk+1]] ⊆ ⊆ [[M [tk+1], M [tk+1]], . . . , M [tk+1]] ⊆ [[L[k+1] n (M ), L[k+1] n (M )], . . . , L[k+1] (M )] n {z n−times 2t−times {z } (M ) = L[k+2] n (cid:3) } } n−times {z complete the proof of the lemma. Further we shall need the following lemma. Lemma 4.4. M nk−k+1 = Lk+1 n (M ). 8 ALICE FIALOWSKI, A.KH. KHUDOYBERDIYEV AND B.A. OMIROV Proof. The proof goes again by induction on k for any n. If k = 1, then M n = [. . . [[M, M ], M ], . . . , M ] = [M, M, . . . , M ]r = L2 n(M ). Applying induction in the equalities n−times {z } M n(k+1)−k−1+1 = M nk−k+1+n−1 = [. . . [[M nk−k+1, M ], M ], . . . , M ] = = [M nk−k+1, M, . . . , M ]r = [Lk+1 n we complete the proof of the lemma. (M ), M, . . . , M ]r = Lk+2 (M ) n−1−times {z n } (cid:3) We denote by R− solvable radical of L, i.e. the maximal solvable ideal of the Leibniz algebra L; Rn− n-solvable radical of L, i.e. the maximal n-solvable ideal of the n-ary algebra Ln(L); N − nilradical of L, i.e. the maximal nilpotent ideal of the Leibniz algebra L; Nn− n-nilradical of L, i.e. the maximal n-nilpotent ideal of the n-ary algebra Ln(L). Proposition 4.5. For a Leibniz algebra L we have R = Rn. Proof. Lemma 4.2 implies that any solvable ideal of L is also n-solvable. Therefore, it is sufficient to prove the inclusion Rn ⊆ R. From Lemma 4.3 it follows that Rn is a solvable subalgebra of L. Thus, we need to prove that Rn is an ideal of L. According to Theorem 2.3, we can write L = S ⊕ R, where S is a simple Lie algebra, R is a solvable ideal. Let π : L → S be the natural quotient map. Since π is a morphism of L, we have π([L, L, . . . , L, Rn]r) = π([[[[L, L], L], . . . , L], Rn]) = = [[[[π(L), π(L)], π(L)], . . . , π(L)], π(Rn)] = [[[[S, S], S], . . . , S], π(Rn)] = [S, π(Rn)]. On the other hand, π([L, L, . . . , L, Rn]r) ⊆ π(Rn). Hence, [S, π(Rn)] ⊆ π(Rn). Taking into account that S is a Lie algebra we obtain [π(Rn), S] ⊆ π(Rn). Therefore, π(Rn) is an ideal of S. Since Rn is an n-solvable ideal of L, π(Rn) is an n-solvable ideal of S, consequently π(Rn) is a solvable ideal (because π(Rn) is an ideal of S). Due to semisimplicity of S we get π(Rn) = 0, which implies Rn ⊆ R. (cid:3) Lemma 4.6. Let I be an ideal of the Leibniz algebra L and d ∈ LDern(L) a Leibniz-derivation for some n ∈ N. Then n (d(I)) ⊆ I + dnk−1 L[k] (L[k] n (I)) for all k ∈ N. Proof. Evidently d(I) ⊆ I + d(I) holds. For k = 2, using (3.1), we have L[2] n (d(I)) = [d(I), d(I), . . . , d(I)]r ⊆ d([I, I, . . . , I]r)+ + X i1 + i2 + · · · + in = n n! i1!i2! . . . in! ∃ij = 0 [di1 (I), . . . , dij −1(I), I, dij +1(I), . . . , din (I)]r ⊆ Assume that L[k] n (d(I)) ⊆ I + dnk−1 (L[k] n (I)). Again using (3.1), we verify the inclusion for k + 1 : ⊆ I + dn(L[2] n (I)). n L[k+1] ⊆ [I + dnk−1 ⊆ I + dnk (d(I)) = [L[k] (L[k] ([L[k] n (I)), I + dnk−1 n (I), L[k] n (d(I)), L[k] (L[k] n (I), . . . , L[k] n (d(I)), . . . , L[k] n (I)), . . . , I + dnk−1 n (I)]r) ⊆ I + dnk (L[k] (L[k+1] n n (d(I))]r ⊆ n (I))]r ⊆ (I))). (cid:3) Theorem 4.7. Let R be the solvable radical of a Leibniz algebra L over a field of characteristic zero. Then d(R) ⊆ R for any d ∈ LDern(L). A CHARACTERIZATION OF NILPOTENT LEIBNIZ ALGEBRAS 9 Proof. Let d be a Leibniz-derivation of order n. Due to Proposition 4.5, R = Rn, so it is enough to prove the assertion of the Theorem for Rn. L[s] Since Rn is a n-solvable radical, there exists s ∈ N such that L[s] n (d(Rn)) ⊆ Rn + dns−1 Further, n (Rn)) = Rn. Thus, we have L[s] (L[s] n (Rn) = 0. Then by Lemma 4.6, n (Rn + d(Rn)) ⊆ Rn. L[2s−1] n (Rn + d(Rn)) ⊆ L[s] n (L[s] n (Rn + d(Rn))) ⊆ L[s] n (Rn) = 0. The n-ideal property of Rn + d(Rn) follows from the following equalities: [l1, . . . , li + d(li), . . . , ln]r = [l1, . . . , li, . . . , ln]r + [l1, . . . , d(li), . . . , ln]r = [l1, . . . , li, . . . , ln]r + d([l1, . . . , li, . . . , ln]r) − n Xj=1,j6=i [l1, . . . , d(lj), . . . , ln]r. Hence Rn + d(Rn) is an n-solvable ideal of the Leibniz algebra L. Since Rn is the n-solvable radical (cid:3) of L, it follows that Rn + d(Rn) ⊆ Rn, therefore d(Rn) ⊆ Rn. Lemma 4.8. Let I be an ideal of the Leibniz algebra L and d ∈ LDern(L) a Leibniz-derivation for some n ∈ N. Then Lk n(d(I)) ⊆ I + dkn−k+1(Lk n(I)) for all k ∈ N. Proof. For k = 1 the assertion of the lemma is obvious. Let k = 2, then using the formula (3.1) we have L2 n(d(I)) = [d(I), d(I), . . . , d(I)]r ⊆ d([I, I, . . . , I]r)+ + X i1 + i2 + · · · + in = n n! i1!i2! . . . in! ∃ij = 0 [di1 (I), . . . , dij −1(I), I, dij −1(I), . . . , din (I)]r ⊆ Assume that Lk n(d(I)) ⊆ I + dkn−k+1(Lk ⊆ I + dn(L2 n(I)). Applying the formula (3.1), we prove the inclusion n(I)). for k + 1 : Lk+1 n (d(I)) = [Lk n(d(I)), (d(I)), . . . , (d(I))]r ⊆ [I + dkn−k+1(Lk n(I)), d(I), . . . , d(I)]r ⊆ ⊆ [I + dkn−k+1+n−1([L[k] n (I), I, . . . , I]r) ⊆ I + d(k+1)n−k(Lk+1 n (I))). (cid:3) Invariant property of nilradical of a Leibniz algebra under a Leibniz-derivation is presented in the following theorem. Theorem 4.9. Let N be the nilradical of a Leibniz algebra L over a field of characteristic zero. Then d(N ) ⊆ N for any d ∈ LDern(L). Proof. The proof is similar to the proof of Theorem 4.7. (cid:3) Next result establish properties of weight spaces with respect to a Leibniz-derivation of a Leibniz algebra. Lemma 4.10. Let L be a complex Leibniz algebra with a given Leibniz-derivation d of order n and L = Lα ⊕ Lβ ⊕ · · · ⊕ Lγ the decomposition of L into weight spaces with respect to d (i.e. Lα = {x ∈ L : (d − α1)kx = 0 for some k}). Then [Lα1 , Lα2 , . . . , Lαn ]r = (cid:26) 0 Lα1+α2+···+αn if α1 + α2 + · · · + αn is not a root of d if α1 + α2 + · · · + αn is a root of d. Proof. First observe that (d − (α1 + α2 + · · · + αn) · 1)[x1, x2, . . . , xn]r = n Xi=1 [x1, . . . , d(xi), . . . , xn]r − n Xi=1 [x1, . . . , αixi, . . . , xn]r = n Xi=1 [x1, . . . , (d − αi · 1)xi, . . . , xn]r. Similarly to Lemma 3.11, by induction on k we get the following equality: (d − (α1 + α2 + · · · + αn) · 1)k[x1, x2, . . . , xn]r = 10 ALICE FIALOWSKI, A.KH. KHUDOYBERDIYEV AND B.A. OMIROV = Xi1+i2+···+in=k k! i1!i2! · · · in! for any xi ∈ Lαi. [(d − α1 · 1)i1 x1, (d − α2 · 1)i2 x2, . . . , (d − αn · 1)in xn]r (4.1) Consider xi ∈ Lαi , 1 ≤ i ≤ n. Then there exist natural numbers ki such that (d − αi · 1)ki(x) = 0. Taking k = n Pi=1 ki in (4.1), we have (d − (α1 + α2 + · · · + αn) · 1)k[x1, x2, . . . , xn]r = 0 which completes the proof. (cid:3) Similarly as in [17] we have the existence of an invertible Leibniz-derivation of nilpotent Leibniz algebra. Proposition 4.11. Every nilpotent Leibniz algebra with nilindex equal to s has an invertible Leibniz- derivation of order [ s 2 ] + 1. Proof. Let L be a Leibniz algebra with nilindex equal to s and set q = [ s subspace W of L complementary to Lq, i.e. L = W + Lq. Define the map P by the following way: 2 ] + 1. Consider the vector P (x) = (cid:26) x if x ∈ W, qx if x ∈ Lq. It is easy to check that P is a Leibniz-derivation for L of order q. (cid:3) Below we present one of the main theorems of the paper. Theorem 4.12. Let L be a complex Leibniz algebra which admits an invertible Leibniz-derivation. Then L is nilpotent. Proof. Let d be an invertible Leibniz-derivation of order n of the Leibniz algebra L and be the decomposition of L into characteristic spaces with respect to d. Let α, β ∈ spec(d). Then by Lemma 4.10 we have L = Lρ1 ⊕ Lρ2 ⊕ · · · ⊕ Lρs [Lα, Lβ, Lβ, . . . , Lβ]r = [. . . [[Lα, Lβ], Lβ, ] . . . , Lβ] ⊆ Lα+(n−1)β. Considering k-times of the n-ary multiplication we have n−1−times {z } n−1−times {z } [. . . [[Lα, Lβ, Lβ, . . . , Lβ]r , Lβ, Lβ, . . . , Lβ]r , . . . , Lβ, Lβ, . . . , Lβ]r = n−1−times {z } n−1−times {z {z k−times } n−1−times {z ⊆ Lα+k(n−1)β. } } = [. . . [[Lα, Lβ], Lβ, ] . . . , Lβ] k(n−1)−times {z } Since for sufficiently large k ∈ N we obtain α + k(n − 1)β 6∈ spec(d), by Lemma 4.10 we obtain [. . . [[Lα, Lβ], Lβ, ] . . . , Lβ] = 0. k(n−1)−times {z } Thus, any operator of right multiplication Rx : L → L, where x ∈ Lβ, is nilpotent and, due to the fact that α, β were taken arbitrary, it follows that every operator from Sk Now from identity (2.1) and Lemma 4.10 it follows that Sk i=1 R(Lρi) is a weekly closed set of an associative algebra End(L). Hence, by Theorem 2.6 it follows that every operator from R(L) is nilpotent. i=1 R(Lρi) is nilpotent. Hence, Rx is nilpotent for any x ∈ L. Now by Engel's Theorem (Theorem 2.7) we conclude that L (cid:3) is nilpotent. Finally from the Theorem 4.12 and Proposition 4.11 we get the analogue of Theorem 1.1 for Leibniz algebras. Theorem 4.13. A Leibniz algebra over a field of characteristic zero is nilpotent if and only if it has an invertible Leibniz-derivation. A CHARACTERIZATION OF NILPOTENT LEIBNIZ ALGEBRAS 11 References [1] Ayupov Sh. A., Omirov B.A. On Leibniz algebras. Algebra and operator theory, Proceedings of the Colloquium in Tashkent 1997. Kluwer Academic Publishers. 1998, p. 1 -- 12. [2] Bajo I. Lie algebras admitting non-singular prederivations. Indag. Mathem., 8(4), 1997, p. 433 -- 437. [3] Barnes D.W. On Levi's theorem for Leibniz algebras. Arxiv. 1109.1060v1., 1 p. [4] Burde D. Lie algebra prederivations and strongly nilpotent Lie algebras. Comm. Algebra, 30(7), 2002, p. 3157 -- 3175. [5] Camacho L.M., Gasas J.M., G´omez J.R., Ladra M., Omirov B.A. On nilpotent Leibniz n-algebras. Journal of Algebra and its Applications. DOI: 10.1142/S0219498812500624, 17 p. [6] Casas J. M., Loday J.-L., Pirashvili T. Leibniz n-algebras, Forum Math., 14, 2002, p. 189 -- 207. [7] Casas J. M., Ladra M., Omirov B.A., Karimjanov I.K. Classification of solvable Leibniz algebras with null-filiform nilradical, arXiv:1202.5275v1, 2012, 13 p. [8] Dixmer J., Lister W.G. Derivations of nilpotent Lie algebras. Proc. Amer. Math. Soc., 8, 1957, p. 155 -- 158. [9] Filippov V. T. n-Lie algebras, Sib. Mat. Zh., 26(6), 1985, p. 126 -- 140. [10] Gago F., Ladra M., Omirov B.A., Turdibaev R.M. Some radicals, Frattini and Cartan subalgebras of Leibniz n− algebras. arXiv:1103.2725v1, 2011, 16 p. [11] Jacobson N. A note on automorphisms and derivations of Lie algebras. Proc. Amer. Math. Soc., 6, 1955, p. 281 -- 283. [12] Ladra M., Rikhsiboev I.M., Turdibaev R.M. Automorphisms and derivations of Leibniz algebras. arXiv:1103.4721v1, 2011, 12 p. [13] Loday J.-L., Cyclic homology, Grundl. Math. Wiss. Bd., 301, Springer-Verlag, Berlin, 1992. [14] Loday J.-L., Une version non commutative des alg`ebres de Lie: les alg`ebres de Leibniz, L'Enseignement Math´ematique, 39, 1993, p. 269 -- 292. [15] Curtis M.L. Abstract Linear Algebra, Springer 1990. [16] Muller D. Isometries of bi-invariant pseudo-Remannian metrics on Lie groups. Geom. Dedicata, 29, 1989, p. 65 -- 96. [17] Moens W.A. A characterisation of nilpotent Lie algebras by invertible Leibniz-derivations. arXiv:1011.6186v1, 2010, 15 p. [18] Omirov B.A. On derivations of filiform Leibniz algebras. Math. Notes, 77(5-6), 2005, p. 677 -- 685. [A. Fialowski] Institute of Mathematics, Eotvos Lor´and University, Budapest, P´azm´any P´eter s´et´any 1/C, 1117 Hungary E-mail address: [email protected] [A.Kh. Khudoyberdiyev -- B.A. Omirov] Institute of Mathematics, 29, Do'rmon yo'li srt., 100125, Tashkent (Uzbekistan) E-mail address: [email protected] --- [email protected]
1602.04134	1	1602	2016-02-04T20:11:31	Parastrophes of quasigroups	[ "math.RA" ]	Parastrophes (conjugates) of a quasigroup can be divided into separate classes containing isotopic parastrophes. We prove that the number of such classes is always 1, 2, 3 or 6. Next we characterize quasigroups having a fixed number of such classes.	math.RA	math	Quasigroups and Related Systems 23 (2015), 221 − 230 Parastrophes of quasigroups Wieslaw A. Dudek Abstract. Parastrophes (conjugates) of a quasigroup can be divided into separate classes con- taining isotopic parastrophes. We prove that the number of such classes is always 1, 2, 3 or 6. Next we characterize quasigroups having a fixed number of such classes. 1. Introduction Denote by SQ the set of all permutations of the set Q. We say that a quasigroup (Q, ·) is isotopic to (Q, ◦) if there are α, β, γ ∈ SQ such that α(x) ◦ β(y) = γ(x · y) for all x, y ∈ Q. The triplet (α, β, γ) is called an isotopism. Quasigroups (Q, ·) and (Q, ◦) for which there are α, β, γ ∈ SQ such that α(x) ◦ β(y) = γ(y · x) for all x, y ∈ Q are called anti-isotopic. This fact is denoted by (Q, ·) ∼ (Q, ◦). In the case when (Q, ·) and (Q, ◦) are isotopic we write (Q, ·) ≈ (Q, ◦). It is clear that the relation ≈ is an equivalence and divides all quasigroups into disjoint classes containing isotopic quasigroups. Each quasigroup Q = (Q, ·) determines five new quasigroups Qi = (Q, ◦i) with the operations ◦i defined as follows: x ◦1 y = z ←→ x · z = y x ◦2 y = z ←→ z · y = x x ◦3 y = z ←→ z · x = y x ◦4 y = z ←→ y · z = x x ◦5 y = z ←→ y · x = z Such defined (not necessarily distinct) quasigroups are called parastrophes or con- jugates of Q. Traditionally they are denoted as Q1 = Q−1 = (Q, \), Q2 = −1Q = (Q, /), Q3 = −1(Q−1) = (Q1)2, Q4 = (−1Q)−1 = (Q2)1 and Q5 = (−1(Q−1))−1 = ((Q1)2)1 = ((Q2)1)2. Each parastrophe Qi can be obtained from Q by the permutation σi, where σ1 = (23), σ2 = (13), σ3 = (132), σ4 = (123), σ5 = (12). Generally, parastrophes Qi do not save properties of Q. Parastrophes of a group are not a group, but parastrophes of an idempotent quasigroup also are idempotent 2010 Mathematics Subject Classification: 20N05, 05B15 Keywords: quasigroup, isotopism, anti-isotopism, conjugate quasigroup, parastrophe. 6 1 0 2 b e F 4 ] . A R h t a m [ 1 v 4 3 1 4 0 . 2 0 6 1 : v i X r a 222 W. A. Dudek quasigroups. Moreover, in some cases (described in [7]) parastrophes of a given quasigroup Q are pairwise equal or all are pairwise distinct (see also [2] and [8]). In [7] it is proved that the number of distinct parastrophes of a quasigroup is always a divisor of 6 and does not depend on the number of elements of a quasigroup. Parastrophes of each quasigroup can be divided into separate classes containing isotopic parastrophes. We prove that the number of such classes is always 1, 2, 3 or 6. The number of such classes depends on the existence of an anti-isotopism of a quasigroup and some parastrophe of it. 2. Classification of parastrophes As it is known (see for example [1]) a quasigroup (Q, ·) can be considered as an algebra (Q, ·, \, /) with three binary operations satisfying the following axioms x(x\z) = z, (z/y)y = z, x\xy = y, xy/y = x, where x\z = y ←→ xy = z and z/y = x ←→ xy = z. We will use these axioms to show the relationship between parastrophes. But let's start with the following simple observation. Lemma 2.1. Let Q be a quasigroup. Then (a) xy = y ◦5 x, x ◦1 y = y ◦3 x, x ◦2 y = y ◦4 x, (b) Q ∼ Q5, Q1 ∼ Q3, Q2 ∼ Q4, (c) xy = yx ←→ Q = Q5 ←→ Q1 = Q3 ←→ Q2 = Q4, (d) Q1 = Q ←→ Q2 = Q3 ←→ Q4 = Q5, (e) Q2 = Q ←→ Q1 = Q4 ←→ Q3 = Q5. To describe the relationship between the parastrophes, we will need these two simple lemmas. Lemma 2.2. Let A, B, C, D be quasigroups. Then (a) A ∼ B, B ∼ C −→ A ≈ C, (b) A ∼ B, B ≈ C −→ A ∼ C, (c) A ≈ B, B ∼ C −→ A ∼ C. Lemma 2.3. Let Q◦ i be the i-th parastrophe of the quasigroup Q◦ = (Q, ◦). Then (a) Q ≈ Q◦ implies Qi ≈ Q◦ i for each i = 1, 2, 3, 4, 5, (b) Qi ≈ Q◦ i for some i = 1, 2, 3, 4, 5 implies Q ≈ Q◦. (c) Moreover, if Q ≈ Q◦, then for each i = 1, . . . , 5 Q ∼ Qi ←→ Q◦ ∼ Q◦ i , and Q ≈ Qi ←→ Q◦ ≈ Q◦ i . Parastrophes of quasigroups 223 Now we will present a series of lemmas about anti-isotopies of quasigroups and their parastrophes. Lemma 2.4. Q ∼ Q ←→ Q ≈ Q5 ←→ Q1 ≈ Q3 ←→ Q2 ≈ Q4. Proof. Indeed, Q ∼ Q ←→ γ(xy) = α(y)β(x) ←→ γ(xy) = β(x) ◦5 α(y) ←→ Q ≈ Q5. Also Q ∼ Q ←→ γ(xy) = α(y)β(x) ←→ γ(z) = α(y)β(z/y) ←→ α(y)\γ(z) = β(z/y). Thus Q ∼ Q ←→ Q1 ∼ Q2. Moreover, Q1 ∼ Q2 ←→ α(y)\γ(z) = β(z/y) ←→ γ(z) = α(y)β(z/y) ←→ β(z/y) = γ(z) ◦4 α(y) ←→ Q2 ≈ Q4. Similarly, for some α′, β ′, γ ′ ∈ SQ we have Q1 ∼ Q2 ←→ γ ′(x/y) = α′(y)/β ′(x) ←→ γ ′(x\y)β ′(x) = α′(y) ←→ γ ′(x\y) = β ′(x) ◦3 α′(y) ←→ Q1 ≈ Q3. This completes the proof. Lemma 2.5. Q ∼ Q1 ←→ Q ∼ Q2 ←→ Q1 ≈ Q2. Proof. Indeed, according to the definition of the operations \ and /, we have γ(x\z) = α(z)β(x) ←→ γ(y) = α(xy)β(x) ←→ α(xy) = γ(y)/β(x). So, Q1 ∼ Q ←→ Q ∼ Q2, which by Lemma 2.2 implies Q1 ≈ Q2. Conversely, if Q1 ≈ Q2, then γ(x\y) = α(x)/β(y), i.e., γ(x\y)β(y) = α(x) for some α, β, γ ∈ SQ. From this, for y = xz, we obtain γ(z)β(xz) = α(x), i.e., β(xz) = γ(z)\α(x). Thus, Q ∼ Q1, and consequently, also Q ∼ Q2. Lemma 2.6. For any quasigroup Q (a) Q1 ∼ Q ←→ Q1 ∼ Q3 ←→ Q ≈ Q3 ←→ Q1 ≈ Q5, (b) Q2 ∼ Q ←→ Q2 ∼ Q4 ←→ Q ≈ Q4 ←→ Q2 ≈ Q5. Proof. Replacing in Lemma 2.5 a quasigroup Q by Q1 we get the first two equi- valences. The third equivalence is a consequence of Lemma 2.3. Similarly, replacing Q by Q2 we obtain (b). Lemma 2.7. Q3 ∼ Q ←→ Q ≈ Q2 ←→ Q1 ≈ Q4 ←→ Q3 ≈ Q5. 224 W. A. Dudek Proof. Obviously Q3 ∼ Q ←→ Q3 ≈ Q5. Moreover, Q3 ∼ Q ←→ γ(xy)α(y) = β(x) ←→ γ(xy) = β(x)/α(y) ←→ Q ≈ Q2. Analogously, xy = z we obtain Q3 ∼ Q ←→ γ(z)α(x\z) = β(x) ←→ Q1 ≈ Q4. This completes the proof. Lemma 2.8. Q4 ∼ Q ←→ Q ≈ Q1 ←→ Q2 ≈ Q3 ←→ Q4 ≈ Q5. Proof. Of course Q4 ∼ Q ←→ Q4 ≈ Q5. Since Q4 ∼ Q ←→ β(x)γ(xy) = α(y), we obtain Q4 ∼ Q ←→ Q ≈ Q1 and Q4 ∼ Q ←→ Q2 ≈ Q3 for x = z/y. Theorem 2.9. All parastrophes of a quasigroup Q are isotopic to Q if and only if Q ∼ Q and Q ∼ Qi for some i = 1, 2, 3, 4. Proof. If Q ∼ Q, then, by Lemma 2.4, we have Q ≈ Q5, Q1 ≈ Q3 and Q2 ≈ Q4. This for Q ∼ Qi, i = 1, 2, 3, 4, by Lemmas 2.6, 2.7 and 2.8, gives Q ≈ Q1 ≈ Q2 ≈ Q3 ≈ Q4 ≈ Q5. So, in this case all parastrophes are isotopic to Q. The converse statement is obvious. Corollary 2.10. If Q ∼ Q and Q ∼ Qi for some i = 1, 2, 3, 4, then also Q ∼ Qi for other i = 1, 2, 3, 4, 5. Theorem 2.11. A quasigroup Q has exactly two classes of isotopic parastrophes if and only if (1) Q 6∼ Q, Q ∼ Q1 and Q 6∼ Qi for i = 2, 3, 4, or equivalently, (2) Q 6∼ Q, Q ∼ Q2 and Q 6∼ Qi for i = 1, 3, 4. In this case Q ≈ Q3 ≈ Q4 and Q1 ≈ Q2 ≈ Q5. Proof. Let Q have exactly two classes of isotopic parastrophes. Then it must be true that Q ≈ Qi for some i = 1, 2, 3, 4, 5 because Q 6≈ Qi for all i = 1, 2, 3, 4, 5 gives Q1 ≈ Qj for some j which by previous lemmas implies Q ≈ Qk for some k. In this case Q2 ≈ Q3 and Q4 ≈ Q5 (Lemma 2.8). So, the Case Q ≈ Q1. following classes of isotopic parastrophes are possible: 1) {Q, Q1, Q2, Q3}, {Q4, Q5}, 2) {Q, Q1, Q4, Q5}, {Q2, Q3}, 3) {Q, Q1}, {Q2, Q3, Q4, Q5}. In the first case from Q1 ≈ Q3, by Lemma 2.4, we conclude Q ≈ Q5 which shows that in this case we have only one class. This contradics our assumption on the number of classes. So, this case is impossible. In the second case, Q ≈ Q5, by the same lemma, implies Q2 ≈ Q4 which (similarly as in previous case) is impossible. Also the third case is impossible because Q2 ≈ Q4 leads to Q1 ≈ Q3. Hence must be Q 6≈ Q1. Parastrophes of quasigroups 225 Case Q ≈ Q2. Then, according to Lemma 2.7, Q1 ≈ Q4 and Q3 ≈ Q5. Thus 1) {Q, Q1, Q2, Q4}, {Q3, Q5}, or 2) {Q, Q2, Q3, Q5}, {Q1, Q4}, or 3) {Q, Q2}, {Q1, Q3, Q4, Q5}. Using the same argumentation as in the case Q ≈ Q1 we can see that the case Q ≈ Q2 is impossible. Case Q ≈ Q3. By Lemmas 2.1, 2.2 and 2.5 only the following classes are In this case Q 6∼ Q (Lemma 2.4) and possible: {Q, Q3, Q4} and {Q1, Q2, Q5}. Q ∼ Q1 (Lemma 2.6). Then also Q ∼ Q2 (Lemma 2.5). Case Q ≈ Q4. Analogously as Q ≈ Q3. Case Q ≈ Q5. Then Q1 ≈ Q3 and Q2 ≈ Q4. Is a similar way as for Q ≈ Q1 we can verify that this case is not possible. So, if Q has exactly two classes of isotopic parastrophes, then Q 6∼ Q and Q ∼ Q1, or Q 6∼ Q and Q ∼ Q2. Conversely, if Q 6∼ Q and Q ∼ Q1, or equivalently, Q 6∼ Q and Q ∼ Q2, then by Lemmas 2.5 and 2.6 we have two classes: {Q, Q3, Q4} and {Q1, Q2, Q5}. Since Q1 6≈ Q3 (Lemma 2.4), these classes are disjoint. Theorem 2.12. A quasigroup Q has exactly three classes of isotopic parastrophes if and only if (1) Q 6∼ Q, Q ∼ Q3 and Q 6∼ Qi for i = 1, 2, 4, or (2) Q 6∼ Q, Q ∼ Q4 and Q 6∼ Qi for i = 1, 2, 3, or (3) Q ∼ Q, Q ∼ Q5 and Q 6∼ Qi for i = 1, 2, 3, 4. In the first case we have {Q, Q2}, {Q1, Q4} and {Q3, Q5}; in the second {Q, Q1}, {Q2, Q3} and {Q4, Q5}; in the third {Q, Q5}, {Q1, Q3} and {Q2, Q4}. Proof. Suppose that a quasigroup Q has exactly three classes of isotopic parastro- phes. From the above lemmas it follows that in this case Q ≈ Qi for some i. Case Q ≈ Q1. Then, by Lemma 2.8, we have three classes {Q, Q1}, {Q2, Q3}, {Q4, Q5} and Q ∼ Q4. Since Q1 6≈ Q3 we also have Q 6∼ Q (Lemma 2.4). Case Q ≈ Q2. In this case {Q, Q2}, {Q1, Q4}, {Q3, Q5} and Q ∼ Q3 (Lemma 2.7). Analogously as in the previous case Q1 6≈ Q3 gives Q 6∼ Q. Case Q ≈ Q3. This case is impossible because by Lemmas 2.5 and 2.6 it leads to two classes. Case Q ≈ Q4. Analogously as Q ≈ Q3. Case Q ≈ Q5. Then Q1 ≈ Q3, Q2 ≈ Q4 and Q ∼ Q. Since classes {Q, Q5}, {Q1, Q3}, {Q2, Q5} are disjoint Q 6∼ Qi for each i = 1, 2, 3, 4. The converse statement is obvious. As a consequence of the above results we obtain 226 W. A. Dudek Corollary 2.13. Parastrophes of a quasigroup Q are non-isotopic if and only if Q 6∼ Q and Q 6∼ Qi for all i = 1, 2, 3, 4. Corollary 2.14. The number of non-isotopic parastrophes of a quasigroup Q is always 1, 2, 3, or 6. Depending on the relationship between parastrophes quasigroups can be di- vided into six types presented below. type classes of isotoipic parastrophes {Q, Q1, Q2, Q3, Q4, Q5} A {Q, Q3, Q4}, {Q1, Q2, Q5} B C {Q, Q2}, {Q1, Q4}, {Q3, Q5} D {Q, Q1}, {Q2, Q3}, {Q4, Q5} {Q, Q5}, {Q1, Q3}, {Q2, Q4} E F {Q}, {Q1}, {Q2}, {Q3}, {Q4}, {Q5} Our results are presented in the following table where "+" means that the corresponding relation holds. The symbol "−" means that this relation has no place. Q ∼ Q + − − − − + − Q ≈ Q5 Q ∼ Q1 + + − − − − − Q ≈ Q3 Q ∼ Q2 + − + − − − − Q ≈ Q4 Q ∼ Q3 + − − + − − − Q ≈ Q2 Q ∼ Q4 + − − − + − − Q ≈ Q1 type A B B C D E F The parastrophe Q5 plays no role in our research since always is Q ∼ Q5. 3. Parastrophes of selected quasigroups In this section we present characterizations of parastrophes of several classical types of quasigroups. We start with parastrophes of IP-quasigroups. As a consequence of our results, we get the following well-known fact (see for example [1]) Proposition 3.1. All parastrophes of an IP -quasigroup are isotopic. Proof. Indeed, in any IP -quasigroup Q there are permutations α, β ∈ SQ such that α(x) · xy = y = yx · β(x) for all x, y ∈ Q. So, Q ≈ Q1 ≈ Q2, i.e., Q is a quasigroup of type A. Parastrophes of quasigroups 227 Corollary 3.2. Parastrophes of a group are isotopic. The same is true for the parastrophes of Moufang quasigroups since groups and Moufang quasigroups are IP-quasigroups. Also parastrophes of T-quasigroups, linear and alinear quasigroups (studied in [3]) are isotopic. This fact follows from more general result proved below. Theorem 3.3. All parastrophes of a quasigroup isotopic to a group are isotopic. Proof. Let G = (G, ◦) be a group. Then ϕ(x ◦ y) = ϕ(y) ◦ ϕ(x) for ϕ(x) = x−1. Since (Q, ·) ≈ (G, ◦), for some α, β, γ we have γ(xy) = α(x) ◦ β(y) = ϕ−1(ϕβ(y) ◦ ϕα(x)) = ϕ−1γ (cid:0)α−1ϕβ(y) · β−1ϕα(x)(cid:1) . Thus γ−1ϕγ(xy) = α−1ϕβ(y) · β−1ϕα(x). So, Q ∼ Q. Moreover, from γ(xy) = α(x) ◦ β(y) for xy = z we obtain α(x)\\γ(z) = β(x\z) and γ(z)//β(y) = α(z/y), where \\ and // are inverse operations in a group G. Thus Q1 ≈ G1 and Q2 ≈ G2. Since G ≈ G1 ≈ G2, also Q ≈ Q1 ≈ Q2. This shows that a quasigroup isotopic to a group is a quasigroup of type A. Hence (Lemma 2.3) all its parastrophes are isotopic. D-loops (called also loops with anti-automorphic property) are defined as loops with the property (xy)−1 = y−1x−1, where x−1 denotes the inverse element [5]. Theorem 3.4. Let Q be a D-loop. Then (1) all parastrophes of Q coincide with Q, or (2) Q has three classes of isotopic parastrophes: {Q, Q5}, {Q1, Q3}, {Q2, Q4}. The second case holds if and only if Q 6∼ Q1 or Q 6≈ Q1. Proof. Let Q be a D-loop. Then Q ∼ Q. Thus all its parastrophes are isotopic to Q or they are divided into three classes {Q, Q5}, {Q1, Q3}, {Q2, Q4} (see Table). By Lemmas 2.6 and 2.8 they are disjoint if and only if Q 6∼ Q1 or Q 6≈ Q1. Corollary 3.5. A D-loop Q has three classes of isotopic parastrophes if and only if Q 6∼ Q2 or Q 6≈ Q2. In [5] is proved that parastrophes of a D-loop Q are isomorphic to one of the quasigroups Q, Q1, Q2. Comparing this fact with our results we obtain Theorem 3.6. For a D-loop Q the following conditions are equivalent: (1) all parastrophes of Q are isomorphic, (2) Q and Q1 are isomorphic, (3) Q and Q2 are isomorphic, (4) Q1 and Q2 are isomorphic. 228 W. A. Dudek Example 3.7. Consider the following three loops. 1 · 1 1 2 2 3 3 4 4 5 5 6 6 2 3 2 3 1 6 6 1 5 2 3 4 4 5 4 5 4 5 5 3 2 4 1 6 6 1 3 2 6 6 4 5 3 2 1 ◦1 1 2 3 4 5 6 1 1 2 3 4 5 6 2 2 1 4 3 6 5 3 4 3 4 5 6 1 5 6 1 2 3 4 2 5 6 5 6 4 3 6 2 2 5 1 4 3 1 ◦2 1 2 3 4 5 6 1 2 1 2 2 1 3 5 4 6 5 4 6 3 3 4 3 4 4 3 1 6 5 1 6 2 2 5 5 6 5 6 6 5 2 4 3 2 1 3 4 1 The first loop is a D-loop, the second and the third are parastrophes of the first. They are not D-loops and are not isotopic to the first. So this D-loop has three classes of isotopic parastrophes. In this case Q = Q5, Q1 = Q3 and Q2 = Q4. 4. Some consequences Note first of all that the proofs of our results remain true also for the case when α = β = γ. In this case an anti-isotopism is an anti-isomorphism and an isotopism is an isomorphism. So, the above results will be true if we replace an anti-isotopism by an anti-isomorphism, and an isotopism by an isomorphism. Moreover, an isotopism of parastrophes can be characterized by the identities: α1(x) · β1(yx) = γ1(y), β2(xy) · α2(x) = γ2(y), β3(yx) · α3(x) = γ3(y), α4(x) · β4(xy) = γ4(y), β5(xy) = γ5(y) · α5(x), (1) (2) (3) (4) (5) where αi, βi, γi are fixed permutations of the set Q. Namely, from our results it follows that Q satisfies (1) ←→ Q1 ∼ Q ←→ Q3 ≈ Q, Q satisfies (2) ←→ Q2 ∼ Q ←→ Q4 ≈ Q, Q satisfies (3) ←→ Q3 ∼ Q ←→ Q2 ≈ Q, Q satisfies (4) ←→ Q4 ∼ Q ←→ Q1 ≈ Q, Q satisfies (5) ←→ Q ∼ Q ←→ Q5 ≈ Q. Lemma 2.3 shows that these identities are universal in some sense, i.e., if one of these identities is satisfied in a quasigroup Q, then in a quasigroup isotopic to Q is satisfied the identity of the same type, i.e., it is satisfied with other permutations. Since Q ∼ Q1 ←→ Q ∼ Q2 we have Parastrophes of quasigroups 229 Proposition 4.1. A quasigroup Q satisfies for some α1, β1, γ1 ∈ SQ the identity (1) if and only if for some α2, β2, γ2 ∈ QS it satisfies the identity (2). As a consequence we obtain the following classification of quasigroups> Theorem 4.2. Let Q be a quasigroup. Then • Q is type A if and only if it satisfies all of the identities (1) − (5), • Q is type B if and only if it satisfies only (1) and (2), • Q is type C if and only if it satisfies only (3), • Q is type D if and only if it satisfies only (4), • Q is type E if and only if it satisfies only (5), • Q is type F if and only if it satisfies none of the identities (1) − (5). If all permutations used in (1) − (5) are the identity permutations, then these equations have of the form: x · yx = y, xy · x = y, yx · x = y, x · xy = y, xy = yx. (6) (7) (8) (9) (10) Basing on our results we conclude that Q satisfies (6) ←→ Q = Q4, Q satisfies (7) ←→ Q = Q3, Q satisfies (8) ←→ Q = Q2, Q satisfies (9) ←→ Q = Q1, Q satisfies (10) ←→ Q = Q5. Since Q satisfies (7) ←→ Q5 = Q2 ←→ ((Q1)2)1 = Q2 ←→ Q1 = ((Q2)1)2 ←→ Q1 = Q5 ←→ Q satisfies (6), we see that identities (7) and (6) are equivalent, i.e., Q satisfies (7) if and only if it satisfies (6). As a consequence we obtain the stronger version of Theorem 4 in [7]. Theorem 4.3. Parastrophes of a quasigroup Q can be characterized by the iden- tities (6) − (10) in the following way: • Q = Qi for 1 6 i 6 5 if and only if it satisfies all of the identities (6) − (10), • Q = Q3 = Q4, Q1 = Q2 = Q5 if and only if Q satisfies only (7) and (6), • Q = Q2, Q1 = Q4, Q3 = Q5 if and only if Q satisfies only (8), • Q = Q1, Q2 = Q3, Q4 = Q5 if and only if Q satisfies only (9), • Q = Q5, Q1 = Q3, Q2 = Q4 if and only if Q satisfies only (10), • Q 6= Qi 6= Qj for all 1 6 i < j 6 5 if and only if Q satisfies none of the identities (6) − (10). 230 W. A. Dudek Corollary 4.4. Parastrophes of a commutative quasigroup Q coincide with Q or are divided into three classes: {Q = Q5}, {Q1 = Q3}, {Q2 = Q4}. Corollary 4.5. For a commutative quasigroup Q the following conditions are equivalent: (1) all parastrophes of Q coincide with Q, (2) Q = Q1, (3) Q = Q2, (4) Q1 = Q2, (5) Q satisfies at least one of the identities (6) − (9). Proof. We prove only the equivalence (1) ←→ (2). Other equivalences can be proved in a similar way. For a commutative Q we have Q = Q5, Q1 = Q3, Q2 = Q4. If Q = Q1, then Q = Q1 = Q3 = Q5. Hence Q1 = Q5 = ((Q2)1)2 which gives (Q1)2 = (Q2)1. So, Q3 = Q4, i.e., (2) implies (1). The converse implication is obvious. Corollary 4.6. Parastrophes of a boolean group coincide with this group. Note finally that identities (6)−(10) can be used to determine some autotopisms of quasigroups [4]. References [1] V.D. Belousov, Foundations of the theory of quasigroups and loops, (Russian), Moscow (1967). [2] G.B. Belyavskaya and T.V. Popovich, Conjugate sets of loops and quasigroups. DC-quasigroups, Bul. Acad. Ştiinţe Repub. Mold. Mat. 1(68) (2012), 21 − 31. [3] G.B. Belyavskaya and A.Kh. Tabarov, Characterization of linear and alinear quasigroups, (Russian), Diskr. Mat. 4 (1992) 142 − 147. [4] A.I. Deriyenko, Autotopisms of some quasigroups, Quasigroups and Related Sys- tems 23 (2015), 217 − 220. [5] I.I. Deriyenko and W.A. Dudek, D-loops, Quasigroups and Related Systems 20 (2012), 183 − 196. [6] A.D. Keedwell and J. Dénes, Latin squares and their applications, Second edi- tion, Elsevier, 2015. [7] C.C. Lindner and D. Steedly, On the number of conjugates of a quasigroups, Algebra Universalis 5 (1975), 191 − 196. [8] T.V. Popovich, On conjugate sets of quasigroups, Bul. Acad. Ştiinţe Repub. Mold. Mat. 3(67) (2011), 69 − 76. Faculty of Pure and Applied Mathematics Wroclaw University of Technology Wyb. Wyspiańskiego 27 50-370 Wroclaw, Poland E-mail: [email protected] Received November 12, 2015
1105.4284	3	1105	2013-03-13T18:25:22	Lie algebras with given properties of subalgebras and elements	[ "math.RA" ]	Results about the following classes of finite-dimensional Lie algebras over a field of characteristic zero are presented: anisotropic (i.e., Lie algebras for which each adjoint operator is semisimple), regular (i.e., Lie algebras in which each nonzero element is regular in the sense of Bourbaki), minimal nonabelian (i.e., nonabelian Lie algebras all whose proper subalgebras are abelian), and algebras of depth 2 (i.e., Lie algebras all whose proper subalgebras are abelian or minimal nonabelian).	math.RA	math	LIE ALGEBRAS WITH GIVEN PROPERTIES OF SUBALGEBRAS AND ELEMENTS PASHA ZUSMANOVICH We study the following classes of Lie algebras: anisotropic (i.e., Lie algebras for which each adjoint operator ad x is semisimple), regular (i.e., Lie algebras in which each nonzero element is regular), minimal nonabelian (i.e., nonabelian Lie algebras all whose proper subalgebras are abelian), and algebras of depth 2 (i.e., Lie algebras all whose proper subalgebras are abelian or minimal nonabelian). All algebras, Lie and associative, are assumed to be finite-dimensional and defined over a fixed field of characteristic zero (though some of the results, in a weaker form or under additional restrictions, will hold also in positive characteristic). We stress that the base field is not assumed to be algebraically closed (all the things considered here are collapsing to vacuous trivialities in the case of an algebraically closed base field). Our notations are standard and largely follow Bourbaki [B]. The symbols ∔, ⊕, and A denote direct sum of vector spaces, direct sum of Lie algebras, and semidirect sum of Lie algebras (the first summand acting on the second), respectively. 1. Anisotropic algebras It is shown in [F1, Propostion 1.2] that any anisotropic solvable Lie algebra is abelian. From this and the Levi -- Malcev decomposition follows that any anisotropic Lie algebra is reductive. Theorem 1. For a reductive Lie algebra L the following are equivalent: (i) L is anisotropic; (ii) all proper subalgebras of L are anisotropic; (iii) all proper subalgebras of L are reductive; (iv) all 2-dimensional subalgebras of L are abelian; (v) L does not contain subalgebras isomorphic to sl(2). Proof. (i) ⇒ (ii). If S is a subalgebra of L, then for any x ∈ S, adS x is a restriction of adL x, hence the semisimplicity of the latter implies the semisimplicity of the former. (ii) ⇒ (iii) follows from the observation above that any anisotropic Lie algebra is reductive. (iii) ⇒ (iv) follows from the obvious fact that a 2-dimensional reductive Lie algebra is abelian. (iv) ⇒ (v) follows from the obvious fact that sl(2) contains a 2-dimensional nonabelian subalgebra. (v) ⇒ (i). Write L as a direct sum L = g ⊕ A, where g is semisimple and A is abelian. Suppose g is not anisotropic. As g contains semisimple and nilpotent components of each of its elements ([B, Chapter I, §6, Theorem 3]), g contains a nonzero nilpotent element, and by the Jacobson -- Morozov theorem ([B, Chapter VIII, §11, Proposition 2]) g contains sl(2) as a (cid:3) subalgebra, a contradiction. Hence g is anisotropic and L is anisotropic. Though the proof is elementary, and all the necessary ingredients are contained in [F1] anyway (in particular, the implication (i) ⇒ (iv) is noted in [F1, §1], and the equivalence (i) ⇔ (v) in the case of semisimple L is proved, with a slightly different argument, in [F1, Date: last revised October 2, 2012. arXiv:1105.4284; Proceedings of the Conference "Algebra - Geometry - Mathematical Physics" (Mulhouse, 2011), Springer Proceedings in Mathematics and Statistics, to appear. 1 2 PASHA ZUSMANOVICH Theorem 2.1]), we find this explicit formulation of Theorem 1 interesting enough. There are many works in the literature devoted to study of minimal non-P Lie algebras, i.e. Lie algebras not satisfying P and such that all their proper subalgebras satisfy P, where P is a certain "natural" property of Lie algebras (abelianity, nilpotency, solvability, simplicity, modularity of the lattice of subalgebras, . . . ). In all the cases studied so far, the class of minimal non-P algebras turns out to be highly nontrivial (without further assumptions about the base field, such as algebraic or quadratic closedness, triviality of the Brauer group, etc.), with lot of simple objects. To the contrary, from the Levi -- Malcev decomposition and Theorem 1 it follows that the class of minimal nonanisotropic Lie algebras is relatively trivial: those are exactly solvable minimal nonabelian Lie algebras. One may ask a "philosophical" question: what makes the condition of being anisotropic different in that regard from other conditions? Where is a borderline for a property P which makes the class of minimal non-P Lie algebras small and "simple" (or even empty)? Corollary. A simple Lie algebra all whose proper subalgebras are not simple, is either min- imal nonabelian, or isomorphic to sl(2). Proof. Let L be a reductive Lie algebra all whose proper subalgebras are not simple. By implication (v) ⇒ (iii) of Theorem 1, either L is isomorphic to sl(2), or all proper subalgebras of L are reductive. As any nonabelian reductive Lie algebra contains a simple subalgebra, in (cid:3) the latter case all proper subalgebras of L are abelian. In [T2, Theorem 2.2] a statement similar to the corollary is proved about simple Lie algebras, all whose proper subalgebras are supersolvable. Theorem 2. Let L be a nonempty class of Lie algebras satisfying the following properties: (i) L is closed with respect to subalgebras; (ii) if each proper subalgebra of a reductive Lie algebra L belongs to L , then L belongs to L ; (iii) solvable Lie algebras belonging to L are abelian. Then L is the class of all anisotropic Lie algebras. Proof. Any class of Lie algebras satisfying conditions (i) and (iii) consists of anisotropic algebras. Indeed, from the Levi -- Malcev decomposition and condition (iii) it follows that any algebra in L is reductive. Then from implication (iii) ⇒ (i) of Theorem 1 and condition (i), it follows that any algebra in L is anisotropic. Now, suppose that there is an anisotropic Lie algebra not belonging to L , and consider such algebra L of the minimal possible dimension. Then all proper subalgebras of L belong (cid:3) to L , and by condition (ii) L itself belongs to L , a contradiction. 2. Regular algebras If N is a nilpotent subalgebra of a Lie algebra L, by L0(N) is denoted the Fitting 0- component with respect to the N-action on L (i.e., the set of all elements of L on which N acts nilpotently). Recall ([B, Chapter VII, §2.2]) that rank rk L of a Lie algebra L is the minimal possible non-vanishing power of the characteristic polynomial of ad x, x ∈ L, and elements of L for which this minimal number is attained are called regular. Another characterization of x ∈ L to be a regular element is the equality dim L0(x) = rk L. If each nonzero element of L is regular, then L itself is called regular. It is clear that any nilpotent Lie algebra is regular, with rank equal the dimension of the algebra. If a regular Lie algebra L is not semisimple, i.e., contains a nonzero abelian ideal I, then for any x ∈ I, (ad x)2 = 0, hence each element in L is nilpotent, and by the Engel theorem L is nilpotent. It is clear also that a regular semisimple Lie algebra is simple (see LIE ALGEBRAS WITH GIVEN PROPERTIES OF SUBALGEBRAS AND ELEMENTS 3 [B, Chapter VII, §2.2, Proposition 7]), and that a regular simple Lie algebra is anisotropic (see [B, Chapter VII, §2.4, Corollary 2]). Theorem 3. For a simple Lie algebra L the following are equivalent: (i) L is regular; (ii) all proper subalgebras of L are regular; (iii) all proper subalgebras of L are either simple, or abelian. Proof. (i) ⇒ (ii) follows from the fact that if S is a subalgebra of L, and x ∈ S is a regular element in L, then x is a regular element in S ([B, Chapter VII, §2.2, Proposition 9]). (ii) ⇒ (iii). By the observation above, any proper subalgebra of L is either simple, or nilpotent. Hence L does not contain a 2-dimensional nonabelian Lie algebra, and by implica- tion (iv) ⇒ (iii) of Theorem 1, all proper subalgebras of L are reductive, and all its nilpotent subalgebras are abelian. (iii) ⇒ (i). By implication (iii) ⇒ (i) of Theorem 1, L is anisotropic. In any Lie algebra, Cartan subalgebras are exactly nilpotent subalgebras N such that L0(N) = N ([B, Chapter VII, §2.1, Proposition 4]). But nilpotent subalgebras of L are abelian, and L0(N) coincides with the centralizer of N, so Cartan subalgebras of L are exactly abelian subalgebras coin- ciding with their own centralizer. For an arbitrary nonzero element x ∈ L, its centralizer ZL(x) cannot be simple, hence it is abelian. But, obviously, ZL(x) coincides with its own centralizer, hence ZL(x) is a Cartan subalgebra of L, dim ZL(x) = dim L0(x) = rk L, and x (cid:3) is regular. Note that similar to the anisotropic case, minimal nonregular Lie algebras are exactly solvable minimal nonnilpotent Lie algebras. Theorem 4. Let L be a nonempty class of Lie algebras satisfying the following properties: (i) L is closed with respect to subalgebras; (ii) if each proper subalgebra of a Lie algebra L belongs to L , then L belongs to L ; (iii) non-semisimple Lie algebras belonging to L are nilpotent. Then L is the class of all regular Lie algebras. Proof. Any class of Lie algebras satisfying conditions (i) and (iii) consists of regular algebras. Indeed, from the Levi -- Malcev decomposition and condition (iii) it follows that any algebra L in L is either semisimple, or nilpotent. In the former case, write L = g1 ⊕ · · · ⊕ gn as the direct sum of simple components. If n > 1, by condition (i) the subalgebra of L of the form g1 ⊕ Kx, where x is an arbitrary nonzero element of g2, belongs to L , and by condition (iii) it is nilpotent, a contradiction. Hence n = 1, that is, L is simple. By conditions (i) and (iii) L does not contain 2-dimensional nonabelian subalgebra, and by implication (iv) ⇒ (iii) of Theorem 1, all subalgebras of L are reductive. This, together with conditions (i) and (iii) again, implies that all subalgebras of L are either simple, or abelian, and by implication (iii) ⇒ (i) of Theorem 3, L is regular. Now, the same elementary reasoning utilizing condition (ii) as at the end of the proof of (cid:3) Theorem 2, shows that any regular Lie algebra belongs to L . 3. Minimal nonabelian algebras It follows from the Levi -- Malcev decomposition that any minimal nonabelian Lie algebra is either simple, or solvable. Solvable minimal nonabelian Lie algebras (even in a slightly more general minimal nonnilpotent setting) were described in [St], [GKM], and [T1]. A simple minimal nonabelian Lie algebra is regular. Simple minimal nonabelian Lie algebras were studied in [F2] and [G], but their full description remains an open problem. Recall that an algebra is called central if its centroid coincides with the base field. For simple algebras this is equivalent to the condition that the algebra remains simple under extension of the base field. 4 PASHA ZUSMANOVICH Theorem 5. There are no central simple minimal nonabelian Lie algebras of types Bl (l ≥ 2), Cl (l ≥ 3, l 6= 2k), Dl (l ≥ 5, l 6= 2k), G2, and F4. Proof. The proof follows from the known classification of central simple Lie algebras of these types (see, for example, [Se, Chapter IV]). Types B -- D. Each central simple Lie algebra of this type (with the exception of D4) is isomorphic to a Lie algebra of J-skew-symmetric elements S−(A, J) = {x ∈ A J(x) = −x}, where A is a central simple associative algebra of dimension n2 > 16 with involution J of the first kind (smaller dimensions of A are covered by "occasional" isomorphisms between "small" algebras of different types, including type A). By a known description of such algebras (see, for example, [J2, Theorem 5.1.23]), A is isomorphic to Mm(D), a matrix algebra of size m×m over a central division algebra D with involution j, and J has the form 1 , . . . , g−1 m ) k,ℓ=1 7→ diag(g1, . . . , gm)(j(dkℓ))⊤ diag(g−1 (dkℓ)m for some g1, . . . , gm ∈ D such that j(gk) = gk, k = 1, . . . , m. If D coincides with the base field, i.e. A is a full matrix algebra, than the Lie algebra S−(A, J) is split and, obviously, contains a lot of proper nonabelian subalgebras. Hence we may assume dim D ≥ 4. From the description above it is clear that, provided m > 1, the subalgebra B of A of all matrices with vanishing last row and column, is isomorphic to Mm−1(D) and is stable under J, hence S−(B, J) is a Lie subalgebra of S−(A, J). Since dim A = m2 dim D ≥ 25, we have dim B = (m − 1)2 dim D = s2 ≥ 9, and this subalgebra is a central simple Lie algebra of dimension s(s−1) . Therefore, if m > 1, S−(A, J) contains proper nonabelian subalgebras, and it remains to consider the case where A = D is a division algebra. or s(s+1) 2 2 Since D has an involution, its exponent is equal to 2, and its dimension n2 is equal to some power of 4. This excludes all the types mentioned in the statement of the theorem. Type G2. Each central simple Lie algebra of this type is a derivation algebra of a 8- dimensional Cayley algebra O. The latter is obtained by the doubling (Cayley -- Dickson) process from the 4-dimensional associative quaternion algebra H, and it is known that each derivation of H can be extended to a derivation of O (see, for example [Sc, Theorem 2]). Thus, Der(O) always contains a 3-dimensional central simple Lie algebra Der(H) as a subalgebra, and hence cannot be minimal nonabelian. Type F4. Each central simple Lie algebra of this type is a derivation algebra of a 27- dimensional exceptional simple Jordan algebra J. It is known that derivations of J mapping a cubic subfield of J to zero form a central simple Lie algebra of type D4 (see, for example, [J1, Chapter IX, §11, Exercise 5]). (cid:3) We conjecture that the remaining types not covered by Theorem 5 -- C2k and D2k -- cannot occur as well. Conjecture. There are no central simple minimal nonabelian Lie algebras of types B -- D (except of D4). Let us provide some evidence in support of this conjecture. Lemma. Let D be a central division algebra of dimension n2 over a field K with involution J of the first kind, such that S−(D, J) is a minimal nonabelian Lie algebra. Then for any J-symmetric or J-skew-symmetric element x in D, not lying in K, one of the following holds: (i) x is J-symmetric and of degree 2; (ii) K[x] is of degree n (iii) K[x] is a maximal subfield of D. 2 , and dimK[x] CD(x) = 4; Proof. The associative centralizer of x in D, CD(x), is a proper simple associative subalgebra of D. By the Double Centralizer Theorem (see, for example, [P, §12.7]), (1) dim K[x] · dim CD(x) = n2, LIE ALGEBRAS WITH GIVEN PROPERTIES OF SUBALGEBRAS AND ELEMENTS 5 and the associative center of CD(x) coincides with K[x]. As CD(x) is stable under J, S−(CD(x), J) is a Lie subalgebra of S−(D, J). If it coincides < 3, with the whole S−(D, J), then S−(D, J) ⊆ CD(x), and by (1), dim K[x] ≤ n2 hence dim K[x] = 2, i.e. K[x] is a quadratic extension of K, the case (i). Note that in this case x cannot be J-skew-symmetric, as otherwise it lies in the Lie center of S−(D, J), a contradiction. n(n−1) 2 If S−(CD(x), J) is a proper subalgebra of S−(D, J), then it is abelian, and by [H, Theorem 2.2], CD(x) is either commutative (i.e., a subfield of D), or is 4-dimensional over its center K[x]. In the former case, since the degree (= dimension) over K of each intermediate field between K and D is ≤ n (actually, a divisor of n), and since K[x] ⊆ CD(x), we have dim K[x] = dim CD(x) = n, and CD(x) = K[x], the case (iii). In the latter case, from (1) we have dim K[x] = n (cid:3) 2 and dim CD(x) = 2n, the case (ii). For example, if the division algebra D is cyclic (what always happens over number fields), then, considering the conditions of the lemma simultaneously for a J-skew-symmetric el- ement x generating a cyclic extension of the base field, and even powers of x (which are J-symmetric), one quickly arrives to a contradiction. For the remaining exceptional types, the question seems to be much more difficult, and it is treated in [GG] using the language and technique of algebraic groups and Galois cohomology. There are central simple minimal nonabelian Lie algebras of types D4 and E8. For types E6 and E7 partial answers are available. Central simple minimal nonabelian Lie algebras of type A of the form D(−)/K1 (i.e., quotient of D, considered as a Lie algebra subject to commutator [a, b] = ab − ba, by the 1-dimensional center spanned by the unit 1 of D), where D is a central division associative algebra, were studied in [G]. A necessary, but not sufficient condition for such Lie algebra to be minimal nonabelian is D to be minimal noncommutative (i.e., all proper subalgebras of D are commutative). In this connection the following observation is of interest: Theorem 6. Let D be a central division associative algebra. Then the Lie algebra D(−)/K1 is regular if and only if D is a minimal noncommutative algebra. Proof. Let the dimension of D over the base field K is equal to n2, so dim D(−)/K1 = n2 − 1. The Lie algebra D(−)/K1 is regular if and only if the Lie centralizer of any nonzero element x ∈ D(−)/K1 is a Cartan subalgebra of dimension n − 1, what, in associative terms, is equivalent to the condition that the associative centralizer CD(x) of any element x ∈ D\K, is a maximal subfield of D of dimension n over K. Taking this into account, the proof is an easy application of the Double Centralizer Theorem, with reasonings similar to those used in the proof of the lemma above. The "only if " part. Suppose that for any x ∈ D\K, CD(x) is a maximal subfield of D. Consider a subfield K[x] ⊆ CD(x) of D. We have CD(x) = CD(K[x]), and by the Double Centralizer Theorem, dim K[x] · dim CD(x) = n2. But the degree (= dimension) over K of each intermediate field between K and D is ≤ n (actually, a divisor of n), hence dim K[x] = dim CD(x) = n, and CD(x) = K[x]. That means that there are no intermediate fields between K and the maximal subfields of D. If A is a noncommutative proper subalgebra of D, then, obviously, A is a division algebra. Its center Z(A), being a field extension of K, either coincides with K, or is a maximal subfield of D. In the former case A is central of dimension m2, where 1 < m < n, and its maximal subfield has degree m over K, a contradiction. In the latter case, we have dim A > dim Z(A) = n. Applying again the Double Centralizer Theorem, we have dim A·dim CD(A) = n2. Since Z(A) ⊆ CD(A), we have dim CD(A) ≥ dim Z(A) = n, a contradiction. The "if " part. Suppose D is minimal noncommutative. For an arbitrary x ∈ D not lying in the base field K, its centralizer CD(x) is a subfield of D. By the Double Centralizer Theorem, CD(CD(x)) is a simple subalgebra of D (and, hence, is also a subfield), and dim CD(x) · 6 PASHA ZUSMANOVICH dim CD(CD(x)) = n2. By the same argument as above about degrees of intermediate fields between K and D, dim CD(x) = dim CD(CD(x)) = n. Since CD(x) ⊆ CD(CD(x)), this (cid:3) implies CD(x) = CD(CD(x)), and CD(x) is a maximal subfield of D. 4. Algebras of depth 2 Define the depth of a Lie algebra in the following inductive way: a Lie algebra has depth 0 if and only if it is abelian, and has depth n > 0 if and only if it does not have depth < n and all its proper subalgebras have depth < n. Thus, minimal nonabelian Lie algebras are exactly algebras of depth 1. Many of the algebras considered below arise as semidirect sums L A V of a Lie algebra L and an L-module module V (in such a situation, we will always assume that V is an abelian ideal: [V, V ] = 0). It is clear that the depth of such semidirect sums is related to depth of L and the maximal length of chains of subspaces of V invariant under action of subalgebras of L, though the exact formulation in the general case seems to be out of reach. In the particular case where L is 1-dimensional, the depth of such semidirect sum is equal to the maximal length of chains in V of invariant subspaces with nontrivial L-action. The following can be considered as an extension of the corresponding results from [St], [GKM], and [T1]. Theorem 7. A non-simple Lie algebra of depth 2 over a field K is isomorphic to one of the following algebras: (i) A 4-dimensional solvable Lie algebra having the basis {x, y, z, t} and the following mul- tiplication table: [x, y] = z, [x, z] = 0, [y, z] = 0, [z, t] = 0, with ad t acting on the space Kx ∔ Ky invariantly, without nonzero eigenvectors, and with trace zero. (ii) A 4-dimensional solvable Lie algebra having the basis {x, y, z, t} and the following mul- tiplication table: [x, y] = z, [x, z] = 0, [y, z] = 0, [z, t] = z, with ad t acting on the space Kx ∔ Ky invariantly, without nonzero eigenvectors, and with trace 1. (iii) A direct sum of a simple minimal nonabelian Lie algebra and 1-dimensional algebra. (iv) A semidirect sum S A V , where S is either the 2-dimensional nonabelian Lie algebra, or a 3-dimensional simple minimal nonabelian Lie algebra, and V is an S-module such that each nonzero element of S acts on V irreducibly. (v) A semidirect sum S A V , where S is an abelian 1- or 2-dimensional Lie algebra, and V is an S-module such that for each nonzero element x ∈ S, the maximal length of chains of x-invariant subspaces of V is equal to 2 (what is equivalent to saying that any proper x-invariant subspace does not contain proper x-invariant subspaces). Proof. It is a straightforward verification that in each of these cases the corresponding Lie algebras have depth 2, so let us prove that each non-simple Lie algebra L of depth 2 has one of the indicated forms. Note that L cannot be semisimple. For, in this case it is decomposed into the direct sum of simple components: L = g1 ⊕ · · · ⊕ gn, n > 1, and any subalgebra of the form g1 ⊕ Kx, x ∈ g2, x 6= 0, is not minimal nonabelian. Suppose that L is non-semisimple and non-solvable, and let L = g A Rad(L) be its Levi -- Malcev decomposition. Then g is minimal nonabelian and hence is simple. Further, Rad(L) abelian, as otherwise g A [Rad(L), Rad(L)] is a proper subalgebra of L which is not minimal nonabelian. Suppose now that rk g > 1, and g acts on Rad(L) nontrivially. Then taking x ∈ g with a nontrivial action on Rad(L), and the Cartan subalgebra H of g of dimension LIE ALGEBRAS WITH GIVEN PROPERTIES OF SUBALGEBRAS AND ELEMENTS 7 > 1 containing x, we get a subalgebra H A Rad(L) of L which is not minimal nonabelian. Hence in the case rk g > 1, Rad(L) is a trivial (and then, obviously, 1-dimensional) g-module, and we arrive at case (iii). If rk g = 1, then g is 3-dimensional. If some nonzero x ∈ g acts on Rad(L) trivially, then so is [x, g], and, since g is generated by the latter subspace, the whole g acts on Rad(L) trivially, a case covered by (iii). Assume that any nonzero x ∈ g acts on Rad(L) nontrivially. The Lie subalgebra Kx A Rad(L) contains, in its turn, a subalgebra Kx A V for any proper ad x-invariant subspace V of Rad(L), what shows that x acts trivially on V . Letting here V to be the Fitting 1-component with respect to the x-action on Rad(L), we see that Rad(L) = V , what means that x acts on Rad(L) nondegenerately, and hence, irreducibly. We arrive at case (iv). It remains to consider the case of L solvable. Take any subspace A of L of codimension 1 containing [L, L], and a complimentary 1-dimensional subspace: (2) L = Kt ∔ A, ad t acts on A. Since A is a proper ideal of L, it is either abelian or minimal nonabelian. In the former case, we arrive at the semidirect sum Kt A A, and it is easy to see that any proper nonabelian subalgebra of L is isomorphic to the semidirect sum Kt A V , where V is a proper ad t-invariant subspace of A. Thus, for L to be of depth 2 is equivalent to the condition described in case (v) (with S 1-dimensional). Suppose now that A is minimal nonabelian. According to [GKM, Theorem 4] (also implicit in [St] and [T1]), each solvable minimal nonabelian Lie algebra is either isomorphic to the 3-dimensional nilpotent Lie algebra, or to the semidirect sum Kx A V such that ad x acts on V irreducibly (in particular, ad xV is nondegenerate). Further, ad t is a derivation of A, and subtracting from t an appropriate element of A, we may assume that either t is central, i.e. (2) is the direct sum of A and 1-dimensional algebra, or ad t is an outer derivation of A. Suppose first that A is 3-dimensional nilpotent, i.e., has a basis {x, y, z} with multipli- cation table [x, y] = z, [x, z] = [y, z] = 0. If t is central, we arrive at a particular case of (i). Straightforward computation shows that each outer derivation of A is equivalent to a derivation d which acts invariantly on the space Kx ∔ Ky, and either dKx∔Ky has trace zero, and d(z) = 0, or dKx∔Ky has trace 1, and d(z) = z. These two cases correspond to the cases (i) and (ii) respectively, with the condition of absence of nonzero eigenvectors to ensure the absence of subalgebras which are not minimal nonabelian. Suppose now that A = Kx A V , ad x acts on V irreducibly. If t is central, L ≃ Kx A (V ∔ Kt) (with ad x acting on t trivially), a case covered by (v) (with S 1-dimensional). Straightforward computation shows that each outer derivation of A is equivalent to a deriva- tion d which acts on V invariantly, and either [ad x, d] = 0 in the Lie algebra gl(V ), and d(x) = 0, or [ad x, d] = ad x and d(x) = x. These two cases correspond to the cases (v) and (iv) respectively (with S 2-dimensional), with the respective conditions to ensure the absence (cid:3) of subalgebras which are not minimal nonabelian. Corollary (to Theorems 1 and 3). A simple Lie algebra of depth 2 is either isomorphic to sl(2), or regular. Proof. It is clear that sl(2) has depth 2. Hence a simple Lie algebra L of depth 2 is either isomorphic to sl(2), or does not contain sl(2) as a proper subalgebra. In the latter case, by implication (v) ⇒ (iii) of Theorem 1, all subalgebras of L are reductive. But as each minimal nonabelian Lie algebra is either simple, or solvable, all subalgebras of L are either simple, or abelian, and by implication (iii) ⇒ (i) of Theorem 3, L is regular. (cid:3) In group theory, a notion analogous to depth in the class of finite p-groups is called An- groups, see [BJ, §65] for their discussion and for a partial description of A2-groups. 8 PASHA ZUSMANOVICH Acknowledgements This work was essentially done more than 20 years ago, while being a student at the Kazakh State University under the guidance of Askar Dzhumadil'daev, and has been reported at a few conferences in the Soviet Union at the end of 1980s. However, I find the results interesting and original enough even today to be put in writing, with some minor additions and modifications. During the final write-up I was supported by grants ERMOS7 (Estonian Science Foundation and Marie Curie Actions) and ETF9038 (Estonian Science Foundation). References [BJ] [B] [F1] [F2] [GG] [G] Y. Berkovich and Z. Janko, Groups of Prime Power Order, Vol. 2, de Gruyter, 2008. N. Bourbaki, Lie Groups and Lie Algebras. Chapters 1-3, Chapters 7-9, Springer, 1989, 2005 (trans- lated from the original French editions: Hermann, 1972, 1975 and Masson, 1982). R. Farnsteiner, On ad-semisimple Lie algebras, J. Algebra 83 (1983), 510 -- 519. , On the structure of simple-semiabelian Lie algebras, Pacific J. Math. 111 (1984), 287 -- 299. S. Garibaldi and P. Gille, Algebraic groups with few subgroups, J. London Math. Soc. 80 (2009), 405 -- 430. A.G. Gein, Minimal noncommutative and minimal nonabelian algebras, Comm. Algebra 13 (1985), 305 -- 328. [GKM] , S.V. Kuznetsov, and Yu.N. Mukhin, On minimal nonnilpotent Lie algebras, Mat. Zapiski Uralsk. Gos. Univ. 8 (1972), N 3, 18 -- 27 (in Russian). I.N. Herstein, Topics in Ring Theory, The Univ. of Chicago Press, 1969. N. Jacobson, Structure and Representations of Jordan Algebras, AMS, 1968. , Finite-Dimensional Division Algebras over Fields, Springer, 1996. R.S. Pierce, Associative Algebras, Springer, 1982. R.D. Schafer, On the algebras formed by the Cayley -- Dickson process, Amer. J. Math. 76 (1954), 435 -- 446. G.B. Seligman, Modular Lie Algebras, Springer, 1967. E.L. Stitzinger, Minimal nonnilpotent solvable Lie algebras, Proc. Amer. Math. Soc. 28 (1971), 47 -- 49. D.A. Towers, Lie algebras all whose proper subalgebras are nilpotent, Lin. Algebra Appl. 32 (1980), 61 -- 73. , Minimal non-supersolvable Lie algebras, Algebras Groups Geom. 2 (1985), 1 -- 9. [H] [J1] [J2] [P] [Sc] [Se] [St] [T1] [T2] Department of Mathematics, Tallinn University of Technology, Ehitajate tee 5, Tallinn 19086, Estonia E-mail address: [email protected]
1603.06257	1	1603	2016-03-20T19:22:16	Symmetric algebras of corepresentations and smash products	[ "math.RA" ]	We investigate Frobenius algebras and symmetric algebras in the monoidal category of right comodules over a Hopf algebra $H$; for the symmetric property $H$ is assumed to be cosovereign. If $H$ is finite dimensional and $A$ is an $H$-comodule algebra, we uncover the connection between $A$ and the smash product $A\# H^*$ with respect to the Frobenius and symmetric properties.	math.RA	math	Symmetric algebras of corepresentations and smash products S. Dascalescu1, C. Nastasescu1,2 and L. Nastasescu1,2 1 University of Bucharest, Faculty of Mathematics and Computer Science, Str. Academiei 14, Bucharest 1, RO-010014, Romania 2 Institute of Mathematics of the Romanian Academy, PO-Box 1-764 e-mail: [email protected], Constantin [email protected], RO-014700, Bucharest, Romania [email protected] Abstract We investigate Frobenius algebras and symmetric algebras in the monoidal category of right comodules over a Hopf algebra H; for the symmetric property H is assumed to be cosovereign. If H is finite dimensional and A is an H-comodule algebra, we uncover the connection between A and the smash product A#H ∗ with respect to the Frobenius and symmetric properties. Mathematics Subject Classification: 16T05, 18D10, 16S40 Key words: Hopf algebra, comodule, monoidal category, Frobenius algebra, symmetric alge- bra, smash product. 1 Introduction and preliminaries We work over a basic field k. A finite dimensional algebra A is called Frobenius if A and its dual A∗ are isomorphic as left (or equivalently, as right) A-modules. Frobenius algebras arise in representation theory, Hopf algebra theory, quantum groups, cohomology of compact oriented manifolds, topological quantum field theory, the theory of subfactors and of extensions of C ∗- algebras, the quantum Yang-Baxter equation, etc., see [13]. A rich representation theory has been uncovered for such algebras, see [14], [19]. It was showed in [1], [2] that A is Frobenius if and only if it also has a coalgebra structure whose comultiplication is a morphism of A, A- bimodules. This equivalent definition of the Frobenius property has the advantage that it makes sense in any monoidal category. The study of Frobenius algebras in monoidal categories was initiated in [15], [20], [21], and such objects have occurred in several contexts, for example in the theory of Morita equivalences of tensor categories, in conformal quantum field theory, in reconstruction theorems for modular tensor categories, see more details and references in [11], [12], [16]; recent developments can be also found in [5]. The representation theory of Frobenius algebras uncovers several symmetry features, for example there is a duality between the categories of left and right finitely generated modules, and the lattices of left and right ideals are anti-isomorphic. Among Frobenius algebras there is a class of objects having even more symmetry. These are the symmetric algebras A, for which A and A∗ are isomorphic as A, A-bimodules. The category of commutative symmetric algebras is equivalent to the category of 2-dimensional topological quantum field theories, see [1]. Symmetric algebras appear in block theory of group algebras in positive characteristic, see [19, Chapter IV]. It is not clear how one could define symmetric algebras in an arbitrary monoidal category. Symmetric algebras in monoidal categories with certain properties were first considered in [10], as 1 structures related to correlation functions in conformal field theories. In [11] symmetric algebras are discussed in sovereign monoidal categories. In this paper we consider the monoidal category MH of right comodules (or corepresentations) over a Hopf algebra H. If A is an algebra in this category, i.e. a right H-comodule algebra, then A ∈ AMH A , i.e. A is a left (A, H)-Doi-Hopf module and a right (A, H)-Doi-Hopf module. On the other hand, A∗ is a right (A, H)-Doi-Hopf module, but not necessarily a left (A, H)-Doi-Hopf module; however A∗ has a natural structure of a left (A(S2), H)-Doi-Hopf module, where A(S2) is the algebra A with the coaction shifted by S2, where S is the antipode of H. If H is cosovereign, i.e. there exists a character u on H such that S2(h) = P u−1(h1)u(h3)h2 for any h ∈ H, then A ≃ A(S2) as comodule algebras, and this induces a structure of A∗ as an object in AMH A , where the left A-action is a deformation of the usual one by u. Then it makes sense to consider when A and A∗ are isomorphic in this category; in this case we say that A is symmetric in MH with respect to u, or shortly that A is (H, u)-symmetric. As MH is a sovereign monoidal category for such H, this is a special case of the general concept of symmetric algebra considered in [11]. In Section 2 we give explicit characterizations of this property in MH. We show that the definition of symmetry depends on the character (i.e. on the associated sovereign structure of MH). Also, we use a modified version of the trivial extension construction to give examples of (H, u)-symmetric algebras of In the case where H is involutory, i.e. S2 = Id, H is cosovereign if we corepresentations. take u = ε, the counit of H, and in this case it is clear that an (H, ε)-symmetric algebra is also symmetric as a k-algebra. However, we show that in general H may be (H, u)-symmetric, without being symmetric as a k-algebra. Given a finite dimensional algebra A in the category MH , where H is a finite dimensional Hopf algebra, one can construct the smash product A#H ∗. Smash products are also called semidirect products, since the group algebra of a semidirect product of groups is just a smash product. Smash product constructions are of great relevance since they describe the algebra structure in a process of bosonization, which associates for instance a Hopf algebra to a Hopf superalgebra. It is proved in [4] that A is Frobenius if and only if so is A#H ∗. On the other hand, we show in an example that such a good connection does not hold for the symmetric property. In Section 3 we show that if A is a Frobenius algebra in MH , then A#H ∗ is a Frobenius algebra in MH ∗, but the converse does not hold. In Section 4 we uncover a good transfer of the symmetry property between A and A#H ∗, more precisely we show that A is (H, α)-symmetric if and only if A#H ∗ is (H ∗, g)-symmetric, where g and α are the distinguished grouplike (or modular) elements of H and H ∗, provided that H is cosovereign by α, and H ∗ is cosovereign by g. For basic concepts and notation on Hopf algebras we refer to [8], [18]. 2 Frobenius algebras and symmetric algebras of corepresenta- tions Let H be a Hopf algebra, and let A be a finite dimensional right H-comodule algebra, with H-coaction a 7→ P a0 ⊗ a1. Then there exists an element Pi ai ⊗ hi ⊗ a∗ i ∈ A ⊗ H ⊗ A∗ such that P a0 ⊗ a1 = Pi a∗ i (a)ai ⊗ hi for any a ∈ A; this element corresponds to the H-comodule structure map of A through the natural isomorphism A ⊗ H ⊗ A∗ ≃ Hom(A, A ⊗ H). A right H-comodule structure is induced on A∗ by a∗ 7→ X a∗(ai)a∗ i ⊗ S(hi), for any a∗ ∈ A∗. i 2 If we consider the left H ∗-actions on A and A∗ associated with these right H-comodule structures, denoted by h∗ · a and h∗ · a∗ for h∗ ∈ H ∗, a ∈ A, a∗ ∈ A∗, we have (h∗ · a∗)(a) = X h∗(S(hi))a∗(ai)a∗ i (a) i = a∗(X h∗(S(hi))a∗ i (a)ai) i = a∗((h∗S) · a) so (h∗ · a∗)(a) = a∗((h∗S) · a) (1) Moreover, A∗ ∈ MH A , with the usual right A-action; this means that the A-module structure of A∗ is right H-colinear. It is known (see [6, Theorem 2.4]) that the following are equivalent: (1) A ≃ A∗ in MH A ; (2) There exists a nondegenerate associative bilinear form B : A × A → k such that B(h∗ · a, b) = B(a, (h∗S) · b) for any a, b ∈ A, h∗ ∈ H ∗; (3) There exists a linear map λ : A → k such that λ(h∗ · a) = h∗(1)λ(a) for any a ∈ A, h∗ ∈ H ∗, and Ker λ does not contain a non-zero right ideal of A; (4) There exists a linear map λ : A → k such that λ(h∗ ·a) = h∗(1)λ(a) for any a ∈ A, h∗ ∈ H ∗, and Ker λ does not contain a non-zero subobject of A in MH A ; (5) A is a Frobenius algebra in the category MH. The connections between an isomorphism θ : A → A∗ as in (1), a bilinear map B as in (2) and a linear map λ as in (3), (4) are given by θ(a)(b) = B(a, b), λ(a) = B(1, a), B(a, b) = λ(ab). On the other hand, A∗ is also a left A-module in a natural way, but in general A∗ is not an object of AMH (with a similar compatibility condition for the A-action and H-coaction). However, A∗ is an object in A(S2)MH , where A(S2) is just the algebra A, with the H-coaction shifted by S2, i.e. a 7→ P a0 ⊗ S2(a1). Assume now that H is a cosovereign Hopf algebra in the sense of [3], i.e. there exists a character u on H (in other words, u is a grouplike element of the dual Hopf algebra H ∗, or equivalently, an algebra morphism from H to k) such that S2(h) = P u−1(h1)u(h3)h2 for any h ∈ H; this is the same with (S2)∗ being an inner algebra automorphism of H ∗ via u. Following [3], we say that u is a sovereign character of H. Then f : A → A(S2), f (a) = u−1 · a = P u−1(a1)a0, is an isomorphism of right H-comodule algebras, and it induces an isomorphism of categories F : A(S2)MH → AMH where for M ∈ A(S2)MH, F (M ) is just M , with the same H-coaction, and A-action ∗ given by a ∗ m = f (a)m, for any a ∈ A and m ∈ M . By restriction, this induces an isomorphism of categories (and we denote it by F , too) Now A∗ ∈ A(S2)MH F : A(S2)MH A , so then F (A∗) ∈ AMH A . A → AMH A Definition 2.1 Let H be a cosovereign Hopf algebra with u as a sovereign character. A finite dimensional right H-comodule algebra A is a symmetric algebra in the category MH with respect to u if F (A∗) ≃ A in the category AMH A . In this case we simply say that A is (H, u)-symmetric. Now we give equivalent characterizations of this property. The next result can be derived from [11, Proposition 4.6], using the structure of duals in a category of corepresentations. In our sketch of proof, explicit description is given for several ways to describe symmetry of algebras of corepresentations. 3 Proposition 2.2 Let A be a right H-comodule algebra, where H is a cosovereign Hopf algebra. Keeping the above notation, the following are equivalent. (1) A is (H, u)-symmetric. (2) There exists a nondegenerate bilinear form B : A × A → k such that B(b, ca) = B(bf (c), a), B(b, a) = B(f (a), b), and B(b, h∗ · a) = B((h∗S) · b, a) for any a, b, c ∈ A, h∗ ∈ H ∗. (3) There exists a linear map λ : A → k such that λ(ba) = λ(af (b)) and λ(h∗ · a) = h∗(1)λ(a) for any a, b ∈ A, h∗ ∈ H ∗, and also Ker λ does not contain a non-zero right ideal of A. (4) There exists a linear map λ : A → k such that λ(ba) = λ(af (b)) and λ(h∗ · a) = h∗(1)λ(a) for any a, b ∈ A, h∗ ∈ H ∗, and also Ker λ does not contain a non-zero subobject of A in MH A . More equivalent conditions can be added if we change right ideal with left ideal in (3), and MH A with AMH in (4). Proof We combine the proof of the equivalent characterizations of a symmetric algebra in the category of vector spaces, see [14, Theorem 16.54], and [6, Theorem 2.4], recalled above. Thus for (1) ⇔ (2), if θ : A → F (A∗) is a linear map, then let B : A × A → k be the bilinear map defined by B(a, b) = θ(b)(a). Then it is straightforward to check that θ is left A-linear if and only if and θ is right A-linear if and only if B(b, ca) = B(bf (c), a) for any a, b, c ∈ A, B(b, ac) = B(cb, a) for any a, b, c ∈ A. We see that if (2) and (3) hold, then B(b, a) = B(b, a1) = B(bf (a), 1) = B(f (a), b), thus B(b, a) = B(f (a), b) for any a, b ∈ A, (2) (3) (4) Moreover, if (2) and (4) hold, then B(b, ac) = B(f (ac), b) = B(f (a)f (c), b) = B(f (a), cb) = B(cb, a), so (3) holds. We have that θ is H-colinear if and only if B(b, h∗ · a) = B((h∗S) · b, a) for any a, b ∈ A, h∗ ∈ H ∗, and θ is bijective if and only if B is non-degenerate, thus (1) ⇔ (2) is clear. For (1) ⇔ (3), λ and B determine each other by the relations λ(a) = B(1, a) for any a ∈ A, ⊓⊔ respectively B(a, b) = λ(ba) for any a, b ∈ A. Example 2.3 1) If H = kG, the group Hopf algebra of a group G, then S2 = Id, so H is cosovereign with ε as a sovereign character. A right H-comodule algebra is just a G-graded algebra A, and A is (H, ε)-symmetric if and only if A is graded symmetric in the sense of [6, Section 5]. 2) More generally, if H is an involutory Hopf algebra, i.e. S2 = Id, then H is obviously cosovereign with ε as a sovereign character. In this case, if A is a finite dimensional algebra in MH, then A∗ ∈ AMH A , so F (A∗) is just A∗, with the usual left and right A-actions. Thus A is (H, ε)-symmetric if and only if A∗ ≃ A in AMH A . Remark 2.4 The definition of symmetry depends on the cosovereign character. Thus it is possible that a cosovereign Hopf algebra H has two sovereign characters u and v, and a right H-comodule algebra A is (H, u)-symmetric, but not (H, v)-symmetric. Indeed, let H = kC2, where C2 = {e, g} is a group of order 2 (e is the neutral element), and the characteristic of k is 6= 2. Let A be a commutative C2-graded division algebra with support C2; for example one can take A = kC2. Then H is involutory, so it is a cosovereign Hopf algebra with two possible sovereign characters ε = pe + pg and u = pe − pg, where {pe, pg} is the basis of H ∗ dual to the basis {e, g} of H. We have u2 = ε, so u−1 = u. 4 It is easy to see that A is (H, ε)-symmetric, i.e. graded symmetric in the terminology of [6], for example by using the results of [7], where the question whether any graded division algebra is graded symmetric is addressed. On the other hand, A is not (H, u)-symmetric. Indeed, let λ : A → k be a linear map such that λ(ab) = λ(b(u · a)) for any a, b ∈ A. We have u · a = a for any a ∈ Ae (the homogeneous component of degree e of A), and u · a = −a for any a ∈ Ag. Then for b = 1 and a ∈ Ag we get λ(a) = 0, thus λ(Ag) = 0. Also, for a, b ∈ Ag we obtain λ(ab) = 0, and since AgAg = Ae, this shows that λ(Ae) = 0. Thus λ must be zero. Remark 2.5 It is obvious that a graded symmetric algebra is symmetric as a k-algebra. More general, if H is involutory, then a (H, ε)-symmetric algebra is symmetric as a k-algebra. How- ever, for an arbitrary cosovereign Hopf algebra H with sovereign character u, if A is (H, u)- symmetric, then A is not necessarily symmetric as a k-algebra, as we show in the following example. Let H = H4, the 4-dimensional Sweedler's Hopf algebra. It is presented by algebra generators c and x, subject to relations c2 = 1, x2 = 0, xc = −cx The coalgebra structure is defined by ∆(c) = c ⊗ c, ∆(x) = c ⊗ x + x ⊗ 1, ε(c) = 1, ε(x) = 0. The antipode S satisfies S(c) = c, S(x) = −cx, thus S2(x) = −x. Apart from ε, H has just one more character α, given by α(c) = −1, α(x) = 0; it acts on the basis elements of H by α · 1 = 1, α · c = −c, α · x = x, α · (cx) = −cx (5) H is cosovereign, and the only sovereign character is α. We show that the linear map λ : H → k, λ(1) = λ(c) = λ(cx) = 0, λ(x) = 1 makes H an (H, α)-symmetric algebra. Indeed, we first see by a straightforward checking that λ is right H-colinear (or equivalently, left H ∗- linear). Next, an easy computation using (5) shows that any element of the form ba − a(α · b) lies in the span of 1, c and cx, thus λ(ba) = λ(a(α · b)) for any a, b ∈ H. Finally, let I be a left ideal of H contained in Kerλ. Let z = δ1 + βc + γcx ∈ I, where δ, β, γ ∈ k. Then cz = δc + β1 + γx ∈ I ⊆< 1, c, cx >, so γ = 0. Then xz = δx − βcx ∈ I ⊆< 1, c, cx >, so δ = 0. Now cxz = −βx ∈ I ⊆< 1, c, cx >, so β must be zero, too. Thus z = 0, and H is (H, α)-symmetric. This will also follow from Proposition 4.4. On the other hand, H is not symmetric as a k-algebra. Indeed, if λ : H → k is a linear map such that λ(ab) = λ(ba) for any a, b ∈ H, then λ(cx) = λ(xc) = −λ(cx), so λ(cx) = 0, and λ(x) = λ(ccx) = λ(cxc) = −λ(x), so λ(x) = 0. Thus the two-sided ideal < x, cx > of H is contained in Kerλ, showing that H is not symmetric. This can also be seen from a general result saying that a Hopf algebra is symmetric as an algebra if and only if it is unimodular (i.e. the spaces of left integrals and right integrals in H coincide) and S2 is inner, see [17]; for H4 the square of the antipode is inner, but the unimodularity condition fails. Now we explain how examples of (H, u)-symmetric algebras in the category MH can be constructed, where H is a cosovereign Hopf algebra with sovereign character u. We recall that for any algebra A (in the category of vector spaces), and any left A, right A-bimodule M , one can construct an algebra structure on the space A ⊕ M with the multiplication defined by (a, m)(a′, m′) = (aa′, am′ + ma′); this is called the trivial extension of A and M . The unit of this algebra is (1, 0). If M = A∗ with the usual A, A-bimodule structure, the trivial extension of A and A∗ is simply called the trivial extension of A, and it is a symmetric algebra, see [14, Example 16.60]. 5 Proposition 2.6 Let A be a right H-comodule algebra, where H is cosovereign with sovereign character u. Then E(A) = A ⊕ F (A∗), with the direct sum structure of a right H-comodule, and the algebra structure obtained by the trivial extension of A and the A, A-bimodule F (A∗), is a right H-comodule algebra which is (H, u)-symmetric. Proof The multiplication of E(A) is given by (a, a∗)(b, b∗) = (ab, a ∗ b∗ + a∗b) = (ab, (u−1 · a)b∗ + a∗b) for any a, b ∈ A and any a∗, b∗ ∈ A∗. We first see that E(A) is a right H-comodule algebra. Indeed, the H-coaction on E(A) is ρ : E(A) → E(A) ⊗ H, given by ρ(a, a∗) = X(a0, 0) ⊗ a1 + X(0, a∗ 0) ⊗ a∗ 1 Then ρ((a, a∗)(b, b∗)) = ρ(ab, a ∗ b∗ + a∗b) = X((ab)0, 0) ⊗ (ab)1 + X(0, (a ∗ b∗)0) ⊗ (a ∗ b∗)1 + X(0, (a∗b)0) ⊗ (a∗b)1 = X(a0b0, 0) ⊗ a1b1 + X(0, a0 ∗ b∗ 0) ⊗ a∗ = (X(a0, 0) ⊗ a1 + X(0, a∗ = ρ(a, a∗)ρ(b, b∗) 0) ⊗ a1b∗ 1)(X(b0, 0) ⊗ b1 + X(0, b∗ 0) ⊗ b∗ 1) 1 + X(0, a∗ 0b0) ⊗ a∗ 1b1 Let λ : E(A) → k be the linear map defined by λ(a, a∗) = a∗(1) for any a ∈ A, a∗ ∈ A∗. Then Now λ(h∗ · (a, a∗)) = (h∗ · a∗)(1) (by (1)) = a∗((h∗S) · 1) = a∗((h∗S)(1)1) = h∗(1)a∗(1) = h∗(1)λ(a, a∗) λ((a, a∗)(u−1 · (b, b∗))) = λ((a, a∗)(u−1 · b, u−1 · b∗)) = λ(a(u−1 · b), (u−1 · a)(u−1 · b∗) + a∗(u−1 · b)) = (u−1 · b∗)(u−1 · a) + a∗(u−1 · b) = b∗((u−1S) · (u−1 · a)) + a∗(u−1 · b) = b∗(u · (u−1 · a)) + a∗(u−1 · b) = b∗(a) + a∗(u−1 · b) = λ(ba, (u−1 · b)a∗ + b∗a) = λ((b, b∗)(a, a∗)) (by (1)) Finally, we see that Kerλ does not contain non-zero right ideals. Indeed, if λ((a, a∗)E(A)) = 0, then b∗(u−1 · a) + a∗(b) = 0 for any b ∈ A, b∗ ∈ A∗. If we take b∗ = 0, we get that a∗(b) = 0 for any b, so a∗ = 0. Then b∗(u−1 · a) = 0 for any b∗, so u−1 · a = 0, showing that a = 0. We conclude that λ makes E(A) an (H, u)-symmetric algebra. ⊓⊔ 6 We note that the previous result shows that any finite dimensional algebra in the category MH, where H is cosovereign via u, is a subalgebra of an (H, u)-symmetric algebra, and also a quotient of an (H, u)-symmetric algebra. The construction in Proposition 2.6 helps us to provide more examples of algebras that are symmetric in categories of corepresentations with respect to certain characters, but which are not symmetric as k-algebras. Example 2.7 Assume that k has characteristic 6= 2, and let H = H4 be Sweedler's Hopf algebra. Let A = k[X]/(X 2), a 2-dimensional algebra, with basis {1, X}, and relation X 2 = 0. Let c and x be the endomorphisms of the space A such that c(1) = 1, c(X) = −X, x(1) = 0, x(X) = 1. Then c2 = Id, x2 = 0 and xc = −cx. Moreover, c is an algebra automorphism of A, and it is easy to check that x(ab) = c(a)x(b) + x(a)b for any a, b ∈ A, thus A is a left H-module algebra with the actions of c and x given by the endomorphisms above, i.e. c · 1 = 1, c · X = −X, x · 1 = 0, x · X = 1. The dual space A∗ has a left H-action given by (h · a∗)(a) = a∗(S(h) · a) for any h ∈ H, a∗ ∈ A∗ and a ∈ A. If we consider the basis {p1, pX } of A∗ dual to the basis {1, X}, this action explicitly writes c · p1 = p1, c · pX = −pX, x · p1 = −pX, x · pX = 0. On the other hand, A∗ has usual left A-module structure given by 1p1 = p1, 1pX = pX, Xp1 = 0, XpX = p1, and usual right A-module structure given by p11 = p1, pX1 = pX, p1X = 0, pXX = p1. A is also a right comodule algebra over the dual Hopf algebra H ∗. Since H ∗ is cosovereign with sovereign character c (via the isomorphism H ≃ H ∗∗), the associated left A-module structure on F (A∗) is given by a ∗ a∗ = (c · a)a∗ for any a ∈ A, a∗ ∈ A∗. Thus 1 ∗ p1 = p1, 1 ∗ pX = pX, X ∗ p1 = 0, X ∗ pX = −p1 The we can consider the algebra E(A) = A ⊕ F (A∗), whose basis is {1, X, p1, pX}, and multipli- cation induced by X 2 = p2 1 = p2 X = p1pX = pXp1 = 0 X ∗ p1 = 0, X ∗ pX = −p1, p1X = 0, pXX = p1 If we denote u = X and v = pX , we can present E(A) by generators u, v, subject to relations u2 = v2 = 0, vu = −uv. The left H-module structure of E(A) is given by c · u = −u, c · v = −v, x · u = 1, x · v = 0. If we denote by {P1, Pc, Px, Pcx} the basis of H ∗ dual to the standard basis {1, c, x, cx} of H, the right H ∗-comodule structure of E(A) is given by u 7→ u ⊗ P1 + c · u ⊗ Pc + x · u ⊗ Px + (cx) · u ⊗ Pcx = u ⊗ (P1 − Pc) + 1 ⊗ (Px + Pcx) v 7→ v ⊗ (P1 − Pc) Since the Hopf algebra H is selfdual, a Hopf algebra isomorphism being given by 1 7→ P1 + Pc, c 7→ P1 − Pc, x 7→ Px − Pcx, cx 7→ −Px − Pcx, we can regard A as a right H-comodule algebra. Summarizing, E(A) is the algebra with generators u, v, relations u2 = v2 = 0, vu = −uv 7 and H-comodule structure given by u 7→ u ⊗ c − 1 ⊗ cx, v 7→ v ⊗ c By Proposition 2.6, E(A) is (H, α)-symmetric, where α = P1 − Pc is the distinguished grouplike element of H ∗. On the other hand, E(A) is not symmetric as a k-algebra. Indeed, if λ : E(A) → k is a linear map such that λ(zz′) = λ(z′z) for any z, z′ ∈ E(A), then λ(uv) = λ(vu) = −λ(uv), thus λ(uv) = 0. But the 1-dimensional space spanned by uv is a two-sided ideal of E(A), so Kerλ contains a non-zero ideal. 3 Frobenius smash products Let A be an algebra in MH, where H is a finite dimensional Hopf algebra. Then A is a left H ∗-module algebra and we can consider the smash product A#H ∗, which is an algebra with multiplication given by (a#h∗)(b#g∗) = X a(h∗ 1 · b)#h∗ 2g∗ It is known that A is a Frobenius algebra if and only if so is A#H ∗, see [4]. On the other hand, A#H ∗ is an algebra in the category MH ∗ , with the H ∗-coaction induced by the comultiplication of H ∗, i.e. a#h∗ 7→ P a#h∗ 2. The aim of this section is to discuss the connection between A being a Frobenius algebra in MH , and A#H ∗ being a Frobenius algebra in MH ∗ . 1 ⊗ h∗ We consider the usual left and right actions of H ∗ on H, h∗ ⇀ h = P h∗(h2)h1 and h ↼ h∗ = P h∗(h1)h2, and the usual left and right actions of H on H ∗, denoted by h ⇀ h∗ and h∗ ↼ h, where h ∈ H and h∗ ∈ H ∗. H also acts on A#H ∗ by h ⇀ (a#h∗) = a#(h ⇀ h∗). We recall that a right (respectively left) integral in H is an element t ∈ H such that th = ε(h)t (respectively ht = ε(h)t) for any h ∈ H, and a left integral on H is an element T ∈ H ∗ such that h∗T = h∗(1)T for any h∗ ∈ H ∗. Theorem 3.1 Let H be a finite dimensional Hopf algebra, and let A be a finite dimensional right H-comodule algebra which is a Frobenius algebra in the category MH. Then the smash product A#H ∗ is a Frobenius algebra in the category MH ∗. Proof Let λ : A → k be a linear map whose kernel does not contain non-zero right ideals of A, and such that λ(h∗ · a) = h∗(1)λ(a) for any h∗ ∈ H ∗ and a ∈ A. Let t be a non-zero right integral in H. Define a linear map λ : A#H ∗ → k such that λ(a#h∗) = λ(a)h∗(t) for any a ∈ A, h∗ ∈ H ∗ We have that λ(h · z) = ε(h)λ(z) for any h ∈ H and z ∈ A#H ∗. Indeed λ(h · (a#h∗)) = X λ(a#h∗ 2(h)h∗ 1) 2(h)λ(a)h∗ 1(t) = X h∗ = λ(a)h∗(th) = λ(a)ε(h)h∗(t) = ε(h)λ(a#h∗) We show that Ker(λ) does not contain non-zero subobjects of A#H ∗ in the category A#H ∗. Let I be a right ideal of A#H ∗ which is also a right H ∗-subcomodule (or equiva- MH ∗ lently, invariant with respect to the induced left H-action on A#H ∗) such that I ⊂ Ker(λ). 8 We know that (A#H ∗)co H ∗ Example 6.4.8]. Then J = I co H ∗ Hopf-Galois extensions shows that the map = A#1 ≃ A and A#H ∗/A is a right H ∗-Galois extension, see [8, is a right ideal of A and the Weak Structure Theorem for J ⊗A (A#H ∗) → I, m ⊗ z 7→ mz is an isomorphism in the category MH ∗ J#H ∗. Since λ(I) = 0, we see that λ(J) = 0, so J = 0. We conclude that I must be zero. A#H ∗, see [8, Theorem 6.4.4]. In particular I = (J#1)(A#H ∗) = ⊓⊔ Corollary 3.2 A finite dimensional Hopf algebra H is Frobenius in the category MH. Proof k is a right H ∗-comodule algebra in a trivial way, and it is clear that k is Frobenius in . Since H ∗∗ ≃ H as Hopf MH ∗ algebras, we see that H is Frobenius in MH. ⊓⊔ . By Theorem 3.1 we get that k#H ∗∗ is Frobenius in MH ∗∗ The following example shows that the converse of Theorem 3.1 is not true. Example 3.3 Let A = A0 ⊕ A1 be a superalgebra, i.e. a C2-graded algebra, which is Frobenius as an algebra, but not graded Frobenius (i.e. it is not a Frobenius algebra in the category MkC2 of supervector spaces). In this example we use the additive notation for the operation of C2. Examples of such A are given in [6, Section 6]; the trivial extension associated to a finite dimensional algebra is one such example. Let µ : A → k be a linear map whose kernel does not contain non-zero left ideals of A. We define λ : A#(kC2)∗ → k, λ(a#px) = µ(a) for any a ∈ A, x ∈ C2 Then λ(y ⇀ (a#px)) = λ(a#px−y) = µ(a) = ε(y)λ(a#px). On the other hand, Kerλ does not contain non-zero left ideals of A#(kC2)∗. Indeed, assume that λ((A#(kC2)∗)z) = 0, where z = a#p0 + b#p1. Since (c#p0)z = ca0#p0 + cb1#p1, we have 0 = µ(ca0) + µ(cb1) = µ(c(a0 + b1)) for any c ∈ A. Thus µ(A(a0 + b1)) = 0, showing that a0 + b1 = 0, and then a0 = b1 = 0. Similarly, since (c#p1)z = ca1#p0 + cb0#p1, we obtain a1 = b0 = 0. Thus z = 0. We conclude that λ makes A#(kC2)∗ a Frobenius algebra in the category M(kC2)∗ . 4 Symmetric smash products The following shows that the good connection between a finite dimensional right H-comodule algebra A and the smash product A#H ∗ being Frobenius does not work anymore for the sym- metric property. Example 4.1 Let C2 =< c >= {e, g} be the cyclic group of order 2, and let A be a C2-graded algebra which is symmetric and such that the homogeneous component Ae is not symmetric. For example one can take the trivial extension A = R ⊕ R∗ of a non-symmetric algebra R, with the grading Ae = R, Ag = R∗. Then A is a right kC2-comodule algebra, so we can consider the smash product A#(kC2)∗. Denote by {pe, pg} the basis of (kC2)∗ dual to the basis {e, g} of kC2. Then 1#pe is an idempotent in A#(kC2)∗ and it is easy to check that (1#pe)(A#(kC2)∗)(1#pe) = Ae#pe ≃ Ae = R Then (1#pe)(A#(kC2)∗)(1#pe) is not a symmetric algebra, so neither is A#(kC2)∗ by [14, Exercise 16.25]. 9 In this section we discuss the connection between A being a symmetric algebra in MH with with respect to respect to some character of H, and A#H ∗ being a symmetric algebra in MH ∗ some character of H ∗ (i.e. a grouplike element of H). Let H be a finite dimensional Hopf algebra. Then there exists a character α ∈ H ∗ such that th = α(h)t for any left integral t in H and any h ∈ H; α is called the distinguished grouplike element of H ∗, and it also satisfies ht′ = α−1(h)t′ for any right integral t′ in H and any h ∈ H, see [18, Section 10.5] or [8, Section 5.5]. Similarly, there exists a distinguished grouplike element g of H, such that T h∗ = h∗(g)T for any left integral T on H and any h∗ ∈ H ∗. We note that in [18], g−1 is called the distinguished grouplike element of H; we prefer the way we defined because g will play the same role for H as α does for H ∗. It is showed in [18, Theorem 10.5.4] that for any left integral t in H Applying this for H ∗ we see that for any left integral T on H one has ∆(t) = X S2(t2)g−1 ⊗ t1 ∆(T ) = X(T2S2)α−1 ⊗ T1 and then for any h ∈ H T ↼ h = X T1(h)T2 = ((T2S2)α−1)(h)T1 = X T2(S2(h1))α−1(h2)T1 = X α−1(h2)(S2(h1) ⇀ T ) Thus for any left integral T on H and any h ∈ H T ↼ h = X α−1(h2)(S2(h1) ⇀ T ) Now if t is a left integral in H, then S(t) is a right integral in H and ∆(S(t)) = X S(t2) ⊗ S(t1) = X S(t1) ⊗ S(S2(t2)g−1) = X S(t1) ⊗ gS2(S(t2)) = S(t)2 ⊗ gS2(S(t)1) ( by (6)) We conclude that for any right integral t in H ∆(t) = X t2 ⊗ gS2(t1) We also see that for a right integral t in H and h ∈ H X t1 ⊗ ht2 = X ε(h1)t1 ⊗ h2t2 = X S(h1)h2t1 ⊗ h3t2 = X S(h1)(h2t)1 ⊗ (h2t)2 = X α−1(h2)S(h1)t1 ⊗ t2 (6) (7) (8) (9) Thus we showed that X t1 ⊗ ht2 = X α−1(h2)S(h1)t1 ⊗ t2 (10) 10 Theorem 4.2 Let H be a finite dimensional Hopf algebra, and let g and α be the distinguished grouplike elements of H and H ∗. We assume that S2(h) = g−1hg = P α−1(h1)α(h3)h2 for any h ∈ H. Then a right H-comodule algebra A is (H, α)-symmetric if and only if A#H ∗ is (H ∗, g)-symmetric. Proof We note that for any h ∈ H. S−2(h) = X α(h1)α−1(h3)h2 (11) Assume that A is (H, α)-symmetric, and let λ : A → k such that λ(ba) = λ(a(α−1 · b)) = λ(aα−1(b1)b0), λ(h∗ · a) = h∗(1)λ(a) for any a, b ∈ A, h∗ ∈ H ∗, and also Ker λ does not contain a non-zero right ideal of A. Let t be a non-zero right integral in H and define λ : A#H ∗ → k, λ(a#h∗) = λ(a)h∗(t) as in the proof of Theorem 3.1. This makes A#H ∗ a Frobenius algebra in the category MH ∗. In order to see that A#H ∗ is symmetric in MH ∗ , it remains to show that λ(zz′) = λ(z′(g−1 ⇀ z)) for any z, z′ ∈ A#H ∗, where g−1 ⇀ (a#h∗) = a#(g−1 ⇀ h∗). Indeed, we see that λ((b#g∗)(g−1 ⇀ (a#h∗)) = λ((b#g∗)(a#(g−1 ⇀ h∗))) 1 · a)#g∗ 1(a1)a0)g∗ 2(g−1 ⇀ h∗)) 2(t1)h∗(t2g−1) = X λ(b(g∗ = X λ(bg∗ = X λ(ba0)g∗(a1t1)h∗(t2g−1) = X λ(a0α−1(b1)b0)g∗(a1t1)h∗(t2g−1) = X λ(a0b0)g∗(a1b1S(b2)t1)α−1(b3)h∗(t2g−1) = X λ(((S(b1)t1) ⇀ g∗) · (ab0))α−1(b2)h∗(t2g−1) = X((S(b1)t1) ⇀ g∗)(1)λ(ab0)α−1(b2)h∗(t2g−1) = X g∗(S(b1)t1)λ(ab0)α−1(b2)h∗(t2g−1) = X λ(ab0)g∗(S(b1)t2)α−1(b2)h∗(gS2(t1)g−1) = X λ(ab0)g∗(S(b1)t2)α−1(b2)h∗(t1) = X λ(ab0)g∗(t2)α−1(b3)h∗(α−1(S(b1))S(S(b2))t1) = X λ(ab0)g∗(t2)α−1(b3)h∗(α(b1)S2(b2)t1) = X λ(ab0)g∗(t2)h∗(S2(α(b1)α−1(b3)b2)t1) = X λ(ab0)g∗(t2)h∗(S2(S−2(b1))t1) = X λ(ab0)g∗(t2)h∗(b1t1) 1(b1)h∗ = X λ(ab0)g∗(t2)h∗ 2g∗)(t) 1 · b))(h∗ = X λ(a(h∗ = X λ(a(h∗ 1 · b)#h∗ 2g∗) = λ((a#h∗)(b#g∗)) by (9) by (10) by (11) 2(t1) Conversely, assume that A#H ∗ is (H ∗, g)-symmetric, and let µ : A#H ∗ → k be a linear map whose kernel does not contain non-zero right ideals of A#H ∗, and such that µ(h ⇀ z) = ε(h)µ(z) 11 and µ(zz′) = µ(z′(g−1 ⇀ z)) for any h ∈ H and any z, z′ ∈ A#H ∗. Let T be a left integral on H and define µ : A → k, µ(a) = µ(a#T ). Let a ∈ A and h∗ ∈ H ∗. We note that g ⇀ T is a right integral on H, see [8, Proposition 5.5.4]. Let z = a#(g ⇀ T ) and z′ = 1#h∗. Then zz′ − z′(g−1 ⇀ z) = (a#(g ⇀ T ))(1#h∗) − (1#h∗)(a#T ) = a#(g ⇀ T )h∗ − X(h∗ = a#h∗(1)(g ⇀ T ) − X(h∗ = h∗(1)a#(g ⇀ T ) − (h∗ · a)#T 1 · a)#h∗ 2T 1 · a)#h∗ 2(1)T Since µ(zz′) = µ(z′(g−1 ⇀ z)), we get µ(h∗ · a) = µ((h∗ · a)#T ) = h∗(1)µ(a#(g ⇀ T )) = h∗(1)ε(g)µ(a#T ) = h∗(1)µ(a) If I is a subobject of A in AMH contained in Kerµ, then µ(I#T ) = 0. But I#T is a left ideal of A#H ∗, so it must be zero. Then I must be zero, too. To show that A is symmetric in MH it only remains to check that µ(ba) = µ(a(α−1 · b)) for any a, b ∈ A. This holds true since µ(ba) = µ(ba#T ) = µ((b#ε)(a#T )) = µ((a#T )(g−1 ⇀ (b#ε))) = µ((a#T )(b#(g−1 ⇀ ε))) = µ((a#T )(b#ε)) = X µ(a(T1 · b)#T2) = X µ(aT1(b1)b0#T2) = X µ(ab0#(T ↼ b1)) = X µ(ab0#α−1(b2)(S2(b1) ⇀ T )) = X ε(S2(b1))α−1(b2)µ(ab0#T ) = X ε(b1)α−1(b2)µ(ab0#T ) = X α−1(b1)µ(ab0#T ) = µ(a(α−1 · b)#T ) = µ(a(α−1 · b)) by (8) Remark 4.3 (1) The conditions on H in Theorem 4.2 are satisfied if H is involutory and unimodular, and H ∗ is unimodular. Indeed, in this case the distinguished grouplike elements are trivial, i.e. α = ε and g = 1. For example, this happens if H = kG, where G is a finite group. Thus a finite dimensional ⊓⊔ 12 with respect to 1. G-graded algebra is graded symmetric if and only if the smash product A#(kG)∗ is symmetric in M(kG)∗ (2) In the case where the characteristic of k is 0, it is known that H is involutory if and only if H is semisimple, if and only if H is cosemisimple, see [18, Theorem 16.1.2], and in this situation H and H ∗ are always unimodular. Thus Theorem 4.2 applies to any semisimple Hopf algebra in characteristic 0. (3) If k has positive characteristic, Theorem 4.2 applies to any semisimple cosemisimple Hopf algebra H. Indeed, it is known that any such H is involutory, see [9, Theorem 3.1]. (4) A Hopf algebra satisfying the conditions of Theorem 4.2 is not necessarily involutory, and it may be not unimodular; take for example Sweedler's 4-dimensional Hopf algebra. As a consequence of Theorem 4.2 we obtain that if a Hopf algebra H is cosovereign by In fact, we can prove that H is α and H ∗ is cosovereign by g, then H is (H, α)-symmetric. (H, α)-symmetric with less assumptions. Proposition 4.4 Let H be a finite dimensional Hopf algebra which is cosovereign with sovereign element α, the distinguished grouplike element of H ∗. Then H is (H, α)-symmetric. Proof In order to use the notation we have already developed, it is more convenient to show that if H ∗ is cosovereign by g, then H ∗ is (H ∗, g)-symmetric. If t is a right integral in H, by the proof of Theorem 3.1 (when we take A = k and identify A#H ∗ with H ∗) we have that the linear map λ : H ∗ → k, λ(h∗) = h∗(t) is H-linear, and its kernel does not contain nonzero subobjects of H ∗ in MH ∗ H ∗ . On the other hand, for any h∗, g∗ ∈ H ∗ λ(g∗(g−1 ⇀ h∗)) = X g∗(t1)h∗(t2g−1) by (9) = g∗(t2)h∗(gS2(t1)g−1) = X h∗(t1)g∗(t2) = (h∗g∗)(t) = λ(h∗g∗) so λ makes H ∗ an (H ∗, g)-symmetric algebra. ⊓⊔ 5 Passing to coinvariants If A is a right H-comodule algebra which is Frobenius (respectively symmetric) as an algebra, it is a natural question to ask whether this property transfers to the subalgebra of coinvariants AcoH. It is easy to see that such a transfer does not hold. Indeed, let A be the algebra from Example 4.1, which is symmetric. A is a kC2-comodule algebra, and its subalgebra of coinvariants is just Ae, which is not even Frobenius. The following shows that a good transfer occurs if A is Frobenius in the category MH, provided H is cosemisimple. Proposition 5.1 Let H be a cosemisimple Hopf algebra. If A is a right H-comodule algebra which is Frobenius in the category MH, then AcoH is a Frobenius algebra. If moreover, H is involutory and A is (H, ε)-symmetric, then AcoH is symmetric. Proof Let i : AcoH → A be the inclusion map, and let i∗ : A∗ → (AcoH )∗ be its dual. Since i is a morphism of AcoH, AcoH- bimodules, then so is i∗. If A is Frobenius in MH, let θ : A → A∗ be A . We show that i∗θi : AcoH → (AcoH )∗ is an isomorphism an isomorphism in the category MH 13 of right AcoH-modules, i.e. AcoH is Frobenius. In fact it is enough to show that i∗θi is injective; since Im θi = (A∗)coH, this is the same with showing that i∗ (A∗)coH is injective. The left H ∗-action on A∗ induced by the right H-coaction is (h∗ · a∗)(a) = P h∗S(a1)a∗(a0), for any A ∈ A. Then a∗ ∈ (A∗)coH if and only if h∗ · a∗ = h∗(1)a∗ for any h∗ ∈ H ∗, and this means that a∗(P h∗S(a1)a0 − h∗(1)a) = 0 for any a ∈ A and any h∗ ∈ H ∗. Since S is bijective (H is cosemisimple), we get that a∗ ∈ (A∗)coH if and only if a∗ vanishes on the subspace V =< h∗(a1)a0 − h∗(1)a h∗ ∈ H ∗, a ∈ A > Since Ker i∗ (A∗)coH = {a∗ ∈ (A∗)coH a∗(AcoH) = 0 } we see that i∗ (A∗)coH is injective if and only if V + AcoH = A. But this is indeed true, since for a left integral T on H such that T (1) = 1, one has a = T ·a−(T ·a−T (1)a). Moreover, T ·a ∈ AcoH, since h∗ · (T · a) = (h∗T ) · a = h∗(1)T a for any h∗ ∈ H ∗, and obviously T · a − T (1)a ∈ V . For the second part we just have to note that i∗θi is a morphism of AcoH, AcoH-bimodules since θ is an isomorphism of A, A-bimodules; now the proof of the first part works also in this ⊓⊔ case. Acknowledgment The research was supported by the UEFISCDI Grant PN-II-ID-PCE- 2011-3-0635, contract no. 253/5.10.2011 of CNCSIS. We thank Daniel Bulacu for very useful discussions about Frobenius algebras. References [1] L. Abrams, Two-dimensional topological quantum field theories and Frobenius algebras, J. Knot Theory Ramifications 5 (1996), 569-587. [2] L. Abrams, Modules, comodules, and cotensor products over Frobenius algebras, J. Algebra 219 (1999), 201-213. [3] J. Bichon, Cosovereign Hopf algebras, J. Pure Appl. Algebra 157 (2001), 121-133. [4] J. Bergen, A note on smash products over Frobenius algebras, Comm. Algebra 21 (1993), 4021-4024. [5] D. Bulacu and B. Torecillas, On Frobenius and separable algebra extensions in monoidal categories. Applications to wreaths, to appear in J. Noncommut. Geom. [6] S. Dascalescu, C. Nastasescu and L. Nastasescu, Frobenius algebras of corepresentations and group graded vector spaces, J. Algebra 406 (2014), 226-250. [7] S. Dascalescu, C. Nastasescu and L. Nastasescu, Are graded semisimple algebras symmet- ric?, preprint, arXiv:1504.04868. [8] S. Dascalescu, C. Nastasescu and S¸. Raianu, Hopf algebras: an introduction, Pure and Applied Math. 235 (2000), Marcel Dekker. [9] P. Etingof and S. Gelaki, On finite dimensional semisimple and cosemisimple Hopf algebras in positive characteristic, Internat. Math. Res. Notices 16 (1998), 851-864. 14 [10] J. Fuchs, I. Runkel, and C. Schweigert, Conformal Correlation Functions, Frobenius Alge- bras and Triangulations, Nucl. Phys. B 624 (2002), 452-468. [11] J. Fuchs and C. Stigner, On Frobenius algebras in rigid monoidal categories, Arab. J. Sci. Eng. Sect. C Theme Issues 33 (2008), no. 2, 175-191. [12] J. Fuchs and C. Schweigert, Hopf algebras and finite tensor categories in conformal field theory, Rev. Un. Mat. Argentina 51 (2010), no. 2, 43-90. [13] L. Kadison, New examples of Frobenius extensions, University Lecture Series 14, American Mathematical Society, Providence, Rhode Island, 1999. [14] T. Y. Lam, Lectures on modules and rings, GTM 189, Springer Verlag, 1999. [15] M. Muger, From subfactors to categories and topology I. Frobenius algebras in and Morita equivalence of tensor categories, J. Pure Appl. Alg. 180 (2003), 81-157. [16] M. Muger, Tensor categories: a selective guided tour, Rev. Un. Mat. Argentina 51 (2010), no. 1, 95-163. [17] U. Oberst and H.-J. Schneider, Uber Untergruppen endlicher algebraischer Gruppen, Manuscripta Math. 8 (1973), 217-241. [18] D. E. Radford, Hopf algebras. Series on Knots and Everything, 49. World Scientific Pub- lishing Co. Pte. Ltd., Hackensack, NJ, 2012. [19] A. Skowro´nski, K. Yamagata, Frobenius algebras I. Basic representation theory, EMS Text- books in Mathematics. European Mathematical Society (EMS), Zrich, 2011. [20] R. Street, Frobenius Monads and Pseudomonoids, J. Math. Phys. 45 (2004), 3930-3948. [21] S. Yamagami, Frobenius algebras in tensor categories ans bimodule extensions, Janelidze, George (ed.) et al., Galois theory, Hopf algebras, and semiabelian categories. Papers from the workshop on categorical structures for descent and Galois theory, Hopf algebras, and semiabelian categories, Toronto, ON, Canada, September 23-28, 2002. Fields Institute Com- munications 43, 551-570 (2004). 15
1708.08312	1	1708	2017-08-28T13:51:39	Monomial bases and pre-Lie structure for free Lie algebras	[ "math.RA" ]	In this paper, we construct a pre-Lie structure on the free Lie algebra L(E) generated by a set E, giving an explicit presentation of L(E) as the quotient of the free pre-Lie algebra generated by E, by some ideal I. The main result in this paper is a description of Gr\"obner bases in terms of trees.	math.RA	math	MONOMIAL BASES AND PRE-LIE STRUCTURE FOR FREE LIE ALGEBRAS MAHDI J. HASAN AL-KAABI, DOMINIQUE MANCHON, AND FR ´ED ´ERIC PATRAS 7 1 0 2 Abstract. In this paper, we construct a pre-Lie structure on the free Lie algebra L(E) generated by a set E, giving an explicit presentation of L(E) as the quotient of the free pre-Lie algebra T E, generated by the (non-planar) E-decorated rooted trees, by some ideal I. The main result in this paper is a description of Grobner bases in terms of trees. g u A 8 2 ] . A R h t a m [ 1 v 2 1 3 8 0 . 8 0 7 1 : v i X r a Contents 1. Introduction 2. Trees 3. Lie and pre-Lie algebras 3.1. Free Lie algebras 3.2. Grobner bases 3.3. Free pre-Lie algebras 4. A pre-Lie structure on free Lie algebras 5. A monomial well-order on the planar rooted trees, and applications 6. A monomial basis for the free Lie algebra References 1 3 5 6 7 9 9 15 21 29 Math. Subject Classification: 05C05, 17D25, 17A50, 17B01. Keywords: pre-Lie algebras, monomial bases, free Lie algebras, rooted trees. 1. Introduction In the spirit of Felix Klein's (1849-1925) "Erlangen Program", any Lie group G is a group of symmetries of some class of differentiable manifolds. The corresponding infinitesimal trans- formations are given by the Lie algebra of G, which is the set of left-invariant vector fields on G. The problem of classification of groups of transformations has been considered by S. Lie (1842-1899) not only for subgroups of GLn, but also for infinite dimensional groups [15]. The problem of classification of simple finite-dimensional Lie algebras over the field of com- plex numbers was solved at the end of the 19th century by W. Killing (1847-1923) and E. Cartan (1869-1951). The central figure of the origins of the theory of the structure of Lie algebras is W. Killing, whose paper in four parts laid the conceptual foundations of the theory. In 1884, 1 2 MAHDI J. HASAN AL-KAABI, DOMINIQUE MANCHON, AND FR ´ED ´ERIC PATRAS Killing introduced the concept of Lie algebra independently of Lie and formulated the prob- lem of determining all possible structures for real, finite dimensional Lie algebras. The joint work of Killing and Cartan establishes the foundations of the theory. Killing's work contained many gaps which Cartan succeeded to fill [14], [15]. W. Killing, H. Cartan, S. Lie, and F. Engel are the main authors of the early development of the theory and some of its various applications. The concept of pre-Lie algebras appeared in many works under various names. E. B. Vinberg and M. Gerstenhaber in 1963 independently presented the concept under two different names; "right symmetric algebras" and "pre-Lie algebras" respectively [25, 13]. Other denominations, e.g. "Vinberg algebras", appeared since then. "Chronological algebras" is the term used by A. Agrachev and R. V. Gamkrelidze in their work on nonstationary vector fields [1]. The term "pre-Lie algebras" is now the standard terminology. The Lie algebra of a real connected Lie group G admits a compatible pre-Lie structure if and only if G admits a left-invariant affine structure [5, Proposition 2.31], see also the work of J. L. Koszul [17] for more details about the pre-Lie structure, in a geometrical point of view. In Sections 2, 3 of this paper, we recall some basics: trees, Lie and pre-Lie algebras, Grobner bases. We construct, in Section 4, a structure of pre-Lie algebra on the free Lie algebra L(E) generated by a set E, and we give the explicit presentation of L(E) as the quotient of the free pre-Lie algebra T E by some ideal. Recall that T E pl is the linear span of the set T E pl of all planar E-decorated rooted trees, which forms together with the left Butcher product ◦ց, and the left grafting ց respectively two mag- matic algebras. In Section 5, we give a tree version of a monomial well-order on T E pl . We adapt the work of T. Mora [20] on Grobner bases to a non-associative, magmatic context, using the descriptions of the free magmatic algebras(cid:0)T E pl , ◦ց(cid:1) and(cid:0)T E pl , ց (cid:1) respectively, following [8]. We split the basis of E- decorated planar rooted trees into two parts O(J′) and T (J′), where J′ is the ideal of T E pl generated by the pre-Lie identity and by weighted anti-symmetry relations: σσ◦ցτ + ττ◦ցσ. Here T (J′) is the set of maximal terms of elements of J′, and its complement O(J′) then defines a basis of L(E). We get one of the important results (Theorem 13), on the description of the set O(J′) in terms of trees. In Section 6, we give a non-planar tree version of the monomial well-order above. We de- scribe monomial bases for the pre-Lie (respectively free Lie) algebra L(E), using the procedures of Grobner bases and our work described in [2], in the monomial basis for the free pre-Lie al- gebra T E. MONOMIAL BASES 3 2. Trees In graph theory, a tree is a undirected connected finite graph, without cycles [10]. A rooted tree is defined as a tree with one designated vertex called the root. The other remaining vertices are partitioned into k ≥ 0 disjoint subsets such that each of them in turn represents a rooted tree, and a subtree of the whole tree. This can be taken as a recursive definition for rooted trees, widely used in computer algorithms [16]. Rooted trees stand among the most important structures appearing in many branches of pure and applied mathematics. In general, a tree structure can be described as a "branching" relationship between vertices, much like that found in the trees of nature. Many types of trees defined by all sorts of constraints on properties of vertices appear to be of interest in combinatorics and in related areas such as formal logic and computer science. A planar binary tree is a finite oriented tree embedded in the plane, such that each internal vertex has exactly two incoming edges and one outgoing edge. One of the internal vertices, called the root, is a distinguished vertex with two incoming edges and one edge, like a tail at the bottom, not ending at a vertex. The incoming edges in this type of trees are internal (connecting two internal vertices), or external (with one free end). The external incoming edges are called the leaves. We give here some examples of planar binary trees: . . . , where the single edge " " is the unique planar binary tree without internal vertices. The degree of any planar binary tree is the number of its leaves. Denote by T bin pl (respectively the linear span) of planar binary trees. (respectively T bin pl ) the set Define the grafting operation "∨" on the space T bin pl to be the operation that maps any planar binary trees t1, t2 into a new planar binary tree t1 ∨ t2, which takes the Y-shaped tree the left (respectively the right) branch by t1 (respectively t2), see the following examples: replacing ∨ = , ∨ = , ∨ = , ∨ = , ∨ = . The number of binary trees of degree n is given by the Catalan number cn = (2n)! (n+1)!n! , where the first ones are 1, 1, 2, 5, 14, 42, 132, . . .. This sequence of numbers is the sequence A000108 in [24]. Let E be a (non-empty) set. The free magma M(E) generated by E can be described as the set of planar binary trees with leaves decorated by the elements of E, together with the "∨" product described above [16, 11]. Moreover, the linear span T bin,E the magma M(E) = T bin,E defined above, equipped with the grafting "∨" is a description of the , generated by the trees of pl pl 4 MAHDI J. HASAN AL-KAABI, DOMINIQUE MANCHON, AND FR ´ED ´ERIC PATRAS free magmatic algebra. For any positive integer n, a rooted tree of degree n, or simply n-rooted tree, is a finite oriented tree together with n vertices. One of them, called the root, is a distinguished vertex without any outgoing edge. Any vertex can have arbitrarily many incoming edges, and any vertex distinct from the root has exactly one outgoing edge. Vertices with no incoming edges are called leaves. A rooted tree is said to be planar, if it is endowed with an embedding in the plane. Otherwise, its called a (non-planar) rooted tree. Let E be a (non-empty) set. An E-decorated rooted tree is a pair (t, d) of a rooted tree t together with a map d : V(t) → E, which decorates each vertex v of t by an element a of E, i.e. d(v) = a, where V(t) is the set of all vertices of t. Here are the planar (undecorated) rooted trees up to five vertices: · · · From now on, we will consider that all our trees are decorated, except for some cases in pl (respectively T E) the set of all planar pl (respectively T E) the linear space pl (respectively T E). Any rooted tree σ with branches σ1, . . . , σk which we will state the property explicitly. Denote by T E (respectively non-planar) decorated rooted trees, and T E spanned by the elements of T E and a root a, can be written as: (1) σ = B+, a(σ1 · · · σk), where B+, a is the operation which grafts a monomial σ1 · · · σk of rooted trees on a common root decorated by an element a in E, which gives a new rooted tree. The planar rooted tree σ in formula (1) depends on the order of the branches, whereas this order is not important for the corresponding (non-planar) tree. Define the (left) Butcher product ◦ց of any planar rooted trees σ and τ by: (2) σ◦ցτ := B+, a(στ1 · · · τk), where τ1, . . . , τk ∈ T E pl, such that τ = B+, a(τ1 · · · τk). It maps the pair of trees (σ, τ) into a new planar rooted tree induced by grafting the root of σ, on the left via a new edge, on the root of τ. The left grafting ց is a bilinear operation defined on the vector space T E pl , such that for any planar rooted trees σ and τ: (3) σ ց τ = Xv vertex o f τ σ ցv τ, MONOMIAL BASES 5 where σ ցv τ is the tree obtained by grafting the tree σ, on the left, on the vertex v of the tree τ, such that σ becomes the leftmost branch, starting from v, of this new tree. For example: ◦ց = , ց = + + . pl is the same than in T bin,E pl The number of trees in T E is given by D. Knuth's rotation correspondence [16] (see subsection 2.1 in [2]). On the other hand, for any homogeneous component T n of (non-planar) undecorated rooted trees of degree "n", for n ≥ 1, the number of trees in T n is given by the sequence: 1, 1, 2, 4, 9, 20, 48, . . . , which is sequence A000081 in [24]. : a one-to-one correspondence between them The two graftings, defined by (2) and (3) above, provide the space T E pl with structures of free magmatic algebras. K. Ebrahimi-Fard and D. Manchon showed, in their joint work (unpub- lished)1, that these two structures on T E pl are isomorphic. In the non-planar case, the usual product ' given by the same formula (2), is known as the Butcher product. It is non-associative permutative (NAP), i.e. it satisfies the following identity: s'(s′ 't) = s′ '(s't), for any (non-planar) trees s, s′, t. The grafting product → defined as a bilinear map on the vector space T E as follows: (4) s → t = Xv ∈V(t) s →v t, for any s, t ∈ T E, where s →v t is the (non-planar) decorated rooted tree obtained by grafting the tree s on the vertex v of the tree t. This product is pre-Lie (see Paragraph 3.3). For the case with one generator, we have: ' = , → = + . 3. Lie and pre-Lie algebras A Lie algebra over a field K is a K-vector space L , with a K-bilinear mapping [· , ·] : L×L → L (the Lie bracket), satisfying the following properties: (5) (6) for all x, y, z ∈ L. [x, y] + [y, x] = 0 (anti-symmetry) [[x, y], z] + [[y, z], x] + [[z, x], y] = 0 (Jacobi identity) 1 More details about this work in [2, subsection 2.1]. 6 MAHDI J. HASAN AL-KAABI, DOMINIQUE MANCHON, AND FR ´ED ´ERIC PATRAS A left pre-Lie algebra is a vector space A over a field K, together with a bilinear operation "⊲" that satisfies: (7) (x ⊲ y) ⊲ z − x ⊲ (y ⊲ z) = (y ⊲ x) ⊲ z − y ⊲ (x ⊲ z), ∀x, y, z ∈ A. The identity (7) is called the left pre-Lie identity, and it can be written as: (8) L[x,y] = [Lx, Ly], ∀x, y ∈ A, where for every element x in A, the linear transformation Lx of the vector space A is defined by Lx(y) = x ⊲ y, ∀y ∈ A, and [x, y] = x ⊲ y − y ⊲ x is the commutator of the elements x and y in A. The usual commutator [Lx, Ly] = LxLy − LyLx of the linear transformations of A defines a structure of Lie algebra over K on the vector space L(A) of all linear transformations of A. For any pre-Lie algebra A, the bracket [·, ·] satisfies the Jacobi identity, hence induces a structure of Lie algebra on A. 3.1. Free Lie algebras. The Lie algebra of Lie polynomials, introduced by E. Witt (1911- 1991), is actually the free Lie algebra. The first appearance of Lie polynomials was at the turn of the century in the work of Campbell, Baker and Hausdorff on the exponential mapping in a Lie group, when the well-known result "Campbell-Baker-Hausdorff formula" appeared. For more details about a historical review of free Lie algebras, we refer the reader to the reference [21] and the references therein. A free Lie algebra is a pair(cid:0)L, i(cid:1), of a Lie algebra L together with a map i : E → L from a (non-empty) set E into L, satisfying the following universal property: for any Lie algebra L′ and any mapping f : E → L′, there is a unique Lie algebra homomorphism f : L → L′ which makes the following diagram commute: E i ❅❅❅❅❅❅❅❅ f L f L′ Figure 1. The universal property of the free Lie algebra. It is unique up to an isomorphism. If L is a K- Lie algebra and E ⊆ L, then we say that E freely generates L if(cid:0)L, i(cid:1) is free, where i is the canonical injection from E to L. Recall [7, 21] that the enveloping algebra U(L) of the free Lie algebra L(E) is the free unital associative algebra on E. The Lie algebra homomorphism ϕ0 : L(E) → U(L) is injective, and ϕ0(L(E)) is the Lie subalgebra of U(L) generated by j(E), where j := ϕ0 ◦ i. / / MONOMIAL BASES 7 3.2. Grobner bases. The theory of Grobner bases was introduced in 1965 by Bruno Buch- berger for ideals in polynomial rings and an algorithm called Buchberger algorithm for their computation. This theory contributed, since the end of the Seventies, in the development of computational techniques for the symbolic solution of polynomial systems of equations and in the development of effective methods in Algebraic Geometry and Commutative Algebra. Moreover, this theory has been generalized to free non-commutative algebras and to various non-commutative algebras of interest in Differential Algebra, e.g. Weyl algebras, enveloping algebras of Lie algebras [20], and so on. The attempt to imitate Grobner basis theory for non-commutative algebras works fine up to the point where the termination of the analogue to the Buchberger algorithm can be proved. Grobner bases and Buchberger algorithm have been extended, for the first time, to ideals in free non-commutative algebras by G. Bergman in 1978. Later, F. Mora in 1986 made precise in which sense Grobner bases can be computed in free non-commutative algebras [20]. The construction of finite Grobner bases for arbitrary finitely generated ideals in non-commutative rings is possible in the class of solvable algebras 2. This class comprises many algebras arising in mathematical physics such as: Weyl algebras, enveloping algebras of finite-dimensional Lie algebras, and iterated skew polynomial rings. Grobner bases were studied, in these algebras, for special cases by Apel and Lassner in 1985, and in full generality by Kandri-Rody and Weispfen- ning in 1990 [3]. Recently, V. Drensky and R. Holtkamp used Grobner theory in their work [8] for a non- associative, non-commutative case (the magmatic case). Whereas, L. A. Bokut, Yuqun Chen and Yu Li, in their work [4], give Grobner-Shirshov basis for a right-symmetric algebra (pre-Lie algebra). The theory of Grobner-Shirshov bases was invented by A. I. Shirshov for Lie algebras in 1962 [23]. We try in this paper, precisely in Section 5, to describe a monomial basis in tree version for the free Lie (respectively pre-Lie) algebras using the procedures of Grobner bases, comparing with the one (i.e. the monomial basis) obtained for the free pre-Lie algebra in our preceding work [2]. We need here to review some basics for the theory of Grobner bases. Definition 1. Let(cid:0)M(E), ·(cid:1) be the free magma generated by E. A total order < on M(E) is said to be monomial if it satisfies the following property: (9) for any x, y, z ∈ M(E), if x < y, then x · z < y · z and z · x < z · y, i.e. it is compatible with the product in M(E). 2For more details about the solvable algebras see [3, Appendix: Non-Commutative Grobner Bases, pages 526- 528]. 8 MAHDI J. HASAN AL-KAABI, DOMINIQUE MANCHON, AND FR ´ED ´ERIC PATRAS This property, in (9), implies that for any x, y ∈ M(E) then x < x · y. An order is called a well-ordering if every strictly decreasing sequence of monomials is finite, or equivalently if every non-empty set of monomials has a minimal element. Let ME be the K-linear span of the free magma M(E), and I be any magmatic (two-sided) λx x (finite sum) in I, define T ( f ) to be the maximal term ideal of ME. For any element f =Px∈M(E) of f with respect to a given monomial order defined on M(E), namely T ( f ) = λx0 x0, with x0 = max{x ∈ M(E), λx , 0}. Denote T (I) := {T ( f ) : f ∈ I} the set of all maximal terms of elements of I. Note that the set T (I) forms a (two-sided) ideal of the magma M(E) [20]. Define the set O(I) := M(E)\T (I). We have that the magma M(E) = T (I) ∪ O(I) is the disjoint union of T (I), O(I) respectively. As a consequence, we get that: (10) ME = SpanK(T (I)) ⊕ SpanK(O(I)). Define a linear mapping ϕ from I into SpanK(T (I)), which makes the following diagram com- mute: I / SpanK(T (I)) ⊕ SpanK(O(I)) i = +❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱ ME ϕ P SpanK(T (I)) Figure 2. Definition of ϕ. where P is the projection map. Then the mapping ϕ is defined by: (11) ϕ( f ) = Xx∈T (I) αx x, for f ∈ I, αx x + corrective term in SpanK(O(I)), and αx ∈ K for all x ∈ T (I). The map ϕ is obviously injective. Indeed, for any f ∈ I and ϕ( f ) = 0, then f ∈ SpanK(O(I)), and from where f = Px∈T (I) Theorem 9, SpanK(O(I))T I = {0}. Also, according to Theorem 9 and by the definition of ϕ in (11), we note that ϕ is surjective. Hence, ϕ is an isomorphism of vector spaces. Thus, we can deduce from the formula (10): (12) ME = I ⊕ SpanK(O(I)). In Section 5, we will give a tree version of the monomial well-ordering with a review of Mora's work [20], in the case of rooted trees. / / + / MONOMIAL BASES 9 3.3. Free pre-Lie algebras. As a particular example of pre-Lie algebras, take the linear space of the set of all (non-planar) E-decorated rooted trees T E which has a structure of pre-Lie alge- bra together with the product "→" defined in (4). Free pre-Lie algebras have been handled in terms of rooted trees by F. Chapoton and M. Livernet [6], who also described the pre-Lie operad explicitly, and by A. Dzhumadil'daev and C. Lofwall independently [9]. For an elementary version of the approach by Chapoton and Livernet without introducing operads, see e.g. [19, Paragraph 6.2]: Theorem 1. Let k be a positive integer. The free pre-Lie algebra with k generators is the vector space T of (non-planar) rooted trees with k colors, endowed with grafting. 4. A pre-Lie structure on free Lie algebras Let L(E) be the free Lie algebra generated by a (non-empty) disjoint union of subsets E = Ei, where Ei is the subset of elements ai of degree i, and #Ei = di. The free Lie 1, . . . , ai di algebra L(E) can be graded, using the grading of E: Fi∈N (13) L(E) =Mi∈N Li, where Li is the subspace of all elements of L(E) of degree i. In particular Ei ⊂ Li. Define an operation ⊲ on L(E) by: (14) for x, y ∈ L(E). x ⊲ y := 1 x [x, y], Proposition 2. The operation ⊲ defined by (14) is a bilinear product which satisfies the pre-Lie identity. Proof. For x, y, z ∈ L(E), we have: (x ⊲ y) ⊲ z − x ⊲ (y ⊲ z) = 1 x [x, y] ⊲ z − 1 y x ⊲ [y, z] [x, [y, z]] + [z, [x, y]] + [y, [z, x]] = 0 (the Jacobi identity) 1 xy [x, [y, z]] 1 xy(cid:0)[[x, y], z] − [y, [z, x]](cid:1), since 1 xy [y, [z, x]] 1 1 = = x(cid:0)x + y(cid:1) [[x, y], z] − x(cid:0)x + y(cid:1) [[x, y], z] − x! y −(cid:0)x + y(cid:1) = 1 y(cid:0)x + y(cid:1) [[y, x], z] − = 1 = (y ⊲ x) ⊲ z − y ⊲ (x ⊲ z). y(cid:0)x + y(cid:1) ! [[x, y], z] + 1 [y, [x, z]] xy 10 MAHDI J. HASAN AL-KAABI, DOMINIQUE MANCHON, AND FR ´ED ´ERIC PATRAS Then L(E) together with ⊲ forms a graded pre-Lie algebra generated by E. (cid:3) A pre-Lie algebra structure can be put this way on any N-graded Lie algebra L such that L0 = {0}. Another pre-Lie bracket, proposed on L by T. Schedler [22]3, is given by: (15) x ◮ y = y x + y [x, y], for any x, y ∈ L. These two constructions are isomorphic, via the linear map: Indeed, α is a bijection, and for any x, y ∈ L we have: (cid:0)L, ◮(cid:1) −→(cid:0)L, ⊲(cid:1), x 7−→ xx. α : α(x ◮ y) = α( [x, y]), (by the definition of ◮ in (15)), y x + y y α([x, y]) = = x + y y x + y (x + y)[x, y], (by the definition of α above), = y[x, y] = xy x [x, y] = (xy)x ⊲ y, (by the definition of ⊲ in (14)), = xx ⊲ yy = α(x) ⊲ α(y). Denote by [·, ·]⊲ the underlying Lie bracket induced by the pre-Lie product ⊲, which is de- fined by: (16) [x, y]⊲ = x ⊲ y − y ⊲ x, for x, y ∈ L. Then the two Lie structures defined on L by the Lie brackets [·, ·], [·, ·]⊲ respectively, are also isomorphic via α. Indeed, by substituting the pre-Lie product ⊲, described in (14), by the Lie bracket [·, ·] in the definition of the Lie bracket [·, ·]⊲ in (16), we get: (17) [x, y]⊲ = 1 x [x, y] − 1 y [y, x] = x + y x y [x, y], for any x, y ∈ L, 3For more details about this construction of pre-Lie algebra see [22, Proposition 3.3.3] and [12]. MONOMIAL BASES 11 but, α(cid:0)[x, y](cid:1) = [x, y] [x, y] =(cid:0)x + y(cid:1)[x, y] = xy[x, y]⊲ = [xx, yy]⊲ = [α(x), α(y)]⊲ (by (17)) (by (4)). For any (non-planar) rooted tree t, we can decorate the vertices of t by elements of E, by means of a map d : V(t) → E, where V(t) is the set of vertices of t. Denote by T E the set of all (non-planar) rooted trees decorated by the elements of E, define the degree t of a decorated tree t in T E by: (18) t := Xv∈V(t) d(v) In particular, there is a unique pre-Lie homomorphism Φ from(cid:0)T E, →(cid:1) onto(cid:0)L(E), ⊲(cid:1), such that: (19) Φ(a) = a for any a ∈ E. If we take t = t1 → (t2 → (· · · → (tk → a) · · · )) ∈ T E, then: (20) Φ(t) = x1 ⊲ (x2 ⊲ (· · · ⊲ (xk ⊲ a) · · · )), with xi = Φ(ti), and ti = xi, ∀i = 1, . . . , k. Let I be the two-sided ideal of T E generated by all elements on the form: (21) s(cid:0)s → t(cid:1) + t(cid:0)t → s(cid:1), for s, t ∈ T E. The ideal I satisfies the following properties: Proposition 3. The quotient L′(E) := T E/I has structures of pre-Lie algebra and Lie algebra, respectively. Proof. Using the pre-Lie grafting → defined on T E, we can define the following operations on L′(E): (22) (23) s ⊲ ∗ t := s → t := s → t, [s, t] := [s, t] := ss → t, for any s, t ∈ T E, where the bar stands for the class modulo I. The product in (22) is pre-Lie by definition. The bracket defined in (23) is well-defined and satisfies the following identities: (1) The anti-symmetry identity: for any s, t ∈ T E, we have [s, t] = −[t, s], since, s(s → t) + t(t → s) ∈ I. 12 MAHDI J. HASAN AL-KAABI, DOMINIQUE MANCHON, AND FR ´ED ´ERIC PATRAS (2) The Jacobi identity: for any s, t, t′ ∈ T E, then [s, [t, t′]] + [[s, t′], t] = st(cid:0)s → (t → t′)(cid:1) + s(s + t′)(cid:0)(s → t′) → t)(cid:1) (using the anti-symmetry identity) −→ = st(cid:0)(s → (t → t′)) − (t → (s → t′))(cid:1) (using the pre-Lie identity) −→ = st(cid:0)(cid:0)(s → t) → t′(cid:1) −(cid:0)(t → s) → t′(cid:1)(cid:1) (using the anti-symmetry identity) −→ = st(cid:0)(s → t + s t = st(cid:0)(s → t − t → s) → t′(cid:1) s → t) → t′(cid:1) (cid:0)(s → t) → t′(cid:1) = s(s + t)(cid:0)(s → t) → t′(cid:1) s + t = st = [[s, t], t′]. t (cid:3) Proposition 4. I = Ker Φ. Let(cid:0)M(E), ·(cid:1) be the free magma generated by E, and let ME be the free magmatic algebra generated by E, i.e. the linear span of the magma M(E). Define a new magmatic product ∗ on M(E) by: (24) x ∗ y := xx · y for any x, y ∈ M(E), and extend bilinearly. We need, to prove Proposition 4, to introduce the following lemmas. Lemma 5. The two magmatic algebras(cid:0)ME, ·(cid:1) and(cid:0)ME, ∗(cid:1) are isomorphic. (cid:0)ME, ·(cid:1) →(cid:0)ME, ∗(cid:1) such that γ(a) = a, for any a ∈ E. For any x, y ∈ ME, we have: Proof. By universal property of the free magmatic algebra, there is a unique morphism γ : (25) γ(x · y) = γ(x) ∗ γ(y) = γ(x) γ(x) · γ(y). Hence one can see, by induction on the degree of elements of the magma M(E), that we have for any z ∈ M(E): (26) γ(z) = f (z) z, where f : M(E) → N is recursively given by: f (a) = 1, for any a ∈ E, and f (x · y) = x f (x) f (y) for x, y ∈ M(E) (for more details about this mapping see Example 1 below). Hence γ is an isomorphism. (cid:3) MONOMIAL BASES 13 Now, let J be the two-sided ideal generated by the the anti-symmetry and the Jacobi identities on(cid:0)ME, ∗(cid:1), and let J′ be the two-sided ideal of(cid:0)ME, ·(cid:1) generated by the pre-Lie identity and the elements on the form: (27) xx · y + yy · x, for x, y ∈ M(E). Lemma 6. J = J′. Proof. Let J′ elements x ∗ y + y ∗ x, for x, y ∈ M(E). We have: 1 be the ideal generated by the elements (27). Equivalently, J′ 1 is generated by the x · (y · z) − (x · y) · z − y · (x · z) + (y · x) · z = 1 xy x ∗ (y ∗ z) − 1 x(x + y) (x ∗ y) ∗ z y ∗ (x ∗ z) + 1 y(x + y) (y ∗ x) ∗ z − 1 xy 1 = = = xy(x + y)(cid:0)(x + y)x ∗ (y ∗ z) − y(x ∗ y) ∗ z − (x + y)y ∗ (x ∗ z) + x(y ∗ x) ∗ z(cid:1) xy(cid:0) − y ∗ (x ∗ z) + (y ∗ x) ∗ z + x ∗ (y ∗ z)(cid:1) 1 (y ∗ x + x ∗ y) ∗ z 1 − x(x + y) 1 xy(cid:0)x ∗ (y ∗ z) + y ∗ (z ∗ x) + z ∗ (x ∗ y)(cid:1) modulo J′ 1, hence x · (y · z) − (x · y) · z − y · (x · z) + (y · x) · z ∈ J. This means J′ ⊂ J. Conversely, (28) x ∗ (y ∗ z) + y ∗ (z ∗ x) + z ∗ (x ∗ y) = xy(cid:0)x · (y · z) − (x · y) · z − y · (x · z) + (y · x) · z(cid:1) modulo J′ hence the left-hand side of (28) belongs to J′, which proves the inverse inclusion. 1, (cid:3) Proof of Proposition 4. The free pre-Lie algebra generated by E is given by T E [6], [9]. Hence, such that the free pre-Lie algebra PL(E) := ME/J′ the quotient L′(E) =(cid:0)ME, ·(cid:1)/J′ = T E/I is a pre-Lie (respectively Lie) algebra. The Lie algebra L(E) = (cid:0)ME, ∗(cid:1)/J carries a pre-Lie algebra structure induced by the product defined in (14), generated by the pre-Lie identity on(cid:0)ME, ·(cid:1), is homomorphic to L(E) by Φ described in (19) and (20), as pre-lie algebras, as in the commutative diagram in Figure 3, where q, q′ therein are quotient maps. From Figure 3 and Lemmas 5, 6, we get that: 2 is the two-sided ideal 2 = T E, where J′ Ker(cid:0)Φ ◦ q′(cid:1) = J′ = J = Ker q, and then Ker Φ = q′(J′) = q′(J) = I, therefore Proposition 4 is proved. (cid:3) 14 MAHDI J. HASAN AL-KAABI, DOMINIQUE MANCHON, AND FR ´ED ´ERIC PATRAS i E "❊❊❊❊❊❊❊❊❊ j Id (cid:0)ME, ·(cid:1) (cid:0)ME, ∗(cid:1) q′ $■■■■■■■■■ $■■■■■■■■■ q T E Φ L(E) Figure 3. Note that the Lie product on L(E) is the image of ∗ by Φ ◦ q′. The pre-Lie product ⊲ is the image of "·" by Φ ◦ q′. Hence, we recover Proposition 2 this way. Example 1. The free magma M(E) can also be identified with the set of all planar binary rooted trees, with leaves decorated by the elements of E, together with the product ∨ defined in section 2. For instance, a b (29) a · b = , (a · b) · c = , a · (b · c) = with x := (a · b) · c, and y := d · e. Then: f (z) = f (x · y) = x f (x) f (y) a b c a b c a b c d e , and z = x · y =(cid:0)(a · b) · c(cid:1) · (d · e) = x y , = x(cid:0)a · b f (a · b) f (c)(cid:1)(cid:0)d f (d) f (e)(cid:1) = x(cid:0)(a + b)(cid:0)d(cid:0)a f (a) f (b) f (c) f (d) f (e)(cid:1)(cid:1)(cid:1) = a d (a + b) (a + b + c) (cid:0) since, f (a) f (b) f (c) f (d) f (e) = 1(cid:1). There is another description of f , detailed as follows: in a planar binary tree, there are two types of edges, going on the left (from bottom to top) or going on the right. Consequently, except the root, there are two types of vertices, the left ones (the incoming edge on the left) and the right ones. Let t be a planar binary tree, with leaves decorated by elements of E, then f (t) is the product over all left vertices v of the sums of the degree of the decorations of the leaves l with a path from v to l. Consequently, from Propositions 2, 3 and 4, we get the following result. Corollary 7. There is a unique pre-Lie (respectively Lie) isomorphism between L′(E) and L(E), such that Φ(a mod.J′) = a mod.J, for any a ∈ E. / / q " $ $ $ ✤ ✤ ✤ $ $ $ MONOMIAL BASES 15 5. A monomial well-order on the planar rooted trees, and applications Let E be a disjoint union E := Fn≥1 subset of all elements of E of degree n. Let us order the elements of E by: En of finite subsets En = {an 1, . . . , an dn }, where En is the (30) a1 1 < · · · < a1 d1 < a2 1 < · · · < a2 d2 < · · · < ai 1 < · · · < ai di < · · · Some particular sets E of generators can be considered: (1) E = Fn≥1 En, where #Ei = 0 or 1. A particular situation is: (a) take E = {a1, . . . , as}, with ai ∈ Ei, and ai = i, for i = 1, . . . , s. (2) E = E1, where #E1 = d1 = d, as a special case: (a) take d1 = d = 2. The set T E pl forms the free magma generated by the set {a : for a ∈ E}, under the left Butcher product ◦ց. Define a total order (cid:22) on T E pl as follows: (31) for any σ, τ ∈ T E pl, then σ (cid:22) τ if and only if (1) σ < τ, or : (2) σ = τ and b(σ) < b(τ), or: (3) σ = τ, b(σ) = b(τ) and (σ1, . . . , σk) (cid:22) (τ1, . . . , τk) lexicographically, where σ = B+,r(σ1 . . . σk), τ = B+,r′(τ1 . . . τk) , or: (4) σ = τ, b(σ) = b(τ), (σ1, . . . , σk) = (τ1, . . . , τk) and the root r of σ is strictly smaller than the root r′ of τ. where k = b(σ) is the number of branches of σ starting from the root. This order depends on an ordering of the generators, here we order them by: (32) a1 1≺ · · · ≺ a1 d1≺ · · · ≺ ai 1≺ · · · ≺ ai di≺ · · · like in (30). The first terms in T E pl, when E = {a1, a2}, are ordered by " ≺ " as follows: 1 ≺ 2 ≺ ≺ 1 2 1 1 2 ≺ 2 1 ≺ 1 1 1 1 ≺ 1 ≺ 1 1 2 1 ≺ 1 1 2 ≺ 2 1 2 2 ≺ 1 1 ≺ 1 1 ≺ 2 2 ≺ 1 ≺ 2 1 1 1 2 2 ≺ 2 2 1 ≺ 2 1 2 1 ≺ 2 2 ≺ 1 ≺ 2 2 2 2 ≺ 2 2 ≺ 1 2 2 2 ≺ · · · , where i is a shorthand notation for ai . Proposition 8. The order (cid:22) defined in (31) is a monomial well-order. pl, such that σ (cid:22) σ′. For any τ ∈ T E pl, we have: τ◦ցσ < τ◦ցσ′, if Proof. Let σ, σ′ ∈ T E If b(σ) < b(σ′), σ < σ′, and they are equal when the degrees of σ and σ′ are equal. then b(τ◦ցσ) < b(τ◦ցσ′). But, if b(σ) = b(σ′) = k, then b(τ◦ցσ) = b(τ◦ցσ′) = k + 1. Lexicographically, (τ, σ1, . . . , σk) (cid:22) (τ, σ′ k). The root 1, . . . , σ′ 1, . . . , σ′ k) when (σ1, . . . , σk) (cid:22) (σ′ 16 MAHDI J. HASAN AL-KAABI, DOMINIQUE MANCHON, AND FR ´ED ´ERIC PATRAS of τ◦ցσ is the root of σ, the same thing for τ◦ցσ′ holds. Then τ◦ցσ (cid:22) τ◦ցσ′. By the same way, one can verify that σ◦ցτ (cid:22) σ′◦ցτ. Hence, the order (cid:22) is a monomial. Obviously, this order is a well-order. (cid:3) In following, we adapt the algorithm of T. Mora [20] to find Grobner bases for the free Lie algebras in tree version. For any element f ∈ T E pl, define T ( f ) to be the maximal term of f with respect to the order (cid:22) defined in (31), and let lc( f ) be the coefficient of T ( f ) in f , for example: 1 1 1 1 + 2 if f = 1 1 + 2 2 , then T ( f ) = 1 1 1 2 , and lc( f ) = 2. Let I be any (two-sided) ideal of T E pl. Define: (33) T (I) :=(cid:8)T ( f ) ∈ T E pl : f ∈ I(cid:9) , O(I) := T E pl\T (I) to be subsets of the magma T E pl, where T (I) forms a (two-sided) ideal of T E pl. Theorem 9. If I is a (two-sided) ideal of T E pl, then: pl = I ⊕ SpanK(O(I)). (1) T E (2) T E, ∗ (3) For each f ∈ T E := T E pl pl/I is isomorphic, as a K-vector space, to SpanK(O(I)). pl there is a unique g := Can( f , I) ∈ SpanK(O(I)), such that f − g ∈ I. Moreover: (a) Can( f , I) = Can(g, I) if and only if f − g ∈ I. (b) Can( f , I) = 0 if and only if f ∈ I. The symbol Can( f , I), which satisfies the identities above, is called the canonical form of f in SpanK(O(I)). Proof. The proof is detailed in [20, Theorem 1.1] in the associative case. The procedure fol- lowed in the proof of (1) consists in the following algorithm: f0 := f, φ0 := 0, h0 := 0, i := 0 while fi , 0 do If T ( fi) /∈ T (I) then φi+1 := φi, hi+1 := hi + lc( fi)T ( fi), fi+1 := fi − lc( fi)T ( fi) else %T ( fi) ∈ T (I)% choose gi ∈ I, such that T (gi) = T ( fi), lc(gi) = 1 φi+1 := φi + lc( fi)gi, hi+1 := hi, fi+1 := fi − lc( fi)gi i := i + 1 φ := φi, h := hi. The correctness of this algorithm is based on the following observations: ∀i : φi ∈ I, hi ∈ SpanK (O(I)), fi+ φi + hi = f. Termination is guaranteed by the easy observation that if fn , 0 then T ( fn) <(cid:22) T ( fn−1) and by the fact that (cid:22) is a well-ordering. (cid:3) Let J′ be the two-sided ideal of T E pl generated by the pre-Lie identity and all elements on the MONOMIAL BASES 17 form: (34) σσ◦ցτ + ττ◦ցσ, for any (non-empty) trees σ, τ ∈ T E pl. 1 3 2 Example 2. In this example we calculate Can( f , J′), where f = ideal defined by (34), using the algorithm described in the proof of Theorem 9 above: + + + 1 3 1 1 1 2 and J′ is the f0 = 1 3 + 1 T ( f0) = 3 1 1 + 2 1 1 + 1 2 , φ0 = 0, h0 = 0 2 ∈ T (J′), choose g0 = 3 2 1 1 1 + 1 2 ∈ J′, lc(g0) = 1 1 1 + 1 2 , h1 = 0, f1 = 1 3 + 3 1 + 2 1 − 3 2 1 1 2 1 φ1 = 3 T ( f1) = 2 1 1 ∈ T (J′), choose g1 = 1 1 2(cid:0) 1 2 + 2 2 1 1 (cid:1) = 1 1 2(cid:0) 2 + 2 2(cid:1)◦ց 1 1 a1∈ J′, lc(g1) = 1 φ2 = 3 2 1 1 1 + 2 − 1 1 3 2 1 2 1 2 − 3 1 , h2 = 0, f2 = 1 3 + 3 1 + 2 1 + 3 2 2 1 T ( f2) = 1 1 2 /∈ T (J′), then: φ3 = 3 2 1 1 1 + 2 − 1 1 3 2 1 2 1 2 − 3 1 , h3 = 1 3 2 1 2 , f3 = 1 3 + 3 1 + 2 1 T ( f3) = 3 1 ∈ T (J′), choose g3 = 1 3 ( 1 3 + 3 3 1 ) ∈ J′ φ4 = 3 2 1 1 1 + 2 − 1 1 3 2 1 2 − 3 2 1 1 + 1 1 3 3 + T ( f4) = 1 3 /∈ T (J′), then: φ5 = 3 2 1 1 1 + 2 − 1 1 3 2 1 2 − 3 2 1 1 + 1 1 3 3 + 3 1 3 1 , h4 = 1 3 2 1 2 , f4 = 1 2 3 3 + , h5 = 1 3 2 1 2 + 1 2 3 3 , f5 = 2 1 2 1 T ( f5) = 2 1 ∈ T (J′), choose g5 = 1 2 ( 1 2 + 2 2 1 ) ∈ J′ φ6 = 3 2 1 1 1 + 2 − 1 1 3 2 1 2 1 2 − 3 1 + 1 1 3 3 + 3 1 + 1 1 2 2 + 2 1 , h6 = 1 3 2 1 2 + 1 2 3 3 , f6 = − 1 1 2 2 18 MAHDI J. HASAN AL-KAABI, DOMINIQUE MANCHON, AND FR ´ED ´ERIC PATRAS T ( f6) = 1 2 /∈ T (J′), then: φ7 = 3 2 1 1 1 + 2 − 1 1 3 2 1 2 − 3 2 1 1 + 1 1 3 3 + then we obtain that Can( f , J′) = 3 2 1 1 2 + 2 3 3 1 1 3 + 1 1 2 2 + 2 1 , h7 = 1 3 2 1 2 + 1 2 3 3 − 1 1 2 2 , f7 = 0, − 1 2 1 2 . One can note that choosing different g's at each step in the procedures above while changing the intermediate computations would not change the final result. Theorem 9 does not describe the contents of each of T (I) and O(I). We try here to get a pl. Let pl generated by the pre-Lie identity with respect to the magmatic description of them, using the magma of planar rooted trees T E J be the (two-sided) ideal of T E product ◦ց. By Theorem 9, we have: pl with its K-linear span T E (35) T E pl = J ⊕ SpanK(O(J)). Proposition 10. O(J) is the set of σ ∈ T E from v are displayed in nondecreasing order from left to right. pl such that for any v ∈ V(σ), the branches starting The following lemma will help us to prove Proposition 10: Lemma 11. Any tree σ in T E T (J). pl which does not verify the condition of Proposition 10 belongs to Proof. Let σ = B+,,r(σ1 · · · σk) be a tree in T E such that σi−1 ≻ σi, for some i = 1, . . . , k − 1. We find that: pl, with k branches for k ≥ 2 starting from the root, (36) f = r σ1 σi−1σi . . . . . . σk − σi−1 σi . . . σk σ1 . . . σi σ1 σi−1 . . . . . . σk σ1 σi σi−1 . . . . . . σk + r − r r is an element in J such that T ( f ) = σ. If the branches start from a vertex v different from the root, the subtree σv, obtained by taking v as a root, is a factor of the tree σ. It is easily seen that σ is the leading term of the element f ∈ J obtained by replacing the factor σv by the corresponding factor given by (36). (cid:3) As a consequence of Lemma 11, we get the following natural result. Proof of Proposition 10. Using the graduation of T E Corollary 12. O(J) is contained in the set(cid:8)σ ∈ T E there is a one-to-one bijection between the subset(cid:8)σ ∈ T E and the the homogeneous component T E n, i.e.: pl : σ has non decreasing branches(cid:9). pl, with respect to the degree of trees therein, pl : σ has non decreasing branches(cid:9)n n of all E-decorated (non-planar) rooted trees of degree #(cid:8)σ ∈ T E pl : σ has non decreasing branches(cid:9)n = #T E n , for all n ≥ 1. But, O(J)n =T E n , for all n ≥ 1, have the same cardinality, hence coincide according to Corol- MONOMIAL BASES 19 lary 12: (37) O(J) =(cid:8)σ ∈ T E pl : σ has non decreasing branches(cid:9). This proves the Proposition 10. (cid:3) In the next Theorem, we describe the set O(J′) for the ideal J′ defined above by (34). Theorem 13. The set O(J′) is a set of ladders, or equivalently, the magmatic ideal T (J′) con- tains all the trees which are not ladders. Proof. We use here the induction on the number n of vertices. Let σ be a tree in T E pl, which is not a ladder, with k branches (starting from the root) and n vertices. Since σ is not a ladder, then n must be greater than or equal to 3. If n = 3, and k = 1 then σ is a ladder. Hence, for k = 2, we have that: y x y is an element of T (J′), since there is f = x r σ = for any x, y, r ∈ E. Also, for any τ ∈ T E is an ideal). x in J′, such that T ( f ) = σ, pl, the elements σ◦ցτ and τ◦ցσ are in T (J′) (since T (J′) x r y +(cid:0)y + r(cid:1) r Suppose that any (no-ladder) tree in T E pl with q vertices, where q < n, is an element in T (J′), r, where σ is not a ladder. Then σ ∈ T (J′) by the pl with n vertices and k branches, which is not a ladder, then: leteσ ∈ T E (1) If k = 1, the tree eσ is written σ◦ց induction hypothesis, henceeσ ∈ T (J′) because T (J′) is an ideal. (2) The case k = 2. This corresponds to the caseeσ = σ◦ցlm , where lm is a ladder in T E with m vertices for m ≥ 2. If σ is an element of T (J′) then so iseσ. If not, σ is a ladder (3) The case k ≥ 3. These are treeseσ =(cid:0)σ◦ցτ(cid:1) where τ ∈ T E ladder. We have theneσ ∈ T (J′) by induction hypothesis. Let us discuss the case (2) when σ is a ladder and the ladder lm does not belong to T (J′). Let by the induction hypothesis. See the discussion below. pl, with k − 1 branches, is not a pl, l1, l2 be ladders in T E pl with n1, n2 vertices respectively, where n1, n2 < n, and let: (38) l1 r l2 = l1 ◦ց(l2 ◦ց r), σ′ = l2 r l1 = l2 ◦ց(l1 ◦ց r). By the pre-Lie identity, with respect to the left Butcher product ◦ց, we find the following ele- ment: (39) l1 f0 = l1 l2 + l2 l1 l2 − l2 − l1 r r r r in J′, such thateσ, σ′ are bigger trees, with respect to the order (cid:22) defined in (31), than the two other trees in f0. Let li = pi, where pi > 0, for i = 1, 2. We have the following cases for pi: eσ = 20 MAHDI J. HASAN AL-KAABI, DOMINIQUE MANCHON, AND FR ´ED ´ERIC PATRAS (1) Either p1 = p2, then in this case we take the elements: l1 l2 g = p2 in J′, where l2 = l(1) 2 l1 + (p1 + r) l1 r l2 , f1 = l1 r r1 (1) l 2 − r1. Then we get the element: r ◦ց r (1) l 2 + r1 f = p2 f0 + g − (p1 + r) f1 ∈ J′, (40) (41) (1) l 2 l1 r r1 − l1 r , (1) l 2 r1 r l1 l1 ≺ (1) l 2 l2 , for the order (cid:22) . such that T ( f ) =eσ, since: (2) Or, p2 < p1, theneσ = T ( f0), where f0 is the element described in (39), henceeσ ∈ T (J′). (3) Or, p1 < p2, here we have thateσ ≺ σ′ and the element f0 described in (39) is an element in J′ such that T ( f0) = σ′, hence σ′ ∈ T (J′). Now, foreσ we can get an element in J′ such that eσ becomes the leading term of this element, as follows: we replace the tree r) in f0 by the tree: σ′ = l2 ◦ց(l1 ◦ց r1 r l1 (42) σ′′ :=(cid:0)l1 ◦ց r(cid:1)◦ցl2 = r l2 , using the element g described in (40). This new tree σ′′ is also greater than eσ with respect to the order (cid:22). By the pre-Lie identity, we can get the element f described in (41) such that: l1 (1) l 2 eσ and σ′ 2 r1 ◦ց r(cid:1)◦ց ◦ց(cid:0)(cid:0)l1 1 := l(1) We verify whether p1 = l1 > l(1) r1(cid:1) = eσ ∈ T (J′). If not, we replace σ′ 1 :=(cid:0)(cid:0)l1 σ′′ If n2 = 1, the tree σ′′ ◦ց (43) l1 r r1 (1) l 2 . = r(cid:1)◦ց r1(cid:1)◦ցl(1) 2 r are the two biggest trees appearing in this element. 2 = p2 − r1, i.e. σ′ 1 in f by the tree: 1 (cid:22) eσ, or not. If so, then 1 is a ladder. If n2 ≥ 2, then σ′′ 1 is not a ladder and is greater than 1 in (43), and eσ. Then we need to apply the pre-Lie identity once again to the tree σ′′ replace it by: l1 σ′ 2 := l(2) 2 , where l2 = (· · · ((l(i) ◦ց(cid:0)(cid:0)(cid:0)l1 r(cid:1)◦ց r1(cid:1)◦ց ◦ց ◦ց 2 2 r2(cid:1) = ri )◦ց Let p(i) 2 = l(i) r r1 , where l(2) 2 ◦ց r2 = l(1) 2 . (2) l 2 r2 ri−1) · · · )◦ց r1, for i ≥ 1. After a finite number s of steps applying the pre-Lie identity in the expression: MONOMIAL BASES 21 σ′′ s :=(cid:0)(cid:0) · · ·(cid:0)(l1 ◦ց r)◦ց r1(cid:1) · · ·(cid:1)◦ց rs−1(cid:1)◦ց(cid:0)l(s) 2 ◦ց l1 r rs−1 rs(cid:1) = l1 r rs−1 (s) l 2 , where σ′ s = (s) l 2 rs rs which can be formulated as: l1 r rs−1 fs = (s) l 2 − rs (s) l 2 l1 r rs−1 (s) l 2 + rs l1 l1 r (s) l 2 rs−1 − rs r rs−1 ∈ J′, rs (44) (45) we can find an element f ∈ J′, such that eσ and σ′ p(s) 2 < p1, i.e. σ′ proved. s ≺ eσ. Hence, eσ described in (38) is in T (J′). Then, Theorem 13 is s become bigger trees of f with 6. A monomial basis for the free Lie algebra The set T E forms the free Non-Associative Permutive (NAP) magma generated by the set {a : for a ∈ E}, under the usual Butcher product '. Corresponding to the total order defined in (31), we can define a non-planar version (cid:22) of this order, as follows: (cid:3) (46) for any s, t ∈ T E, then s (cid:22) t if and only if (1) s < t, or : (2) s = t and b(s) < b(t), or: (3) s = t, b(s) = b(t) = k and s = B+,r(s1 . . . sk), t = B+,r′(t1 . . . tk) such that ∃ j ≤ k, with si = ti, for i < j, s j (cid:22) t j where s1 (cid:22) · · · (cid:22) sk, t1 (cid:22) · · · (cid:22) tk are the branches of s, t respectively, or: (4) s = t, b(s) = b(t) = k, sl = tl, for all l = 1, . . . , k and r ≤ r′, where r (respectively r′ ) is the root of s (respectively t). By the same way as in Proposition 8, we observe that the order (cid:22) defined in (46) is a mono- mial well order. The space T E forms with the Butcher product the free NAP algebra generated by E [18]. The first author introduced in [2]4 a section S from the NAP algebra (T E, ') into the magmatic algebra (T E pl, ◦ց): (T E pl, ◦ց) π S / (T E, '). Here, we choose S (t) = S min(t) := Min(cid:22)(cid:8)τ ∈ T E means that we choose the minimal element τ in T E pl : π(τ) = t(cid:9), for any t ∈ T E, where Min(cid:22){−} pl with respect to the order "(cid:22)" with π(τ) = t. Proposition 14. The section map S min defined above is an increasing map. 4See subsection 2.2 in [2]. / / / o o 22 MAHDI J. HASAN AL-KAABI, DOMINIQUE MANCHON, AND FR ´ED ´ERIC PATRAS Proof. Take two trees s and t in T E with s (cid:22) t. The section S min, obviously, respects the degree and the number of branches of the trees. Hence, we can suppose s = t and b(s) = b(t) = l. We have then: (47) s = B+,r(s1, . . . , sl), t = B+,r′(t1, . . . , tl), with s1 (cid:22) · · · (cid:22) sl , t1 (cid:22) · · · (cid:22) tl. Condition (3) of the definition, in (46), of the order (cid:22) exactly means that the l-tuple of branches of S min(s) is lexicographically smaller than the l-tuple of branches of S min(t). If s and t have the same branches and s (cid:22) t, we also have S min(s) (cid:22) S min(t), as one can see by comparing the roots. This proves Proposition 14. (cid:3) Proposition 15. The section map S min on T E is a bijection onto O(J), where J is the (two-sided) ideal generated by the pre-Lie identity in(cid:0)T E pl, ◦ց(cid:1). Proof. Clear from Proposition 10. (cid:3) Define a relation R on T E as follows: (48) sRs′ if and only if there are t, t′ ∈ T E and v, w ∈ V(t′) such that s = t →v t′, s′ = t →w t′ for s, s′ ∈ T E, and w is related with v by an edge with w above v. Let # be the transitive closure of the relation R defined in (48), i.e. for s, s′ ∈ T E, we say that s#s′ if and only if there is s1, . . . , sl ∈ T E such that sRs1R . . . RslRs′. v w Lemma 16. Let s, s′, t ∈ T E, if s′ (cid:22) s then s′ →v t (cid:22) s →v t, for v ∈ V(t). Proof. Immediate from the definition (46) of the order (cid:22). (cid:3) Lemma 17. Let s, s′ ∈ T E, if s#s′ then s′ ≺ s. Proof. For s, s′ ∈ T E, if sRs′, then by definition of the relation R in (48), there are t, t′ ∈ T E and v, w ∈ V(t) such that s = t →v t′, s′ = t →w t′, and an edge in t′. Obviously, the tree obtained by grafting t on the tree t′ at v is greater, with respect to the order (cid:22), than the tree deduced by grafting t on t′ at w, i.e s′ ≺ s. The passage from R to # is obvious. (cid:3) w v Proposition 18. The Butcher product ' is compatible with the relation R, i.e. for s, s′, t ∈ T E, if sRs′ then (s't)R(s′ 't′), for t′ ∈ T E. 't) and (t's)R(t's′). Also, if sRs′ and tRt′ then (s't)#(s′ Proof. For any s, s′, t, t′ ∈ T E, if sRs′ and tRt′, then by definition of R we have: MONOMIAL BASES 23 s = s1 →v s2, s′ = s1 →w s2, for v, w ∈ V(s2), with w v in s2, and t = t1 →u t2, t′ = t1 →w′ t2, for u, w′ ∈ V(t2), with w′ u in t2. Let: s't = (s1 →v s2)'(t1 →u t2) = s1 →v s′′, for v ∈ V(s′′), where s′′ = s2't, and s′ 't = (s1 →w s2)'(t1 →u t2) = s1 →w s′′, for w v in s′′, then: (49) s't = (s1 →v s′′)R(s1 →w s′′) = s′ 't. Also, for s′ V(s′′′), and s′ 't′ = (s1 →w s2)'(t1 →w′ t2) = t1 →w′ s′′′, where s′′′ = s′ 't2, with w′ ∈ V(t2) ⊂ 't = (s1 →w s2)'(t1 →u t2) = t1 →u s′′′, for u ∈ V(s′′′). Then we have: (50) s′ 'tRs′ 't′. One can verify that s'tRs't′ by following the same steps as above. So, from (49) and (50), we obtain that s't#s′ 't′. (cid:3) For any t ∈ T E, define a class of t with respect to # by: (51) [t]# := {s ∈ T E : t#s}. This class has the following properties: (1) t is maximal among the representative elements in the class [t]#, i.e. for any s ∈ [t]# then s (cid:22) t. This property is deduced from Lemma 16. (2) For any s ∈ [t]#, then [s]# ⊂ [t]#. Lemma 19. For any t ∈ T E, then: (52) eΨS min(t) = Xs∈[t]# Ψ (T E pl, ◦ց) βS min(s, t)s, (T E pl, ց) Smin π π (T E, ') eΨS min (T E, →) Figure 4. where the mapeΨS min and the coefficients βS min(s, t) are described in [2, Corollary 5]. / / / / O O 24 MAHDI J. HASAN AL-KAABI, DOMINIQUE MANCHON, AND FR ´ED ´ERIC PATRAS Proof. We prove this Lemma by the induction on the number of vertices of the tree. Suppose that (52) is realized for any tree in T E with a number of vertices less than or equal to n. Take t ∈ T E be a tree, such that #V(t) = n + 1 and t = t1't2, where t1 is the minimal branch of t with respect to the order (cid:22). Then we have: eΨS min(t) =eΨS min(t1't2) = Ψ ◦ S min(t1't2) = Ψ(cid:0)S min(t1)◦ցS min(t2)(cid:1) = π(cid:0)Ψ ◦ S min(t1) ց Ψ ◦ S min(t2)(cid:1) =eΨS min(t1) → eΨS min(t2) = Xs′∈[t1]# βS min(s′, t1)s′ → Xs′′∈[t2]# = Xs′ ∈[t1]# s′′ ∈[t2]# βS min(s′, t1) βS min(s′′, t2) s′ → s′′ . βS min(s′′, t2)s′′ From Proposition 18, we have that: (53) t = t1't2 # s := s′ 's′′ # s′ →v s′′, for v ∈ V(s′′). Let sv be the smallest branch of the tree s, defined above in (53), starting from v, and sv be the corresponding trunk (what remains when the branch sv is removed). Then we have: (54) βS min(s, t) = Xv∈V(s) βS min(sv, t1) βS min(sv, t2). The formula (54) above is induced by the formula 2.15 and the definition of the coefficients βS min(s, t) described in [2]. Hence, we get: eΨS min(t) = Xs∈[t]# βS min(s, t)s. (cid:3) Corollary 20. Let t ∈ T E, then the maximal term T(cid:0)eΨS min(t)(cid:1), with respect to the order defined in (46), ofeΨS min(t) is the tree t itself. From [2], we have that the set B = neΨS min(t) : t ∈ T Eo forms a monomial basis for the free pre-Lie algebra(cid:0)T E, →(cid:1). Let I be the (two-sided) ideal generated by the elements on the form described in (21), then we have the following commutative diagram: MONOMIAL BASES 25 q (cid:0)L′(E), ⊲ ∗(cid:1) e= qqqqq (cid:0)T E, →(cid:1) (cid:0)L(E), ⊲(cid:1) Φ xqqqqq Figure 5. where L′(E) = T E/I, and the product ⊲∗ is the pre-Lie product defined in (22). L(E) is the free Lie algebra generated by E which carries the pre-Lie algebra structure by the product ⊲ defined in (14). The restriction of Φ to SpanK(O(I)) is an injective map. Indeed, for any h1, h2, ∈ SpanK(O(I)), Φ(h1) = Φ(h2) ⇒ Φ(h1 − h2) = 0 ⇒ (h1 − h2) ∈ Ker Φ = I ⇒ (h1 − h2) ∈ SpanK(O(I)) ∩ I =(cid:8)0(cid:9) ⇒ h1 − h2 = 0 ⇒ h1 = h2. Also, since Φ : T E −→ L(E) is a surjective map, then we have: L(E) = Φ(cid:0)T E(cid:1) = Φ(cid:0)I ⊕ SpanK(O(I))(cid:1) = Φ(cid:0)SpanK(O(I))(cid:1) ( by Theorem 9 ) , since Ker Φ = I and Φ(cid:0)I(cid:1) =(cid:8)0(cid:9). Hence, Φ : SpanK(O(I)) −→ L(E) is a surjective and an injective map. Then it is an isomor- phism of vector spaces. Theorem 21. For any t ∈ O(I), we have: (55) eΨS min(t) = Can(cid:0)eΨS min(t), I(cid:1) = t. Moreover, the set eB := {Φ(t) : t ∈ O(I)} is a monomial basis for the pre-Lie algebra(cid:0)L(E), ⊲(cid:1). Proof. The property (55) is induced from Theorem 13 and the definition ofeΨS min. We obviously is an isomorphism of vector spaces, eB := Φ(B′) forms a basis for the pre-Lie algebra(cid:0)L(E), ⊲(cid:1). have that the set B′ = O(I) is a basis for SpanK(O(I)). Therefore, as Φ : SpanK(O(I)) −→ L(E) This basis is monomial thanks to (55), such that: this proves Theorem 21. (cid:3) Φ(t) = Φ(cid:0)eΨS min(t)(cid:1), for all t ∈ O(I), / / / / x 26 MAHDI J. HASAN AL-KAABI, DOMINIQUE MANCHON, AND FR ´ED ´ERIC PATRAS Consequently, we get the following immediate result. Corollary 22. The set eB := {Φ(t) : t ∈ O(I)} is a monomial basis for the free Lie algebra (cid:0)L(E), [·, ·](cid:1). Examples 23. Here,we calculate few first bases eBn for homogeneous components Ln of the free Lie algebra L(E) up to n = 4, using Corollary 22, as follows: (1) As a particular case, take E = {ai : i ∈ N}, such that ai = i, for all i ∈ N, with total order a1 < a2 < · · · < as < · · · on the generators. From [2], we have: B(cid:0)T E B(cid:0)T E B(cid:0)T E B(cid:0)T E 1(cid:1) = { 2(cid:1) = { 3(cid:1) = { 4(cid:1) = { a1 : a1 ∈ E} . a1 a1 a2 : a2 ∈ E} ⊔(cid:26) a1 a3 : a3 ∈ E} ⊔(cid:26) a2 a4 : a4 ∈ E} ⊔(cid:26) a3 a1 a2 a1 , , : a1 ∈ E(cid:27) . a2 a1 a3 , a1 a1 + a1 , a1 a1 , a1 a1  a2  a1 a1 , a1 : a1, a2 ∈ E(cid:27) ⊔ a1 a2 : a1, a2, a3 ∈ E(cid:27) ⊔ a1 a2 a1 a1 a1 a1 a1 + a1 a1 : a1 ∈ E . a2 , a1 a1 , a1 a1 a1 a2 a2 a1 a1 , a1 a1 a1 a2 a2 + a1 a1 a1 a1 a1 + a1 a1 a1 a1 , a1 + a1 + a1 a1 , : a1, a2 ∈ E a1 ⊔ a1 a1 a1 a1 a1 a1 + 3 a1 + a1 + a1 a1 a1 a1 a1 a1 a1 a1 a1 a1 a1 a1 : a1 ∈ E .  Then, we get the following monomial bases eBn for Ln , up to n = 4: eB1 = {a1} . eB2 = {a2} . eB3 =(cid:8)a3, [a1, a2](cid:9) . eB4 = {a4, [a1, a3],(cid:2)[a1, a2], a1(cid:3)(cid:9) . MONOMIAL BASES 27 (2) Let us take E = {x, y} ordered by x < y, such that x = y = 1. Denote by the vertex the vertex decorated by y, such that decorated by x, and defined in (46), we arrange the first terms of T E as follows: < . Using the order 1 < < < < < < < < < < < < < < < < < < < < < · · · . Also, we calculate here the monomial bases for the homogeneous components T E n = 4: n up to B(cid:0)T E 1(cid:1) =n , o . B(cid:0)T E B(cid:0)T E 2(cid:1) =n 3(cid:1) =n , , , , , , o. , , , , , + , + , + , + , + , + o. B(cid:0)T E 4 (cid:1) =n , , , , . . . , {z } 16 terms + , + , . . . , + , {z } 12 terms + , + + , + + , . . . , + + , {z } 16 terms + + , + 3 + + , . . . , {z } 8 terms + + + 3 o. 28 MAHDI J. HASAN AL-KAABI, DOMINIQUE MANCHON, AND FR ´ED ´ERIC PATRAS Hence, we have: eB1 = E . eB2 =n[x, y] : x, y ∈ Eo . eB3 =n(cid:2)[x, y], x(cid:3) ,(cid:2)[x, y], y(cid:3) : x, y ∈ Eo . eB4 =n(cid:2)[[x, y], x], x(cid:3) , (cid:2)[[x, y], x], y(cid:3) , (cid:2)[[x, y], y], y(cid:3) : x, y ∈ Eo . Remark 24. In the monomial basis eB4 for L4, calculated in (2) above, we observe the follow- ing: the tree is not in O(I), since there is an element f = − that belongs to I such that T ( f ) = ∈ T (I). Indeed, from the pre-Lie identity, and the so-called weighted anti-symmetry identity described in (21), we have, drawing non-planar trees explicitly: f1 = Ψ(cid:16)(cid:0) ◦ց (cid:1)◦ց − ◦ց −(cid:0) ◦ց (cid:1)◦ց + ◦ց(cid:0) ◦ց (cid:1)(cid:17) = Ψ(cid:16) − − + (cid:17), f2 = Ψ(cid:16)(cid:0) ◦ց + 2 ◦ց (cid:1)◦ց (cid:17) = Ψ(cid:16) + 2 (cid:17), and f3 = Ψ(cid:16) ◦ց + 3 ◦ց (cid:17) = Ψ(cid:16) + 3 (cid:17) are elements in I, hence f4 = f1 + f2 − f3 = Ψ(cid:16)3 − 3 − (cid:17) ∈ I. But, f5 = Ψ(cid:16) (cid:17) ∈ I, hence f = f4 + f5 = 3 − 3 ∈ I. Then, we have: Φ( f ) = 3(cid:0)(x ⊲ y) ⊲ x(cid:1) ⊲ y − 3(cid:0)(x ⊲ y) ⊲ y(cid:1) ⊲ x − (x ⊲ y) ⊲ (y ⊲ x) =(cid:2)[[x, y], x], y(cid:3) −(cid:2)[[x, y], y], x(cid:3) +(cid:2)[[x, y], [x, y](cid:3) = 0 , MONOMIAL BASES 29 and then, (cid:2)[[x, y], x], y(cid:3) =(cid:2)[[x, y], y], x(cid:3). Acknowledgments. The authors thank K. Ebrahimi-Fard and L. Foissy for valuable discussions and comments. References [1] A. Agrachev, R. Gamkrelidze, Chronological algebras andnonstationary vector fields, J. Sov. Math. 17 Nol, 1650-1675 (1981). [2] Mahdi J. Hasan Al-Kaabi, Monomial Bases for Free Pre-Lie Algebras, S´eminaire Lotharingien de Combinatoire 71 (2014), Article B71b. [3] T. Becker, V. Weispfenning, H. Kredel, Grobner Bases, A Computational Approach to Com- mutative Algebra, Springer-Verlag New York, Inc (1993). [4] L. A. Bokut, Yuqun Chen, Yu Li, Grobner-Shirshov bases for Vinberg-Koszul-Gerstenhaber right-symmetric algebras, Fundamental and Applied Math. 14(8), 55-67 (2008). [5] D. Burde, Leftsymmetric algebras, orpre-Lie algebras ingeometry andphysics, Cent. Eur. J. Math. 4(3), 323-357 (2006). [6] F. Chapoton, M. Livernet, Pre-Lie algebras and the rooted trees operad, Internat. Math. Res. Notice 8, 395-408 (2001), arXiv: math/0002069. [7] J. Dixmier, Alg`ebres Enveloppantes, Paris: Gauthier-Villars, France (1974). [8] V. Drensky and R. Holtkamp, Planar trees, free non-associative algebras, invariants, and el- liptic integrals , Algebra Discrete Math., No. 2, 1-41 (2008). [9] A. Dzhumadil'daev, C. L ofwall, Trees, free right-symmetric algebras, free Novikov algebras and identities, Homology, Homotopy and Appl. 4(2), 165-190 (2002). [10] K. Ebrahimi-Fard, D. Manchon, On an extension of Knuth's rotation correspondence to re- duced planar trees , Journal of Non-commutative Geometry Volume 8, Issue 2, pp 303-320 (2014). [11] Ph. Flajolet, T. Sedgewick, Analytic Combinatorics, Cambridge Univ. Press (2009). [12] L. Foissy, Alg`ebres pr´e-Lie et alg`ebres de Lie tordues, Expos´e au 14 mars 2011, Groupe de Travail Interuniversitaire en Alg`ebre, Paris. [13] M. Gerstenhaber, The cohomology structure of an associative ring, Ann. Math. 78, No. 2, 267-288 (1963). [14] T. Hawkins, Wilhelm Killing and the structure of Lie algebras, Springer, Archive for History of Exact Sciences 5. VII., Volume 26, Issue 2, pp 127-192 (1982). [15] V. G. Kac, Infinite dimensional Lie algebras, third edition, Cambridge University Press (1990). [16] D. E. Knuth, The art of computer programming I. Fundamental algorithms, Addison-Wesley (1968). [17] J. L. Koszul, Domaines born´es homog`enes et orbites de groupes de transformations affines, Bull. Soc. Math. France 89, No. 4, 515-533 (1961). [18] M. Livernet, Arigidity theorem for pre-Lie algebras, J. Pure Appl. Alg. 207 (1), 1-18 (2006), arXiv: hep-th/0010059. 30 MAHDI J. HASAN AL-KAABI, DOMINIQUE MANCHON, AND FR ´ED ´ERIC PATRAS [19] D. Manchon, Algebraic Background for Numerical Methods, Control Theory and Renormal- ization, Proc. Combinatorics and Control, Benasque, Spain, 2010, arXiv:1501.07205. [20] T. Mora, An introduction to commutative and non-commutative Grobner bases, Theoretical Computer Science 134, 131-173 (1994). [21] Ch. Reutenauer, Free Lie algebras, Oxford University Press, New York(US) (1993). [22] T. Schedler, Connes-Kreimer quantizations and PBW theorems for pre-Lie algebras,SMF S´eminaires et Congr`es 26, p 223-251 (2013). [23] A. I. Shirshov, Some Algorithmic Problems for Lie Algebras, Siberian Math. J. 3, 292-296 (1962). [24] N. J. A. Sloane, The On-Line Encyclopedia of Integer Sequences, oeis.org. [25] E. B. Vinberg, The Theory of homogeneous convex cones, Transl. Moscow Math. Soc. 12, 340-403 (1963). Mathematics Department, College of Science, Al-Mustansiriya University, Palestine Street, P.O.Box 14022, Baghdad, IRAQ. [email protected] LMBP, CNRS-UMR6620, Universit´e Clermont-Auvergne, 3 place Vasar´ely, CS 60026, F63178 Aubi`ere, CEDEX, France. [email protected] Laboratoire J.A.Dieudonn´e, UMR CNRS-UNS N7351 Universit´e de Nice Sophia-Antipolis, Parc Valrose, 06108 NICE Cedex 2. [email protected]
1203.0765	1	1203	2012-03-04T19:30:20	Atomistic subsemirings of the lattice of subspaces of an algebra	[ "math.RA" ]	Let A be an associative algebra with identity over a field k. An atomistic subsemiring R of the lattice of subspaces of A, endowed with the natural product, is a subsemiring which is a closed atomistic sublattice. When R has no zero divisors, the set of atoms of R is endowed with a multivalued product. We introduce an equivalence relation on the set of atoms such that the quotient set with the induced product is a monoid, called the condensation monoid. Under suitable hypotheses on R, we show that this monoid is a group and the class of k1_A is the set of atoms of a subalgebra of A called the focal subalgebra. This construction can be iterated to obtain higher condensation groups and focal subalgebras. We apply these results to G-algebras for G a group; in particular, we use them to define new invariants for finite-dimensional irreducible projective representations.	math.RA	math	ATOMISTIC SUBSEMIRINGS OF THE LATTICE OF SUBSPACES OF AN ALGEBRA DANIEL S. SAGE Abstract. Let A be an associative algebra with identity over a field k. An atomistic subsemiring R of the lattice of subspaces of A, endowed with the natural product, is a subsemiring which is a closed atomistic sublattice. When R has no zero divisors, the set of atoms of R is endowed with a multivalued product. We introduce an equivalence relation on the set of atoms such that the quotient set with the induced product is a monoid, called the condensation monoid. Under suitable hypotheses on R, we show that this monoid is a group and the class of k1A is the set of atoms of a subalgebra of A called the focal subalgebra. This construction can be iterated to obtain higher condensation groups and focal subalgebras. We apply these results to G-algebras for G a group; in particular, we use them to define new invariants for finite-dimensional irreducible projective representations. 1. Introduction Let A be an associative algebra with identity over a field k, and let S(A) be the complete lattice of subspaces of A. The algebra multiplication on A induces a product on S(A) given by EF = span{ef e ∈ E, f ∈ F }. The lattice S thus becomes an additively idempotent semiring, with {0} and k = k1A (which we will often denote by 0 and 1) as the additive and multiplicative identities. Let R be a closed sublattice of S(A) which is also a subsemiring, i.e., R contains 0 and k and is closed under arbitrary sums and intersections and finite products. (We do not require the maximum element of R to be A.) A nonzero element X ∈ R is called decomposable (or join-reducible) if there exists U, V ( R such that X = U + V and indecomposable otherwise. It is immediate that the multiplication in R is determined by the product of indecomposable elements. In other words, the semiring structure is determined by the structure constants cW U,V for U, V, W ∈ R indecomposable, where cW U,V is 1 if W ⊂ U V and 0 otherwise. In this paper, we consider subsemirings R whose product is determined by its minimal nonzero elements -- the atoms of the lattice. This means that the indecom- posable elements of R are precisely the atoms, so that every nonzero element is a join of atoms, i.e., R is an atomistic lattice1. Definition 1.1. A subsemiring R ⊂ S(A) is called an atomistic subsemiring of S(A) if it is also a closed atomistic sublattice. Note that k is always an atom in R. 2000 Mathematics Subject Classification. Primary: 16Y60, 20N20; Secondary: 16W22. The author gratefully acknowledges support from NSF grant DMS-0606300 and NSA grant H98230-09-1-0059. 1In the usual definition, every nonzero element of an atomistic lattice is a finite join of atoms. In this paper, we allow arbitrary joins of atoms. 1 2 DANIEL S. SAGE Example 1.2. For any A, S(A) and {0, 1} are atomistic subsemirings. Example 1.3. Let X be any proper subspace with X + k1A = A. Then R = {0, 1, X, A} is an atomistic subsemiring if and only if X 2 ∈ R. All four possible values for X 2 can occur. Indeed, if we let X = k¯t in the three two-dimensional algebras k[t]/(t2), k[t]/(t2 − 1), and k[t]/(t2 − t), we obtain X 2 equal to 0, 1, and X respectively. On the other hand, if X = span(¯t, ¯t2) in A = k[t]/(t3 − 1), then X 2 = A. (Note that there are never any atomistic subsemirings of size 3.) Example 1.4. Let V be a vector space with dim V ≥ 2, and suppose (char k, dim V ) = 1. Let A = End(V ), and let X = {x ∈ End(V ) tr(x) = 0}. Then R = {0, 1, X, A} is atomistic with X 2 = A. To see this, simply note that every matrix unit lies in X 2: Eii = Eij Eji and Eij = Eij (Eii − Ejj ) where i 6= j. Our primary motivation for considering atomistic subsemirings comes from rep- resentation theory. Let G be a group which acts on A by algebra automorphisms. This means that A is a k[G]-module such that g · 1A = 1A and g · (ab) = (g · a)(g · b) for all g ∈ G and a, b ∈ A. We let SG(A) ⊂ S(A) be the set of all k[G]-submodules of A. This set, called the subrepresentation semiring of A, is simultaneously a subsemiring and complete sublattice of S(A); such semirings were introduced and studied in [9, 10]. If A is a completely reducible representation, i.e., a direct sum of irreducible representations, then SG(A) is an atomistic subsemiring. For example, this occurs when G is finite, A is finite-dimensional, and k has characteristic zero. When G = SU(2) (or more generally, G is a quasi-simply reducible group), then the subrepresentation semirings for the G-algebras End(V ) (with V a representation of G) have had important applications in materials science and physics [5, 4, 9]. The structure of such semirings is intimately related to the theory of 6j-coefficients from the quantum theory of angular momentum [9, 10, 11, 6]. Our goal in this paper is to study the set of atoms Q(R) of an atomistic sub- semiring and to use it to define new invariants for appropriate R -- the condensation group, the focus, the focal subalgebra, and higher analogues. Our methods are motivated by the theory of hypergroups. We now give a brief outline of the contents of the paper. In Section 2, we define a multivalued product on the set Q(R) of atoms of an atomistic subsemiring R. In the next section, we introduce an equivalence relation ζ ∗ on Q(R). We show that if R has no zero-divisors, then the quotient set Q(R)/ζ ∗ is naturally a monoid (called the condensation monoid) while if R is weakly reproducible, the condensation monoid is in fact a group. In Section 4, we define the focus R ⊂ Q(R) and focal subalgebra F (R) ⊂ A of R. The main result is Theorem 4.3, which states that if R is weakly reproducible of finite length, then [0, F (R)] is an atomistic subsemiring with the same properties and whose set of atoms is R. This allows us to iterate our construction to obtain higher order versions of our invariants. In Section 5, we prove Theorem 4.3 by analyzing complete subsets of Q(R). We apply our results to G-algebras in the final section. In particular, we show how to associate new invariants to irreducible projective representations. 2. A hyperproduct on the set of atoms From now on, R will always be an atomistic subsemiring of S(A). Let Q(R) denote the set of atoms of R. If R = SG(A) for a G-algebra A, we write QG(A) ATOMISTIC SUBSEMIRINGS 3 instead of Q(SG(A)). We make the notational convention that, unless otherwise specified, capital letters towards the end of the alphabet will denote atoms. There is a natural operation Q(R) × Q(R) → P(Q(R)) given by X ◦ Y = {Z ∈ Q(R) Z ⊂ XY }. Our first goal is to find a natural equivalence relation on Q(R) (for appropriate R) for which ◦ induces a monoid (or group) structure on the set of equivalence classes. Before proceeding, we need to recall some definitions from the theory of hyper- groups. A set H is called a hypergroupoid if it is endowed with a binary operation ◦ : H × H → P∗(H), where P∗(H) is the set of nonempty subsets of H. If this operation is associative, then H is called a semihypergroup; if H also satisfies the reproductive law H ◦ x = H = x ◦ H for all x ∈ H, then H is called a hypergroup. (For more details on hypergroups, see the books by Corsini [1] and Vougiouklis [12].) An element e of the hypergroupoid H is called a scalar identity if e ◦ x = {x} = x◦e for all x ∈ H; if a scalar identity exists, it is unique. For later use, we introduce a weak version of the reproductive law. A hypergroupoid with scalar identity e satisfies the weak reproductive law if for any x ∈ H, there exists u, v ∈ H such that e ∈ x ◦ u ∩ v ◦ x. Note that a semihypergroup that satisfies the weak reproductive law is a hypergroup. Indeed, given y ∈ H, y ∈ y ◦ e ⊂ y ◦ (v ◦ x) = (y ◦ v) ◦ x, so there exists w ∈ y ◦ v such that y ∈ w ◦ x. Similarly, there exists w′ such that y ∈ x ◦ w′. In general, Q(R) is not even a hypergroupoid. However, we have the following result: Proposition 2.1. Let R be an atomistic subsemiring. Then (Q(R), ◦) is a hyper- groupoid if and only if R is an entire semiring (i.e., R has no left or right zero divisors). Proof. Suppose R is entire. If X, Y ∈ Q(R), then the nonzero subspace XY must contain an atom, so X ◦ Y 6= ∅. Conversely, if E, F are nonzero elements of R such that EF = 0, then choosing X, Y ∈ Q(R) such that X ⊂ E and Y ⊂ F implies that XY = 0, i.e., X ◦ Y = ∅. (cid:3) In particular, if A has zero divisors, then Q(S(A)) is not a hypergroupoid. We will only be interested in atomistic subsemirings R for which Q(R) is a hypergroupoid, so, from now on, we assume that R is entire, unless otherwise specified. Note that k is a scalar identity for Q(R). We begin by considering a motivating example. We need to recall some basic properties of semisimple, multiplicity-free representations. This class of G-modules is closed under taking submodules and quotients. Any such representation V is the direct sum of its irreducible submodules, and this is the only way of decomposing V as the internal direct sum of irreducible submodules. Moreover, there is a bijection between the power set of the set of irreducible submodules of V and the set of subrepresentations of V given by J 7→PX∈J X. It follows that if {Vi i ∈ I} is a collection of submodules of V and W =Pi∈I Vi, then for X irreducible, X ⊂ W if and only if X ⊂ Vj for some j ∈ I. Proposition 2.2. Let A be a multiplicity-free G-algebra with no proper, nontrivial left (or right) invariant ideals. Then QG(A) is a hypergroup. Proof. First, we show that the multiplication on QG(A) is associative. Fix X, Y, Z ∈ QG(A). Since A is multiplicity-free, XY =Pj∈J Uj, where X ◦ Y = {Uj j ∈ J}. 4 DANIEL S. SAGE As discussed above, an irreducible submodule W lies in (XY )Z = P UjZ if and only if it is contained in UiZ for some i, i.e., W ∈ Ui◦Z. We thus see that (X ◦Y )◦Z is the set of irreducible submodules of XY Z. A similar argument shows that the same holds for X ◦ (Y ◦ Z). Next, we show that X ◦ Y 6= ∅ for any X, Y ∈ QG(A). It suffices to show that XY 6= 0 for all X, Y . Let Y ⊥ = {a ∈ A ay = 0 for all y ∈ Y }. The subspace Y ⊥ is clearly a left ideal. Moreover, it is a subrepresentation: given g ∈ G, u ∈ Y ⊥, (g · a)u = g · (a(g−1 · u)) = g · 0 = 0. Since Y ⊥ 6= A, our hypothesis on invariant left ideals implies that Y ⊥ = 0 and XY 6= 0 for all X. Finally, we show that X ◦ QG(A) = QG(A) = QG(A) ◦ X for any X. The subspace AX is a nonzero left ideal which is obviously a subrepresentation, so AX = A. Writing A as a sum of irreducible submodules A = Pi∈I Ui, we have A =P UiX. The usual multiplicity-free argument shows that each Uj lies in some Uij X, so Uj ∈ Uij ◦ X. The other equality uses the condition on invariant right ideals. (cid:3) Matrix algebras give an important class of examples. If V is a finite-dimensional vector space and End(V ) is a G-algebra, then V is naturally a projective represen- tation of G [8]. It was further shown in [8] that End(V ) for such representations has no proper, nontrivial invariant left or right ideals if and only if V is irreducible. Hence, we obtain: Corollary 2.3. If V is a finite-dimensional irreducible projective representation of a group such that End(V ) is multiplicity free, then QG(End(V )) is a hypergroup. This corollary applies, for example, to every irreducible complex representation of SU(2). The importance of Proposition 2.2 stems from the fact that there is a group nat- urally associated to every hypergroup. More generally, let H be a semihypergroup. Consider the relation β defined by x β y if and only if there exists z1, . . . , zn ∈ H such that x, y ∈ z1 ◦ · · · ◦ zn. Koskas showed that if β∗ is the transitive closure of β, then the induced multiplication makes H/β∗ into a semigroup, and β∗ is the largest equivalence relation on H with this property [7]. If H is a hypergroup, then Freni proved that β is automatically transitive [2]; thus, H/β is a group. We are led to the following provisional definition. Definition 2.4. Let A be a G-algebra satisfying the hypotheses of Proposition 2.2. The group QG(A) = QG(A)/β is called the condensation group of A. We will generalize this definition to a much broader class of atomistic subsemir- ings below. However, before continuing we provide a few examples. Example 2.5. If k denotes the trivial G-algebra, then QG(k) is the trivial group. Example 2.6. If V is any irreducible representation of SU(2), then QSU(2)(End(V )) = 1. The proof is a special case of Theorem 6.6 below. Example 2.7. Let V be the standard representation of S3 over the complex numbers. The corresponding S3-algebra decomposes as End(V ) = C ⊕ σ ⊕ V , where σ is the sign representation. Since σ2 = C, σV = V σ = V , and V 2 = C ⊕ σ, we see that the classes of β are {C, σ} and {V }; hence, the condensation group has order 2. Example 2.8. If F is a finite Galois extension of k with abelian Galois group G, then QG(F ) = G. ATOMISTIC SUBSEMIRINGS 5 We remark that if the relation β is replaced by Freni's relation γ [3], one gets an abelian group canonically related to any hypergroup. However, we will not attempt to generalize the abelian group QG(A)/γ to other atomistic subsemirings in this paper. 3. The equivalence relation ζ ∗ It is not true in general that the hypergroupoid Q(R) is a hypergroup or even a semihypergroup. For example, the binary operation on QA4 (End(W )) is not associative, where W is the three-dimensional irreducible representation of A4. Moreover, the reproductive law is not satisfied. (See Example 6.5 below.) We can thus no longer use the relation β∗ to associate a monoid or group to R. Instead, we will do so by introducing a new relation ζ. This relation will coincide with β in the situation of Proposition 2.2. Definition 3.1. The relation ζ on Q(R) is defined by X ζ Y if and only if there i=1 Zi. We let ζ ∗ denote the transitive exists Z1, . . . , Zn ∈ Q(R) such that X, Y ⊂Qn closure of ζ. It is obvious that ζ ∗ is an equivalence relation. We will let ¯X denote the equiv- alence class of X ∈ Q(R). Remark 3.2. If Z is an atom contained in the ◦ product of Z1, . . . , Zn with any i=1 Zi. In fact, the relation β can be defined for hypergroupoids, and this observation just says that β ⊂ ζ. However, the set of β∗-equivalence classes is not necessarily a monoid. choice of parentheses, then Z ⊂ Qn Definition 3.3. Let R be an entire, atomistic subsemiring of S(A). (1) R is called weakly reproducible if the hypergroupoid Q(R) satisfies the weak reproductive law, i.e., for all X ∈ Q(R), there exists Y, Z ∈ Q(R) such that k ∈ X ◦ Y ∩ Z ◦ X. (2) R is called reproducible if Q(R) satisfies the reproductive law, i.e., for all X ∈ Q(R), Q(R) ◦ X = Q(R) = X ◦ Q(R). Remark 3.4. One can define an atomistic subsemiring R to be weakly reproducible without the assumption that R is entire. However, R is then entire automatically. Indeed, if XY = 0 for X, Y ∈ Q(R), then weak reproducibility implies the existence of Z such that k ⊂ ZX, so Y = kY ⊂ ZXY = 0, a contradiction. Theorem 3.5. Let R be an entire, atomistic semiring of S(A). Then (1) The induced multiplication on classes makes Q(R) (2) If R is weakly reproducible, then Q(R) is a group. def = Q(R)/ζ ∗ into a monoid. Definition 3.6. The monoid Q(R) is called the condensation monoid (or group) of R. The following lemma shows that this terminology does not conflict with our previous definition. Lemma 3.7. If A satisfies the hypotheses of Proposition 2.2, then β and ζ coincide on QG(A). Proof. A similar argument to that used to demonstrate the associativity of QG(A) i=1 Zi, so β = ζ. (cid:3) shows that Z1 ◦· · ·◦Zn is the set of irreducible submodules ofQn 6 DANIEL S. SAGE Recall that an equivalence relation ∼ on a hypergroupoid H is called strongly regular if, for any x, y such that z ∼ y and any w ∈ H, then u ∈ x ◦ w and v ∈ y ◦ w (resp. u ∈ w ◦ x and v ∈ w ◦ y) implies that u ∼ v. It is a standard fact that for such ∼, ◦ induces a binary operation on H/∼ via ¯x ◦ ¯y = ¯z, where z ∈ x ◦ y [1]. Indeed, strong regularity implies that the set {¯z z ∈ x′ ◦ y′ for some x′ ∈ ¯x, y′ ∈ ¯y} is a singleton. Lemma 3.8. The equivalence relation ζ ∗ is strongly regular. Proof. First, suppose that X ζ Y , so X, Y ⊂Qn and V ∈ Y ◦ W , then U ⊂ XW and V ⊂ Y W . Thus, U, V ⊂ (Qn i=1 Zi for some Zi's. If U ∈ X ◦ W i=1 Zi)W , i.e., U ζ W . If X ζ ∗ Y , then there exists X0, . . . , Xs ∈ Q(R) with X = X0, Y = Xs, and Xi ζ Xi+1 for all i. Taking Ui ∈ Xi ◦ W with U = U0 and V = Us, the previous case shows that Ui ζ Ui+1 for all i, i.e., U ζ ∗ V . The opposite direction in the definition of strong regularity is proved similarly. (cid:3) We now verify that the induced binary operation makes Q(R) into a monoid. The identity is given by ¯k; indeed, this follows immediately from the fact that k◦X = X = X ◦k. Next, we check that ( ¯X ◦ ¯Y )◦ ¯Z = ¯X ◦( ¯Y ◦ ¯Z). Choose U ∈ X ◦Y and V ∈ U ◦Z, so that ¯V = ( ¯X ◦ ¯Y )◦ ¯Z. Since U ⊂ XY , V ⊂ U Z ⊂ XY Z. Similarly, choosing T ∈ Y ◦ Z and W ∈ X ◦ T gives ¯W = ¯X ◦ ( ¯Y ◦ ¯Z) and W ⊂ XT ⊂ XY Z. By definition, V ζ W , so Q(R) is associative. Remark 3.9. If we allow R to be an atomistic hemiring of S(A), i.e., we do not require that k ∈ R, then the same argument shows that Q(R) is a semigroup. Finally, assume that R is weakly reproducible. Given X ∈ Q(R), choose Y, Z such that k ⊂ XY ∩ ZX. By definition of the product on Q(R), we obtain ¯X ◦ ¯Y = ¯k = ¯Z ◦ ¯X, so ¯X is left and right invertible. This shows that Q(R) is a group and finishes the proof of the theorem. Remark 3.10. Any monoid can be realized as the condensation monoid of an atom- istic subsemiring. Indeed, given a monoid M , let kM be the corresponding monoid algebra over k with basis elements {exx ∈ M }. Let R = {span{ex x ∈ F } F ⊂ M }. This is an entire atomistic subsemiring of S(kM ) with Q(R) = {kex x ∈ M }. It is now easy to see that Q(R) = M . 4. The focus and the focal subalgebra Recall that if H is a hypergroup, the heart ωH of H is the kernel of the canonical homomorphism φ : H → H/β∗; it is a subhypergroup of H. Returning to the context of Proposition 2.2, let A be a multiplicity-free G-algebra with no proper, nonzero left or right invariant ideals. We may then use the heart ω of the hypergroup QG(A) to define an invariant subalgebra with the same properties. Proposition 4.1. Let A be a multiplicity-free G-algebra with no proper, nontrivial one-sided invariant ideals. Then B = P{X X ∈ ω} is a multiplicity-free G- subalgebra with no proper one-sided invariant ideals. Proof. It is trivial that B is a multiplicity-free subrepresentation that contains k. Moreover, if X, Y ∈ ω and Z ⊂ XY is irreducible, then φ(Z) = φ(X)φ(Y ) = 1, i.e., Z ∈ ω. This means that Z and hence XY are subspaces of B. It remains to show that BX = B = XB for any X ∈ ω. Choose Z ∈ ω. Since QG(A) is a hypergroup, ATOMISTIC SUBSEMIRINGS 7 there exists Y irreducible such that Z ∈ Y ◦ X. Since 1 = φ(Z) = φ(Y )φ(X) = φ(Y ), we see that Y ∈ ω, so Z ⊂ BX. The proof that Z ⊂ XB is similar. (cid:3) This result allows us to iterate the construction of the condensation group. In- deed, the hypergroup structure on QG(B) = ω gives rise to the group QG(B) and an invariant subalgebra B′ ⊂ B such that QG(B′) is again a hypergroup. See Section 6 for examples. Motivated by this situation, we make the following definitions. Definition 4.2. Let R be an entire atomistic subsemiring. (1) The focus R of R is the kernel of the homomorphism ψR : Q(R) → Q(R). Equivalently, it is the equivalence class of k. (2) The subspace F (R) = P{X X ∈ R} ∈ R is called the focal subalgebra associated to R. We can now state one of the main results of the paper. Theorem 4.3. Let R be an entire atomistic subsemiring of S(A). (1) The focal subspace F (R) is a unital subalgebra of A. (2) The sublattice [0, F (R)] ⊂ R is an entire atomistic subsemiring of S(F (A)) with R ⊂ Q([0, F (R)]). (3) If R is weakly reproducible and has finite length, then Q([0, F (R)]) = R. (4) If R is weakly reproducible (resp. reproducible) of finite length, then the same holds for [0, F (R)]. We remark that part (3) is very useful in computations as it is often easier to calculate F (R) than to compute R directly. We will only prove the first two parts of the theorem now. The proof of the other parts requires a more detailed study of the relation ζ ∗ and will be given at the end of Section 5. Proof of parts (1) and (2). If X, Y ∈ R and Z ∈ X ◦ Y , then 1 = ψ(X)ψ(Y ) = ψ(Z). This means that Z ∈ , so XY ⊂ F (R). Since k ⊂ F (R), F (R) is a subalgebra. This implies that F (R)2 = F (R), so if E, E′ ∈ [0, F (R)], then E + E′ ⊂ F (R) and EE′ ⊂ F (R). Thus, the closed sublattice [0, F (R)] ⊂ R is a subsemiring of R, and it is immediate that it is entire and atomistic. The atoms of [0, F (R)] are precisely the atoms of R which are contained in F (R), so R ⊂ Q([0, F (R)]). (cid:3) Corollary 4.4. If R is weakly reproducible and has finite length, then Q(R) = 1 if and only if F (R) is the maximum element of R, i.e., [0, F (R)] = R. Proof. If Q(R) = 1, then R = Q(R). Thus, F (R) contains every atom in R, hence is the maximum element of R. Conversely, if F (R) is the maximum of R, then part (3) of the theorem implies that R = Q(R). This gives Q(R) = 1. (cid:3) Remark 4.5. The forward implication in the corollary holds for any entire atomistic subsemiring. The theorem shows that we can iterate the construction of the invariants asso- ciated to R. Definition 4.6. The higher foci, focal subalgebras, and condensation monoids (or groups) for R are defined recursively as follows: 8 DANIEL S. SAGE • 1 • n+1 R = R, F 1(R) = F (R), and Q1(R) = Q(R); R = [0,F n(R)], F n+1(R) = F ([0, F n(R)]), and Qn+1(R) = Q([0, F n(R)]). We observe that if R is weakly reproducible and has finite length, then Qn(R) is a group for all n. 5. Complete subsets of Q(R) In order to prove Theorem 4.3, we need a better understanding of the equivalence relation ζ ∗. In this section, we define complete subsets of Q(R) and use them to investigate the ζ ∗-equivalence classes. Our analysis of ζ ∗ follows a similar pattern to that of β∗ carried out by Corsini and Freni [1, 2]. In the end, we will show that if R is weakly reproducible, then every element of R is ζ-related (and not just ζ ∗-related) to k; this will be the key ingredient in the proof of Theorem 4.3. Definition 5.1. exists X ∈ E such that X ⊂Qn (1) A subset E ⊂ Q(R) is called complete if for all X1, . . . , Xn ∈ Q(R), if there i=1 Xi, Y ∈ E. (2) If E is a nonempty subset of Q(R), then the intersection of all complete subsets containing E is denoted by C(E); it is called the complete closure of E. i=1 Xi, then for any Y ⊂Qn It is obvious that C(E) is the smallest closed subset containing E. Remark 5.2. This is not the usual notion of a complete subset of a semihyper- group [1, 7], though it coincides in the context of Proposition 2.2. In this paper, we only consider completeness in the sense given above. The basic examples of closed subsets are the ζ ∗-equivalence classes. Proposition 5.3. Any ζ ∗-equivalence class is closed. Proof. Consider the class of Z. Suppose that X ζ ∗ Z and X, Y ⊂ Qn Y ζ X, so Y ζ ∗ Z. i=1 Xi. Then (cid:3) The complete closure may be computed inductively. Indeed, given E 6= ∅, define a sequence of subsets κn(E) ⊂ Q(R) recursively as follows: κ1(E) = E and κn+1(E) = {X ∈ Q(R) ∃Y1, . . . , Ys ∈ Q(R) and Y ∈ κn(E) such that X, Y ⊂ Set κ(E) = ∪n≥1κn(E). Proposition 5.4. For any nonempty E ⊂ Q(R), C(E) = κ(E). Yi}. s Yi=1 i=1 Yi. Then X ∈ κn+1(E) ⊂ κ(E), so κ(E) is complete. Since E ⊂ κ(E), C(E) ⊂ κ(E). Conversely, suppose that F ⊃ E and F is complete. We show inductively that κn(E) ⊂ F . This is obvious for n = 1. Suppose κn(E) ⊂ F . If X ∈ κn+1(E), then we can find i=1 Yi. Completeness of F (cid:3) Proof. Suppose Y ∈ κ(E), say Y ∈ κn(E), and X, Y ⊂ Qs Y1, . . . , Ys ∈ Q(R) and Y ∈ κn(E) such that X, Y ⊂ Qs now shows that X ∈ F as desired. We can now give a new characterization of ζ ∗. Define a relation κ on Q(R) by X κ Y if and only if X ∈ C(Y ), where C(Y ) = C({Y }). Theorem 5.5. The relations κ and ζ ∗ coincide. ATOMISTIC SUBSEMIRINGS 9 Before beginning the proof, we will need a lemma. Lemma 5.6. (1) For any X ∈ Q(R) and n ≥ 2, κn+1(E) = κn(κ2(X)). (2) For X, Y ∈ Q(R), X ∈ κn(Y ) if and only if Y ∈ κn(X). Y ∈ κn−1(κ2(X)) such that Y, Z ⊂Qs Proof. Note that κn(κ2(X)) consists of those atoms Z for which there exists Yi's and i=1 Yi. If n = 2, then κn−1(κ2(X)) = κ2(X), and this is precisely the defining property of κ3(X). If n > 2, then κn−1(κ2(X)) = κn(X) by inductive hypothesis, and we see that such atoms are precisely the ele- ments of κn+1(X). This proves part (1). The second assertion is also proven by induction. Suppose X ∈ κ2(Y ). Then i=1 Yi, so Y ∈ κ2(X). Next, assume that the i=1 Yi for some Yi's and Z ∈ κn(Y ). By definition, Z ∈ κ2(X), and Y ∈ κn(Z) by induction. Hence, Y ∈ κn(κ2(X)) = κn+1(X). (cid:3) there exist Yi's such that X, Y ⊂ Qs statement holds for n. If X ∈ κn+1(Y ), then X, Z ⊂ Qs Proof of Theorem 5.5. First, we show that κ is an equivalence relation. It is clear that κ is reflexive. If X κ Y and Y κ Z, then X ∈ C(Y ) and Y ∈ C(Z). Since C(Z) is complete and contains Y , C(Y ) ⊂ C(Z), so X ∈ C(Z), i.e., X κ Z. Finally, if X κ Y , then Proposition 5.4 implies that X ∈ κn(Y ) for some n. By the lemma, Y ∈ κn(X) ⊂ κ(X), and another application of Proposition 5.4 gives Y κ X. i=1 Xi for some Xi's, so X κ Y . Since κ is an equivalence relation, it follows that ζ ∗ ⊂ κ. Next, suppose that X ζ Y . Then X, Y ⊂Qs such that X, X1 ⊂Qs1 choose Xr+1 ∈ κn−r(Y ) satisfying Xr, Xr+1 ⊂ Qsr+1 Conversely, assume that X κ Y , say X ∈ κn+1(Y ). Set X0 = X. We recursively construct Xj ∈ κn+1−j(Y ) for 0 ≤ j ≤ n satisfying Xj κ Xj+1. Choose X1 ∈ κn(Y ) i=1 Y1,i for some Y1,i's. This means that X ζ X1. Suppose that we have constructed the desired atoms up through Xr with r < n. Again, we can i=1 Yr+1,i for some Yr+1,i's, and this gives Xr ζ Xr+1. Note that Xn ∈ κ1(Y ) = {Y }, i.e., Xn = Y . We conclude that X ζ ∗ Y as desired. (cid:3) Corollary 5.7. For any E ⊂ Q(R) nonempty, ψ−1(ψ(E)) = ∪X∈E C(X) = C(E). In particular, the ζ ∗-equivalence class of X is C(X). Proof. The set ψ−1(ψ(E)) consists of those atoms equivalent to an atom in E, hence is the union of the ζ ∗ = κ equivalence classes of atoms in E. This gives the first equality. The second follows immediately from the fact that a union of closed subsets is closed. (cid:3) To proceed further, we need to impose additional conditions on R. Proposition 5.8. (1) If R is reproducible, then for all X ∈ Q(R), C(X) = R ◦ X = X ◦ R. In particular, the subhypergroupoid R satisfies the reproductive law. (2) If R is weakly reproducible, then R satisfies the weak reproductive law. Proof. First, assume that R is reproducible. Suppose that Y ∈ C(X), so Y ζ ∗ X. By reproducibility, there exist U such that Y ⊂ XU , i.e., Y ∈ X ◦ U . Hence, ψ(Y ) = ψ(X)ψ(U ), so ψ(U ) = 1. This shows that U ∈ R, giving Y ∈ X ◦ R. Conversely, if Z ∈ X ◦ R, then ψ(Z) = ψ(X). This means that Z ∈ C(X). The 10 DANIEL S. SAGE equality C(X) = R ◦ X is proved in the same way. When X ∈ R, the first statement says that R = R ◦ X = X ◦ R, which is the reproductive law. If R is weakly reproducible, then the argument given above (with Y = k and X ∈ R) shows that there exists U, V ∈ R such that k ⊂ U ◦ X ∩ X ◦ V as desired. (cid:3) Given Z ∈ Q(R), define M (Z) = {X ∈ Q(R) ∃Y1, . . . , Ys ∈ Q(R) such that X, Z ⊂ Yi}. s Yi=1 Lemma 5.9. If R is reproducible (resp. weakly reproducible), then M (Z) is a complete part for all Z (resp. for Z = k). X ⊂ n Yj=1 j=1 Zj also. Then i=1 Yi for j=1 Zj. By reproducibility, choose V, W such that Proof. Assume that R is reproducible. Take Y ∈ M (Z), so Y, Z ⊂ Qs some Yi's. Suppose that Y ⊂ Qn Z ⊂ Y V and Zn ⊂ W Z. Now, suppose that X ⊂Qn Zj Zj Zj  W Y V ⊂  W Z ⊂  W s Yi=1   Zj Zj Zj  W ZV ⊂  V ⊂ Yi! V.  W s Yi=1 Yj=1 Yj=1   Zj ⊂  Z ⊂ Y V ⊂  On the other hand, Yi! V. n−1 Yj=1 n−1 Yj=1 n−1 Yj=1 n Yj=1 n−1 n−1 Thus, Y ∈ M (Z), so M (Z) is complete. If R is weakly reproducible, then the same argument works with Z = k. Indeed, we need only set W = Zn and use weak reproducibility to choose V such that k ⊂ Y V . (cid:3) Corollary 5.10. If R is reproducible (resp. weakly reproducible), then M (Z) = R for any Z ∈ R (resp. for Z = k). Proof. Suppose that Z ∈ R. If X ∈ M (Z), then X ζ Z by definition, so X ∈ R. Thus, M (Z) ⊂ R. Conversely, the lemma shows that, under the hypothesis on R, M (Z) is a complete subset containing Z, so C(Z) = R ⊂ M (Z). (cid:3) Theorem 5.11. (1) If R is reproducible, then ζ is transitive. (2) If R is weakly reproducible, then X ζ ∗ k implies that X ζ k. Proof. Assume that R is reproducible, and take X ζ ∗ Y . By Proposition 5.8, there exists U ∈ R such that Y ∈ X ◦ U . Since M (k) = R, Corollary 5.10 implies the i=1 Yi ⊃ Xk = X, (cid:3) so Y ζ X. If R is weakly reproducible, the same argument works for Y = k. existence of Yi's such that U, k ⊂Qs i=1 Yi. Thus, Y ⊂ XU ⊂ XQs We are now ready to return to the proof of Theorem 4.3. We first state a proposition. Proposition 5.12. Let R be weakly reproducible of finite length. Then there exists X1, . . . , Xn ∈ Q(R) such that F (R) =Qn i=1 Xi. ATOMISTIC SUBSEMIRINGS 11 i=1 Xi ⊂ F (R). Indeed, if Z ⊂ i=1 Xi is the sum of the atoms it contains. i=1 Xi, then Qn Proof. First, note that if k ⊂ Qn Qn i=1 Xi is an atom, then Z ζ k, so Z ⊂ F (R). The claim follows because Qn Choose E =Qn that k, Y ⊂ E′ =Qm i=1 Xi containing k such that [E, F (R)] has minimal length. If this length is 0, then E = F (R), so suppose it is positive, i.e., E ( F (R). Take Y ∈ R such that Y ( E. By Theorem 5.11, Y ζ 1, so there exist Y1, . . . , Ym ∈ Q(R) such j=1 Yj. This implies that E = Ek ⊂ EE′ and Y = kY ⊂ EE′, and the previous paragraph shows that EE′ ⊂ F (R). We obtain E ( EE′ ⊂ F (R), contradicting the minimality of the length of [E, F (R)]. (cid:3) We apply the proposition to prove part (3) of the theorem. Indeed, if Z ∈ Q(R) i=1 Xi, then Z ζ 1 by definition. This means that Z ∈ R as and Z ⊂ F (R) = Qn desired. Finally, we prove part (4). Suppose that R is reproducible of finite length, and X, Y ⊂ F (R) are atoms. By part (3), X, Y ∈ R. By hypothesis, there exists Z ∈ Q(R) such that Y ∈ X ◦ Z. Since 1 = ψ(Y ) = ψ(X)ψ(Z) = ψ(Z), Z ⊂ F (R) and so Q([0, F (R)]) = X ◦ Q[0, F (R)]). Similarly, one shows Q([0, F (R)]) = Q[0, F (R)])◦X, so Q([0, F (R)]) is reproducible. The same argument applies when R is weakly reproducible; here, one takes Y = k. This completes the proof of Theorem 4.3. 6. Applications to G-algebras In this section, we apply our results on atomistic semirings to subrepresen- tation semirings. We assume throughout that A is a G-algebra which is com- pletely reducible as a representation. We write QG(A) (resp. FG(A)) instead of Q(SG(A)) (resp. F (SG(A))). We will now be able to generalize our earlier results on multiplicity-free G-algebras. Proposition 6.1. Let A be a G-algebra in which the trivial representation has multiplicity one. Then A has no proper, nontrivial one-sided invariant ideals if and only if SG(A) is weakly reproducible. Indeed, if Remark 6.2. Note that both conditions imply that SG(A) is entire. the condition on invariant ideals holds, then the argument given in the proof of Proposition 2.2 shows that SG(A) is entire. The analogous statement for weak reproducibility was shown in Remark 3.4. Proof. Assume that A has no proper, nontrivial invariant ideals. Fix X ∈ QG(A), and express A as a direct sum of irreducible subrepresentations A =Li∈I Yi. The subspace AX is a nonzero invariant left ideal, so we obtain A = AX =Pi∈I XYi. The trivial representation must accordingly be an irreducible component of some XYj. The fact that the trivial representation has multiplicity one in A implies that k ⊂ XYj as desired. Similarly, since A = XA, there exists Yl such that k ⊂ YlX. Conversely, suppose that 0 6= L 6= A is an invariant left ideal. Let X ⊂ L be an irreducible submodule. For any Y ∈ QG(A), we have Y X ⊂ L; since k ∩ L = 0, SG(A) is not weakly reproducible. A similar argument works for right ideals. (cid:3) Corollary 6.3. The atomistic semiring SG(A) is weakly reproducible of finite length if (1) A is a finite Galois extension of k and G is the Galois group; or (2) A = End(V ) is a finite-dimensional G-algebra whose underlying projective representation V is irreducible. 12 DANIEL S. SAGE Proof. Schur's lemma shows that End(V ) contains the trivial representation with multiplicity one, and the statement about invariant ideals was proved in [8, Theorem 5.2]. The analogous verifications for the other case are obvious. (cid:3) We are thus able to define invariants for any G-algebra satisfying the conditions of Proposition 6.1, without our earlier assumption that the G-algebra is multiplicity- free. In particular, our results determine two new sequences of invariants associ- ated to any irreducible projective representation, namely, the condensation groups Qn G(End(V )). G(End(V )) and the focal subalgebras F n The focal subalgebras F n G(A) are a decreasing sequence of invariant subalgebras (i.e., subalgebras which are also subrepresentations) of A. This is particularly interesting for A = End(V ) with V irreducible and k algebraically closed because in this case, there is a complete classification of such invariant subalgebras in terms of representation-theoretic data [8, Theorem 3.23]. For the rest of the paper, we assume that either G is finite and k is algebraically closed of characteristic zero or G is a compact group and k = C. We let V be an irreducible (linear) representation of G, and set A = End(V ). (We make these assumptions on G and k to guarantee complete reducibility of End(V ); the classi- fication of invariant subalgebras described below holds in general.) An invariant subalgebra of A is determined by data consisting of a quadruple (H, W, U, U ′); here, H is a finite index subgroup of G, W is a linear representation of H such that V = IndG H (W ), and U, U ′ are a pair of projective representations of H such that W ∼= U ⊗ U ′. More precisely, there is a bijection between invariant algebras and equivalence classes of such quadruples under conjugation by G. In particular, there are a finite number of invariant subalgebras. gives a direct sum decomposition V =Ln H (End(W )) def= Ln Given such a quadruple (H, W, U, U ′), we construct the corresponding invariant subalgebra as follows: Let g1 = e, g2, . . . , gn be a left transversal for H in G. This i=1 giW and an associated block diagonal invariant subalgebra IndG i=1 End(giW ). As an algebra, this is just the direct product of n copies of End(W ). Next, the isomorphism W ∼= U ⊗ U ′ shows that the endomorphism algebra factors (as H-algebras) into the tensor product End(W ) ∼= End(U ) ⊗ End(U ′). It is now immediate that End(U ) ⊗ k is an H-invariant subalgebra of End(W ). Finally, we obtain the invariant algebra for the quadruple: IndG H (End(U ) ⊗ k). We remark that the two obvious invariant subalgebras k and End(V ) correspond to (G, V, k, V ) and (G, V, V, k) respectively. It now follows that the sequence of focal subalgebras associated to the irreducible representation V gives rise to a sequence of such quadruples. The classification of invariant subalgebras can be very helpful for computing the F n G(End(V )). For example, suppose that End(V ) has no nontrivial invariant subalgebras, so that any irreducible representation generates End(V ). In order to show that FG(End(V )) = End(V ), it is only necessary to check that FG(End(V )) contains a nonscalar matrix. However, it should be noted that computing the invariant subalgebras is not necessarily straightforward. Even when G is finite, it is not determined by the character table of G. In general, one needs to know the character tables of a covering group for every subgroup of G whose index divides dim(V ). Example 6.4. Let V be the standard representation of S3. We have already seen that Q1 G(End(V )) = C ⊕ σ is isomorphic G(End(V )) = Z2. The focal subalgebra F 1 ATOMISTIC SUBSEMIRINGS 13 to C ⊕ C as an algebra; it comes from the quadruple (A3, χ, χ, C), where χ is either nontrivial character of A3. Since C and σ are not ζ ∗-equivalent in QG(F 1 G(End(V ))), we have Q2 G (End(V )) = C (corresponding to (S3, V, C, V )). Finally, Qn G(End(V )) = Z2 and for m ≥ 2, F m G(End(V )) = 1 for n ≥ 3, Example 6.5. Let W be the three-dimensional irreducible representation of A4. We will show that QA4(End(W )) is not associative and does not satisfy the reproductive law. We have the direct sum decomposition End(W ) = C ⊕ Z ⊕ Z ′ ⊕ X ⊕ Y , where Z and Z ′ correspond to the two nontrivial characters of A4 and X and Y are isomorphic to W . We can choose a basis for W with respect to which Y (resp. X) consists of the skew-symmetric (resp. off-diagonal symmetric) matrices and the diagonal T is the direct sum of C, Z, and Z ′. There are an infinite number of atoms isomorphic to W , parameterized by [a : b] ∈ P1(C); we set U[a:b] = span{(a+b)E23 +(a−b)E32, (a−b)E13 +(a+b)E31, (a+b)E12 +(a−b)E21}. In this notation, X = U[1:0] and Y = U[0:1]. Let P = U[1:1]. It is easily checked that P 2 = U[1:−1]. We now calculate that P C = P Z = P Z ′ = P , P (P 2) = T , and P U[a:b] = T ⊕ P 2 for [a : b] 6= [1 : ±1]. We thus see that QA4 (End(W )) does not satisfy the reproductive law; if [a : b] 6= [1 : ±1], there is no V for which U[a:b] ∈ P ◦ V . To verify that the associative law does not hold, note that X ∈ (P ◦ P 2) ◦ X = {C, Z, Z ′} ◦ X. However, P ◦ (P 2 ◦ X) = P ◦ {C, Z, Z ′, P } = {P, P 2} does not contain X. Since P X = T ⊕ P 2, we have C, Z, Z ′, P 2 ∈ . Also, P 2X = T ⊕ P , so A4 (End(W )) = End(W ) for all n, and by Corollary 4.4, Q ∈ . This implies that F n Qn A4(End(W )) = 1 for all n. The only nontrivial invariant subalgebra of End(W ) is T . (It corresponds to (H, χ, χ, C), where H ∼= Z2 × Z2 is the subgroup of order 4 and χ is any nontrivial character of H.) Thus, one knows that FA4 (End(W )) = End(W ) as soon as one know that P 2 ∈ . We conclude by computing the condensation groups and focal subalgebras of endomorphism algebras for simple compact Lie groups. Theorem 6.6. Let V be an irreducible representation of the simple compact Lie group G. Then Qn G(End(V )) = End(V ) for all n. G(End(V )) = 1 and F n Proof. If V = C, the statement is trivial. Any other V has dimension at least 2. By Corollary 4.4, it suffices to show that FG(End(V )) = End(V ). Moreover, by [8, Theorem 4.3], the only proper invariant subalgebra of End(V ) is C. Hence, we need only show that FG(End(V )) contains a nonscalar matrix. Let λ be the highest weight of V . The highest weight of the dual representation V ∗ is −w0λ, where w0 is the longest element in the Weyl group. The representation End(V ) ∼= V ⊗ V ∗ has a unique irreducible submodule X with highest weight λ − w0λ. We can write down a highest and lowest weight vector in X explicitly. Let vλ (resp. wλ) be a highest (resp. lowest) weight vector in V . (The highest and lowest weights are different since dim V ≥ 2.) Extend the set {vλ, wλ} to a basis of weight vectors for V , and let v∗ λ be the corresponding dual basis vectors in V ∗. λ) is a highest (resp. lowest) weight vector in V ∗. It follows that Then w∗ vλ ⊗ w∗ λ) is a highest (resp. lowest) weight vector in X. λ (resp. v∗ λ (resp. wλ ⊗ v∗ λ, w∗ 14 DANIEL S. SAGE Multiplying, we obtain z = (vλ ⊗ w∗ λ ∈ X 2. The matrix z has rank one, so is not a scalar matrix. Thus, X 2 6= C. However, tr(z) = 1, so z is not orthogonal to C. This implies that C ⊂ X 2. We conclude that contains at least two elements, so FG(End(V )) 6= C. (cid:3) λ)(wλ ⊗ v∗ λ) = vλ ⊗ v∗ References [1] P. Corsini, Prolegomena of hypergroup theory, Supplement to Riv. Mat. Pura. Appl., Aviani Editore, Tricesimo, 1993. [2] D. Freni, Une note sur le coeur d'un hypergroupe et sur la cloture transitive β ∗ de β, Riv. Mat. Pura Appl. 8 (1991), 153 -- 156. [3] D. Freni, A new characterization of the derived hypergroup via strongly regular equivalences, Comm. Algebra 32 (2002), 3977 -- 3989. [4] Y. Grabovsky, G. Milton, and D. S. Sage, Exact relations for effective tensors of polycrystals: Necessary and sufficient conditions, Comm. Pure. Appl. Math. 53 (2000), 300 -- 353. [5] Y. Grabovsky and D. S. Sage, Exact relations for effective tensors of polycrystals. II: Appli- cations to elasticity and piezoelectricity, Arch. Rat. Mech. Anal. 143 (1998), 331 -- 356. [6] N. Kwon and D. S. Sage, Subrepresentation semirings and an analogue of 6j-coefficients, J. Math. Phys. 49 (2008), 063503. [7] M. Koskas, Groupoıdes, demi-hypergroupes et hypergroupes, J. Math. Pures Appl. 49 (1970), 155 -- 192. [8] D. S. Sage, Group actions on central simple algebras, J. Algebra 250 (2002), 18 -- 43. [9] D. S. Sage, Racah coefficients, subrepresentation semirings, and composite materials, Adv. App. Math 34 (2005), 335 -- 357. [10] D. S. Sage, Quantum Racah coefficients and subrepresentation semirings, J. Lie Theory 15 (2005), 321 -- 333. [11] D. S. Sage, Subrepresentation semirings and 3nj-symbols for simply reducible groups, preprint. [12] T. Vougiouklis, Hyperstructures and their representations, Hadronic Press, Inc., Palm Har- bor, Fl, 1994. Department of Mathematics, Louisiana State University, Baton Rouge, LA 70803 E-mail address: [email protected]
1511.08860	1	1511	2015-11-28T01:34:32	Hopf-Galois objects of Calabi-Yau Hopf algebras	[ "math.RA" ]	By using the language of cogroupoids, we show that Hopf-Galois objects of a twisted Calabi-Yau Hopf algebra with bijective antipode are still twisted Calabi-Yau, and give their Nakayama automorphism explicitly. As applications, cleft Galois objects of twisted Calabi-Yau Hopf algebras and Hopf-Galois objects of the quantum automorphism groups of non-degenerate bilinear forms are proved to be twisted Calabi-Yau.	math.RA	math	HOPF-GALOIS OBJECTS OF CALABI-YAU HOPF ALGEBRAS XIAOLAN YU Abstract. By using the language of cogroupoids, we show that Hopf- Galois objects of a twisted Calabi-Yau Hopf algebra with bijective antipode are still twisted Calabi-Yau, and give their Nakayama automorphism ex- plicitly. As applications, cleft Galois objects of twisted Calabi-Yau Hopf algebras and Hopf-Galois objects of the quantum automorphism groups of non-degenerate bilinear forms are proved to be twisted Calabi-Yau. Introduction We work over a fixed field k. Let E ∈ GLm(k) with m > 2 and let B(E) be the algebra presented by generators (uij)16i,j6m and relations E−1utEu = Im = uE−1utE, where u is the matrix (uij)16i,j6m, ut is the transpose of u and Im is the identity matrix. The algebra B(E) is a Hopf algebra and was defined by Dubois-Violette and Launer [8] as the quantum automorphism group of the non-degenerate bilinear form associated to E. It is not difficult to check that B(Eq) = Oq(SL2(k)) (the quantised coordinate algebra of SL2(k)), where Eq = 0 −q−1 0 ! . 1 So the Hopf algebras B(E) are generalizations of Oq(SL2(k)). For any E ∈ GLm(k), in [4], Bichon constructed a free Yetter-Drinfeld reso- lution of the trivial module over B(E). Consequently, B(E) is smooth of di- mension 3 and satisfies a Poincar´e duality between Hochschild homology and cohomology. Namely, B(E) is a twisted Calabi-Yau (CY for short) algebra of dimension 3 (Definition 1.6). CY algebras were introduced by Ginzburg in 2006 [9]. They are studied in recent years because of their applications in algebraic geometry and mathematical physics. Twisted CY algebras process similar ho- mological properties as the CY algebras and include CY algebras as a subclass. 2000 Mathematics Subject Classification. 16E65, 16E40, 16W30, 16W35. Key words and phrases. Hopf-Galois object; Calabi-Yau algebra; Quantum group. 1 2 XIAOLAN YU They are the natural algebra analogues of the Bieri-Eckmann duality groups [6]. Associated to a twisted CY algebra, there exists a so-called Nakayama automorphism. The Nakayama automorphism is the tool that enables one to describe the duality between Hochschild cohomology and homology. This au- tomorphism is unique up to an inner automorphism. A twisted CY algebra is CY if and only if its Nakayama automorphism is an inner automorphism. Aubriot gave the classification of Hopf-Galois objects (Definition 1.1) of the algebras B(E) in [1]. Let E ∈ GLm(k) and F ∈ GLn(k) with m, n > 2. Define B(E, F ) to be the algebra with generators uij, 1 6 i 6 m, 1 6 j 6 n, subject to the relations: F −1utEu = In; uF −1utE = Im. On one hand, if tr(E−1Et) = tr(F −1F t), then B(E, F ) is a left B(E)-Galois object. On the other hand, if A is a left B(E)-Galois object, then there exists F ∈ GLn(k) satisfying tr(E−1Et) = tr(F −1F t), such that A ∼= B(E, F ) as Galois objects. The algebras B(E) are twisted CY algebras. One naturally asks whether B(E, F ) are twisted CY as well. More generally, the CY property of Hopf algebras was discussed in [7] and [12], we would like to know whether a Hopf-Galois object of a twisted CY Hopf algebra is still a twisted CY algebra. Hopf-Galois objects can be described by the language of cogroupoids. A cogroupoid C consists of a set of objects ob(C) such that for any X, Y ∈ ob(C), C(X, Y ) is an algebra. Moreover, for any X, Y, Z ∈ ob(C), there are morphisms ∆Z X,Y , εX and SX,Y . They satisfy certain axioms. If C(X, Y ) 6= 0 for any ob- jects X, Y , then C is called connected. The details about cogroupoids can be found in Subsection 1.2. Let H be a Hopf algebra and let A be a left H-Galois object. Then there exists a connected cogroupoid C with two objects X, Y such that H = C(X, X) and A = C(X, Y ). Therefore, the theory of Hopf- Galois objects is actually equivalent to the theory of connected cogroupoids. The following is the main result of this paper. Theorem 0.1. Let C be a connected cogroupoid satisfying that SX,Y is bijective for any X, Y ∈ ob(C). Let X be an object in C such that C(X, X) is a twisted C(X,X) = kξ, where ξ : C(X, X) → k is an algebra homomorphism. Then for any Y ∈ ob(C), C(X, Y ) is twisted CY of dimension d with Nakayama automorphism µ defined as CY algebra of dimension d with homological integral R l µ(x) = ξ(xX,X 1 )SY,X(SX,Y (xX,Y 2 )), for any x ∈ C(X, Y ). HOPF-GALOIS OBJECTS OF CALABI-YAU HOPF ALGEBRAS 3 The homological integral of a twisted CY algebra will be explained in Sub- section 1.4. Sweedler's notation for cogroupoids will be recalled in Subsection 1.2. The assumption that SX,Y is bijective for any X, Y ∈ ob(C) is to make sure that SY,X ◦ SX,Y is an algebra automorphism. This assumption is not so restrictive, the examples of cogroupoids mentioned in [5] all satisfy this as- sumption. Actually, if C is a connected cogroupoid such that for some object X, C(X, X) is a Hopf algebra with bijective antipode, then SX,Y is bijective for any objects X, Y . This will be explained in Remark 2.6. As a direct corol- lary of the main theorem, we obtain the following description about the CY property of Hopf-Galois objects. Corollary 0.2. Let H be a twisted CY Hopf algebra of dimension d with bijective antipode. Then any Hopf-Galois object A over H still is a twisted CY algebra of dimension d. This result generalizes a number of previous results in the literature, such as those of [12] for Sridharan algebras, or [19]. An important class of Hopf-Galois objects are cleft Galois objects. Let H be a Hopf algebra. A left H-cleft object is a left H-Galois object A, such that A ∼= H as left H-comodules. Left H-cleft objects are just the algebras Hσ−1, where σ is a 2-cocycle on H (the definition of the algebras Hσ−1 can be found in Subsection 3.1). The algebras Hσ−1 are a part of the so-called 2-cocycle cogroupoid. As an application of Theorem 0.1, we obtain that if H is a Hopf algebra with bijective antipode, such that it is twisted CY, then its left cleft Galois objects are all twisted CY. The dual of this result coincides with Theorem 2.23 in [19]. However, the proof in this paper is more direct. As another application, we obtain that Hopf-Galois objects of the algebras B(E), namely the algebras B(E, F ), are all twisted CY. Theorem 0.3. Let E ∈ GLm(k), F ∈ GLn(k) with m, n > 2, and tr(E−1Et) = tr(F −1F t). The algebra B(E, F ) is a twisted CY algebra with Nakayama au- tomorphism µ defined by µ(u) = (Et)−1Eu(F t)−1F. This paper is organized as follows. In Section 1, we recall necessary prelimi- naries and fix some notations. Our main result is proved in Section 2. In the final Section 3, we give two applications of the main theorem. 4 XIAOLAN YU 1. Preliminaries Throughout this paper, the unadorned tensor ⊗ means ⊗k and Hom means Homk. Given an algebra A, we write Aop for the opposite algebra of A and Ae for the enveloping algebra A ⊗ Aop. The category of left (resp. right) A-modules is denoted by Mod-A (resp. Mod-Aop). An A-bimodule can be identified with an Ae-module, that is, an object in Mod-Ae For an A-bimodule M and two algebra automorphisms µ and ν, we let µM ν denote the A-bimodule such that µM ν ∼= M as vector spaces, and the bimodule structure is given by a · m · b = µ(a)mν(b), for all a, b ∈ A and m ∈ M . If one of the automorphisms is the identity, we will A as A-bimodules, and that Aµ ∼= A as omit it. It is well-known that Aµ ∼= µ−1 A-bimodules if and only if µ is an inner automorphism. 1.1. Hopf-Galois objects. We recall the definition of Hopf-Galois objects. Definition 1.1. Let H be a Hopf algebra. A left H-Galois object is a left H-comodule algebra A 6= (0) such that if α : A → H ⊗ A denotes the coaction of H on A, the linear map κl : A ⊗ A α⊗1A−−−→ H ⊗ A ⊗ A 1H ⊗m −−−−→ H ⊗ A is an isomorphism. A right H-Galois object is a right H-comodule algebra A 6= (0) such that if β : A → A ⊗ H denotes the coaction of H on A, the linear map κr : A ⊗ A 1A⊗β −−−→ A ⊗ A ⊗ H m⊗1H−−−−→ A ⊗ H is an isomorphism. If L is another Hopf algebra, an H-L-bi-Galois object is an H-L-bicomodule algebra which is both a left H-Galois object and a right L-Galois object. 1.2. Cogroupoid. Definition 1.2. A cocategory C consists of: • A set of objects ob(C). • For any X, Y ∈ ob(C), an algebra C(X, Y ). • For any X, Y, Z ∈ ob(C), algebra homomorphisms ∆Z XY : C(X, Y ) → C(X, Z) ⊗ C(Z, Y ) and εX : C(X, X) → k HOPF-GALOIS OBJECTS OF CALABI-YAU HOPF ALGEBRAS 5 such that for any X, Y, Z, T ∈ ob(C), the following diagrams commute: C(X, Y ) ∆Z X,Y−−−−→ C(X, Z) ⊗ C(Z, Y ) ∆T X,Yy C(X, T ) ⊗ C(T, Y ) 1⊗∆Z −−−−−→ C(X, T ) ⊗ C(T, Z) ⊗ C(Z, Y ) T ,Y ∆T X,Z ⊗1y C(X, Y ) C(X, Y ) P P P P P P P P P P P P ∆Y X,Y P P P P P P P P P P P P 1⊗εY ◗ ◗ ◗ ◗ ◗ ◗ ◗ ◗ ◗ ◗ ◗ ◗ ∆X X,Y ◗ ◗ ◗ ◗ ◗ ◗ ◗ ◗ ◗ ◗ ◗ ◗ εX ⊗1 C(X, Y ) ⊗ C(Y, Y ) / C(X, Y ) C(X, X) ⊗ C(X, Y ) / C(X, Y ). Thus a cocategory with one object is just a bialgebra. A cocategory C is said to be connected if C(X, Y ) is a non zero algebra for any X, Y ∈ ob(C). Definition 1.3. A cogroupoid C consists of a cocategory C together with, for any X, Y ∈ ob(C), linear maps SX,Y : C(X, Y ) −→ C(Y, X) such that for any X, Y ∈ C, the following diagrams commute: C(X, X) εX / k u / C(X, Y ) ∆Y X,X m C(X, Y ) ⊗ C(Y, X) 1⊗SY,X / C(X, Y ) ⊗ C(X, Y ) C(X, X) εX / k u / C(Y, X) ∆Y X,X m C(X, Y ) ⊗ C(Y, X) SX,Y ⊗1 / C(Y, X) ⊗ C(Y, X) From the definition, we can see that C(X, X) is a Hopf algebra for each object X ∈ C. We use Sweedler's notation for cogroupoids. Let C be a cogroupoid. For aX,Y ∈ C(X, Y ), we write ∆Z X,Y (aX,Y ) = aX,Z 1 ⊗ aZ,Y 2 . Now the cogroupoid axioms are (∆T X,Z ⊗ 1) ◦ ∆Z 3 = (1 ⊗ ∆Z T,Y ) ◦ ∆Z X,Y (aX,Y ); X,Y (aX,Y ) = aX,T )aX,Y εX (aX,X 1 ⊗ aT,Z 2 ⊗ aZ,Y 2 = aX,Y = aX,Y 1 1 εY (aY,Y 2 ); / / / / / O O / / / O O 6 XIAOLAN YU SX,Y (aX,Y 1 )aY,X 2 = εX (aX,X 1 )1 = aX,Y 1 SY,X(aY,X 2 ). The following is Proposition 2.13 in [5]. tipodes". It describes properties of the "an- Lemma 1.4. Let C be a cogroupoid and let X, Y ∈ ob(C). (1) SY,X : C(Y, X) → C(X, Y )op is an algebra homomorphism. (2) For any Z ∈ ob(C) and aY,X ∈ C(Y, X), ∆Z X,Y (SY,X(aY,X)) = SZ,X(aZ,X 2 ) ⊗ SY,Z(aY,X 1 ). For other basic properties of cogroupoids, we refer to [5]. The following theorem shows the relation between Hopf-Galois objects and cogroupoids. Theorem 1.5. Let H be a Hopf algebra and let A be a left H-Galois object. Then there exists a connected cogroupoid C with two objects X, Y such that H = C(X, X) and A = C(X, Y ). Proof. This is Theorem 2.11 in [5] (see also [11]). Thus the theory of Hopf-Galois objects is actually equivalent to the theory of connected cogroupoids. In what follows, without otherwise stated, we assume that the cogroupoids mentioned are connected. 1.3. Calabi-Yau algebras. Definition 1.6. An algebra A is called a twisted Calabi-Yau algebra of dimen- sion d if (i) A is homologically smooth, that is, A has a bounded resolution by finitely generated projective Ae-modules; (ii) There is an automorphism µ of A such that (1) as Ae-modules. Exti Ae(A, Ae) ∼=  0, i 6= d Aµ, i = d If such an automorphism µ exists, it is unique up to an inner automorphism and is called the Nakayama automorphism of A. A Calabi-Yau algebra is a twisted Calabi-Yau algebra whose Nakayama automorphism is an inner automorphism. In what follows, Calabi-Yau is abbreviated to CY for short. HOPF-GALOIS OBJECTS OF CALABI-YAU HOPF ALGEBRAS 7 CY algebras were first introduced by Ginzburg in [9]. Twisted CY algebras include CY algebras as a subclass. However, the twisted CY property of non- commutative algebras has been studied under other names for many years, even before the definitions of CY algebras. Rigid dualizing complexes of non- commutative algebras were studied in [17]. The twisted CY property was called "rigid Gorenstein" in [7] and was called "skew Calabi-Yau" in a recent paper [15]. 1.4. Homological integral. A standard tool in the theory of the finite di- mensional Hopf algebras is the notion of integrals. Homological integrals for an AS-Gorenstein Hopf algebra introduced in [13] is a generalization of this notion. The concept was further extended to a general AS-Gorenstein algebra in [7]. The following lemma is probably well-known. example, in [19, Lemma 2.15]. Its proof can be found, for Lemma 1.7. Let A be an augmented algebra such that A is a twisted CY algebra of dimension d with Nakayama automorphism µ. Then A is of global dimension d. Moreover, there is an isomorphism of right A-modules Exti A(Ak, AA) ∼=  0, i 6= d; kξ, i = d, where ξ : A → k is the homomorphism defined by ξ(h) = ε(µ(h)) for any h ∈ A. In a similar way, the following isomorphisms of left A-modules holds: Exti A(kA, AA) ∼=  0, i 6= d; η k, i = d, where η : A → k is the homomorphism defined by η = ε ◦ µ−1. Therefore, if A is a twisted CY augmented algebra, then A has finite global dimension and satisfy the AS-Gorenstein condition. However, A is not nec- essarily Noetherian. It is not AS-regular in the sense of [7, Definition. 1.2]. We still call the 1-dimensional right A-module Exti A(Ak, AA) right homological A. Similarly, we call the 1-dimensional left A. integral of A and denoted it by R l A(kA, AA) right homological integral of A and denoted it by R r A-module Extd 8 XIAOLAN YU 2. The CY property of Hopf-Galois objects The aim of this paper is to discuss the CY property of Hopf-Galois objects. Be- cause of the relation between Hopf-Galois objects and connected cogroupoids, this is equivalent to discuss the CY property of connected cogroupoids. We luckily find that the techniques in [7] can be used. Let C be a cogroupoid and X, Y ∈ ob(C). Both the morphisms ∆Y X,X : C(X, X) → C(X, Y ) ⊗ C(Y, X) and SY,X : C(Y, X) → C(X, Y )op are algebra homomorphisms, so (2) (id ⊗SY,X) ◦ (∆Y X,X) : C(X, X) → C(X, Y )e(= C(X, Y ) ⊗ C(X, Y )op) is an algebra homomorphism. This induces a functor L : Mod-C(X, Y )e → Mod-C(X, X). Let M be a C(X, Y )-bimodule, that is, an object in Mod- C(X, Y )e. The C(X, X)-module structure of L(M ) is given by x → m = xX,Y 1 mSY,X(xY,X 2 ), for any m ∈ M and x ∈ C(X, X). In this paper, we view C(X, Y )e as a left and right C(X, Y )e-module respec- tively in the following ways: (3) and (4) (a ⊗ b) → (x ⊗ y) = a (cid:5) x ⊗ y (cid:5) b, (x ⊗ y) ← (a ⊗ b) = x (cid:5) a ⊗ b (cid:5) y. for any x ⊗ y, a ⊗ b ∈ C(X, Y )e. Then L(C(X, Y )e) is a left C(X, X)-module with the module structure given by a → (x ⊗ y) = aX,Y 1 (cid:5) x ⊗ y (cid:5) SY,X(aY,X 2 ) for any a ∈ C(X, X) and x ⊗ y ∈ C(X, Y ). L(C(X, Y )e) is actually a C(X, X)- C(X, Y )e-bimodule. Moreover, the right C(X, Y )e-module structure defined in (4) and the algebra homomorphism (2) make C(X, Y )e a right C(X, X)-module. That is, (5) (x ⊗ y) ← a = x (cid:5) aX,Y 1 ⊗ SY,X(aY,X 2 ) (cid:5) y for any x ∈ C(X, X) and a ⊗ b ∈ C(X, Y )e. Lemma 2.1. (i) L is an exact functor. (ii) L(C(X, Y )e) is a free left (right) C(X, X)-module. (iii) L preserves injectivity. HOPF-GALOIS OBJECTS OF CALABI-YAU HOPF ALGEBRAS 9 Proof. (i) This is clear. (ii) We define the following morphism: φ : C(X, X) ⊗ C(X, Y ) −→ L(C(X, Y )e) x ⊗ y 7−→ xX,Y 1 ⊗ y (cid:5) SY,X(xY,X 2 ). The morphism φ is a left C(X, X)-module homomorphism, where the left C(X, X)-module structure of C(X, X)⊗C(X, Y ) is induced from the left module structure of the factor C(X, X). We check it in detail as follows: 1 (cid:5) xX,Y φ(zx ⊗ y) = zX,Y = zX,Y (cid:5) xX,Y = z → (xX,Y = z → φ(x ⊗ y). 1 (cid:5) xY,X 1 ⊗ y (cid:5) SY,X(zY,X ) (cid:5) SY,X(zY,X 1 ⊗ y (cid:5) SY,X(xY,X 1 ⊗ y (cid:5) SY,X(xY,X )) ) 2 2 2 2 2 ) The first and the second equation hold since ∆Y X,X : C(X, X) → C(X, Y ) ⊗ C(Y, X) and SY,X : C(Y, X) → C(X, Y )op are algebra homomorphisms. More- over, φ is an isomorphism with inverse L(C(X, Y )e) −→ C(X, X) ⊗ C(X, Y ) x ⊗ y 7−→ xX,X 1 ⊗ y (cid:5) SY,X(SX,Y (xX,Y 2 )). Therefore, L(C(X, Y )e) is isomorphic to C(X, X) ⊗ C(X, Y ), it is a free left C(X, X)-module. Similarly, we have the following isomorphism of right C(X, X)-modules: ϕ : C(X, X) ⊗ C(X, Y ) −→ L(C(X, Y )e) x ⊗ y 7−→ xX,Y 1 ⊗ SY,X(xY,X 2 ) (cid:5) y. The right C(X, X)-module structure of C(X, X) ⊗ C(X, Y ) is induced from the one of the factor C(X, X). The inverse of ϕ is given by L(C(X, Y )e) −→ C(X, X) ⊗ C(X, Y ) x ⊗ y 7−→ xX,X 1 ⊗ xX,Y 2 (cid:5) y. Therefore, the module L(C(X, Y )e) is also a free right C(X, X)-module. C(X,X), M ) ∼= L(M ) as left (iii) We first show that HomC(X,Y )e(C(X,Y )eC(X, Y )e C(X, X)-modules for any C(X, Y )-bimodule M . Let ψ be a morphism defined as ψ : HomC(X,Y )e(C(X,Y )e(C(X, Y )e)C(X,X), M ) −→ L(M ) f 7−→ f (1 ⊗ 1). 10 XIAOLAN YU We only need to show that ψ is a left C(X, X)-module homomorphism. For any x ∈ C(X, X), the following equations hold: ψ(x · f ) = (x · f )(1 ⊗ 1) = f ((1 ⊗ 1) ← x) 1 ⊗ SY,X(xY,X = f (xX,Y 1 = x → f (1 ⊗ 1) = x → ψ(f ). )) = xX,Y 2 (cid:5) f (1 ⊗ 1) (cid:5) SY,X(xY,X 2 ) Now let M be an injective left C(X, Y )e-module. We have HomC(X,X)(−, L(M )) ∼= HomC(X,X)(−, HomC(X,Y )e(C(X,Y )e(C(X, Y )e)C(X,X), M )) ∼= HomC(X,Y )e(C(X,Y )e(C(X, Y )e)C(X,X) ⊗C(X,X) −, M ). The last isomorphism follows from the Hom-⊗ adjunction. M is an injective C(X, Y )e-module and C(X,Y )e(C(X, Y )e)C(X,X) is a free right C(X, X)-module by (ii). So the functor HomC(X,Y )e(C(X,Y )e(C(X, Y )e)C(X,X) ⊗C(X,X) −, M ) is exact. Consequently, the functor HomC(X,X)(−, L(M ) is exact, that is, L(M ) is an injective C(X, X)-module. (cid:3) The following is the key lemma of this paper. Lemma 2.2. Let C be a cogroupoid, and X, Y ∈ ob(C). The algebra C(X, X) is a Hopf algebra, then we have the trivial C(X, X)-module k. Let M be a C(X, Y )-bimodule. (i) HomC(X,X)(k, L(M )) = HomC(X,Y )e(C(X, Y ), M ). (ii) Exti C(X,Y )e(C(X, Y ), M ) ∼= Exti C(X,X)(k, L(M )), for all i > 0. Proof. This lemma can be obtained by combining [5, Theorem 7.1] and [7, Lemma 2.4]([10]). Here we give its proof for the sake of completeness. (i) By definition, HomC(X,X)(k, L(M )) = {m ∈ M x → m = ε(x)m, for any x ∈ C(X, X)} and HomC(X,Y )e(C(X, Y ), M ) = {m ∈ M ym = my, for any y ∈ C(X, Y )}. If ym = my for all y ∈ C(X, Y ), then for any x ∈ C(X, X), x → m = xX,Y = xX,Y = (xX,Y = ε(x)m. 1 mSY,X(xY,X 2 (SY,X(xY,X (cid:5) SY,X(xY,X 1 1 2 2 ) )m) ))m HOPF-GALOIS OBJECTS OF CALABI-YAU HOPF ALGEBRAS 11 Conversely, if x → m = ε(x)m for all x ∈ C(X, X), then for any y ∈ C(X, Y ), my = ε(yX,X )myX,Y 2 1 = (yX,Y = yX,Y = yX,Y = ym. 2 1 mSY,X(yY,X 1 m(SY,X(yY,X 1 mε(yY,Y ) 2 2 3 ))yX,Y ) (cid:5) yX,Y 3 ) Hence, the proof is complete. (ii)Let I• be an injective C(X, Y )e-resolution of M . Exti C(X,Y )e(C(X, Y ), M ) are the homologies of the complex HomC(X,Y )e(C(X, Y ), I•). We know from Lemma 2.1 that L is exact and preserve injectivity. Hence L(I•) is an injective resolution of L(M ). So Exti C(X,Y )(k, L(M )) are the homologies of the complex HomC(X,Y )(k, L(I•)). Now, as a consequence of (i), we obtain that Exti C(X,Y )e(C(X, Y ), M ) ∼= Exti C(X,X)(k, L(M )), for all i > 0. (cid:3) Until the end of the section, we assume any cogroupoid C mentioned satisfies that SX,Y is bijective for any X, Y ∈ ob(C). Under this assumption, the morphism SY,X ◦ SX,Y will be an algebra automorphism of C(X, Y ). Let C be a cogroupoid. For an object X ∈ C, C(X, X) is a Hopf algebra. Let ξ : C(X, X) → k be an algebra homomorphism, then we have the right C(X, X)- module kξ. Let Y be another object in C, L(C(X, Y )e) is a C(X, X)-C(X, Y )e- bimodule, the right C(X, Y )e-module structure is given by the equation (4). Then the tensor product kξ ⊗C(X,X) L(C(X, Y )e) is a right C(X, Y )e-module. We have the following lemma. Lemma 2.3. Let C be a cogroupoid, and X, Y ∈ ob(C). Let ξ : C(X, X) → k be an algebra homomorphism. Then we have kξ ⊗C(X,X) L(C(X, Y )e) ∼= (C(X, Y ))µ as right C(X, Y )e-modules, where µ is the algebra automorphism of C(X, Y ) defined by µ(x) = ξ(xX,X 1 )SY,X(SX,Y (xX,Y 2 )) for any x ∈ C(X, Y ). Proof. Define a morphism ϕ : kξ ⊗C(X,X) L(C(X, Y )e) → (C(X, Y ))µ by ϕ(1 ⊗ x ⊗ y) = ξ(xX,X )) and a morphism ψ : (C(X, Y ))µ → )y (cid:5) SY,X(SX,Y (xX,Y 1 2 12 XIAOLAN YU kξ ⊗C(X,X) L(C(X, Y )e) by ψ(x) = 1 ⊗ 1 ⊗ x. We first show that ϕ is well- defined. We have ϕ(1 ⊗ z → (x ⊗ y)) 1 1 1 1 (cid:5) xX,X (cid:5) xX,X (cid:5) xX,X ε(zX,X = ϕ(1 ⊗ (zX,Y = ξ(zX,X = ξ(zX,X = ξ(zX,X = ξ(zX,X 2 = ξ(z)ξ(xX,X = ϕ(ξ(z) ⊗ x ⊗ y). 1 1 1 1 1 ))) 3 (cid:5) x ⊗ y (cid:5) SY,X(zY,X )y (cid:5) SY,X(zY,X )y (cid:5) SY,X(zY,X )y (cid:5) SY,X(SX,Y (zX,Y ) (cid:5) xX,X 3 1 )y (cid:5) SY,X(SX,Y (xX,Y )) 2 2 ) (cid:5) SY,X(SX,Y (zX,Y ) (cid:5) SY,X(SX,Y (zX,Y 2 2 2 )) (cid:5) xX,Y )) (cid:5) SY,X(SX,Y (xX,Y 2 )) ) (cid:5) zY,X )y (cid:5) SY,X(SX,Y (xX,Y 3 2 ) (cid:5) SY,X(SX,Y (xX,Y )) 2 )) 2 Similar calculations show that ϕ and ψ are right C(X, Y )e-module homomor- phisms and they are inverse to each other. (cid:3) Lemma 2.4. Let C be a cogroupoid, and X, Y ∈ ob(C). If C(X, X) is homo- logically smooth, then so is C(X, Y ). Proof. The algebra C(X, X) is homologically smooth, then C(X, X) admits a bounded resolution of finitely generated projective C(X, X)e-modules, say Pd → Pd−1 → · · · → P1 → P0 → C(X, X) → 0. By tensoring the functor k⊗C(X,X) −, we get a projective right C(X, X)-module resolution k⊗C(X,X) Pd → k⊗C(X,X) Pd−1 → · · · → k⊗C(X,X) P1 → k⊗C(X,X) P0 → k → 0 of the trivial C(X, X)-module k (cf. [2, Lemma 2.4]). The module L(C(X, Y )e) is C(X, X)-C(X, Y )e-bimodule, and is free as left C(X, X)-module by Lemma 2.1 (ii), hence the functor − ⊗C(X,X) L(C(X, Y )e), from Mod-C(X, X) to Mod- (C(X, Y )e) is exact and preserves projectivity. Therefore, we obtain a bounded resolution of projective right (C(X, Y )e)-modules: k ⊗C(X,X) Pd ⊗C(X,X) L(C(X, Y )e) → · · · → k ⊗C(X,X) P0 ⊗C(X,X) L(C(X, Y )e) → k ⊗C(X,X) L(C(X, Y )e) → 0 Moreover, each term k ⊗C(X,X) Pi ⊗C(X,X) L(C(X, Y )e) is a finitely generated C(X, Y )e-module, since Pi is a finitely generated C(X, X)e-module. It follows from Lemma 2.3 that k ⊗C(X,X) L(C(X, Y )e) ∼= C(X, Y )µ as right C(X, Y )e- modules, where µ = SY,X ◦SX,Y . In conclusion, we obtain a bounded resolution of C(X, Y )µ by finitely generated projective right C(X, Y )e-modules: k ⊗C(X,X) Pd ⊗C(X,X) L(C(X, Y )e) → · · · → k ⊗C(X,X) P0 ⊗C(X,X) L(C(X, Y )e) → C(X, Y )µ → 0. HOPF-GALOIS OBJECTS OF CALABI-YAU HOPF ALGEBRAS 13 For each term k⊗C(X,X) Pi⊗C(X,X)L(C(X, Y )e), it is well-known that if we twist its right C(X, Y )e-module structure by an automorphism of C(X, Y )e, then it is still a projective right C(X, Y )e-module. Therefore, in the above resolution, if we twist the right module structure of each term by the automorphism µ−1 ⊗1, then we obtain a desired bounded resolution of C(X, Y ) by finitely generated projective C(X, Y )e-modules. The algebra C(X, Y ) is homologically smooth. (cid:3) Now we are ready to provide our main result. Theorem 2.5. Let C be a cogroupoid and let X ∈ ob(C) such that C(X, X) C(X,X) = kξ, where ξ : C(X, X) → k is an algebra homomorphism. Then for any Y ∈ ob(C), C(X, Y ) is twisted CY with Nakayama automorphism µ defined as is a twisted CY algebra with homological integral R l µ(x) = ξ(xX,X 1 )SY,X(SX,Y (xX,Y 2 )), for any x ∈ C(X, Y ). Proof. By Lemma 2.2, for all i > 0, Exti C(X,Y )e(C(X, Y ), C(X, Y )e) ∼= Exti C(X,X)(k, L(C(X, Y )e)). In the proof of Lemma 2.1 (ii), we obtain a left C(X, X)-module isomorphism: φ : C(X, X) ⊗ C(X, Y ) −→ L(C(X, Y )e) x ⊗ y 7−→ xX,Y 1 ⊗ y (cid:5) SY,X(xY,X 2 ). This isomorphism is also an isomorphism of right C(X, Y )e-modules if we en- dow a right C(X, Y )e-module structure on C(X, X) ⊗ C(X, Y ) as follows: (a ⊗ b) ← (x ⊗ y) = axX,X 1 ⊗ y (cid:5) b (cid:5) SY,X(SX,Y (xX,Y 2 )), for any a ⊗ b ∈ C(X, X) ⊗ C(X, Y ) and x ⊗ y ∈ C(X, Y )e. Now we obtain the following right C(X, Y )e-module isomorphisms: Exti ∼= Exti ∼= Exti ∼= Exti ∼= Exti C(X,X)(k, L(C(X, Y )e)) C(X,X)(k, C(X, X) ⊗ C(X, Y )) C(X,X)(k, C(X, X)) ⊗ C(X, Y ) C(X,X)(k, C(X, X)) ⊗C(X,X) C(X, X) ⊗ C(X, Y ) C(X,X)(k, C(X, X)) ⊗C(X,X) L(C(X, Y )e). The algebra C(X, X) is a twisted CY algebra with homological integralR l C(X,X)(k, C(X, X)) = 0 for i 6= d and Extd C(X,X) = C(X,X)(k, C(X, X)) = kξ. Therefore, Exti 14 XIAOLAN YU kξ. We obtain that, Exti C(X,Y )(C(X, Y ), C(X, Y )e) = 0 for i 6= d and C(X,Y )e(C(X, Y ), (C(X, Y ))e) ∼= kξ ⊗C(X,X) L(C(X, Y )e) Extd ∼= C(X, Y )µ, where µ is the algebra automorphism of C(X, Y ) defined by µ(x) = ξ(xX,X 1 )SY,X(SX,Y (xX,Y 2 )). The second isomorphism follows from Lemma 2.3. In conclusion, we have that Exti C(X,Y )e(C(X, Y ), C(X, Y )e) ∼=  0 i 6= d; C(X, Y )µ i = d. Moreover, by Lemma 2.4, the algebra C(X, Y ) is homologically smooth. Now the proof is complete. (cid:3) Remark 2.6. If C is a cogroupoid such that C(X, X) has bijective antipode for some object X, then SX,Y is bijective for any objects X, Y in C. Indeed, let C be a cogroupoid such that C(X, X) has bijective antipode for some object X, then the tensor category of finite dimensional right comodules over C(X, X) has right duals, and so has the category of finite dimensional comodules over C(Y, Y ) for any object Y . Then one can use the right duality to construct inverses to all the antipodes SX,Y , similarly to the proof of Proposition 2.23 in [5]. Combining Theorem 1.5, we obtain the following direct corollary. Corollary 2.7. Let H be a twisted CY Hopf algebra of dimension d with bijective antipode. Then any Hopf-Galois object A over H still is a twisted CY algebra of dimension d. As a consequence of Theorem 2.5, we obtain the following duality between Hochschild homology and cohomology. Corollary 2.8. Let C be a cogroupoid and let X ∈ ob(C) such that C(X, X) is H = kξ, where ξ : C(X, X) → k is an algebra homomorphism. Let Y be any object in C. For any C(X, Y )-bimodule M and for all 0 6 i 6 d, a twisted CY algebra of dimension d with homological integral R l Hi(C(X, Y ), M ) ∼= Hd−i(C(X, Y ), M µ), where µ is the algebra automorphism of C(X, Y ) defined as µ(x) = ξ(xX,X 1 )SY,X(SX,Y (xX,Y 2 )) for any x ∈ C(X, Y ). HOPF-GALOIS OBJECTS OF CALABI-YAU HOPF ALGEBRAS 15 Proof. By Theorem 2.5, the algebra C(X, Y ) is twisted CY of dimension d. Now this corollary is a consequence of [18, Theorem 1]. 3. Applications 3.1. Cleft Hopf-Galois objects. An important class of Hopf-Galois objects are cleft Galois objects. In this subsection, we discuss their CY property. To define cleft Galois objects, we first recall the concept of a 2-cocycle. Let H be a Hopf algebra. A (right) 2-cocycle on H is a convolution invertible linear map σ : H ⊗ H → k satisfying σ(h1, k1)σ(h2k2, l) = σ(k1, l1)σ(h, k2l2); σ(h, 1) = σ(1, h) = ε(h) for all h, k, l ∈ H. The set of 2-cocycles on H is denoted Z 2(H). The convolution inverse of σ, denote σ−1, satisfies σ−1(h1k1, l)σ−1(h2, k2) = σ−1(h, k1l1)σ−1(k2, l2) σ−1(h, 1) = σ−1(1, h) = ε(h) for all h, k, l ∈ H. Let H be a Hopf algebra and let σ be a 2-cocycle on H. The algebra σH is defined to be the vector space H together with the multiplication given by h (cid:5) k = σ(h1, k1)h2k2, for any h, k ∈ H. σH is a right H-comodule algebra with ∆ : σH → σH ⊗ H as a coaction, and is a right H-Galois object. Similarly we have the algebra Hσ−1. As a vector space Hσ−1 = H, its multi- plication is given by h (cid:5) k = h1k1σ−1(h2, k2), for any h, k ∈ H. Hσ−1 is a left H-comodule algebra with coaction ∆ : Hσ−1 → H ⊗ Hσ−1, and Hσ−1 is a left H-Galois object. The following theorem can be found in [14]. Theorem 3.1. Let H be a Hopf algebra and let A be a left H-Galois object. The following assertions are equivalent: (i) There exists a 2-cocycle σ on H such that A ∼= Hσ−1 as left comodule algebras. (ii) A ∼= H as left H-comodules. (iii) There exists a convolution invertible H-colinear morphism φ : H → A. 16 XIAOLAN YU A left H-Galois object is said to be cleft if it satisfies the above equivalent conditions. There is a similar result for right cleft H-Galois objects. Right cleft H-Galois objects are just the algebras σH, where σ is a 2-cocycle. The algebras Hσ−1 and σH are part of the 2-cocycle cogroupoid. Let H be a Hopf algebra and let σ, τ ∈ Z 2(H). The algebra H(σ, τ ) is defined to be the vector space H together with the multiplication given by (6) h (cid:5) k = σ(h1, k1)h2k2τ −1(h3, k3), for any h, k ∈ H. It is easy to see that H(1, 1) (Here 1 stands for ε ⊗ ε) is just the algebra H. For a 2-cocycle σ, H(1, σ) is the algebra Hσ−1 and H(σ, 1) is the algebra σH. Now we recall the necessary structural maps for the 2-cocycle cogroupoid of H. For any σ, τ, ω ∈ Z 2(H), define the following maps: (7) (8) (9) ∆ω σ,τ = ∆ : H(σ, τ ) −→ H(σ, ω) ⊗ H(ω, τ ) h 7−→ h1 ⊗ h2. εσ = ε : H(σ, σ) −→ k. Sσ,τ : H(σ, τ ) −→ H(τ, σ) h 7−→ σ(h1, S(h2))S(h3)τ −1(S(h4), h5). If H has bijective antipode, then Sσ,τ is bijective for any σ, τ ∈ Z 2(H). The inverse of Sσ,τ is given as: (10) S−1 σ,τ : H(τ, σ) −→ H(σ, τ ) x 7−→ σ−1(x5, S−1(x4))S−1(x3)τ (S−1(x2), x1). The 2-cocycle cogroupoid of H, denoted by H, is the cogroupoid defined as follows: (i) ob(H) = Z 2(H). (ii) For σ, τ ∈ Z 2(H), the algebra H(σ, τ ) is the algebra H(σ, τ ) defined in (6). (iii) The structural maps ∆• •,•, ε• and S•,• are defined in (7), (8) and (9) respectively. [5, Lemma 3.13] shows that the maps ∆• •,•, ε• and S•,• indeed satisfy the conditions required for a cogroupoid. It is clear that the 2-cocycle cogroupoid is connected. As an application of Theorem 2.5, we obtain the following theorem. HOPF-GALOIS OBJECTS OF CALABI-YAU HOPF ALGEBRAS 17 Theorem 3.2. Let H be a Hopf algebra with bijective antipode, such that it is H = kξ, where ξ : H → k is an algebra homomorphism. Let σ be a 2-cocycle on H, then the left cleft Galois object Hσ−1 is twisted CY with Nakayama automorphism µ defined by twisted CY with homological integral R l µ(x) = ξ(x1,1 1 )Sσ,1(S1,σ(x1,σ 2 )), for any x ∈ Hσ−1. Proof. The 2-cocycle cogroupoid H is connected. Moreover, Sσ,τ is bijective for any σ, τ ∈ Z 2(H), since H has bijective antipode. The algebra H is just the algebra H(1, 1) and the algebra Hσ−1 is just the algebra H(1, σ). Now this theorem is a direct consequence of Theorem 2.5. (cid:3) Remark 3.3. Under the assumption of Theorem 3.2, dually, we obtain that a right cleft Galois object σH is twisted CY with Nakayama automorphism ν defined by ν(x) = S−1 2 ). This coincides with Theorem 2.23 in [19]. 1 ))ξS(x1,1 σ,1(S−1 1,σ(xσ,1 3.2. Quantum groups associated to the bilinear form. Let E ∈ GLm(k). Recall that the algebra B(E) [8] is the algebra presented by generators (uij)16i,j6m and the relations E−1utEu = Im = uE−1utE, where u is the matrix (uij)16i,j6n, ut is the transpose matrix and Im is the identity matrix. It is a Hopf algebra with the following structures: ∆(uij) = n Xk=1 uik ⊗ akj, ε(uij) = δij , S(u) = E−1utE. The following lemma describes the CY property of the algebras B(E). Lemma 3.4. Let E ∈ GLm(k). The algebra B(E) is a twisted CY algebra of dimension 3 with Nakayama automorphism µ defined by µ(u) = (Et)−1Eu(Et)−1E. Its left homological integral is given by R l homomorphism defined by ξ(u) = (Et)−1E(Et)−1E. B(E) = kξ, where ξ is the algebra Proof. The fact that B(E) is a twisted CY algebra was shown in [16]. Here we show that it can be obtained directly from Theorem 6.1 and Corollary 6.3 in [4]. By [4, Theorem 6.1], the algebra B(E) is homologically smooth and Exti B(E)e(B(E), B(E)e) = 0 for i > 4. 18 XIAOLAN YU B(E)e(B(E), B(E)e) = TorB(E)e [4, Corollary 6.3] shows that Exti for i = 0, 1, 2, 3. where σ is the algebra automorphism given by σ(u) = It is easy to see that σ(B(E)e) ∼= B(E)e as left B(E)e- E−1EtuE−1Et. modules. So we have TorB(E)e (B(E), σ(B(E)e)) = 0 unless i = 0. However, TorB(E)e . The automor- phism σ−1 is just the automorphism µ defined in this lemma. In conclusion, (B(E), σ(B(E)e)) = B(E) ⊗B(E)e σ(B(E)e) ∼= B(E)σ−1 (B(E), σ(B(E)e)) 3−i 0 i Exti B(E)e(B(E), B(E)e) =  0 i 6= 3; B(E)µ i = 3. Now we conclude that B(E) is a twisted CY algebra of dimension 3. It follows from Lemma 2.2 that R l (Et)−1E(Et)−1E. B(E) = kξ, where ξ = ε ◦ µ. That is, ξ(u) = (cid:3) Remark 3.5. (i) In [4], the author worked over the field C. However, the results in that paper, except those in Subsection 6.4 do not depend on the base field. (ii) The homological integral of B(E) can also be obtained by [4, Proposition B(E)(kB(E), B(E)) ∼= η k, where η is the algebra homomorphism defined by η(u) = E−1EtE−1Et. Con- B(E) = Ext3 6.2]. From that proposition, we have R r sequently, R l B(E) = kξ, where ξ = η ◦ S. That is, ξ(u) = (Et)−1E(Et)−1E. The aim of this subsection is to discuss the CY property of the Hopf-Galois objects of the algebras B(E). To describe the Hopf-Galois objects, we recall a cogroupoid B. The algebras B(E) is part of this cogroupoid. Let E ∈ GLm(k) and let F ∈ GLn(k). The algebra B(E, F ) is defined to be the algebra with generators uij, 1 6 i 6 m, 1 6 j 6 n, subject to the relations: (11) F −1utEu = In; uF −1utE = Im. It is clear that B(E) = B(E, E). Now we recall the structural maps for the cogroupoid B. For any E ∈ GLm(k), F ∈ GLn(k) and G ∈ GLp(k), define the following maps: (12) (13) (14) ∆G E,F : B(E, F ) −→ B(E, G) ⊗ B(G, F ) uij 7−→ Pp εE : B(E) −→ k k=1 uik ⊗ ukj, uij 7−→ δij , SE,F : B(E, F ) −→ B(F, E)op u 7−→ E−1utF. HOPF-GALOIS OBJECTS OF CALABI-YAU HOPF ALGEBRAS 19 It is clear that SE,F is bijective. Lemma 3.2 in [5] ensures that with these morphisms we have a cogroupoid. The cogroupoid B is defined as follows: (i) ob(B) = {E ∈ GLm(k), m > 1}. (ii) For E, F ∈ ob(B), the algebra B(E, F ) is the algebra defined as in (11) (iii) The structural maps ∆• •,•, ε• and S•,• are defined in (12), (13) and (14) respectively. The cogroupoid B is not necessarily connected. However, we have the following lemma. Lemma 3.6. ([3],[5, Lemma 3.4]) Let E ∈ GLm(k), F ∈ GLn(k) with m, n > 2. Then B(E, F ) 6= (0) if and only if tr(E−1Et) = tr(F −1F t). This lemma induces the following corollary. Corollary 3.7. Let λ ∈ k. Consider the full subcogroupoid Bλ of B with objects ob(Bλ) = {E ∈ GLn(k), m > 2, tr(E−1Et) = λ}. Then Bλ is a connected cogroupoid. The classification of the Hopf-Galois objects of the Hopf algebras B(E) is given in [1]. Lemma 3.8. Let E ∈ GLm(k) with m > 2. (i) If F ∈ GLn(k) satisfies n > 2 and tr(E−1Et) = tr(F −1F t), then B(E, F ) is a left B(E)-Galois object; (ii) Let A be a left B(E)-Galois object. Then there exists F ∈ GLn(k) with n > 2 and tr(E−1Et) = tr(F −1F t), such that A ∼= B(E, F ); (iii) For F ∈ GLn(k) and G ∈ GLp(k), the algebras B(E, F ) and B(E, G) are isomorphic as left B(E)-comodule algebras if and only if n = p and there is P ∈ GLn(k) such that F = P GP t. Now we give the CY property of the algebras B(E, F ). Theorem 3.9. Let E ∈ GLm(k), F ∈ GLn(k) with m, n > 2 and tr(E−1Et) = tr(F −1F t). The algebra B(E, F ) is a twisted CY algebra with Nakayama au- tomorphism µ defined by µ(u) = (Et)−1Eu(F t)−1F . Proof. E and F are objects in the connected subcogroupoid Bλ, where λ = tr(E−1Et) = tr(F −1F t). Moreover, for any E′, F ′ ∈ ob(Bλ), the morphism 20 XIAOLAN YU SE ′,F ′ is bijective. The Hopf algebra B(E) is twisted CY by Lemma 3.4. Now this theorem is a consequence of Theorem 2.5. Acknowledgement. The author sincerely thanks the referee for his/her valu- able comments and suggestions that helped her to improve the paper quite a lot. This work is supported by grants from NSFC (No. 11301126, No.11571316), ZJNSF (No. LQ12A01028, No. LY16A010003). References [1] T. Aubriot, On the classification of Galois objects over the quantum group of a nonde- generate bilinear form, Manuscripta. Math. 122 (2007), 119 -- 135. [2] R. Berger and R. Taillefer, Poincar´e-Birkhoff-Witt deformations of Calabi-Yau algebras, J. Noncommut. Geom. 1 (2007), no. 2, 241 -- 270. [3] J. Bichon, The representation category of the quantum group of a non-degenerate bilinear form, Comm. Algebra 31 (2003), no. 10, 4831 -- 4851. [4] J. Bichon, Hochschild homology of Hopf algebras and free Yetter-Drinfeld resolutions of the counit, Compos. Math. 149 (2013), no. 4, 658C678. [5] J. Bichon, Hopf-Galois objects and cogroupoids, Rev. Un. Mat. Argentina, 55 (2014), no. 2, 11-69. [6] R. Bieri, B. Eckmann, Groups with homological duality generalizing Poincar´e duality, Invent. Math. 20 (1973), 103 -- 124. [7] K. A. Brown and J. J. Zhang, Dualizing complexes and twisted Hochschild (co)homology for Noetherian Hopf algebras, J. Algebra 320 (2008), no. 5, 1814 -- 1850. [8] M. Dubois-Violette, G. Launer, The quantum group of a non-degenerate bilinear form, Phys. Lett. B 245 (1990), 175 -- 177. [9] V. Ginzburg, Calabi-Yau algebras, arXiv:AG/0612139. [10] V. Ginzburg and S. Kumar, Cohomology of quantum groups at roots of unity, Duke Math. J. 69 (1993), no. 1, 179 -- 198. [11] C. Grunspan, Hopf-Galois systems and Kashiwa algebras, Comm. Algebra 32 (2004), 3373-3389. [12] J. W. He, F. Van Oystaeyen and Y. H. Zhang, Cocommutative Calabi-Yau Hopf algebras and deformations, J. Algebra 324 (2010), no. 8, 1921 -- 1939. [13] D. M. Lu, Q. S. Wu and J. J. Zhang, Homological integral of Hopf algebras, Trans. Amer. Math. Soc. 359 (2007), 4945 -- 4975. [14] S. Montgomery, Hopf algebras and their actions on rings, CBMS Lecture Notes 82, Amer. Math. Soc., 1993. [15] M. Reyes, D. Rogalski and J.J. Zhang, Skew Calabi-Yau algebras and Homological iden- tities, Adv. Math. 264 (2014), 308 -- 354. [16] C. Walton, X. Wang, On quantum groups associated to non-Noetherian regular algebras of dimension 2, arXiv:1503.09185. [17] M. Van den Bergh, Existence theorems for dualizing complexes over non-commutative graded and filtered rings, J. Algebra 195 (1997), 662 -- 679. HOPF-GALOIS OBJECTS OF CALABI-YAU HOPF ALGEBRAS 21 [18] M. Van den Bergh, A relation between Hochschild homology and cohomology for Goren- stein rings, Proc. Amer. Math. Soc. 126 (1998), 1345 -- 1348. Erratum: Proc. Amer. Math. Soc. 130 (2002), 2809 -- 2810. [19] X. L. Yu, F. Van Oysteayen, Y. H. Zhang Cleft extensions of Koszul twisted Calabi-Yau algebras, to appear in Isr. J. Math.. Xiaolan YU Department of Mathematics, Hangzhou Normal University, Hangzhou, Zhejiang 310036, China E-mail address: [email protected]
1106.6086	1	1106	2011-06-30T00:01:06	Rational curves and ruled orders on surfaces	[ "math.RA", "math.AG" ]	We study ruled orders. These arise naturally in the Mori program for orders on projective surfaces and morally speaking are orders on a ruled surface ramified on a bisection and possibly some fibres. We describe fibres of a ruled order and show they are in some sense rational. We also determine the Hilbert scheme of rational curves and hence the corresponding non-commutative Mori contraction. This gives strong evidence that ruled orders are examples of the non-commutative ruled surfaces introduced by Van den Bergh.	math.RA	math	Rational curves and ruled orders on surfaces DANIEL CHAN 1 , KENNETH CHAN University of New South Wales e-mail address:[email protected],[email protected] Abstract We study ruled orders. These arise naturally in the Mori program for orders on projective surfaces and morally speaking are orders on a ruled surface ramified on a bisection and possibly some fibres. We describe fibres of a ruled order and show they are in some sense rational. We also determine the Hilbert scheme of rational curves and hence the corresponding non-commutative Mori contraction. This gives strong evidence that ruled orders are examples of the non-commutative ruled surfaces introduced by Van den Bergh. Throughout, we work over an algebraically closed base field k of characteristic zero. 1 Introduction Over the last decade, orders on projective surfaces have proved to be a fruitful area of research. As sheaves of algebras on surfaces, they provide examples of non-commutative surfaces which can be studied using the vast arsenal of techniques in algebraic geometry. Notable advances include a non-commutative version of Mori's minimal model program for orders on projective surfaces [CI], as well as an Enriques style classification of them [CI],[CK05]. On a coarse level, the classification gives a trichotomy for the so-called terminal orders without "exceptional curves" into: i) del Pezzo orders, ii) (half)-ruled orders and iii) minimal models. Del Pezzo orders have been studied in the generic case since they are examples of Sklyanin algebras (see [ATV90],[ATV91],[A92]), whilst some exotic examples have been studied via the cyclic covering construction (see [CK11] and Lerner's thesis [Ler]). In this paper, we wish to study ruled orders which can be described as follows. We start with a smooth projective surface Z ruled over a curve C say by f : Z −→ C. Let A be a maximal order of rank e2 on Z such that k(A) := A ⊗Z k(Z) is a division ring. In keeping with the philosophy of the minimal model program, we shall further assume that A is terminal (see §2 for a definition), a condition on ramification data which ensures in particular that locally at any closed point, A has global dimension two. If the ramification locus consists of a bisection D together with some fibres, then we will say that A is a ruled order. Unlike the complete local structure of orders on surfaces, which has been well-studied (see [A86], [RV], [CI], [CHI]), little is known about the global structure of orders on projective surfaces and we wish in particular to address this lacuna. To guide our study of ruled orders we pose the motivating question, "In what sense is a ruled order A ruled?". The approach in [CI] essentially corresponds to the idea that the canonical divisor of a ruled order is positive on fibres. However, it frequently occurs that equivalent concepts in commutative algebraic geometry are inequivalent in non-commutative algebraic geometry. Another incarnation of ruledness comes from joint work of the first author with Nyman [CN]. For the purposes of this introduction, a rational curve will be a quotient map A −→ M where M is an A-module supported on a smooth rational curve F , such that as a sheaf on F , i) M is locally free of rank e and ii) h0(M ) = 1, h1(M ) = 0. This generalises the notion of line modules on the quantum projective plane (see [ATV91], [A92]). Naturally, we would like to show that ruled orders have lots of rational curves and compute the moduli of them. Furthermore, interest in these rational curves stems from a non-commutative Mori contraction theorem which takes the following form. Given such a 1This project was supported by the Australian Research Council, Discovery Project Grant DP0880143. 1 rational curve M which is K-non-effective and has self-intersection zero (see §2 for definitions), there is a morphism of non-commutative schemes SpecZ A −→ Y where Y is the Hilbert scheme of deformations of A −→ M (see theorem 2.2 for an explanation of terms and details). Furthermore, Y is a generically smooth (commutative) curve. A natural question thus is Question 1.1 Does a ruled order have a K-non-effective rational curve with self-intersection zero? If so, what are the corresponding component of the Hilbert scheme and non-commutative Mori contraction? A natural place to look for rational curves is to consider fibres F of the fibration f : Z −→ C and to consider rank e locally free quotients of ¯A := A ⊗Z OF with the desired cohomology. This suggests the next Question 1.2 Describe the fibres ¯A of A. In particular, is ¯A "rational" in the sense that H 1( ¯A) = 0? Now if F is not a ramification curve, then ¯A is an order and so embeds in a maximal order, which by Tsen's theorem has the form End F V for some vector bundle on V on F . Thus one need only determine possibilities for V and subalgebras of End F V . Question 1.2 is nevertheless fairly subtle. Indeed, ramification data, and hence the notion of a ruled order is Morita invariant whilst cohomology is not. To select the best "model" in the Morita equivalence class, we will assert the minimality of the second Chern class c2(A). A key tool is the notion of Morita transforms, already introduced by Artin-de Jong in the case of Azumaya algebras, to construct Morita equivalences. The answer to question 1.2 is nicest when F is a fibre which meets the ramification divisor transver- sally (necessarily then in two distinct points). The following comes from theorem 5.2. Theorem 1.3 Let A be a ruled order with minimal c2 in its Morita equivalence class. Let F ⊂ Z be a fibre which meets the ramification locus transversally in two distinct points. Then ¯A := A ⊗Z OF ≃ O  O(−1) ... O(−1)  O(−1) O . . . ··· ··· . . . . . . O(−1)  O(−1) ... O(−1) O .  In particular, H 1( ¯A) = 0. This theorem shows there are e rational curves which arise as quotients of ¯A, namely, the columns of the matrix form above. The natural way to enumerate them is as the points of Spec H 0( ¯A). That suggests that one can get a handle on the Hilbert scheme of rational curves by considering the following setup. Note that by proposition 2.3, f∗A is a commutative coherent sheaf of algebras on C. There is also a natural multiplication map Ψ : A ⊗Z f ∗f∗A −→ A which respects the left A-module structure and right f ∗f∗A-modules structure. Theorem 1.4 Let A be a ruled order with minimal c2 in its Morita equivalence class. Suppose further that every ramified fibre has ramification index equal to deg A (e.g. when deg A is prime). Then i. C ′ := SpecC f∗A is a smooth projective curve. ii. C ′ is a component of the Hilbert scheme Hilb A and Ψ : A⊗Z f ∗f∗A −→ A is the universal rational curve. iii. The non-commutative Mori contraction is given by the "morphism of algebras" f∗A −→ A. This is the main theorem and the proof involves analysing the fibres ¯A of A case by case. The paper has been organised as follows. Section 2 contains background material. We deal with the generic fibre 2 in section 3. To obtain nice results at other fibres, we will have to impose minimality of the second Chern class and so in section 4, we study the procedure of Morita transforms for constructing Morita equivalences. In particular, we compute formulas for how c2 changes with Morita transforms. In sec- tion 5, we study hereditary fibres, namely, those where F intersects the ramification locus transversally in 2 distinct points. In particular, we prove theorem 1.3. To study the other fibres, we embed them in a minimal hereditary order ¯A1 and in section 6, we give a description of ¯A1 under some hypotheses. Sec- tions 8,9,10 deal with the three possible non-hereditary fibres, namely, i) F is tangential to the bisection D, ii) F contains a node of D and iii) F is a ramification curve. The key tool to analysing these cases is the notion of rational filtrations introduced in section 7. The construction of the requisite rational filtration involves the over-order ¯A1. Our analysis shows that with the hypotheses of the theorem, we do indeed have H 1( ¯A) = 0. Finally in section 11, we give the proof of theorem 1.4. Presumably the results of theorem 1.4 hold for all ruled orders and extending results to the general setting is the subject of ongoing work. The problem with ramified fibres is that in this case, ¯A is not an order so our methods do not apply. 2 Background and Setup The theory of the birational classification of orders on projective surfaces developed in [CI] mirrors that of commutative surfaces. We recall it here in this section as well as the theory of non-commutative Mori contractions in [CN]. Let Z be a smooth projective surface. An order on Z is a torsion-free coherent sheaf of algebras A on Z such that k(A) := A ⊗Z k(Z) is a central simple k(Z)-algebra. We can thus consider orders as subalgebras of a central simple algebra, and so order them by inclusion. The maximal orders (with respect to inclusion) correspond to the normal surfaces. Orders are Azumaya on some dense open subset U ⊆ Z. Its complement Z − U is called the ramification locus. It is a union of ramification curves (see [CI, section 2.2] for more details on ramification). The role of smooth models in the birational classification of orders on surfaces is played by terminal orders, which can be defined as follows. We say a maximal order A on Z is terminal if i. the ramification locus is a normal crossing divisor, and ii. if p is a node of the ramification locus lying in the intersection of ramification curves C1, C2, then the ramification of A at Ci is itself totally ramified at p for at least one of the curves C1 or C2. Further details on condition ii) of the definition can be found in [CI, section 2.2]. The complete local structure of terminal orders has been determined in [CI, section 2.3]. To describe this we use the skew-power series ring kζ[[x, y]] := khhx, yii/(yx − ζxy) where ζ ∈ k is a root of unity. Theorem 2.1 Let A be a terminal order on Z and p ∈ Z be a closed point. Then identifying OZ,p with k[[u, v]] appropriately, we have a k[[u, v]]-algebra isomorphism of A ⊗Z OZ,p with a full matrix algebra in for some primitive e-th root of unity ζ. Here xe = u, ye = v and the ramification locus is uv = 0. The ramification indices along u = 0 and v = 0 are ne and e respectively. If one wants, one can use this description as a definition for terminal orders. That is, a maximal order on a smooth projective surface is terminal if and only if complete locally at any closed point, it has the form in the theorem. 3 kζ [[x, y]]   (x) ... (x) kζ[[x, y]] ··· . . . ··· kζ[[x, y]] ··· . . . . . . (x) kζ[[x, y]] ... ...   ⊆ Mn(kζ [[x, y]]) To simplify the treatment from now on, we will almost exclusively consider orders in a division ring only, that is, those such that k(A) is a division ring. We will justify this assumption shortly. One of the main results in [CI], is a theory of Mori contractions for terminal orders on projective surfaces. These Mori contractions were classified and described in complete analogy with the commu- tative case. In particular, the Mori contraction corresponding to a ruled surface has non-commutative counterparts dubbed ruled orders. To define them, we start with a ruled surface f : Z −→ C and let A be a terminal order on Z in a k(Z)-central division ring k(A). We say that A is a ruled order if furthermore, the ramification of A is the union of a bisection D of f with some fibres. In this case, the degree say e := √rank Z A of A is the same as the ramification index of A on D by the Artin-Mumford sequence (see [AM, theorem 1]). Following [AZ94], we consider a non-commutative scheme to be a k-linear abelian category C together with a distinguished object O called the structure sheaf. Given an order A we can obtain such a non- commutative scheme by first setting C = A − Mod the category of quasi-coherent A-modules. For the choice of the structure sheaf, we usually pick O = A itself. This means that the cohomology of A-modules will be their usual Zariski cohomology as sheaves on Z. We will allow ourselves to change O to some local pro-generator P which has the same effect as replacing our order A with the Morita equivalent one A′ := End Z P and "restoring" the structure sheaf to O = A′. Now any terminal order in a central simple k(Z)-algebra is Morita equivalent to a terminal order in a division ring, which explains why we restricted our attention to this special case. We will have further need of Morita equivalences later. We will sometimes write SpecZ A for the non-commutative scheme (C,O) = (A − Mod, A). The notion of non-commutative Mori contractions in [CN] is very different from the Mori contrac- tions for terminal orders in [CI]. To describe the former, we need to introduce some terminology. As motivation, recall that a ruled surface corresponds to the Mori contraction of a K-negative rational curve with self-intersection zero. Let A be a terminal order on a smooth projective surface Z. A ratio- nal curve on SpecZ A is a cyclic (left) A-module M which is pure of dimension one and has cohomology H 0(M ) = k, H 1(M ) = 0. We say that the rational curve M is K-non-effective if H 0(ωA ⊗A M ) = 0 where ωA is the A-bimodule Hom Z (A, ωZ) (see [CK03, section 3, proposition 5] for more information about ωA). Finally we say that the rational curve M has self-intersection zero if the Euler form χ(M, M ) = 2 Xi=0 dimk Exti A(M, M ) =: −M 2 = 0. Recall from [AZ01, section E], that the Hilbert scheme Hilb A parametrising quotients of A is a separated locally finite type scheme which is a countable union of projective schemes. We may thus consider the component Y of the Hilbert scheme containing the point corresponding to a rational curve M and let M be the universal family on Y . We call Y the Hilbert scheme of deformations of M . We summarise some results from [CN] in the special case of terminal orders on surfaces. Below, a 1-critical module is a 1-dimensional module all of whose non-trivial quotients have dimension < 1. Theorem 2.2 Let A be a terminal order on a smooth projective surface Z and M a 1-critical K-non- effective rational curve with self-intersection zero. Let Y be the Hilbert scheme of deformations of M and M be the universal family. Then i. Y is a projective curve which is smooth at the point corresponding to M . ii. If πY : Y × Z −→ Y, πZ : Y × Z −→ Z denote projection maps then the Fourier-Mukai type transform πZ∗(M ⊗Y ×Z π∗ is an exact functor with a right adjoint. Y (−)) : QCoh(Y ) −→ A − Mod 4 Remark: A morphism of non-commutative schemes is just a pair of adjoint functors so we will refer to the Fourier-Mukai transform in ii) as the non-commutative Mori contraction contracting M . Proof. One checks easily that SpecZ A is a non-commutative smooth proper surface in the sense of [CN, § 3] (the Serre functor is given by ωA ⊗A − and the dimension function is the usual dimension of coherent sheaves). We may thus invoke [CN, corollary 9.7] to obtain i) and [CN, theorem 10.2] to obtain ii). We would like to show that ruled orders give examples of non-commutative Mori contractions and examine closely the nature of these contractions. The following elementary result is crucial. Proposition 2.3 Let A be a ruled order on the ruled surface f : Z −→ C. Then the OC -algebra f∗A is commutative and in fact, is a sheaf of integral domains on C. Proof. First note that f∗A is a torsion-free coherent sheaf on C so the subring f∗A ⊗C k(C) ⊂ k(A) is a division ring. Tsen's theorem shows that in fact f∗A ⊗C k(C) is a field so in particular f∗A is commutative. We may thus consider the finite cover C ′ := SpecC f∗A of C. Note that Z ′ := C ′×C Z = SpecZ f ∗f∗A and that A is an (A, f ∗f∗A)-bimodule and hence, also can be considered as a family of A-modules over C ′. Furthermore, there is a natural map Ψ : A ⊗Z f ∗f∗A −→ A fl ⊆ C ′ is the locus where induced by multiplication which is surjective since f ∗f∗A ⊃ OZ. Hence if C ′ A is flat, then there is an induced map h : C ′ fl = C ′ and h is an isomorphism of C ′ with a Hilbert scheme of deformations of a rational curve. The former would follow easily if we knew a priori that C ′ were smooth, but unfortunately, we will only be able to prove smoothness by first showing the Hilbert scheme is smooth and h is well-defined on all of C ′. fl −→ Hilb A. We wish of course to show that C ′ 3 Generic behaviour Consider a degree e ruled order A and the map Ψ : A ⊗Z f ∗f∗A −→ A introduced in section 2. In this section, we study the behaviour of Ψ generically on C. We start by studying f∗A and more generally Rif∗A. First note that A is locally free as a sheaf on Z so is flat over C. The following result has already been noted by Artin and de Jong in [AdJ, proposition 4.3.1]. Proposition 3.1 We have rank f∗A = e and R1f∗A is a torsion sheaf. Proof. Let η ∈ C be the generic point and Aη be the pullback of A to Zη := Z ×C η. Since cohomology commutes with the flat base change η ֒→ C, it suffices to show that dimk(C) H 0(Aη) = e, H 1(Aη) = 0. We first use a formula of Artin-de Jong for the Euler characteristic (see [AdJ, (4.1.2)] or [Chan]) which is easily derived from examining the ´etale local forms of an order. χ(Aη) (deg Aη)2 = χ(OZη ) − 1 2 s Xi=1 (1 − 1 ei ) where e1, . . . , es are the ramification indices of Aη written with multiplicity. Here the multiplicity In of the ramification index ei corresponding to the ramification point pi ∈ Zη is [k(pi) : k(C)]. our case, A is ramified on a bisection union some fibres, so either Aη is ramified on two distinct points with multiplicity one, or one point with multiplicity two. In either case, the ramification indices with multiplicity are e1 = e2 = e. Also, Z is ruled so χ(OZη ) = 1. Thus χ(Aη) = e. Note that H 0(Aη)(z) ≃ H 0(Aη)k(Z) ⊂ k(A) is a commutative subalgebra of k(A) so dimk(C) H 0(Aη) ≤ e. The only possibility is dimk(C) H 0(Aη) = e, H 1(Aη) = 0. 5 It will become apparent later, that R1f∗A is not necessarily zero. The vanishing locus of R1f∗A will be the locus where the map Ψ is well-behaved. Let c ∈ C be a closed point and Ac denote the restriction of A to the fibre f −1(c) ⊂ Z. By cohomology and base change results [Hart, theorem III.12.11], we have R1f∗A ⊗C k(c) ≃ H 1(Ac) and if this is zero, we also have f∗A ⊗C k(c) ≃ H 0(Ac) which we note is commutative. Suppose that the fibre F := f −1(c) is not a ramification curve. Then Ac is an order on F ≃ P1. If furthermore the fibre intersects the ramification locus transversally, then by the local structure theory of terminal orders given in theorem 2.1, Ac is an hereditary order ramified at two points with ramification indices both e. The splitting theorem of [C05], ensures that the left Ac-module Ac is the direct sum of e locally projective Ac-modules, say P1, . . . , Pe, each of rank e. The image of the identity 1 ∈ Ac in each summand Pi is an idempotent. This gives the useful Lemma 3.2 An hereditary order ¯A on P1 which is ramified at two or fewer points has a Peirce decom- position e ¯A = Mi,j=1 OP1 (pij ) for some divisors pij ∈ Div P1. We may now describe the behaviour of Ψ : A ⊗Z f ∗f∗A −→ A for generic values of c. Theorem 3.3 Let c ∈ C be a closed point such that f −1(c) intersects the ramification locus of A transversally and furthermore, H 1(Ac) = 0. Also, let C0 ⊆ C be the dense open set of such points. Then i. We have Ac ≃   O O(−1) O(−1) ... O(−1) O . . . ··· ··· . . . . . . O(−1)  O(−1) ... .  O(−1) O ii. π : C ′ = SpecC f∗A −→ C is ´etale over C0. iii. Suppose c′ ∈ C ′ is a point in the pre-image of c ∈ C0. Then A ⊗C ′ k(c′) ≃ OP1 ⊕ OP1 (−1)⊕(e−1). In particular, A is flat over π−1(C0). iv. For c′ ∈ π−1(C0), the A-module A ⊗C ′ k(c′) is a 1-critical K-non-effective rational curve with self-intersection zero. Proof. To see part i), note that H 1(Ac) = 0 implies deg pij ≥ −1 for all i, j. Also, pii = 0 so H 0(Ac) = e forces in fact deg pij = −1 for all i 6= j. Let ε1, . . . , εe be the e diagonal idempotents in i). Then f∗A ⊗C k(c) = H 0(Ac) = kε1 × . . . × kεe. Thus the pre-image of c in C ′ consists of e distinct points so π is ´etale over C0. To prove iii), note that c′ corresponds to an idempotent of H 0(Ac), say εi. Then Ψ ⊗C ′ k(c′) is just the multiplication map Ac ⊗k kεi −→ Acεi. Now Acεi is just the i-th column of Ac so is isomorphic to OP1 ⊕ OP1 (−1)⊕(e−1). Flatness now follows from the fact that the Hilbert polynomial is constant (see [Pot, theorem 4.3.1] or [Hart, theorem III.9.9]). We prove iv). Let M = A ⊗C ′ k(c′) which we note is certainly 1-critical. Also h0(M ) = 1 and h1(M ) = 0 so M is indeed a rational curve. Now if D denotes the bisection that A is ramified on, then [CK, § 3, proposition 5] shows that we can write the canonical bimodule ωA in the form I⊗ZO(F ) where 6 F is a linear combination of fibres and I is an ideal of A such that I e = A⊗ZO(−D−F ′) for some F ′ ≥ 0 a linear combination of fibres. It follows that ωA ⊗AM ≃ IM . Note that M is generated as an A-module by the unique copy of O in M , so if h0(IM ) 6= 0 then IM = M , a contradiction since I eM ≃ M (−D). This shows that M is K-non-effective. We need to show that the Euler form χ(M, M ) = 0. Now HomA(M, M ) = k and Serre duality gives Ext2 A(M, M ) = k. Note first that for c1 ∈ C distinct from c we have χ(M, Ac) = χ(M, Ac1 ) = 0 since M and Ac1 have disjoint support. Serre duality shows that Ext2 A(M, Ac) = 0 while direct computation from the matrix form in i) shows that HomA(M, Ac) = k. Hence Ext1 A(M, Ac) = k. We are thus reduced to showing that dimk Ext1 A(M, M ) ≥ 1. This follows since the family A/C ′ gives non-trivial flat deformations of M . This completes the proof of the theorem. A(M, M ) = 0 so we need only show that Ext1 A(M, M ) is a direct summand of Ext1 The above proof can now be used to cover some cases where f −1(c) does not intersect the ramification locus transversally. Proposition 3.4 Suppose c ∈ C is a closed point with H 1(Ac) = 0 and c′ ∈ C ′ lies over c. Let M = A ⊗C ′ k(c′) = Ac ⊗H 0(Ac) k(c′). If M is locally free of rank e and χ(M ) = 1 then A/C ′ is flat at c′ and M is a K-non-effective rational curve with self-intersection zero. Proof. We know that M is a quotient of Ac so H 1(Ac) = 0 implies that H 1(M ) = 0. Thus H 0(M ) = k so M is indeed a rational curve. In fact, M ≃ OP1 ⊕ OP1(−1)⊕(e−1) so as before, we see that A/C ′ is flat over c′. Continuity of Euler characteristic shows that M has self-intersection zero. The proof of K-non-effectivity in the theorem shows that M is also K-non-effective. It will be convenient to introduce the following Definition 3.5 We say that the closed fibre F = f −1(c) ⊂ Z is good if H 1(Ac) = 0 and the con- clusions of theorem 3.3iii),iv) hold. By proposition 3.4, F is good if H 1(Ac) = 0 and for each c′ ∈ π−1(c) ⊂ C ′ we have that A ⊗C ′ k(c′) is a 1-critical A-module whose underlying sheaf is iso- morphic to OP1 ⊕ OP1(−1)⊕(e−1). The good locus is the set Cgood ⊆ C of such points c where f −1(c) is good. fl denotes the locus where A is flat over C ′. The next result shows how good fibres are Recall that C ′ indeed good for our purposes. Proposition 3.6 The dense open set C0 ⊆ C of theorem 3.3 is contained in Cgood so in particular Cgood is dense open. Also, π−1(Cgood) ⊆ C ′ fl. The map Ψ : A ⊗Z f ∗f∗A −→ A induces an open immersion h : π−1(Cgood) −→ Hilb A. Furthermore, π−1(Cgood) is smooth. Proof. The theorem shows C0 ⊆ Cgood. Note that A is flat over π−1(Cgood) since the Hilbert polyno- mial is constant so π−1(Cgood) ⊆ C ′ fl and the map h is well-defined. Furthermore, theorem 2.2 ensures that Hilb A is smooth 1-dimensional at all points of im h. It thus suffices now to show that h is gener- ically injective. The easiest way to see this is to consider a point c ∈ C0 and the matrix form for Ac given in theorem 3.3. Then the points of π−1(c) correspond to the columns of this matrix form which correspond to distinct points of Hilb A. Naturally, we would like to show that all fibres are good. 4 Morita transforms The previous section shows that the condition H 1(Ac) = 0 is extremely important to make proofs work. To attain this condition, we pass to a Morita equivalent order. The question of which order is a little delicate. One way to restrict the choices is to insist that c2(A) is minimal. We note that by results of Artin-de Jong [AdJ], there is a lower bound for c2 for maximal orders in a fixed division ring. In the case of terminal orders, any two such orders are Morita equivalent by [CI, corollary 2.13]. 7 We describe here the method of Morita transforms for changing to a Morita equivalent order in the same central simple algebra. This was called "elementary transformations" by Artin and de Jong, a term we have avoided since it may have an alternate meaning when dealing with ruled orders. Artin-de Jong's treatment was restricted to Azumaya algebras (see [AdJ, § 8.2] or [dJ, section 2]). proof. We thank Michael Artin and Johan de Jong for allowing us to include the following result and their Proposition 4.1 Let Z be a smooth projective surface and Q be a k(Z)-central division ring of degree e. Given these data, there is a constant γ such that for any maximal order A in Q we have χ(A) ≤ γ. In particular, there is a lower bound on the second Chern class c2(A) depending only on Z and Q. Proof. We quote almost verbatim from [AdJ, § 7.1]. From the ´etale local form of maximal orders at codimension one points, c1(A) is independent of the choice of maximal order (see [AdJ, § 7.1] or [Chan]). Hence we need only bound the Euler characteristic. Consider the reduced determinant map det : A −→ OZ. If L is a line bundle then this map extends to a morphism det : A ⊗Z L −→ L⊗e, and hence gives a map of affine varieties H 0(A ⊗Z L) −→ H 0(L⊗e). Now det is multiplicative and Q a division ring, so the fibre det−1(0) consists of exactly one point, namely 0. Chevalley's theorem then shows dimk H 0(A ⊗Z L) ≤ dimk H 0(L⊗e) (1) In particular, h0(A) ≤ h0(OZ). We now bound h2(A). By Serre duality we have h2(A) = h0(A∗⊗Z ωZ). We may embed A∗ in D (for example using the reduced trace map) and so assume that A∗ ⊂ A⊗ZOZ (∆) for a sufficiently ample divisor ∆. Then equation (1) shows that h2(A) ≤ h0(ω⊗e Z (e∆)). The result follows. Remark: From a strictly logical point of view, we will only need the result concerning the upper bound on χ(A). The result on c2(A) has been included since it provides the best context for presenting results. Consider now a terminal order A on a smooth surface Z and F ⊂ Z a smooth curve. Let ¯A := AF and I < A be a left ideal containing A(−F ) such that M := A/I is an ¯A-module which is pure of dimension one in the sense that no submodules are supported at closed points. Thus I is reflexive. In fact more can be said. Proposition 4.2 Let A be a degree d terminal order on a quasi-projective surface Z. Let P be an A- module which is reflexive as a sheaf on Z and with the same rank as A. Then P is a local progenerator which induces a Morita equivalence between A and the maximal order A′ := End A P . In fact, for any closed p ∈ Z we have an isomorphism of completions Pp ≃ Ap. Proof. First note that Z is smooth so P is locally free as a sheaf on Z. Hence P is locally projective by [Ram, proposition 3.5]. Note that k(Z) ⊗Z P ≃ k(A) so that A′ is also an order in k(A). Now P is a reflexive sheaf so the same is true of A′. By Auslander-Goldman's criterion [AG, theorem 1.5], maximality of A′ will follow from maximality at all codimension one points. Let C be j a prime divisor. We check maximality at C by computing the discriminant ideal m C there, where mC ⊳ OZ,C is the maximal ideal of the local ring at the generic point of C. Let e be the ramification 8 index of A at C, f = d/e and p ∈ C be a general point so that either A is Azumaya at p or p lies on the smooth locus of the ramification. Then by theorem 2.1 we know Ap ≃ Sf ×f where S ≃   k[[x, y]] (x) ... (x) . . . ··· ··· ··· k[[x, y]] k[[x, y]] ... ... . . . (x) k[[x, y]] .   Here, we may identify k[[x, y]] with the complete local ring OZ,p and x = 0 is a local equation for C. Note that there are e distinct indecomposable projective Ap-modules, say P1, . . . , Pe corresponding to the columns of S above, and that Ap ≃ P f 1 ⊕ . . . ⊕ P f e . We seek to show that Pp ≃ Ap from which it will follow that the discriminant ideals of A and A′ are equal and hence, A′ will also be maximal. Now we can write Pp ≃ P f1 where P fi = d. The discriminant ideal of A′ is m 1 ⊕ . . . ⊕ P fe e j C where j = 1 2hd(d − 1) − e Xi=1 fi(fi − 1)i = 1 2(cid:0)d2 − e Xi=1 f 2 i (cid:1). Now j is maximised precisely when all the fi equal f . On the other hand, m discriminant ideal of the maximal order containing A′. Hence all the fi are equal and Pp ≃ Ap. To finish the proof of the proposition, we need only show that Pp ≃ Ap even for points p at the nodes of the ramification locus. This follows using the same computation as above, but using the complete local structure of terminal orders at such nodes. j C is contained in the We call A′ := End A I the Morita transform of A associated to M . We wish to examine how c2 changes with Morita transforms. Following Artin-de Jong, we consider the subalgebra B = A ∩ A′ which we write as End A(I ⊂ A), the sheaf of endomorphisms of A which stabilise I. From this point of view, we obtain the exact sequence 0 −→ B −→ A φ −→ Hom A(I, M ). Now right multiplication by any section of A clearly sends A(−F ) into A(−F ) so φ factors through A −→ Hom ¯A(M ′, M ) where M ′ := I/A(−F ). We have thus an exact sequence 0 −→ B −→ A −→ Hom ¯A(M ′, M ) −→ T −→ 0 for some sheaf T supported on F . We can also write B = End A(A(−F ) ⊂ I) from which we see there is also an exact sequence of the form for some sheaf T ′ supported on F . 0 −→ B −→ A′ −→ Hom ¯A(M (−F ), M ′) −→ T ′ −→ 0 Proposition 4.3 With the above notation we have i. T = Ext 1 ¯A(M, M ), T ′ = Ext 1 ¯A(M ′, M ′). 9 ii. c2(A′) − c2(A) = χ(Hom ¯A(M ′, M ))) − χ(Hom ¯A(M (−F ), M ′)) − χ(T ) + χ(T ′). Proof. We calculate T ′ as follows. Let ¯I = I ⊗Z OF which we note is a locally projective ¯A-module. We have the exact sequence 0 −→ M (−F ) −→ ¯I −→ M ′ −→ 0. Now the map A′ −→ Hom ¯A(M (−F ), M ′) factors through the natural quotient map The formula for T ′ follows now from the exact sequences A′ −→ A′ ⊗Z OF = End ¯A ¯I. ¯A( ¯I, M (−F )) = 0 ¯A(M ′, M ′) −→ Ext 1 The computation for T is easier and uses the same method so will be omitted. ¯I −→ Hom ¯A( ¯I, M ′) −→ Ext 1 Hom ¯A( ¯I, M ′) −→ Hom ¯A(M (−F ), M ′) −→ Ext 1 End ¯A ¯A( ¯I, M ′) = 0. Part ii) follows from the fact that the rank and first Chern classes of A and A′ are the same so Riemann-Roch (see [Pot, §91, p.154]) gives c2(A′) − c2(A) = χ(A) − χ(A′). Proposition 4.4 Suppose further that F is not a ramification curve so that ¯A is an order. Then i. With the above notation, χ(T ) = χ(T ′). ii. If A has minimal c2 in its Morita equivalence class, then χ(Hom ¯A(M ′, M ))) ≥ χ(Hom ¯A(M (−F ), M ′)). In view of proposition 4.3, it suffices to prove part i). Now M, M ′ are torsion-free sheaves Proof. on F so are locally projective ¯A-modules everywhere except possibly where the ramification locus of A intersects F with multiplicity ≥ 2. We may thus calculate the Ext-sheaves T, T ′ by going to the complete local rings at such points. We apply two long exact sequences in cohomology related to the exact sequence 0 −→ M ′ −→ ¯A −→ M −→ 0. One application shows that Ext 1 quence ¯A(M ′, M ′) ≃ Ext 2 ¯A(M, M ′). Another application gives the exact se- Ext 1 ¯A(M, ¯A) −→ Ext 1 ¯A(M, M ) −→ Ext 2 ¯A(M, M ′) −→ Ext 2 ¯A(M, ¯A). Now for any p ∈ F , the completion A ⊗Z OZ,p is regular of dimension two [CI, theorem 2.12] so ¯Ap := ¯A⊗F OF,p has injective dimension one. Thus Ext 2 ¯A(M, ¯A) = 0. Also, M is torsion-free and hence, a first syzygy so also Ext 1 Proposition 4.5 Let A be a degree e ruled order with minimal c2 in its Morita equivalence class. Let F ⊂ Z be a fibre of the ruling which is not a ramification curve and consider an exact sequence of the form ¯A(M, ¯A) = 0. This completes the proof of the proposition. 0 −→ M ′ −→ ¯A −→ M −→ 0 where M is locally free as a sheaf on F . Then 2χ(M ) ≥ e + ν(M ) where ν(M ) := χ(Hom ¯A(M, M )) − χ(Hom ¯A(M ′, M ′)) − χ(Ext 1 ¯A(M, M )) + χ(Ext 1 ¯A(M, M ′)). 10 Proof. We use the long exact sequences 0 −→ Hom ¯A(M, M ) −→ M −→ Hom ¯A(M ′, M ) −→ Ext 1 0 −→ Hom ¯A(M, M ′) −→ M ′ −→ Hom ¯A(M ′, M ′) −→ Ext 1 ¯A(M, M ) −→ 0 ¯A(M, M ′) −→ 0. From proposition 4.4ii), we obtain the inequality 0 ≤ χ(Hom ¯A(M ′, M ))) − χ(Hom ¯A(M, M ′)) = χ(M ) − χ(Hom ¯A(M, M )) + χ(Ext 1 ¯A(M, M )) − χ(M ′) + χ(Hom ¯A(M ′, M ′)) − χ(Ext 1 ¯A(M, M ′)) The proposition now follows on noting that χ(M ′) + χ(M ) = χ( ¯A) = e by continuity of Euler charac- teristic and theorem 3.3. Remark: As will be seen in lemma 8.4, one can compute ν(M ) locally. 5 Hereditary fibres We return to the study of our degree e ruled order A on the ruled surface f : Z −→ C. In this section, we examine goodness of fibres F of f : Z −→ C which meet the ramification locus transversally. This corresponds to the case where ¯A := AF is an hereditary order. We start with a lemma we will use more generally. Lemma 5.1 Let ¯A be a degree e hereditary order on F = P1 which is totally ramified at some point p (that is, the ramification index of ¯A at p is e). Suppose that ¯A has a complete set of e idempotents and that the corresponding Peirce decomposition has the form ¯A = e Mi,j=1 OP1(pij ). Then for i 6= j we have deg pij + deg pji < 0. Proof. Let dij = deg pij. Note that closure under multiplication ensures that dij + dji ≤ 0. We wish to derive a contradiction from the assumption dij + dji = 0. Let ε ∈ H 0( ¯A) be the idempotent which has a 1 in the i and j-th diagonal entry and 0s elsewhere. Now ε ¯A ε is the Azumaya algebra End(O ⊕ O(dij )) so ε( ¯A⊗F k(p))ε ≃ k2×2. If J is the Jacobson radical of the ring ¯A⊗F k(p), then since ¯A is totally ramified at p, we know that ¯A⊗F k(p)/J ≃ ke. However, the nilpotent ideal εJε ⊳ ε( ¯A⊗F k(p))ε must be 0. This contradicts the fact that ¯A⊗F k(p)/J is commutative. The lemma is proved. Theorem 5.2 Let A be a degree e ruled order such that c2(A) is minimal in its Morita equivalence class. Then for every fibre F which meets the ramification locus transversally, we have H 1( ¯A) = 0 so the results of theorem 3.3 apply. In particular, F is good. Proof. From lemma 3.2, there is a Peirce decomposition ¯A = e Mi,j=1 OP1 (pij ) for some divisors pij ∈ Div P1 of degree dij. Note that pii = 0 for all i. Let Pj be the locally projective ¯A-module corresponding to the j-th column ⊕i O(pij). Note that Hom ¯A(Pi, Pj) = O(pij ). We wish first to show that dij < 0 for i 6= j. To this end, let A′ be the Morita transform of A associated to the quotient Pj . We use proposition 4.3 with M = Pj, M ′ = ⊕l6=jPl both locally projective to see 0 ≤ c2(A′) − c2(A) = Xl6=j dlj −Xl6=j djl. 11 Adding djj to both sums shows that Pl dlj ≥ Pl djl. We can also apply a Morita transform to A associated to the quotient ⊕l6=jPl to obtain the reverse inequality so Xl dlj = Xl djl. We argue by contradiction and assume without loss of generality that d12 ≥ 0 so by closure under multiplication we have dl2 ≥ dl1, d1l ≥ d2l for any l. Then d1l ≥ Xl dl2 ≥ Xl dl1 = Xl Xl d2l = Xl dl2. Hence equality must hold throughout, which in turn implies that dl2 = dl1, d1l = d2l for every l. In particular, d12 = 0 = d21. However, ¯A is totally ramified at two points (since A has ramification index e along the bisection D) so the lemma gives a contradiction. We have thus proved that dij < 0 for i 6= j so H 0( ¯A) = e. Furthermore, A is flat over C so χ( ¯A) = e showing H 1( ¯A) = 0. This completes the proof of the theorem. 6 Minimal hereditary orders containing ¯A Our approach to analysing the non-hereditary fibres ¯A of A is to embed them in an hereditary order ¯A1 which is close to ¯A and then bound the possibilities for ¯A1. Our setup is as follows. As usual, let A be a degree e ruled order on the ruled surface f : Z −→ C with minimal c2 in its Morita equivalence class. Pick a closed fibre F which intersects the ramification divisor in exactly one point p (necessarily then with multiplicity 2). Suppose there is an hereditary order ¯A1 containing ¯A := AF which is totally ramified at p. By lemma 3.2, there is a Peirce decomposition ¯A1 = ⊕ O(pij) and the indecomposable summands P1, . . . , Pe of ¯A1 correspond to columns of this matrix form. The following result is standard though we do not know a suitable reference for it. Proposition 6.1 Let P, P ′ be locally projective ¯A1-modules of rank e. i. If d, d′ are the degrees of the invertible OF -modules Hom ¯A1 (P, P ′),Hom ¯A1(P ′, P ), then d + d′ = 0 if P ⊗F OF,p ≃ P ′ ⊗F OF,p and is −1 otherwise. ii. If χ(P ) = χ(P ′), then P ≃ P ′. Proof. We prove part i) first by analysing the algebra End ¯A1 (P ⊕ P ′). Since ¯A1 is totally ramified at p, we have ¯A1p ≃ k[[z]] (z) ... (z)   ··· k[[z]] . . . ··· ··· k[[z]] . . . (z) k[[z]] ... ...   Suppose P ⊗F OF,p, P ′ ⊗F OF,p are isomorphic to the i-th and j-column of the matrix form above. If i = j then End ¯A1(P ⊕ P ′) ⊗F OF,p is the full matrix algebra in OF,p. If i 6= j then let ε ∈ ¯A1p be the idempotent with 1's in the i-th and j-th diagonal entry and 0's elsewhere. Then End ¯A1 (P ⊕ P ′)⊗F OF,p is the hereditary algebra ε ¯A1p ε with discriminant ideal (z). Looking globally we see that End ¯A1 (P ⊕ P ′) ≃ (cid:18) OF OF (q) OF (q′) OF (cid:19) 12 for some divisors q, q′ of degrees d, d′. Conjugating the above matrix if necessary, we may assume that q = dp. Furthermore, End ¯A1 (P ⊕ P ′) is Azumaya away from p since the same is true of ¯A1. Hence we may also assume q′ = d′p. The local computations in the previous paragraph now show that d + d′ = 0 if i = j and d + d′ = −1 otherwise. We now prove part ii). From part i), we may assume without loss of generality that d ≥ 0, so Hom ¯A1(P, P ′) 6= 0. Then any non-zero morphism P −→ P ′ must be an isomorphism, since both are 1-critical and have the same Euler characteristic. We now consider Morita transforms arising from the following setup. Let M1 be a locally projective quotient of ¯A1 and M be the image of ¯A under ¯A1 −→ M1. We complete the following commutative diagram with exact rows. 0 0 / M ′ / ¯A / M / M ′ 1 / ¯A1 / M1 / 0 / 0 In the next lemma we use the number ν(M ) = χ(Hom ¯A(M, M )) − χ(Hom ¯A(M ′, M ′)) − χ(Ext 1 ¯A(M, M )) + χ(Ext 1 ¯A(M, M ′)) introduced in proposition 4.5. Lemma 6.2 We assume the above setup, i. Suppose that when M1 = Pi, we have ν(M ) ≥ 2 − e. Then χ(Pi) ≥ 1. ii. Suppose that χ(Pi) ≥ 1 for all i. Then   O ¯A1 ≃ ··· O ··· ... O(−1) O ... . . . ··· O(−1) O O(−1) . . . ...  .  Proof. To see i), note that when M1 = Pi, the assumption on ν and proposition 4.5, show that χ(Pi) ≥ χ(M ) ≥ 1. To prove ii), we first show that the components O(pij ) of the Peirce decomposition for ¯A1 satisfy Claim 6.3 Let dij := deg pij. Then for distinct i, j we have {dij, dji} = {−1, 0}. Proof. Complete locally at p, the Pi are the e distinct indecomposable projective ¯A1p-modules, so by proposition 6.1i), it suffices to show that dij ≤ 0. Now if dij ≥ 1, then χ(Pj ) ≥ χ(Pi) + e so it in turn suffices to show that for all i we have χ(Pi) ≤ e. Proposition 6.1 ensures that χ(P1), . . . , χ(Pe) are distinct so the assumption on the χ(Pi) forces χ( ¯A1) = e Xi=1 χ(Pi) ≥ 1 + 2 + . . . + e = 1 2 e(e + 1) with equality if and only if {χ(P1), . . . , χ(Pe)} = {1, . . . , e}. But we may also compute χ( ¯A1) as follows. Since ¯A1 is totally ramified at p, we know from the local structure theory of hereditary orders, that we may embed ¯A1 in a maximal order ¯A2 so that χ( ¯A2 / ¯A1) = 1 2 e(e − 1). Also, Tsen's theorem tells us that ¯A2 is trivial Azumaya so χ( ¯A2) = e2 giving χ( ¯A1) = 1 Returning to the proof of the lemma, we wish to show that after conjugating ¯A1 by some permutation matrix, we have dij = 0 for i ≤ j and dij = −1 for i > j. To find this permutation matrix, note that 2 e(e + 1). This proves the claim. 13 / / / / / / / / we can partially order the set of columns {Pi}e i=1 by Pi ≤ Pj if dij = 0. Indeed, the relation is reflexive since dii = 0, anti-symmetric by the claim and transitive by closure under multiplication. Furthermore, the claim shows that the order on {Pi} is in fact a total order. Conjugating by a permutation matrix, we may thus assume that Pi ≤ Pj if and only if i ≤ j which proves the lemma. 7 Rational filtrations and goodness In this section, we introduce the notion of rational filtrations to give a criterion for goodness of a fibre. We let ¯A be a coherent sheaf of algebras on F = P1 which is locally free of rank e2 as a sheaf. Suppose further that k( ¯A) := ¯A ⊗F k(F ) satisfies Hypothesis 7.1 All simple modules have dimension e over k(F ). In particular, any ¯A-module which is rank e torsion-free is 1-critical. Note that any order in the full matrix algebra Me(k(F )) satisfies the hypothesis. We first make the following Definition 7.2 An ¯A-module N is said to be rational if it is torsion-free of rank e and has cohomology h0(N ) = 1, h1(N ) = 0. Equivalently, we have the isomorphism of sheaves N ≃ O ⊕ O(−1)⊕e−1. A filtration of N is said to be rational if all the factors N i+1/N i are rational or zero, in which case we also say that N is rationally filtered. 0 ≤ N 1 ≤ N 2 ≤ . . . ≤ N r = N Proposition 7.3 Any rationally filtered module N satisfies H 1(N ) = 0, rank N = eh0(N ). Proof. As a sheaf on F , N is just the direct sum of the factors in a rational filtration. The proposition follows. Rationally filtered modules enjoy the following nice properties. Lemma 7.4 Let φ : N1 −→ N2 be a non-zero morphism of ¯A-modules where N1 is rationally filtered. Then i. H 1(im φ) = 0. ii. if N2 is rational, then φ is surjective whenever it is non-zero. iii. if N2 is rationally filtered then so are ker φ and coker φ. Proof. Part i) follows from proposition 7.3 and the long exact sequence in cohomology. To prove part ii), suppose now that N2 is rational and φ is non-zero so that im φ has rank e too. Grothendieck's splitting theorem and part i) ensure im φ ≃ ⊕e i=1 O(di) where the di ≥ −1. If φ is not surjective then we must have im φ ≃ O(−1)⊕e. We derive a contradiction as follows. Let N ′ 1 be a rational submodule of N1. The restricted map φ′ : N ′ 1 −→ N2 is not injective since the summand O maps to zero. Hence φ′ = 0. Continuing inductively up the filtration we see that φ = 0. Finally we prove part iii) and so assume there are rational filtrations 0 < N 1 0 < N 1 1 < N 2 2 < N 2 1 < . . . < N r 2 < . . . < N s 1 = N1 2 = N2 We argue by induction on s and assume first that s = 1. We may as well assume that φ is non-zero and so can pick i maximal such that φ(N i 1 −→ N2 is thus an isomorphism 1) = 0. The induced map N i+1 /N i 1 14 by part ii), and its inverse yields a splitting of the map N1/N i N1/N i+1 1 so is rationally filtered. This completes the case s = 1. by N i 1 1 −→ N2. Hence ker φ is an extension of in which case is surjective by part ii). We may thus consider the following 2 For general s, we may assume by induction that φ does not factor through N s−1 the composite N1 −→ N2 −→ N2/N s−1 2 commutative diagram with exact rows 0 0 / K φ′ N1 φ N2/N s−1 2 / N s−1 2 π / N2 / N2/N s−1 2 / 0 / 0 where π is the natural quotient map and K is the appropriate kernel. The s = 1 case shows that K is also rationally filtered. Now ker φ = ker φ′, coker φ = coker φ′ and both are rationally filtered by the inductive hypothesis. The lemma is proved. Theorem 7.5 Let A be a degree e ruled order on a ruled surface Z and F be a ruling of Z such that ¯A := AF satisfies hypothesis 7.1 and is rationally filtered as an ¯A-module. Then F is a good fibre. Proof. By proposition 3.4 and our hypothesis 7.1, we need only show that for any algebra quotient map H 0( ¯A) −→ k we have ¯A⊗H 0( ¯A)k ≃ O ⊕ O(−1)⊕e−1. Let J ⊳ H 0( ¯A) be the Jacobson radical. Now H 1( ¯A) = 0 so cohomology commutes with base change and we see that H 0( ¯A) is commutative. Hence H 0( ¯A)/J ≃ Qr l=1 kεl for some idempotents εl. We need to show that the direct summand ( ¯A / ¯A J)εl of ¯A / ¯A J is rational. First note that ¯A / ¯A J is rationally filtered by lemma 7.4iii) applied to ¯A⊗kJ −→ ¯A, so the same is true of ( ¯A / ¯A J)εl. Also, ( ¯A / ¯A J)εl 6= 0 since it contains εl. It suffices now to check its rank. Now H 1( ¯A J) = 0 since ¯A J is a quotient of a direct sum of dim J copies of ¯A. Hence rank ( ¯A / ¯A J) = eh0( ¯A / ¯A J) = eh0( ¯A) − eh0( ¯A J) ≤ eh0( ¯A) − e dim J = e dim H 0( ¯A)/J = er. It follows that the summand ( ¯A / ¯A J)εl must have rank e and so be rational. This proves the theorem. 8 Fibres tangential to D We assume as usual that A is a degree e ruled order with minimal c2 in its Morita equivalence class. In this section, we consider a fibre F which is tangential to the ramification curve D and let p be the intersection point F ∩ D. We use a subscript p to denote the completion at p, so for example ¯Ap := ¯A⊗F OF,p. Theorem 2.1 can be used to determine ¯Ap. We will pick an hereditary over-order ¯A1 of ¯A so that the inclusion ¯Ap ⊂ ¯A1p is given by k[[z]] (z2) ... (z2)   ··· k[[z]] . . . ··· ··· k[[z]] k[[z]] ... ...   ⊂   (z) ... (z) . . . (z2) k[[z]] ··· k[[z]] . . . ··· ··· k[[z]] . . . (z) k[[z]] ... ...   where we have used the isomorphism OF,p ≃ k[[z]]. Note that ¯A1 is an hereditary order which is totally ramified at the point p, so we are in a good position to use the results of sections 6 and 7. Since ¯Ap has a complete set of idempotents, the rank e torsion-free ¯Ap-modules are easily computed to be (up to isomorphism), those of the form Pij :=   I1 ... Ie   , i, j ∈ {1, . . . , e} 15 / / / / / / / / / / where I1 = ··· = Ii = k[[z]], Ii+1 = ··· = Ij = (z), Ij+1 = ··· = Ie = (z2). The projective ¯Ap-modules are those of the form Pii which correspond to the columns of the matrix form. The columns of ¯A1p are those of the form Pie. First note that we have the following non-split exact sequences 0 −→ Prs −→ Pjs ⊕ Pir −→ Pij −→ 0, for i ≤ r < j ≤ s 0 −→ Prs −→ Pis ⊕ Prj −→ Pij −→ 0, for r < i ≤ s < j We can now classify the indecomposable torsion-free ¯Ap-modules. (2) (3) Proposition 8.1 i. Ext1 ¯Ap (Pij , Prs) = k if i ≤ r < j ≤ s or r < i ≤ s < j and is zero otherwise. ii. Every indecomposable torsion-free ¯Ap-module is isomorphic to Pij for some i, j. Proof. i) Equation (2) gives in particular a partial projective resolution 0 −→ Pij −→ Pii ⊕ Pjj −→ Pij −→ 0 from which we can calculate the ext groups. ii) Let N be a torsion-free ¯Ap-module. We show it is the direct sum of rank e torsion-free modules by induction on rank. Pick a surjection N −→ Pij with j − i minimal. Such surjections exist since ¯Ap is an order in a full matrix algebra over OF,p. We may assume the surjection is not split, otherwise we are done by induction. Consider the corresponding exact sequence E : 0 −→ K −→ N −→ Pij −→ 0 which we view as a non-zero extension E ∈ Ext1 (Pij , K). By induction, K = ⊕lPrlsl . Pick some l ¯Ap such that for r = rl, s = sl, the image of E in Ext1 (Pij , Prs) is non-zero. We obtain thus a morphism ¯Ap of extensions 0 0 / K N Pij / Prs / N ′ / Pij / 0 / 0 Now from part i) we know Ext1 (Pij , Prs) = k and either i ≤ r < j ≤ s or r < i ≤ s < j. In the first ¯Ap case, we have as in equation (2) that N ′ ≃ Pjs ⊕ Pir. This is a contradiction since we then obtain a surjection N −→ N ′ −→ Pir and r − i < j − i. In the second case, we have as in equation (3) that N ′ ≃ Pis ⊕ Prj and again obtain a similar contradiction. We omit the elementary proof of the next presumably well-known result. Lemma 8.2 Let R be a noetherian ring with Jacobson radical J and P be a finitely generated projective R-module. Suppose that R/J is artinian. Given any submodule P ′ < P and surjective homomorphism φ : P −→ P/P ′ we have ker φ ≃ P ′. The possibilities for the exact sequence 0 −→ M ′ −→ ¯A −→ M −→ 0 complete locally at p are If ε1, . . . , εe ∈ ¯Ap are the standard given in the next result. We introduce the following notation. diagonal idempotents with exactly one non-zero entry, then the dimension vector of any ¯Ap-module N is defined to be (dimk ε1N, . . . , dimk εeN ). For example, the dimension vector of Pij/(rad ¯Ap)Pij is (0, . . . , 0, 1, 0, . . . , 0, 1, 0 . . . , 0) where the 1s occur in the i-th and j-th co-ordinate. Proposition 8.3 Let P be a torsion-free ¯Ap-module and φ : ¯Ap −→ P be a surjective morphism. Then there is a partition of {1, . . . , e} into 4 disjoint subsets {i1, . . . , im},{j1, . . . , jm}, Λ, Λ′ such that 16 / / / / / / / / / / i. P ≃ Lλ∈Λ Pλλ ⊕Ll Piljl . ii. ker φ ≃ Lλ∈Λ′ Pλλ ⊕Ll Piljl . Proof. Part i) follows from proposition 8.1ii) and the fact that ¯Ap /rad ¯Ap has dimension vector (1, 1, . . . , 1). We now prove part ii). By lemma 8.2, we may assume that φ is a direct sum of surjections of the form Pilil ⊕ Pjljl −→ Piljl , Pλλ −→ Pλλ for λ ∈ Λ and Pλλ −→ 0 for λ ∈ Λ′. The proposition follows now from equation (2). We wish to use lemma 6.2 to show H 1( ¯A1) = 0 first. We need to calculate ν(M ) = χ(Hom ¯A(M, M )) − χ(Hom ¯A(M ′, M ′)) − χ(Ext 1 ¯A(M, M )) + χ(Ext 1 ¯A(M, M ′)) which is possible since it depends only on local data. Lemma 8.4 Consider an exact sequence of ¯A-modules with M torsion-free. Then 0 −→ M ′ −→ ¯A −→ M −→ 0 ¯A(M, M )) = χ(Ext 1 i. χ(Ext 1 ii. χ(Hom ¯A(M, M )) − χ(Hom ¯A(M ′, M ′)) = 1 ¯A(M, M ′)), and e (rank M − rank M ′). In particular, ν(M ) = 1 e (rank M − rank M ′). Proof. To prove i), note that M is locally projective as an ¯A-module away from p, so the ext groups can be computed after completing at p. Now proposition 8.1i) shows Ext1 (Pij , Prr) = 0 for all i, j, r ¯Ap while proposition 8.3 shows that Mp and M ′ p only differ by projectives so part i) holds. We now verify ii). Suppose that M has rank re. We can embed Hom ¯A(M, M ) in a maximal order B where locally Bp ≃ Or×r F,p . Now Hom ¯A(M, M ) is maximal everywhere except at p so χ(Hom ¯A(M, M )) = χ(B) − colength Hom ¯Ap (Mp, Mp) = r2 − colength Hom ¯Ap (Mp, Mp) where the colength is computed as a subsheaf of Bp. To compute the colength term, we use proposi- tion 8.3 to write m m Mp ≃ Mλ∈Λ Pλλ ⊕ Ml=1 Piljl , M ′ p ≃ Mλ∈Λ′ Pλλ ⊕ Ml=1 Piljl for a partition {i1, . . . , im},{j1, . . . , jm}, Λ, Λ′ of {1, . . . e}. The direct sum decomposition for Mp gives a Peirce decomposition for Hom ¯Ap (Mp, Mp). If we use an isomorphism OF,p ≃ k[[z]], we can write some of these Peirce components as follows. i. Hom ¯Ap (Pλλ, Pρρ) ⊕ Hom ¯Ap (Pρρ, Pλλ) ≃ k[[z]] ⊕ (z2) for λ 6= ρ. ii. Hom ¯Ap (Pλλ, Pij) ⊕ Hom ¯Ap (Pij , Pλλ) ≃ (z) ⊕ (z) if i < λ ≤ j and ≃ k[[z]] ⊕ (z2) otherwise. In either case, their colength (in some appropriate copy of O2 ⊂ Or×r) is 2. We thus find χ(Hom ¯A(M, M )) = r2 − colength Hom ¯Ap (Mp, Mp) = r − colength End ¯Ap P ′ + m2 − m where P ′ = Pi1j1 ⊕ . . . ⊕ Pimjm . Similarly, if M ′ has rank r′e we find χ(Hom ¯A(M ′, M ′)) = r′ − colength End ¯Ap P ′ + m2 − m which completes the proof of the lemma. 17 Theorem 8.5 ¯A1 ≃ O ... ···  O ··· ... O(−1) O ... . . . ··· O(−1) O O(−1)  . . . .   Proof. This follows from lemma 6.2 and the ν computations in lemma 8.4. We wish now to show that ¯A is rationally filtered. To this end, let Pi ≃ O⊕i ⊕ O(−1)⊕e−i be the indecomposable summand of ¯A1 corresponding to the i-th column of the matrix form in theorem 8.5. We work a little more generally and consider an arbitrary order B in ¯A1. We define Bi := B ∩ P>e−i, Si := coker(Bi ֒→ P>e−i) where P>e−i = Pe−i+1 ⊕ . . . ⊕ Pe. We thus obtain a filtration 0 < B1 < . . . < Be := B. When B = ¯A, we wish to apply proposition 4.5 in the following situation. Let φ : ¯A ֒→ ¯A1 −→ P≤e−i := P1 ⊕ . . . ⊕ Pe−i be the composite of the natural inclusion with the natural projection. Then we can i apply a Morita transform with respect to M = im φ and note that M ′ = ker φ = ¯A . Lemma 8.6 χ( ¯Ai) = i so χ( ¯Ai+1 / ¯Ai) = 1. Proof. Using the notation above, we see from lemma 8.4 and proposition 4.5, that χ(M ) ≥ 1 e rank M . i Hence χ( ¯A = i. We proceed to prove the reverse inequality. Now Si is both a quotient of P>e−i ⊗F k(p), which has dimension vector (i, i, . . . , i), and a submodule of ¯A1 / ¯A, which has dimension vector (0, 1, 2, . . . , e − 1). Hence e rank M = 1 ) ≤ e − 1 e rank ¯A i χ(Si) ≤ 0 + 1 + 2 + . . . + i + i + . . . + i where the number of i's on the right hand side is e− i. Also χ(P>e−i) = e + (e− 1) + . . . + (e− i + 1) so This completes the proof of the lemma. χ( ¯Ai) = χ(P>e−i) − χ(Si) ≥ i. Theorem 8.7 Let B ⊂ ¯A1 be an order with χ(Bi) = i. Then the filtration {Bi}e particular, ¯A is rationally filtered and F is a good fibre. i=1 is rational. In Proof. Note that χ(Bi+1/Bi) = 1 so in particular, the injection B/Be−1 ֒→ P1 must be an iso- morphism. We proceed by downward induction on i. Let i be maximal such that Bi+1/Bi ≃ (Bi+1 + P>e−i)/P>e−i is not rational. Then we must have h0(Bi+1/Bi) ≥ 2. Also, P>e−i is a direct sum of line bundles isomorphic to O and O(−1) so we may lift sections to find s ∈ H 0(Bi+1 + P>e−i) such that in the matrix form of theorem 8.5, s is zero in the first e − i − 1 columns and in the (e − i)-th column is i) zero on or below the main diagonal and ii) non-zero strictly above the main diagonal. An elementary matrix computation shows that right multiplication by s induces a non-zero morphism σ : B/Bi+1 −→ (Bi+1 + P>e−i)/P>e−i. Now B/Bi+1 is rationally filtered so lemma 7.4 shows that H 1(im σ) = 0. Also, the cokernel of im σ −→ (Bi+1 + P>e−i)/P>e−i has finite length, so H 1(Bi+1/Bi) = 0 too. This gives a contradiction as then χ(Bi+1/Bi) = h0(Bi+1/Bi) > 1. Thus {Bi} is a rational filtration and the rest of the theorem follows from lemma 8.6 and theorem 7.5. 18 9 Fibres intersecting a node of D Let A be a ruled order of degree e on f : Z → C with minimal c2 in its Morita equivalence class. Let F be a fibre of f which intersects a node p of the ramification divisor D. Following the notation of the previous section, a subscript p will denote completion at p and we fix an isomorphism OF,p ≃ k[[z]]. Let ¯A = A ⊗Z OF and k( ¯Ap) be the central simple k((z))-algebra ¯Ap ⊗F k(F ). By theorem 2.1, we have the following isomorphism ¯Ap ≃ k[[z]]hx, yi (xe − z, ye − z, xy − ζyx) (4) where ζ is a primitive e-th root of unity. The Jacobson radical J of ¯Ap is generated by x and y, so ¯Ap /J ≃ k and ¯Ap is a local ring. Following section 6, we wish to study ¯Ap by embedding it in an hereditary order. To do this we first determine the torsion-free ¯Ap-modules of rank e up to isomorphism. Lemma 9.1 Let I ⊆ J be an ideal of ¯Ap such that z ¯Ap ⊆ I e, and Q be a torsion-free ¯Ap-module of rank e. Then the I-adic filtration on Q has 1-dimensional quotients. In particular, the I-adic and J-adic filtrations on Q are equal. Proof. Note that I Q is torsion-free of rank e, so it suffices by induction to show that Q/I Q ≃ k. Since z ¯Ap ⊆ I e we have a surjection Q/z Q → Q/I eQ, so dimk(Q/I eQ) 6 dimk(Q/z Q) = e. Nakayama's lemma gives I kQ/I k+1Q 6= 0, and since dimk(Q/I eQ) = Pe−1 k=0 dimk(I kQ/I k+1Q), we have Q/I Q ≃ k. From this it follows that I Q = J Q, so the I-adic and J-adic filtrations on Q are equal. Proposition 9.2 Let Qi denote the quotient ¯Ap / ¯Ap ξi where ξi = x − ζi−(e−1)/2y. Then Qi is a torsion-free ¯Ap-module of rank e. Moreover J Qi ≃ Qi+1. Proof. Since x, y skew-commute, one readily computes that Qi is generated by 1, y, . . . , ye−1 as a k[[z]]- module. We wish to show they are free generators so suppose this is not the case and that rank Qi < e. Now ¯Ap is a degree e order so Qi must be torsion and ξi must be invertible in k( ¯Ap). However, a direct computation shows ξe i = 0, and this contradiction shows that Qi is torsion-free of rank e. From lemma 9.1 we know y Qi = J Qi so is the cyclic module generated by ¯y := y + ¯Ap ξi. Now ξi+1y = ζ yξi so ξi+1 annihilates ¯y and there is a natural surjection Qi+1 −→ JQi. It is an isomorphism since Qi+1 and J Qi are free k[[z]]-modules of the same rank. Proposition 9.3 Let Q be a torsion-free ¯Ap-module of rank e. Then Q ≃ Qi for some i = 1, . . . , e. Proof. Note that left multiplication by x or y induces a k[[z]]-module isomorphism Q → J Q. Hence y−1x induces a k[[z]]-module automorphism ϕ of Q and consequently, also of Q/JQ. Let ¯v ∈ Q/J Q be an eigenvector of ϕ. Its eigenvalue must have the form ζi−(e−1)/2 since (y−1x)e = ζ −e(e−1)/2. We lift ¯v to an eigenvector v ∈ Q so that ξiv = 0. By Nakayama's lemma and lemma 9.1, v generates Q so there is a surjection ψ : Qi −→ Q. Rank considerations show ψ is an isomorphism. We define the order ¯A1p as the subring of Endk[[z]](Qe) consisting of all endomorphisms which stabilise J iQe for all i = 1, . . . , e − 1. Since J iQe is an ¯Ap-module for all i > 0, we have ¯Ap ⊆ ¯A1p. The decomposition Qe = k[[z]]¯xe−1 ⊕ k[[z]]¯xe−2 ⊕ ··· ⊕ k[[z]]¯x ⊕ k[[z]]¯1 19 induces an isomorphism Endk[[z]](Qe) ≃ k[[z]]e×e. By lemma 9.1, we have J iQe = xiQe, so we can identify ¯A1p as the subring k[[z]] (z) ... (z)   ··· . . . . . . ··· k[[z]] ··· . . . . . . (z) k[[z]] ... ...   ⊆ k[[z]]e×e. (5) It is clear from this description that ¯A1p is an hereditary order. Furthermore, we can identify ¯Ap as the subring of ¯A1p generated by the following matrices x = 0 ... ... 0 z   1 . . . . . . . . . 0 0 . . . . . . . . . ··· ··· . . . . . . . . . ··· 0 ... 0 1 0   and y = 0 ... ... 0 a1z   a2 . . . . . . . . . 0 0 . . . . . . . . . ··· ··· . . . . . . . . . ··· 0 ... 0 ae 0   (6) where ai = ζi+(e−1)/2. Hence the matrix (ξrs) corresponding to ξj is also zero away from the s − r ≡ 1 mod e diagonal and ξe−j−1,e−j = 0. Let ε1, . . . , εe denote the standard matrix idempotents in ¯A1p. Proposition 9.4 We have an isomorphism of ¯A1p-modules ¯A1p εj ≃ Qe−j. Proof. Note that the composition ϕ : ¯Ap ֒→ ¯A1p → ¯A1p εj is surjective. Now ξe−jεj = 0, so ¯Ap ξe−j ⊆ ker ϕ and we obtain an isomorphism ¯Ap / ¯A ξe−j = Qe−j → ¯A1p εj of ¯Ap-modules. Proposition 9.5 Let ¯Ap(cid:10)x−1y(cid:11) be the smallest subalgebra of k( ¯Ap) containing ¯Ap and x−1y. Then we have ¯A1p = ¯Ap(cid:10)x−1y(cid:11). Moreover, as k-vector spaces, we have ¯A1p = ¯Ap ⊕(cid:16)M k xiyjz−1(cid:17) (7) where the direct sum ranges over 0 < i, j < e and i + j > e. In particular, dimk( ¯A1p / ¯Ap) = e(e − 1)/2. Proof. Firstly x−1y is a diagonal matrix by (6), so ¯Ap(cid:10)x−1y(cid:11) ⊆ ¯A1p. In fact, the entries of ζ −(e−1)/2x−1y are the e e-th roots of unity so k[x−1y] is the subalgebra of ¯A1p consisting of k's along the diagonal in the matrix form (5). This shows that ¯Ap(cid:10)x−1y(cid:11) contains all the standard matrix idempotents ε1, . . . , εe. Now εl−mxmεl generate εl−m ¯A1p εl as a k[[z]]-module, so ¯A1p = ¯Ap(cid:10)x−1y(cid:11). Note that (x−1y)i is a scalar multiple of xe−iyiz−1, so ¯A1p = ¯Ap +   Mi+j>e 0≤i,j<e k[[z]] xiyjz−1  . But ¯Ap = L0≤i,j<e k[[z]]xiyj so the proposition follows. Remark 9.6 One can show, using the criteria in [HN, Section 0.1], that the local ring ¯Ap does not have finite representation type if e > 3. This is in contrast with the situation in section 8. 20 Define the order ¯A1 on F by ¯A1 U = ¯AU where U = F − p and ( ¯A1)p = ¯A1p. Since ¯A is maximal away from p, it is clear that ¯A1 is an hereditary order totally ramified at p. This puts us in a good position to use the results of sections 6 and 7. We wish first to show that H 1( ¯A1) = 0 by invoking lemma 6.2. As in section 8, we will need to carry out some local computations of ext groups and endomorphism rings in preparation for this. Let Ki = ¯Ap ξi, K1i = ¯A1p ξi and J1 be the Jacobson radical of ¯A1p. Note that J1 = x ¯A1p = ¯A1p x. Proposition 9.7 We have a natural isomorphism ¯A1p ≃ Q1 ⊕ ··· ⊕ Qe and under this isomorphism, K1i embeds as Lj6=i J1Qj ⊆ Le Proof. The first isomorphism follows from ¯A1p = Le ¯A1p εj and proposition 9.4. Note that εjξi = ξiεj+1 so K1i = Le ¯A1p ξiεj+1. Moreover ξiεj+1 is a scalar multiple of xεj+1 if j 6= e − i − 1, and ξiεe−i = 0. Since ¯A1p x = J1, we have K1i = Lj6=e−1−i J1 ¯A1p εj+1. Finally, we see that Ki = ker( ¯Ap −→ Qi) ⊇ K1i ∩ ¯Ap while the reverse inclusion is clear. j=1 Qj. Furthermore, Ki = K1i ∩ ¯Ap. j=1 j=1 We omit the proofs of the next two elementary results. Lemma 9.8 Let Si denote the simple ¯A1p-module Qi/J1Qi. Then dimk Ext1 ¯A1p (Si, Qj) = δi+1,j. Lemma 9.9 Let M be a torsion-free ¯Ap-module and N be a torsion-free ¯A1p-module. Then there is a natural isomorphism Hom ¯A1p ( ¯A1p M, N ) ≃ Hom ¯Ap (M, N ). Proposition 9.10 With the above notation, we have Ext1 ¯Ap any j. (Qj, Qj) ≃ k and Ext1 ¯Ap (Qj, Kj) ≃ k for (Qj, Qj) ≃ coker(Hom ¯Ap ( ¯Ap, Qj) → Proof. We first show that Ext1 ¯Ap Hom ¯Ap (Kj, Qj)) and by lemma 9.9, this is isomorphic to coker(Hom ¯A1p ( ¯A1p, Qj) → Hom ¯A1p (K1j, Qj)). This shows that (Qj, Qj) ≃ k. Firstly Ext1 ¯Ap Ext1 ¯Ap (Qj, Qj) ≃ Ext1 ¯A1p (cid:0) ¯A1p /K1j, Qj(cid:1) . By proposition 9.7, we have ¯A1p /K1j ≃ Qj ⊕Li6=j Si and the result follows from lemma 9.8. (Qj, Kj), we use the following exact sequence To compute Ext1 ¯Ap Hom ¯Ap (Qj, ¯Ap) ρj−→ k[[z]] −→ Ext1 ¯Ap (Qj, Kj) −→ Ext1 ¯Ap (Qj, ¯Ap). The right most term is zero since ¯Ap has injective dimension one, while Qj is torsion-free and hence is a first syzygy. The image of ρj contains ρj(Hom ¯Ap (Qj, z ¯A1p)) = Hom ¯A1p (Qj, zQj) = (z). Also, the morphism ρj cannot be surjective for otherwise, any preimage of the identity map idQj in Hom ¯Ap (Qj, ¯Ap) would split the projection map ¯Ap → Qj. This is impossible since ¯Ap is local. Hence Ext1 (Qj, Kj) ≃ k[[z]]/(z) = k. ¯Ap Proposition 9.11 End ¯Ap (K1i) = End ¯A1p (K1i) is naturally isomorphic to the non-unital subring of ¯A1p obtained by replacing the (e− i)-th row and column of the matrix form in (5) with 0s. In particular, it is a (unital) subring of k[[z]](e−1)×(e−1) with colength (e − 1)(e − 2)/2. 21 Proof. This follows from proposition 9.7 and the natural isomorphism Hom ¯A1p (Qi, Qj) ∼−→ Hom ¯A1p (J1Qi, J1Qj) : φ 7→ φJ1Qi . Proposition 9.12 There is an exact sequence 0 −→ End ¯Ap (Ki) −→ End ¯A1p (K1i) −→ K1i/Ki −→ 0. Moreover, K1i/Ki is a k-vector space of dimension (e − 1)(e − 2)/2. Proof. We first establish the exact sequence above by considering the following exact sequence 0 −→ End ¯Ap (Ki) −→ Hom ¯Ap (Ki, K1i) α −→ Hom ¯Ap (Ki, K1i/Ki). Note that by lemma 9.9, the middle term is isomorphic to End ¯A1p (K1i). Now the surjection ¯Ap → ¯Ap ξi = Ki induces an injection β : Hom ¯Ap (Ki, K1i/Ki) −→ Hom ¯Ap ( ¯Ap, K1i/Ki) ≃ K1i/Ki. so it suffices to show surjectivity of α ◦ β. module, we are reduced to showing We know from proposition 9.7 that K1i/Ki ≃ (K1i + ¯Ap)/ ¯Ap so viewing K1i as a right End ¯A1p (K1i)- ξi End ¯A1p (K1i) + ¯A1p = K1i + ¯A1p (8) Matrix computations using propositions 9.7 and 9.11 show that ξi End ¯A1p (K1i) = Mj6=e−i−1 εjK1i =: K ′ 1i where εj are the standard diagonal matrix idempotents. Furthermore, xξi, . . . , xe−1ξi ∈ ¯Ap ∩K1i gen- erate K1i/K ′ 1i as a k[[z]]-module so equation (8) follows. This establishes the exact sequence of the proposition. We now compute dimk K1i/Ki. From proposition 9.7 we see there is an exact sequence of the form 0 −→ K1i/Ki −→ ¯A1p /J −→ ¯A1p /(K1i + J) −→ 0. Now the natural map J −→ JQi is surjective so proposition 9.7 also shows that K1i + J = J ¯A1p. Thus dimk ¯A1p /(K1i + J) = e. On the other hand dimk ¯A1p /J = e(e − 1)/2 + 1 by proposition 9.5 so dimk K1i/Ki = (e − 1)(e − 2)/2 as was to be shown. Theorem 9.13 ¯A1 ≃   O O(−1) ... ··· . . . . . . O ··· ... . . . ... . . . ··· O(−1) O   (9) O(−1) Proof. We consider an exact sequence of the form 0 −→ K −→ ¯A −→ P −→ 0 where P is a torsion-free ¯A-module of rank e. Note that proposition 9.4 shows that Pp ≃ Qi for some i, while Kp ≃ Ki by lemma 8.2. Hence lemma 6.2 reduces the proof to showing that ν(P ) > 2 − e where ν(P ) = χ(End ¯A(P )) − χ(End ¯A(K)) − dimk Ext1 ¯Ap (Qi, Qi) + dimk Ext1 ¯Ap (Qi, Ki). 22 The dimensions of the ext terms above cancel by proposition 9.10. To calculate χ(End ¯A(K)), we let ϕ : End ¯A(K) → B be an embedding into a maximal order B. Then χ(End ¯A(K)) = χ(B) − length(coker ϕ). Since χ(B) = (e − 1)2 we conclude from proposition 9.12 and proposition 9.11 that χ(End ¯A(K)) = (e − 1)2 − (e − 1)(e − 2) = e − 1. Finally we have End ¯A(P ) = OF , so ν(P ) = 2 − e and the theorem is proved. We next show that ¯A is rationally filtered. Let Pi be the i-th column of (9), so that χ(Pi) = i, and let ¯Ai = ¯A ∩P>e−i as in section 8. This gives a filtration 0 < ¯A1 < ··· < ¯Ae−1 < ¯Ae = ¯A . Theorem 9.14 The filtration { ¯A j Proof. By theorem 8.7, it suffices to prove that χ( ¯A }e j=1 is rational and F is a good fibre. j j ) = j. We calculate χ( ¯A ) using the exact sequence j 0 −→ ¯A −→ ¯A ψ −→ P6e−j −→ Cj −→ 0 where Cj = coker ψ. Note that Cj is supported at p so we may compute χ(Cj ) locally. Let Q = (P6e−j )p. Since Q is a summand of ¯A1p, we have Q ≃ Qi1 ⊕ ··· ⊕ Qie−j where the indices are pairwise distinct. Let θ = ψ ⊗F OF,p : ¯Ap → Q, and note that im θ generates Q as an ¯A1p-module. Let θs : J s/J s+1 → J sQ/J s+1Q denote the map induced by θ. We need two lemmas. Lemma 9.15 The map θs is injective if 0 6 s < e − j. Furthermore, θe−j−1 is an isomorphism. Proof. We first show that θs is injective for 0 ≤ s < e − j. Let θ(1) = Pe−j m=1 am where am is the appropriate generator of Qim. Now {xs, xs−1y, . . . , ys} gives a k-basis for J s/J s+1 so it suffices to show that their images in J sQ/J s+1Q remain linearly independent. Now lemma 9.1 shows that {xsa1, . . . , xsae−j} gives a basis of J sQ/J s+1Q so the Xm=1 ζ −l(im+(e−1)/2)xsam + J s+1Q xs−lylam + J s+1Q = e−j e−j Xm=1 are linearly independent for l = 0, 1, . . . , s. This proves injectivity of θs. Furthermore, dimk J e−j−1/J e−j = e − j = dimk J e−j−1Q/J e−jQ so θe−j−1 is an isomorphism. The following is standard so we omit the proof. Lemma 9.16 Let ϕ : M ′ → M be a morphism of finite length modules over an Artin ring R. Denote by J the Jacobson radical of R. Suppose that for all k > 0, the induced map ϕk : J kM ′/J k+1M ′ → J kM/J k+1M is injective, then ϕ is injective. Rest of proof of theorem 9.14. By lemma 9.15, θe−j−1 is an isomorphism so θJ e−j−1 is surjective by Nakayama's lemma. Thus θ induces a map of left modules θ′ : ¯Ap /J e−j−1 → Q/J e−j−1Q with coker θ′ = Cj. Furthermore, θ′ is injective by lemmas 9.16 and 9.15, so length(Cj ) = length(Q/J e−j−1Q) − length( ¯Ap /J e−j−1) (e − j)(e − j − 1) 2 = (e − j)(e − j − 1) − = (e − j)(e − j − 1) 2 23 and j χ( ¯A ) = χ( ¯A) − χ(P6e−j ) + χ(Cj) + (e − j)(e − j + 1) 2 = e − = j. (e − j)(e − j − 1) 2 10 Ramified fibres Let A be a degree e ruled order. Suppose now that F is a ramification curve. We will only consider the case where F is totally ramified, that is, the ramification index of A at F is e. Because of this, we will not need to consider the condition that c2 is minimal. Note that if e is prime then any ramification curve is totally ramified. This time ¯A := A ⊗Z OF is no longer an order so we analyse it as follows. Let η be the generic point of F and mη ⊳ OZ,η be the maximal ideal of the local ring at η. The localisation Aη of A at η is a maximal order in a division ring so we have complete information about it. In particular, its Jacobson radical Jη is a principal ideal satisfying J e η = mη Aη and it has residue ring Aη/Jη the degree e field extension of k(F ) which is ramified at the two points D ∩ F where D is the bisection that A is ramified on. We define J := A ∩ Jη. We need Proposition 10.1 For i ∈ N we have, i) J i is an invertible bimodule, ii) J i J i = A ∩ J i η. Proof. Note that J is saturated in A in the sense that A/J embeds in A/J ⊗Z OZ,η. Hence J is reflexive and thus an invertible bimodule. This proves i). Also, Jη = J ⊗Z OZ,η so ii) follows. Finally, we see that J i is reflexive so also saturated. Thus iii) follows from ii). η = J i ⊗Z OZ,η and, iii) Proposition 10.2 The residue ring A/J ≃ OG where G is the smooth rational curve which is the e-fold cover of F ramified at the two points of D ∩ F . Proof. We know that A/J ֒→ Aη/Jη and that A/J ⊗F k(F ) ≃ Aη/Jη. Hence A/J ≃ OG where G is a projective model for the field extension of k(F ) totally ramified at D ∩ F with ramification index e. It suffices to show that G is smooth. This can be done by local computations using theorem 2.1. We need to introduce some notation from [AV]. Given a sheaf F on a scheme X, and an automor- phism σ of X we obtain an OX-bimodule Fσ whose left module structure is given by F but the right module structure is σ∗ F . Lemma 10.3 The OG-bimodule J/J 2 is isomorphic to (OG)σ where σ ∈ Aut G is the automorphism given by ramification data. Proof. Ramification theory tells us that Jη/J 2 OG. Hence deg J/J 2 = 0 as a sheaf on G. This proves the lemma. η ≃ k(G)σ. Now (J/J 2)e ≃ J e/J e+1 = A(−F )/J(−F ) ≃ Even though ¯A is not an order, ¯A⊗F k(F ) still satisfies hypothesis 7.1 so we are in a good position to apply the results of section 7. Theorem 10.4 ¯A is rationally filtered so in particular, the fibre F is good. Furthermore, if c ∈ C is the point corresponding to the fibre F , then the map C ′ −→ C is totally ramified at c. 24 Proof. We consider the filtration of ¯A 0 < J e−1/A(−F ) < . . . < J/A(−F ) < ¯A . The factors are all isomorphic to OG ≃ OF ⊕ OF (−1)e−1 so the filtration is rational and F is good by theorem 7.5. Also J 2/A(−F ) is rationally filtered so we may lift a non-zero global section of J/J 2 to δ ∈ H 0(J/A(−F )). One sees that H 0( ¯A) = k[δ]/(δe). Thus C ′ −→ C is totally ramified at c. 11 The non-commutative Mori contraction In this section, we collate the results of the previous sections to prove theorem 1.4. Proposition 11.1 Let A be a ruled order with minimal c2 in its Morita equivalence class. Suppose further that every fibre is good. Then i. C ′ := SpecC f∗A is a smooth projective curve. ii. C ′ is a component of the Hilbert scheme Hilb A and Ψ : A⊗Z f ∗f∗A −→ A is the universal rational curve. iii. The non-commutative Mori contraction is given by the "morphism of algebras" f∗A −→ A. Proof. Parts i) and ii) are essentially a restatement of proposition 3.6. The morphism ψ : f∗A −→ A determines a pull-back functor ψ∗ : f∗A − Mod −→ A − Mod as follows. Given a quasi-coherent f∗A- module L then f ∗L is an f ∗f∗A-module so we may consider the A-module ψ∗(L) := A ⊗f ∗f∗A f ∗L. Since the universal rational curve is given by Ψ, we see that this coincides with the Fourier-Mukai transform of theorem 2.2ii). Part iii) follows immediately. To see theorem 1.4, it suffices to show that all fibres are good. This has been checked in theorems 5.2, 8.7, 9.14 and 10.4. References [A86] M. Artin, "Maximal orders of global dimension and Krull dimension two", Invent. Math., 84, (1986), p.195-222 [A92] M. Artin, "Geometry of quantum planes", Contemp. Math., 124, (1992), p.1-15 [AdJ] M. Artin, A. J. de Jong, "Stable Orders over Surfaces", manuscript available at www.math.lsa.umich.edu/courses/711/ordersms-num.pdf [AG] M. Auslander, O. Goldman, "Maximal orders", Trans. of the AMS 97, (1960), p.1-24 [AM] M. Artin, D. Mumford, "Some elementary examples of unirational varieties which are not ratio- nal", Proc. LMS 3, (1972), p.75-95 [ATV90] M. Artin, J. Tate, M. Van den Bergh, "Some algebras associated to automorphisms of curves", Grothendieck Festschrift, Birkhauser, Basel (1990), p.33-85 [ATV91] M. Artin, J. Tate, M. Van den Bergh, "Modules over regular algebras of dimension 3", Invent. Math. 106, (1991), p.335-89 [AV] M. Artin, M. Van den Bergh, "Twisted Homogeneous Coordinate Rings", J. of Algebra, 133, (1990) , p.249-271 25 [AZ94] M. Artin, J. Zhang, "Noncommutative projective schemes", Adv. Math. 109 (1994), p.228-87 [AZ01] M. Artin, J. Zhang, "Abstract Hilbert Schemes", Alg. and Repr. Theory 4 (2001), p.305-94 [C05] D. Chan, "Splitting bundles over hereditary orders", Comm. Alg. 333 (2005), p.2193-9 [Chan] K. Chan, "Log terminal orders are numerically rational", http://arxiv.org/abs/0911.5545 [CHI] D. Chan, P. Hacking, C. Ingalls, "Canonical singularities for orders over surfaces", Proc. LMS 98, (2009), p.83-115 [CI] D. Chan, C. Ingalls, "Minimal model program for orders over surfaces" Invent. Math. 161 (2005) p.427-52 [CK03] D. Chan, R. Kulkarni, "Del Pezzo Orders on Projective Surfaces", Advances in Mathematics 173 (2003), p.144-177 [CK05] D. Chan, R. Kulkarni, "Numerically Calabi-Yau orders on surfaces" J. London Math. Soc. 72 (2005), p.571-584 [CK11] D. Chan, R. Kulkarni, "Moduli of bundles on exotic del Pezzo orders", Amer. J. Math. 133, (2011), p.273-93 [CN] D. Chan, A. Nyman, "Non-commutative Mori contractions and P1-bundles", http://arxiv.org/abs/0904.1717 [dJ] A.J. de Jong, "The period-index problem for the Brauer group of an algebraic surface", Duke Math. J. 123, (2004), p.71-94 [Ler] B. Lerner, PhD thesis UNSW, in preparation [HN] H. Hijikata and K. Nishida, "Primary orders of finite representation type", J. Algebra, 192 (1997) p.592-640 [Pot] J. le Potier, "Lectures on vector bundles", Cambridge studies in advanced mathematics 54, Cam- bridge University Press, Cambridge, (1997) [Ram] M. Ramras, "Maximal orders over regular local rings of dimension two" Trans. of the AMS, 142 (1969) p.457-79 [RV] I. Reiten, M. Van den Bergh, "Two-dimensional Tame and Maximal Orders of Finite Represen- tation Type", Mem. of the AMS Vol. 80 No. 408, July 1989 26
1810.00032	1	1810	2018-09-28T18:30:11	Orthomodular lattices can be converted into left residuated l-groupoids	[ "math.RA", "math.CO" ]	We show that every orthomodular lattice can be considered as a left residuated l-groupoid satisfying divisibility, antitony, the double negation law and three more additional conditions expressed in the language of residuated structures. Also conversely, every left residuated l-groupoid satisfying the mentioned conditions can be organized into an orthomodular lattice.	math.RA	math	Orthomodular lattices can be converted into left residuated l-groupoids∗ Ivan Chajda and Helmut Langer Abstract We show that every orthomodular lattice can be considered as a left residuated l-groupoid satisfying divisibility, antitony, the double negation law and three more additional conditions expressed in the language of residuated structures. Also con- versely, every left residuated l-groupoid satisfying the mentioned conditions can be organized into an orthomodular lattice. AMS Subject Classification: 06C15, 06A11 Keywords: orthomodular lattice, left residuated l-groupoid, divisibility, antitony, double negation law It is well-known that residuated structures form an algebraic axiomatization of fuzzy log- ics, see e. g. [1] for an overview. The reader can find necessary concepts and definitions concerning residuated structures in [5], however this paper is self-contained. Orthomod- ular lattices were introduced by G. Birkhoff and J. von Neumann as an algebraic axiom- atization of the logic of quantum mechanics, see e. g. [4], [6] or [2] for details. Hence it is a natural question if these two concepts have a common base, i. e. if orthomodular lattices can be considered as residuated structures and hence as an axiomatization of certain fuzzy logic and, conversely, if certain residuated structures can be converted into orthomodular lattices, i. e. if the logic of quantum mechanics can be considered as a kind of fuzzy logic. For the theory of orthomodular lattices cf. the monographs [6] and [2] as well as the paper [3]. We start with the definition of an orthomodular lattice. Definition 1. An orthomodular lattice is an algebra L = (L, ∨, ∧,′ , 0, 1) of type (2, 2, 1, 0, 0) satisfying (i) -- (v) for all x, y ∈ L: (i) (L, ∨, ∧, 0, 1) is a bounded lattice. (ii) x ∨ x′ = 1 1Support of the research of both authors by the bilateral project entitled "New perspectives on residuated posets", supported by the Austrian Science Fund (FWF), project I 1923-N25, and the Czech Science Foundation (GA CR), project 15-34697L, as well as by OAD, project 04/2017, is gratefully acknowledged. ∗Preprint of an article published by Miskolc University Press in Miskolc Mathematical Notes It is available online at: DOI: 10.18514/MMN.2017.1730. 18 (2017), No. 2, pp. 685 -- 689. mat76.mat.uni-miskolc.hu/mnotes. 1 (iii) x ≤ y implies y ′ ≤ x′. (iv) (x′)′ = x (v) x ≤ y implies y = x ∨ (y ∧ x′). Remark 2. In every lattice (L, ∨, ∧) with a unary operation ′ satisfying (iii) and (iv) the so-called de Morgan laws (x ∨ y)′ = x′ ∧ y ′ and (x ∧ y)′ = x′ ∨ y ′ hold. Remark 3. According to the de Morgan laws condition (v) can be replaced by (vi) x ≤ y implies x = y ∧ (x ∨ y ′). Now we introduce left residuated l-groupoids. Definition 4. A left residuated l-groupoid is an algebra A = (A, ∨, ∧, ⊙, →, 0, 1) of type (2, 2, 2, 2, 0, 0) satisfying (i) -- (iii) for all x, y, z ∈ A: (i) (A, ∨, ∧, 0, 1) is a bounded lattice. (ii) x ⊙ 1 = 1 ⊙ x = x. (iii) x ⊙ y ≤ z if and only if x ≤ y → z. Condition (iii) is called left adjointness. A is said to satisfy divisibility if for all x, y ∈ A. We define a unary operation ′ on A by (x → y) ⊙ x = x ∧ y for all x ∈ A. A is said to satisfy antitony if x′ := x → 0 x ≤ y implies y ′ ≤ x′ for all x, y ∈ A and A is said to satisfy the double negation law if for all x ∈ A. (x′)′ = x Example 5. If A := {0, a, a′, b, b′, 1}, (A, ∨, ∧, 0, 1) denotes the bounded lattice with the Hasse diagram a ✑✑ ◗◗ s ✑ a′ ◗ ◗ b ✑ ◗◗ ✑✑ s b′ s 1 ◗ ✑ ❆ ✁ ◗ ✑ ❆ ✁ ❆ ✁ ❆ ✁ ❆ ✁ ◗ ✑ ❆ ◗ ✁ ✑ 0 s s s 2 and the binary operations ⊙ and → are defined by the tables b ⊙ 0 a a′ b′ 0 0 0 0 0 0 a 0 a 0 b a′ 0 0 a′ b 0 a a′ b 0 b b′ 0 a a′ 0 b′ 1 0 a a′ b′ 1 0 b′ a b′ a′ b b′ 1 b b 1 a a′ 1 1 b′ 1 → 0 1 0 1 1 a′ 1 a′ a′ a′ 1 a 1 a 1 a′ a a b′ 1 b′ b′ b′ b 1 1 b′ b b b 1 0 b′ 1 a a′ a 1 b b then (A, ∨, ∧, ⊙, →, 0, 1) is a left residuated l-groupoid satisfying divisibility, antitony and the double negation law. The mentioned lattice is the smallest orthomodular lattice which is not a Boolean algebra and it is usually denoted by MO2. The following theorem says that to every orthomodular lattice there can be assigned a left residuated l-groupoid in a natural way. Theorem 6. Let L = (L, ∨, ∧,′ , 0, 1) be an orthomodular lattice and define binary oper- ations ⊙ and → on L by the following formulas: x ⊙ y = (x ∨ y ′) ∧ y, x → y = (y ∧ x) ∨ x′. (1) (2) Then A(L) = (L, ∨, ∧, ⊙, →, 0, 1) is a left residuated l-groupoid satisfying divisibility, antitony, the double negation law as well as the following identity: x ⊙ (x ∨ y) = x. (3) Moreover, x′ = x → 0 for all x ∈ L. Proof. Let a, b ∈ L. We have Of course, (L, ∨, ∧, 0, 1) is a bounded lattice. Moreover, a → 0 = (0 ∧ a) ∨ a′ = 0 ∨ a′ = a′. a ⊙ 1 = (a ∨ 1′) ∧ 1 = (a ∨ 0) ∧ 1 = a ∧ 1 = a and 1 ⊙ a = (1 ∨ a′) ∧ a = 1 ∧ a = a. If a ⊙ b ≤ c then (a ∨ b′) ∧ b ≤ c and hence a ≤ a ∨ b′ = ((a ∨ b′) ∧ b) ∨ b′ = (((a ∨ b′) ∧ b) ∧ b) ∨ b′ ≤ (c ∧ b) ∨ b′ = b → c. If, conversely, a ≤ b → c then a ≤ (c ∧ b) ∨ b′ and hence a ⊙ b = (a ∨ b′) ∧ b ≤ (((c ∧ b) ∨ b′) ∨ b′) ∧ b = ((c ∧ b) ∨ b′) ∧ b = c ∧ b ≤ c. Now, using orthomodularity (i. e. (v) of Definition 1), we have (a → b) ⊙ a = (((b ∧ a) ∨ a′) ∨ a′) ∧ a = ((b ∧ a) ∨ a′) ∧ a = a ∧ b. In view of Definition 1, a ≤ b implies b′ ≤ a′ and we have (a′)′ = a. Finally, by applying (1) and (vi) of Definition 1 we obtain a ⊙ (a ∨ b) = (a ∨ (a ∨ b)′) ∧ (a ∨ b) = a. 3 Remark 7. The operation x ⊙ y := (x ∨ y ′) ∧ y is called the Sasaki projection of x onto y (cf. [6] and [2]). Conversely, certain left residuated l-groupoids give rise to an orthomodular lattice. Theorem 8. Let A = (A, ∨, ∧, ⊙, →, 0, 1) be a left residuated l-groupoid satisfying an- titony, the double negation law as well as identities (1) and (3) of Theorem 6. Moreover, define x′ := x → 0 for all x ∈ A. Then L(A) = (A, ∨, ∧,′ , 0, 1) is an orthomodular lattice. Proof. Let a, b ∈ A. Clearly, (A, ∨, ∧, 0, 1) is a bounded lattice and a′ = a → 0. Using antitony we see that a ≤ b implies b′ ≤ a′. Moreover, we have (a′)′ = a according to the double negation law. Finally, if a ≤ b then, using (3) and (1), we have b = (b′)′ = (b′ ⊙ (b′ ∨ a′))′ = (b′ ⊙ a′)′ = ((b′ ∨ a) ∧ a′)′ = a ∨ (b ∧ a′) and hence a ∨ a′ = a ∨ (1 ∧ a′) = 1. Finally, we prove that the correspondence described in the last two theorems is one-to-one. Theorem 9. We have L(A(L)) = L for every orthomodular lattice L and A(L(A)) = A for every left residuated l-groupoid satisfying antitony, the double negation law as well as identities (1) -- (3) of Theorem 6. Proof. If L = (L, ∨, ∧,′ , 0, 1) is an orthomodular lattice, A(L) = (L, ∨, ∧, ⊙, →, 0, 1) and L(A(L)) = (L, ∨, ∧,∗ , 0, 1) then x∗ = x → 0 = (0 ∧ x) ∨ x′ = 0 ∨ x′ = x′ for all x ∈ L, therefore we obtain L(A(L)) = L. Conversely, if A = (A, ∨, ∧, ⊙, →, 0, 1) is a left residuated l-groupoid satisfying divisibility, antitony, the double negation law as well as identities (1) -- (3) of Theorem 6, L(A) = (A, ∨, ∧,′ , 0, 1) and A(L(A)) = (A, ∨, ∧, ◦, ⇒, 0, 1) then x ◦ y = (x ∨ y ′) ∧ y = x ⊙ y and x ⇒ y = (y ∧ x) ∨ x′ = x → y for all x, y ∈ A, therefore we obtain A(L(A)) = A. Remark 10. We have shown that orthomodular lattices can be considered as special residuated lattices and hence the logic of quantum mechanics axiomatized by them has a common base with a certain fuzzy logic axiomatized just by means of residuated lattices as pointed out in [1]. This sheds a new light on the logic of quantum mechanics and yields new tools for its investigation. References [1] R. Belohl´avek, Fuzzy Relational Systems. Foundations and Principles. Kluwer, New York 2002. ISBN 0-306-46777-1/hbk. 4 [2] L. Beran, Orthomodular Lattices. Algebraic Approach. Kluwer, Dordrecht 1985. [3] I. Chajda, R. Halas and J. Kuhr, Many-valued quantum algebras. Algebra Universalis 60 (2009), 63 -- 90. [4] A. Dvurecenskij and S. Pulmannov´a, New Trends in Quantum Structures. Kluwer, Dordrecht 2000. ISBN 0-7923-6471-6. [5] N. Galatos, P. Jipsen, T. Kowalski and H. Ono, Residuated Lattices. An Algebraic Glimpse at Substructural Logics. Elsevier, Amsterdam 2007. ISBN 978-0-444-52141- 5/hbk. [6] G. Kalmbach, Orthomodular Lattices. Academic Press, London 1983. Authors' addresses: Ivan Chajda Palack´y University Olomouc Faculty of Science Department of Algebra and Geometry 17. listopadu 12 771 46 Olomouc Czech Republic [email protected] Helmut Langer TU Wien Faculty of Mathematics and Geoinformation Institute of Discrete Mathematics and Geometry Wiedner Hauptstrasse 8-10 1040 Vienna Austria [email protected] 5
1603.05377	3	1603	2016-04-06T00:53:16	A Lie algebra related to the universal Askey-Wilson algebra	[ "math.RA" ]	Let $\mathbb{F}$ denote an algebraically closed field. Denote the three-element set by $\mathcal{X}=\{A,B,C\}$, and let $\mathbb{F}\left<\mathcal{X}\right>$ denote the free unital associative $\mathbb{F}$-algebra on $\mathcal{X}$. Fix a nonzero $q\in\mathbb{F}$ such that $q^4\neq 1$. The universal Askey-Wilson algebra $\Delta$ is the quotient space $\mathbb{F}\left<\mathcal{X}\right>/\mathbb{I}$, where $\mathbb{I}$ is the two-sided ideal of $\mathbb{F}\left<\mathcal{X}\right>$ generated by the nine elements $UV-VU$, where $U$ is one of $A,B,C$, and $V$ is one of \begin{equation} (q+q^{-1}) A+\frac{qBC-q^{-1}CB}{q-q^{-1}},\nonumber \end{equation} \begin{equation} (q+q^{-1}) B+\frac{qCA-q^{-1}AC}{q-q^{-1}},\nonumber \end{equation} \begin{equation} (q+q^{-1}) C+\frac{qAB-q^{-1}BA}{q-q^{-1}}.\nonumber \end{equation} Turn $\mathbb{F}\left<\mathcal{X}\right>$ into a Lie algebra with Lie bracket $\left[ X,Y\right] = XY-YX$ for all $X,Y\in\mathbb{F}\left<\mathcal{X}\right>$. Let $\mathcal{L}$ denote the Lie subalgebra of $\mathbb{F}\left<\mathcal{X}\right>$ generated by $\mathcal{X}$, which is also the free Lie algebra on $\mathcal{X}$. Let $L$ denote the Lie subalgebra of $\Delta$ generated by $A,B,C$. Since the given set of defining relations of $\Delta$ are not in $\mathcal{L}$, it is natural to conjecture that $L$ is freely generated by $A,B,C$. We give an answer in the negative by showing that the kernel of the canonical map $\mathbb{F}\left<\mathcal{X}\right>\rightarrow\Delta$ has a nonzero intersection with $\mathcal{L}$. Denote the span of all Hall basis elements of $\mathcal{L}$ of length $n$ by $\mathcal{L}_n$, and denote the image of $\sum_{i=1}^n\mathcal{L}_i$ under the canonical map $\mathcal{L}\rightarrow L$ by $L_n$. We study some properties of $L_4$ and $L_5$.	math.RA	math	A Lie algebra related to the universal Askey-Wilson algebra Rafael Reno S. Cantuba Mathematics Department De La Salle University Manila 2401 Taft Ave., Manila, 1004 Philippines rafael [email protected] Abstract Let F denote an algebraically closed field. Denote the three-element set by X = {A, B, C}, and let F hX i denote the free unital associative F-algebra on X . Fix a nonzero q ∈ F such that q4 6= 1. The universal Askey-Wilson algebra ∆ is the quotient space F hX i /I, where I is the two-sided ideal of F hX i generated by the nine elements U V − V U , where U is one of A, B, C, and V is one of (q + q−1)A + (q + q−1)B + (q + q−1)C + qBC − q−1CB q − q−1 qCA − q−1AC q − q−1 qAB − q−1BA q − q−1 , , . Turn F hX i into a Lie algebra with Lie bracket [X, Y ] = XY − Y X for all X, Y ∈ F hX i. Let L denote the Lie subalgebra of F hX i generated by X , which is also the free Lie algebra on X . Let L denote the Lie subalgebra of ∆ generated by A, B, C. Since the given set of defining relations of ∆ are not in L, it is natural to conjecture that L is freely generated by A, B, C. We give an answer in the neg- ative by showing that the kernel of the canonical map F hX i → ∆ has a nonzero intersection with L. Denote the span of all Hall basis elements of L of length n by Ln, and denote the image of Pn i=1 Li under the canonical map L → L by Ln. We show that the simplest nontrivial Lie algebra relations on L occur in L5. We exhibit a basis for L4, and we also exhibit a basis for L5 if q is not a sixth root of unity. Keywords: universal Askey-Wilson algebra; Hall basis; Lie algebra relations 1 Introduction Let F be an algebraically closed field and fix a nonzero q ∈ F such that q4 6= 1. Given a, b, c ∈ F, the Askey-Wilson algebra with parameters a, b, c is the unital associative F-algebra AW := AWq(a, b, c) defined as having generators A, B, C and relations A + B + C + qBC − q−1CB q2 − q−2 qCA − q−1AC q2 − q−2 qAB − q−1BA q2 − q−2 = = a q + q−1 , b q + q−1 , = c q + q−1 . The algebra AW was introduced in [8] in order to desrcibe the Askey-Wilson polynomials [2]. A wide range of applications of the Askey-Wilson algebra is discussed in [7, Section 1]. These applications include 1 integrable systems, quantum mechanics, the theory of quadratic algebras, Leonard pairs and Leonard triples, and quantum groups. A central extension of the Askey-Wilson algebra AW is introduced in [7], which is called the universal Askey-Wilson algebra. Definition 1.1 ([7, Definition 1.2]). The universal Askey-Wilson algebra is the unital associative F-algebra, which we denote by ∆, defined as having generators A, B, C, and relations which assert that the following are central in ∆: A + B + C + qBC − q−1CB q2 − q−2 qCA − q−1AC q2 − q−2 , , qAB − q−1BA q2 − q−2 , (1) (2) (3) where q is a nonzero scalar that is not a fourth root of unity. Our main object of study is the Lie subalgebra L of ∆ generated by A, B, C. We show that a set of defining relations for ∆ cannot be expressed in terms of Lie algebra operations only, and yet this does not imply that L is freely generated by A, B, C. Denote the free unital associative F-algebra on the three-element set X = {A, B, C} by F hX i, and the free Lie algebra on X by L. Recall that L is the Lie subalgebra of F hX i generated by A, B, C. We use the basis of L which was introduced by Hall [5]. Let us call the images of the Hall basis elements under the canonical map L → L as the standard Lie monomials of L. We show that the kernel of the canonical map F hX i → ∆ has a nonzero intersection with L. The generators A, B, C are the standard Lie monomials of length 1. The standard Lie monomials of lengths ≥ 1 are constructed according to some rules, which we shall discuss in later sections. We show that the simplest Lie algebra relations on L occur at length 5, and we determine a maximal linearly independent set of standard Lie monomials of length at most 5. 2 Preliminaries Let F be an algebraically closed field. Throughout, by an F-algebra we mean a unital associative F-algebra. Let A be an F-algebra. Recall that an anti-automorphism of A is a bijective F-linear map ψ : A → A such that ψ(f g) = ψ(g)ψ(f ) for all f, g ∈ A. We turn A into a Lie algebra with Lie bracket [f, g] = f g − gf for f, g ∈ A. Let N = {0, 1, 2, . . .} denote the set of natural numbers. Given a nonzero n ∈ N, let X denote an n- element set. We shall refer to any element of X as a letter. For t ∈ N, by a word of length t on X we mean a sequence of the form X1X2 · · · Xt, (4) where Xi ∈ X for 1 ≤ i ≤ t. Given a word W on X , denote the length of W by W . The word of length 0 will be denoted by 1. Let hX i denote the set of all words on X . Given words X1X2 · · · Xs and Y1Y2 · · · Yt on X , their concatenation product is X1X2 · · · XsY1Y2 · · · Yt. We now recall the free F-algebra F hX i. The F-vector space F hX i has basis hX i. Multiplication in the F-algebra F hX i is the concatenation product. We endow F hX i with a symmetric bilinear form ( , ) with respect to which hX i is an orthonormal basis. For any f ∈ F hX i and any word W , the coefficient of W in f is (f, W ). Given n ∈ N, the subspace of F hX i spanned by all the words of length n is the n-homogenous component of F hX i. Observe that F hX i is the direct sum of all the n-homogenous components for n ∈ N. If f is an element of the m-homogenous component and g is an element of the n-homogenous component, then f g is an element of the (m + n)-homogenous component. It follows that the set of all n-homogenous components of F hX i for all n ∈ N is a grading of F hX i. 2 The following notation will be useful. Let W = X1X2 · · · Xt denote a word on X . We define W ∗ to be the word XtXt−1 · · · X1 on X . Let θ denote the F-linear map θ : F hX i → F hX i , W 7→ (−1)W W ∗, (5) for any word W . By [6, p. 19], the map θ is the unique anti-automorphism of the F-algebra F hX i that sends X to −X for any letter X. Let L denote the Lie subalgebra of the Lie algebra F hX i generated by X . Following [6, Theorem 0.5], we call L the free Lie algebra on X . Proposition 2.1 ([6, Lemma 1.7]). For f ∈ L, we have θ(f ) = −f . We now recall the notion of a Lie monomial on X . The set of all Lie monomials on X is the minimal subset of F hX i that contains X and is closed under the Lie bracket. Observe that 0 is a Lie monomial. Let U be a Lie monomial. Then U is an element of some n-homogenous component of F hX i. We define the length of the Lie monomial U to be n. Observe that 0 has length n for any n ∈ N. Any nonzero Lie monomial has a unique length. Observe that the set of all Lie monomials of length 1 is X . We now consider an ordering of Lie monomials. Definition 2.2 ([4, p. 581]). Fix an ordering < on X . Suppose that the set of all Lie monomials of lengths 1, 2, . . . , t − 1 have been ordered such that U < V if the length of U is strictly less than that of V . If U, V both have length t, and can be written as U = [X1, Y1] , V = [X2, Y2], then we compare U, V using the following rules: (i) If Y1 6= Y2, then U < V iff Y1 < Y2. (ii) If Y1 = Y2, then U < V iff X1 < X2. We now introduce a basis for L consisting of Lie monomials. Proposition 2.3 ([5, Theorem 3.1]). Let H be the set of Lie monomials such that X ⊂ H, and that for any U, V ∈ H, the Lie monomial [U, V ] is also in H whenever the following conditions hold. (i) U > V . (ii) If U = [X, Y ] for some Lie monomials X, Y , then Y ≤ V . Then H is a basis for L, often referred to as the Hall basis of L. Example 2.4. Suppose X = {A, B, C} and A < B < C. Then the elements of H of length at most 4 are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bserve that the above Lie monomials are listed according to the ordering in Definition 2.2. Given a Lie algebra L and x, y ∈ L, recall the adjoint linear map ad x : L → L that sends y 7→ [x, y]. Denote an arbitrary word on X by W = X1X2 · · · Xt. The Lie bracketing from left to right is the linear map F hX i → L that sends 1 7→ 0 and sends the word W into a Lie monomial according to the following rules: 3 (i) If W = 1, then W 7→ W . (ii) Suppose that the images of all words of length < W have been defined. Denote the image of X1X2 · · · Xt−1 by V . Then W 7→ (−ad Xt) (V ) = [V, Xt] . That is, X1X2 · · · Xt 7→ [[[X1, X2] , · · · ] , Xt] for t ≥ 2. A Lie monomial that is an image of some word under Lie bracketing from left to right is said to be left-normed. Notation 2.5. Given a word W , we denote the image of W under Lie bracketing from left to right by [W ]. Example 2.6. With reference to Example 2.4, we rewrite (6) using Notation 2.5. A, B, C, [BA] , [CA] , [CB] , (cid:2)BA2(cid:3) , (cid:2)CA2(cid:3) , [BAB] , [CAB] , (cid:2)CB2(cid:3) , [BAC] , [CAC] , [CBC] , (cid:2)BA3(cid:3) , (cid:2)CA3(cid:3) , (cid:2)BA2B(cid:3) , (cid:2)CA2B(cid:3) , (cid:2)BAB2(cid:3) , (cid:2)CAB2(cid:3) , (cid:2)CB3(cid:3) , (cid:2)BA2C(cid:3) , (cid:2)CA2C(cid:3) , [BABC] , [CABC] , (cid:2)CB2C(cid:3) , (cid:2)BAC2(cid:3) , (cid:2)CAC2(cid:3) , (cid:2)CBC2(cid:3) , [[CA] , [BA]] , [[CB] , [BA]] , [[CB] , [CA]] . (7) Throughout, by an ideal of an F-algebra A we mean a two-sided ideal of A. By a Lie ideal of a Lie algebra L we mean an ideal of L under the Lie algebra structure. We now recall the notion of algebras having generators and relations (i.e., having a presentation). Denote the elements of X by G1, G2, . . . , Gn. Let f1, f2, . . . , fm ∈ F hX i and let I be the ideal of F hX i generated by f1, f2, . . . , fm. We define F hX i /I as the F-algebra with generators G1, G2, . . . , Gn and relations f1 = 0, f2 = 0, . . . , fm = 0. The Lie subalgebra of F hX i /I generated by X is L/ (I ∩ L). Let g1, g2, . . . , gm ∈ L and let J be the Lie ideal of L generated by g1, g2, . . . , gm. We define L/J as the Lie algebra with generators G1, G2, . . . , Gn and relations g1 = 0, g2 = 0, . . . , gm = 0. Suppose L is a Lie algebra (over F) generated by X . Then there exists an ideal K of L such that L = L/K. Let φ : L → L/K be the canonical Lie algebra homomorphism. Then the following span L: φ(U ), for U ∈ H. (8) We call (8) the standard Lie monomials of the Lie algebra L. Observe that the list of the standard Lie monomials of L is identical to the list of elements of H. This is because the Lie algebra homomorphism φ fixes generators. We order the list of standard Lie monomials of L in a manner analogous to that given in Definition 2.2. 3 The universal Askey-Wilson algebra Hereon, let F be an algebraically closed field, and fix a nonzero q ∈ F such that q4 6= 1. We fix X = {A, B, C}. Let F hX i be the free associative algebra on X . We use the ordering A < B < C to construct the Hall basis H of the free Lie algebra L on X . Define the following elements of the free algebra F hX i. α := (q + q−1)A + β := (q + q−1)B + γ := (q + q−1)C + qBC − q−1CB q − q−1 qCA − q−1AC q − q−1 qAB − q−1BA q − q−1 , , . 4 (9) (10) (11) We also define the following Lie products in F hX i. r0 := [A, α] , r1 := [B, β] , r2 := [C, γ] , r3 := [B, α] , r4 := [C, β] , r5 := [A, γ] , r6 := [C, α] , r7 := [A, β] , r8 := [B, γ] . Define I as the ideal of F hX i generated by r0, r1, . . . , r8. With reference to Definition 1.1, we express ∆ as a quotient space of F hX i, and as a consequence make explicit the defining relations of ∆. Proposition 3.1. ∆ = F hX i /I. Proof. Recall ∆ has relations which assert that each of (1),(2),(3) commutes with every element of ∆. Equivalently, each of (1),(2),(3) commutes with every generator A, B, C. Observe that each of α, β, γ is a scalar multiple of (1),(2),(3), respectively. Then it suffices to define ∆ as having nine defining relations of the form [X, δ], where X ∈ {A, B, C} and δ ∈ {α, β, γ}. By the definition of I, we get the desired result. (cid:4) We denote the images of α, β, γ under the canonical map F hX i → ∆ by the same symbols. Proposition 3.2. r0, r1, . . . , r8 /∈ L. Proof. Let θ denote the F-linear map in (5). It is routine to show that in the free algebra F hX i, we have θ(ri) + ri 6= 0 for 0 ≤ i ≤ 8. Use Proposition 2.1. (cid:4) By a a word in ∆ we mean the image of an element of hX i under the canonical map F hX i → ∆. Observe that the list of all words in ∆ is identical to the list of all the words on X in the free algebra F hX i since the canonical map F hX i → ∆ is an F-algebra homomorphism that fixes generators. We also preserve the ordering of generators A < B < C in ∆. By a ∆-word, we mean all elements of ∆ of the form W αrβsγt (12) where W is a word in ∆, and r, s, t ∈ N. We now recall some properties of ∆ as studied in [7]. Let U = X1X2 · · · Xt be a ∆-word, where Xi is either a generator of ∆ or one of α, β, γ for 1 ≤ i ≤ t. Without loss of generality, we assume U is of the form (12) since α, β, γ are central in ∆. By an inversion for W we mean an ordered pair (j, k) ∈ N2 such that 1 ≤ j < k ≤ t and Xj, Xk ∈ {A, B, C} such that Xj > Xk. Any ∆-word with no inversions is said to be irreducible. For instance, CABA has 4 inversions and CB2A has 5, while the ∆-words A2BC, AB2C are irreducible. The shortest words for which inversions exist are BA, CA, CB and using (9) to (11), the following hold in both F hX i and ∆. BA = q2AB + q(q + q−1)(q − q−1)C − q(q − q−1)γ, CA = q−2AC − q−1(q + q−1)(q − q−1)B + q−1(q − q−1)β, CB = q2BC + q(q + q−1)(q − q−1)A − q(q − q−1)α. (13) (14) (15) Consider the word CABA, one of the 4 inversions in which is caused by the first two letters C, A. Substituting for CA using (14), the result is a linear combination of ACBA, B2A, BAβ, each having fewer inversions than CABA. Remark 3.3. Following [7, p. 7] and [3, Theorem 1.2], for any ∆-word W , there exists a finite number of steps of substituting for inversions using (13) to (15) such that the final result is a unique linear combination of irreducible ∆-words. It follows that a basis for ∆ consists of the vectors AiBjC kαrβsγt, i, j, k, r, s, t, ∈ N. (16) Given subspaces H, K of ∆, define HK := Span {hk h ∈ H, k ∈ K}. If K is a subspace of H, we say that a subspace K ′ of H is a complement of K in H whenever H = K + K ′. (direct sum) 5 We now recall a filtration for ∆ as given in [7, Section 5]. This filtration is a sequence {∆n}n∈N of subspaces of ∆ defined by ∆0 ∆1 ∆n := F1, := ∆0 + Span {A, B, C, α, β, γ}, := ∆1∆n−1, n ≥ 1, and has the following properties for all i, j ∈ N. ∆i ⊆ ∆i+1, ∆ = [n∈N ∆n, ∆i∆j = ∆i+j . Given n ∈ N, a basis for ∆n consists of the vectors AiBjC kαrβsγt, i, j, k, r, s, t, ∈ N, i + j + k + r + s + t ≤ n, while the following vectors form a basis for a complement of ∆n in ∆n+1 AiBjC kαrβsγt, i, j, k, r, s, t, ∈ N, i + j + k + r + s + t = n + 1. We denote the span of the vectors (20) by ∆c n. By [7, Lemma 6.1], the following elements of ∆ coincide and are central. qABC + q2A2 + q−2B2 + q2C2 − qAα − q−1Bβ − qCγ, qBCA + q2A2 + q2B2 + q−2C2 − qAα − qBβ − q−1Cγ, qCAB + q−2A2 + q2B2 + q2C2 − q−1Aα − qBβ − qCγ, q−1CBA + q−2A2 + q2B2 + q−2C2 − q−1Aα − qBβ − q−1Cγ, q−1ACB + q−2A2 + q−2B2 + q2C2 − q−1Aα − q−1Bβ − qCγ, q−1BAC + q2A2 + q−2B2 + q−2C2 − qAα − q−1Bβ − q−1Cγ. (17) (18) (19) (20) (21) (22) (23) (24) (25) (26) Denote this element by Ω, which is called in [7] as the Casimir element of ∆. As shown in [7, Section 7], we have other bases for ∆, ∆n (for n ∈ N) that involve Ω. First, the following vectors form a basis for ∆. AiBjC kΩlαrβsγt, i, j, k, l, r, s, t ∈ N, ijk = 0. (27) Given n ∈ N, a basis for ∆n consists of the vectors AiBjC kΩlαrβsγt, i, j, k, l, r, s, t ∈ N, ijk = 0, i + j + k + 3l + r + s + t ≤ n, (28) while the following vectors form a basis for a complement of ∆n in ∆n+1. AiBjC kΩlαrβsγt, i, j, k, l, r, s, t ∈ N, ijk = 0, i + j + k + 3l + r + s + t = n + 1. Recall that ∆ is a Lie algebra with Lie bracket [X, Y ] := XY − Y X for X, Y ∈ ∆. Denote the derived algebra of ∆ by [∆, ∆], and the ideal of ∆ generated by [∆, ∆] by ∆ [∆, ∆] ∆. It follows that the Lie subalgebra of ∆ generated by A, B, C is L := L/(I ∩ L). Given nonzero n ∈ N, denote the span of all Hall i=1 Li under the canonical map L → L by Ln. basis elements of L of length n by Ln. Denote the image of Pn It follows that all standard Lie monomials of L of length at most n span Ln. Proposition 3.4. L ⊆ FA + FB + FC + ∆ [∆, ∆] ∆. Proof. By the definition of L, we have L ⊆ FA + FB + FC + [∆, ∆]. Since ∆ has a multiplicative identity, we have [∆, ∆] ⊆ ∆ [∆, ∆] ∆. From these we get the desired set inclusion. (cid:4) Proposition 3.5. If q is not a root of unity, then L has zero center. 6 Proof. Let q be not a root of unity. Suppose that Z(L) has a nonzero element f . Since the generators A, B, C of ∆ are also in L, we have Z(L) ⊆ Z(∆). By [7, Corollary 8.3], Z(∆) is generated by α, β, γ, Ω. Observe that there exists a filtration subspace ∆n such that f ∈ ∆n, and that ∆n ∩ Z(∆) has a basis consisting of the vectors Since f is nonzero, there exists a nonzero c ∈ F and a vector Ωwαxβyγz in (29) such that Ωlαrβsγt, l, r, s, t ∈ N, 3l + r + s + t ≤ n. f − cΩwαxβyγz = g, (29) (30) algebra of polynomials in three mutually commuting indeterminates ¯A, ¯B, ¯C, with coefficients from F. As where g is a linear combination of vectors in (29) other than Ωwαxβyγz. Let F(cid:2) ¯A, ¯B, ¯C(cid:3) denote the F- shown in [7, p. 17], there exists a unique surjective F-algebra homomorphism Ψ : ∆ → F(cid:2) ¯A, ¯B, ¯C(cid:3) with kernel ∆ [∆, ∆] ∆ that sends (31) Under this homomorphism, denote the images of α, β, γ, Ω by ¯α, ¯β, ¯γ, ¯Ω, respectively. As shown in [7, Lemma 11.3], we have A 7→ ¯A, B 7→ ¯B, C 7→ ¯C. ¯α = (q + q−1) ¯A + ¯B ¯C, ¯β = (q + q−1) ¯B + ¯A ¯C, ¯γ = (q + q−1) ¯C + ¯A ¯B, ¯Ω = (q + q−1) ¯A ¯B ¯C − ¯A2 − ¯B2 − ¯C2. It is routine to show that the vectors ¯A, ¯B, ¯C, ¯Ωl ¯αr ¯βs¯γt, l, r, s, t ∈ N, 3l + r + s + t ≤ n, are linearly independent in F(cid:2) ¯A, ¯B, ¯C(cid:3). Observe also that by Proposition 3.4, f ∈ FA + FB + FC + ker Ψ. Applying Ψ to both sides of (30), we have c1 ¯A + c2 ¯B + c3 ¯C − c ¯Ωw ¯αx ¯βy ¯γz = ¯g, (32) (33) (34) (35) (36) (37) where ¯g is a linear combination of the vectors in (36) except ¯Ωw ¯αx ¯βy ¯γz. We get a contradiction from (37). Therefore, Z(L) = 0. (cid:4) We end this section by discussing some properties of ∆ related to the group P SL2(Z). We denote by P SL2(Z) the free product of the cyclic group of order two and the cyclic group of order three [1]. Let ρ, σ denote the generators of P SL2(Z) such that ρ3 = 1 and σ2 = 1. By [7, Theorem 3.1], the group P SL2(Z) acts faithfully on ∆ as a group of automorphisms in the following way: u ρ(u) σ(u) A B B B C A C A C + (q − q−1)−1[A, B] γ α β β α γ β α γ By [7, Theorem 6.4], Ω is fixed by ρ, σ. It is routine to show that given n ∈ N, the filtration subspace ∆n is invariant under ρ. Proposition 3.6. The Lie algebra L is P SL2(Z)-invariant. Proof. Let τ ∈ {ρ, σ}. It suffices to argue in the following way. Show that the images of the generators A, B, C under τ are in L, and show that if the images of f, g ∈ L under τ are in L, then so is the image of [f, g] under τ . By the above table, we are done with the first step. For the second step, assume that the images of f, g ∈ L under τ are in L. Since τ is an F-algebra automorphism, we have τ ([f, g]) = τ (f g − gf ) = [τ (f ), τ (g)] ∈ L. (cid:4) 7 4 L is not free In this section, all computations are done in the free algebra F hX i. Our goal is to show that I ∩ L contains a nonzero element. Proposition 4.1. In the free algebra F hX i, [BA] q(q − q−1) [CA] q−1(q − q−1) − (q + q−1)C = AB − γ, − (q + q−1)B = AC − β. Proof. Use (13),(14) to get (38),(39), respectively. Proposition 4.2. In the free algebra F hX i, (cid:2)BA2(cid:3) q2(q − q−1)2 − (q + q−1) [CA] q(q − q−1) [BAC] = A2B + (q + q−1)AC − Aγ + r5 q(q − q−1) , (q + q−1)(q − q−1)2 = B2 − A2 + Aα − Bβ + r1 q + q−1 . (38) (39) (cid:4) (40) (41) Proof. Apply −ad A to both sides of (38). The linear combination in the right side of the resulting equation contains ABA which can be further simplified using (13). From this, we get (40). We show (41) holds. Apply −ad C to both sides of (38). The result involves ACB in the right side, which can be further simplified using (15). From this, we get (41). (cid:4) Definition 4.3. We define the following elements of L. H0 := [[CB] , [BA]] − [BABC] (q + q−1)2(q − q−1)2 + (cid:2)BA2(cid:3) q − q−1 − 2 [CA] I0 := [H0, [BA]] Lemma 4.4. In the free algebra F hX i, [BA] α (q + q−1)2 = H0 + r1B − Ar3 (q + q−1)2 + q(q − q−1)r5 q + q−1 . (42) (43) (44) Proof. Apply −ad B to both sides of (41). The left side involves BA2 which can be written uniquely as a linear combination of A2B, AC, Aγ, B, β by repeated use of the relations (13),(14). We get [BACB] q2(q + q−1)2(q − q−1)3 = A2B + q−3(q4 + 1)AC − Aγ −q−2(q + q−1)(q − q−1)B + q−2(q − q−1)β + r5 q(q + q−1) − [BA] α + Ar3 − r1B q2(q + q−1)2(q − q−1) (45) In (45), eliminate A2B, Aγ using (40), and eliminate AC, B, β using (39). In the resulting equation, express all Lie monomials in terms of Hall basis elements. From this, we get (44). (cid:4) Lemma 4.5. In the free algebra F hX i, I0 = [BA] [r0, B] − [r3, A] (q + q−1)2 − q(q − q−1) [r5, [BA]] q + q−1 + [Ar3 − r1B, [BA]] (q + q−1)2 . Proof. Apply −ad [BA] to both sides of (44). The resulting left side is [[BA] α, [BA]] (q + q−1)2 , 8 (46) (47) which can be simplified into [BA] [r0, B] − [r3, A] (q + q−1)2 using r0 = [A, α] , r3 = [B, α]. The result is (46). (cid:4) Theorem 4.6. The Lie algebra L = L/(I ∩ L) is not freely generated by A, B, C. Proof. Observe that if we write I0 in terms of Hall basis elements, we have I0 = [[[CB] , [BA]] , [BA]] − [[BABC] , [BA]] (q + q−1)2(q − q−1)2 + (cid:2)(cid:2)BA2(cid:3) , [BA](cid:3) q − q−1 − 2 [[CA] , [BA]] , which, by the linear independence of the Hall basis elements, implies that I0 6= 0. But by (46), we have I0 ∈ I ∩ L. Therefore, L/(I ∩ L) 6= L. (cid:4) 5 Properties of some standard Lie monomials of L We discuss properties of some standard Lie monomials of L in relation to the filtration {∆n}n∈N of ∆. Proposition 5.1. For any i, j ∈ N, the following hold in ∆. (cid:2)BAi(cid:3) − qi(q − q−1)iAiB ∈ ∆i, (cid:2)CAi(cid:3) − (−1)iq−i(q − q−1)iAiC ∈ ∆i, (cid:2)CBj(cid:3) − qj(q − q−1)jBjC ∈ ∆j. (cid:2)BAi−1(cid:3) − qi−1(q − q−1)i−1Ai−1B ∈ ∆i−1. Proof. We show (48) holds by induction on i. The case i = 0 is trivial. Suppose that for some i ∈ N, we have Denote the element in (51) by X. By the properties of the filtration {∆n}n∈N, we have [X, A] ∈ ∆i. Using (48), we further obtain (48) (49) (50) (51) (cid:4) (52) (53) (54) (55) (56) (57) [X, A] + qi(q − q−1)iAi−1(cid:0)(q + q−1)C − γ(cid:1) = (cid:2)BAi(cid:3) − qi(q − q−1)iAiB, which proves (48). The relations (49) and (50) are proven similarly. Proposition 5.2 ([7, Lemma 8.1]). Let i, j, k ∈ N. Then the following hold in ∆. (cid:2)A, AiBjC k(cid:3) − (cid:0)1 − q2(j−k)(cid:1) Ai+1BjC k ∈ ∆i+j+k, (cid:2)B, AiBjC k(cid:3) − (cid:0)q2i − q2k(cid:1) AiBj+1C k (cid:2)C, AiBjC k(cid:3) − (cid:0)q2(j−i) − 1(cid:1) AiBj C k+1 ∈ ∆i+j+k. ∈ ∆i+j+k, Proposition 5.3. For nonzero i, j, k ∈ N, the following hold in ∆. (cid:2)BAiBj(cid:3) − (−1)jqi(q2i − 1)j(q − q−1)iAiBj+1 ∈ ∆i+j, (cid:2)CAiC k(cid:3) − (−1)iq−i(2k+1)(q2i − 1)k(q − q−1)iAiC k+1 ∈ ∆i+k, (cid:2)CBj C k(cid:3) − (−1)kqj(q2j − 1)k(q − q−1)jBjC k+1 ∈ ∆j+k. Proof. To show (55), use the relation (48), the relation (53) with k set to zero, and induction on j. The relations (56) and (57) are proven similarly. (cid:4) Proposition 5.4. The complement ∆c 1 of ∆1 in ∆2 contains [CAB] and [BAC]. 9 Proof. Use the canonical map F hX i → ∆ on (41) in order to obtain [BAC] (q + q−1)(q − q−1)2 = B2 − A2 + Aα − Bβ q + q−1 ∈ ∆c 1. Apply −ρ2 to both sides of (58). We get [CAB] (q + q−1)(q − q−1)2 = C2 − A2 + Aα − Cγ q + q−1 ∈ ∆c 1. (cid:4) Proposition 5.5. For nonzero i, j, k ∈ N with i ≥ 2, the following hold in ∆. (cid:2)BABjC k(cid:3) − (−1)j+kq−j(q2j − 1)k(q − q−1)j+1Bj C k−1Ω ∈ ∆j+k+1, (cid:2)CAiBj(cid:3) − (−1)i+jq1−i(q2(i−1) − 1)j(q − q−1)iAi−1Bj−1Ω ∈ ∆i+j , (cid:2)CAiBC k(cid:3) − (−1)i+1q(1−i)(1+2k)(q2(i−1) − 1)k+1(q − q−1)iAi−1C kΩ ∈ ∆i+k+1. Proof. We show (59) holds. We first consider the case k = 1. By setting i = 1 in (55), we get Apply −ad C to the element in (62). We get (cid:2)BABj(cid:3) − (−1)jqj+1(q − q−1)j+1ABj+1 ∈ ∆j+1. In (54) set i, j, k to 1, j + 1, 0, respectively, and combine with (63). We have (cid:2)BABj C(cid:3) − (−1)jqj+1(q − q−1)j+1(cid:2)ABj+1, C(cid:3) ∈ ∆j+2. (cid:2)BABj C(cid:3) − (−1)j+1qj+1(q2j − 1)(q − q−1)j+1ABj+1C ∈ ∆j+2. Using (13), it is routine to show that ABn − q−2nBnA ∈ ∆n, for n ∈ N. Set n = j + 1 in (65) and multiply the element by C from the right. We get ABj+1C − q−2(j+1)Bj+1AC ∈ ∆j+2. From (64) and (66), we get (cid:2)BABj C(cid:3) − (−1)j+1q−(j+1)(q2j − 1)(q − q−1)j+1Bj+1AC ∈ ∆j+2. Using the fact that Ω is equal to (26), we have Bj+1AC − qBjΩ ∈ ∆j+2. From (67) and (68), we get (cid:2)BABj C(cid:3) − (−1)j+1q−j(q2j − 1)(q − q−1)j+1BjΩ ∈ ∆j+2, from which we see that (59) holds for k = 1 and for nonzero j ∈ N. Using (54),(69) and induction on k, we find that (59) holds for nonzero j, k ∈ N. We now show (60) holds. Since i ≥ 2, we can rewrite (59) changing the exponents j, k to i − 1, j, respectively. (58) (59) (60) (61) (62) (63) (64) (65) (66) (67) (68) (69) (70) (cid:2)BABi−1C j(cid:3) − (−1)i+j−1q1−i(q2(i−1) − 1)j(q − q−1)iBi−1C j−1Ω ∈ ∆i+j Denote the element in (70) by X. Since ∆i+j is invariant under ρ, we have −ρ2(X) ∈ ∆i+j , where −ρ2(X) is the element in (60). Thus, (60) holds for nonzero i, j ∈ N. Finally, we show (61) holds. Set j = 1 in (60). (cid:2)CAiB(cid:3) − (−1)i+1q1−i(q2(i−1) − 1)(q − q−1)iAi−1Ω ∈ ∆i+1 (71) 10 Since Ω is central, if we apply −ad C to the element in (71), we get From (54) we obtain (cid:2)CAiBC(cid:3) − (−1)i+1q1−i(q2(i−1) − 1)(q − q−1)i(cid:2)Ai−1, C(cid:3) Ω ∈ ∆i+2. From (71) and (73), (cid:2)Ai−1, C(cid:3) Ω − q−2(i−1)(q2(i−1) − 1)Ai−1CΩ ∈ ∆i+2. (cid:2)CAiBC(cid:3) − (−1)i+1q(1−i)·3(q2(i−1) − 1)2(q − q−1)iAi−1CΩ ∈ ∆i+2, from which we see that (61) holds for k = 1. Using (54),(74) and induction on k, we find that (61) holds for nonzero i, k ∈ N with i ≥ 2. (cid:4) Proposition 5.6. The following hold in ∆. (cid:2)CA2B(cid:3) + (q − q−1)3AΩ ∈ ∆3, [BABC] − (q − q−1)3BΩ ∈ ∆3, [[CB] , [CA]] − (q − q−1)3CΩ ∈ ∆3. (75) (76) (77) Proof. The relations (75),(76) follow from (60),(59), respectively. We show (77) holds. Let V := −(cid:2)BAC2(cid:3)+ [CABC]. By Proposition 5.4, we have V ∈ ∆3. Denote the element in (76) by X. Using the fact that ∆3 is invariant under ρ, we have ρ(X) = [CBCA] − (q − q−1)3CΩ ∈ ∆3. Using the Jacobi identity to express [CBCA] in terms of standard Lie monomials, we further have and it follows that ρ(X) = [[CB] , [CA]] − V − (q − q−1)3CΩ ∈ ∆3, [[CB] , [CA]] − (q − q−1)3CΩ = V + ρ(X) ∈ ∆3. (cid:4) 6 The standard Lie monomials of L of length at most 4 Recall that the span of the standard Lie monomials of L of length at most n is Ln. Our goal in this section is to show that the standard Lie monomials of L of length at most 4 are linearly independent, and hence form a basis for L4. Proposition 6.1. For nonzero j, k ∈ N, the following hold in ∆. (72) (73) (74) (78) (79) (80) (81) (82) (83) (cid:2)BABj(cid:3) − (−1)jq(j+1)(q − q−1)j+1ABj+1 ∈ ∆j+1, (cid:2)CAC k(cid:3) + q−(k+2)(q − q−1)k+1AC k+1 ∈ ∆k+1, (cid:2)CBC k(cid:3) − (−1)kq(k+1)(q − q−1)k+1BC k+1 ∈ ∆k+1, (cid:2)BA2Bj(cid:3) − (−1)jq2(j+1)(q + q−1)j(q − q−1)j+2A2Bj+1 ∈ ∆j+2, (cid:2)CA2C k(cid:3) − q−2(k+1)(q + q−1)k(q − q−1)k+1A2C k+1 ∈ ∆k+2, (cid:2)CB2C k(cid:3) − (−1)kq2(k+1)(q + q−1)k(q − q−1)k+2B2C k+1 ∈ ∆k+2. Proof. Set i = 1, 2 in (55) to get (78),(81). Do similarly to (56) and (57) in order to show the other relations. (cid:4) Lemma 6.2. Fix a nonzero n ∈ N. The following vectors are linearly independent in ∆ for any i, j, k ∈ N such that 1 ≤ i, j, k ≤ n. 1, A, B, C, [CAB] , [BAC] , (cid:2)CA2B(cid:3) , [BABC] , [[CB] , [CA]] , (cid:2)BAi(cid:3) ,(cid:2)BABj(cid:3) ,(cid:2)BA2Bj(cid:3) , (cid:2)CAi(cid:3) ,(cid:2)CAC k(cid:3) ,(cid:2)CA2C k(cid:3) , (cid:2)CBj(cid:3) ,(cid:2)CBC k(cid:3) ,(cid:2)CB2C k(cid:3) . 11 (84) (85) (86) (87) (88) (89) Proof. Fix n ∈ N. It suffices to show that there exists an upper triangular transition matrix from the above vectors to a subset of the basis of ∆ consisting of the vectors in (27): AiBjC kΩlαrβsγt, i, j, k, l, r, s, t ∈ N, ijk = 0. Let i, j, k ∈ N such that 1 ≤ i, j, k ≤ n. From Propositions 5.1, 5.4, 5.6, 6.1 we have the following data: [CAB] − c1C2 − d1A2 − d2Cγ − d3Aα ∈ ∆0, [BAC] − c2B2 − d4A2 − d5Bβ − d6Aα ∈ ∆0, (cid:2)CA2B(cid:3) − c3AΩ ∈ ∆3, [BABC] − c4BΩ ∈ ∆3, [[CB] , [CA]] − c5CΩ ∈ ∆3, (cid:2)BAi(cid:3) − eiAiB ∈ ∆i, (cid:2)CAi(cid:3) − fiAiC ∈ ∆i, (cid:2)CBj(cid:3) − gjBjC ∈ ∆j, jABj+1 ∈ ∆j+1, kAC k+1 ∈ ∆k+1, kBC k+1 ∈ ∆k+1, j A2Bj+1 ∈ ∆j+2, k A2C k+1 ∈ ∆k+2, k B2C k+1 ∈ ∆k+2, (cid:2)BABj(cid:3) − e′ (cid:2)CAC k(cid:3) − f ′ (cid:2)CBC k(cid:3) − g′ (cid:2)BA2Bj(cid:3) − e′′ (cid:2)CA2C k(cid:3) − f ′′ (cid:2)CB2C k(cid:3) − g′′ (90) (91) (92) (93) (94) (95) (96) (97) (98) (99) (100) (101) (102) (103) where the small letters (other than i, j, k) denote scalars. Each of (92) to (103) is of the form M − aV ∈ ∆m, where M is a Lie monomial, a ∈ F, and V is an element of the basis of ∆ consisting of the vectors in (27), and V /∈ ∆m. Call V the leading term of M . For (90),(91), define the leading terms of [CAB], [BAC] by C2, B2, respectively. Observe that no two distinct Lie monomials found in (90) to (103) have the same leading terms. This yields a transition matrix from the vectors (84) to (89) to some of the vectors in (27) such that all entries below the main diagonal are zero, and that the diagonal entries are c1, . . . , c5, ei, fi, gj, e′ j, f ′ k, g′ k, e′′ j , f ′′ k , g′′ k . By Propositions 5.1, 5.4, 5.6 and 6.1, all such scalars are nonzero. Hence, the transition matrix is upper triangular. (cid:4) Notation 6.3. Let In denote the set consisting of all the linearly independent vectors in Lemma 6.2. Lemma 6.4. Fix nonzero m, n ∈ N. The vectors Xαrβsγt are linearly independent in ∆ for any X ∈ In and any r, s, t ∈ N such that r + s + t ≤ m. Proof. The proof is similar to that of Lemma 6.2, but with (90) to (103) modified as follows. For each of (90) to (103), multiply the element by αrβsγt and add r + s + t to the index of the filtration subspace. Based on these new data, an upper triangular transition matrix can be constructed. (cid:4) Notation 6.5. Let Im Observe that the vectors n denote the set consisting of all the linearly independent vectors in Lemma 6.4. are in I 1 3 . Let I ∗ denote the set obtained from I 1 3 by replacing the vectors in (104) by the vectors [CB] γ, [BA] β, [CA] α, [CB] β, [CA] γ, [BA] α, (cid:2)CAB2(cid:3) , (cid:2)BA2C(cid:3) , [CABC] , (cid:2)BAC2(cid:3) , [[CA] , [BA]] , [[CB] , [BA]] . 12 (104) (105) Proposition 6.6. The following hold in ∆. q − q−1 − 2 [CA] , [[CB] , [BA]] − [BABC] [BA] α (q + q−1)2 = (q + q−1)2 = (cid:2)BAC2(cid:3) − [CABC] [CB] β [CA] γ (q + q−1)2 = − [BA] β (q + q−1)2 = [CB] γ (q + q−1)2 = [CA] α (q + q−1)2 = (q + q−1)2(q − q−1)2 + (cid:2)BA2(cid:3) (q + q−1)2(q − q−1)2 + (cid:2)CB2(cid:3) [[CA] , [BA]] +(cid:2)CA2B(cid:3) (q + q−1)2(q − q−1)2 + [[CA] , [BA]] −(cid:2)BA2C(cid:3) (q + q−1)2(q − q−1)2 − − [[CB] , [BA]] + [BABC] −(cid:2)CAB2(cid:3) (q + q−1)2(q − q−1)2 − (cid:2)CA2(cid:3) (q + q−1)2(q − q−1)2 − [CABC] q − q−1 + 2 [BA] , [CAC] q − q−1 − 2 [CB] , [BAB] q − q−1 + 2 [CB] , q − q−1 − 2 [BA] . − [CBC] q − q−1 − 2 [CA] , Proof. Apply the canonical map F hX i → ∆ to both sides of (44) to get (106). Apply ρ, ρ2 to both sides of (106) to get (107),(108), respectively. To get (109), apply σ to both sides of (106). Apply ρ, ρ2 to both sides of (109) to get (110),(111), respectively. (cid:4) Lemma 6.7. The vectors in I ∗ are linearly independent in ∆. Proof. We use (106) to (111) to construct a transition matrix from the elements of I 1 I ∗. Denote such transition matrix by T . Order the rows of T such that the last 17 correspond to 3 to the elements of (106) (107) (108) (109) (110) (111) (112) (113) (114) [BA] , [CA] , [CB] , (cid:2)BA2(cid:3) , (cid:2)CA2(cid:3) , [BAB] , (cid:2)CB2(cid:3) , [CAC] , [CBC] , (cid:2)CA2B(cid:3) , [BABC] , (cid:2)CAB2(cid:3) , (cid:2)BA2C(cid:3) , [CABC] , (cid:2)BAC2(cid:3) , [[CA] , [BA]] , [[CB] , [BA]] , in that order, while order the columns of T such that the last 17 correspond to the vectors in (112),(113) together with [CB] γ, [BA] β, [CA] α, [CB] β, [CA] γ, [BA] α. (115) Observe that all the vectors in I 1 3 to be replaced to form I ∗ are in (115), all the replacements are in (114), and all the other vectors that appear in (106) to (111) (which we use to construct the transition matrix) appear in (112),(113). Then T is of the form 0 T ′ (cid:21) T = (cid:20) I M where I is an identity matrix, M is some matrix with 7 columns, and T ′ is a 7 × 7 matrix which has the following properties. All diagonal entries of T ′ are nonzero. Denote the ij-entry of T ′ by T ′ ij. All entries of T ′ below the main diagonal and all entries in the first two rows are zero except the ones that appear below: 1 (q − q−1)2 = −T ′ 11 = T ′ 31 = −T ′ 71 = −T ′ 22 = T ′ 62 6= 0. By these observations about T ′, we find that T is invertible. This implies that the vectors in I ∗ are linearly independent. (cid:4) Theorem 6.8. The standard Lie monomials of L of length at most 4 form a basis for L4. Proof. All such vectors are in I ∗. Use Lemma 6.7 and the fact that the vectors in the statement span L4. (cid:4) 13 7 The standard Lie monomials of L of length at most 5 In this section, we show the Lie algebra relations that hold in L5. We also exhibit a basis for L if q is not a sixth root of unity. Lemma 7.1. The following hold in ∆. [CA] β + [BA] γ = −(cid:2)(cid:2)BA2(cid:3) , [CA](cid:3) +(cid:2)(cid:2)CA2(cid:3) , [BA](cid:3) (q − q−1)3 − (cid:2)CA2C(cid:3) +(cid:2)BA2B(cid:3) (q − q−1)2 − (q + q−1)2 ([BAC] − [CAB]) q − q−1 [CB] α − [BA] γ = − [[BAB] , [CB]] +(cid:2)(cid:2)CB2(cid:3) , [BA](cid:3) (q − q−1)3 − (cid:2)CB2C(cid:3) −(cid:2)BA2B(cid:3) (q − q−1)2 − (q + q−1)2 [CAB] q − q−1 (116) (117) Proof. In view of Remark 3.3, write each of the left and right sides of (116) as a linear combination of irreducible ∆-words. This yields the same linear combination of the basis vectors (16) of ∆. Apply ρ to both sides of (116) to get (117). (cid:4) Theorem 7.2. The following relations hold in L. (cid:2)BA2BC(cid:3) q − q−1 = − −(2q2 + 1)(q2 + 2)(cid:0)(cid:2)(cid:2)BA2(cid:3) , [CB](cid:3) + [[BAB] , [CA]](cid:1) 2q2(q + q−1)2(q − q−1) (cid:2)(cid:2)CB2(cid:3) , [CA](cid:3) 2(q + q−1)2(q − q−1) = (cid:2)(cid:2)CA2(cid:3) , [CB](cid:3) 2(q + q−1)2(q − q−1) = − (q4 + 3q2 + 1) ([[BAC] , [BA]] − 2 [[CAB] , [BA]]) 2q2(q + q−1)2(q − q−1) −(cid:2)BAC2(cid:3) + 2 [CABC] −(cid:2)BAB2(cid:3) +(cid:2)BA3(cid:3) , −(3q4 + 5q2 + 3) [[BAC] , [CB]] 2q2(q + q−1)2(q − q−1) + − (q4 + 3q2 + 1) [[CAB] , [CB]] 2q2(q + q−1)2(q − q−1) (2q4 + 3q2 + 2) [[CBC] , [BA]] 2q2(q + q−1)2(q − q−1) q − q−1 −(cid:2)BABC2(cid:3) −(cid:2)CAB2C(cid:3) +(cid:2)CBC2(cid:3) +(cid:2)BA2C(cid:3) −(cid:2)CB3(cid:3) +(cid:2)CA2B(cid:3) , −2 [[BAC] , [CA]] + [[CAB] , [CA]] 2q2(q + q−1)2(q − q−1) + (2q2 + 1)(q2 + 2) [[CAC] , [BA]] 2q2(q + q−1)2(q − q−1) [[CAC] , [CB]] = − [[BAB] , [CB]] + [[CBC] , [CA]] −(cid:2)(cid:2)BA2(cid:3) , [CA](cid:3) q − q−1 − 2 [[CB] , [BA]] −(cid:2)CAC2(cid:3) +(cid:2)CA2BC(cid:3) +2 [BABC] −(cid:2)CAB2(cid:3) +(cid:2)CA3(cid:3) , +(cid:2)(cid:2)CB2(cid:3) , [BA](cid:3) +(cid:2)(cid:2)CA2(cid:3) , [BA](cid:3) . 14 (118) (119) (120) (121) Proof. To show (118), we first show that the equation (cid:2)BA2BC(cid:3) (q + q−1)2(q − q−1)2 = q − q−1 q2(q − q−1) −(cid:2)BAC2(cid:3) + 2 [CABC] (q4 + 1)(cid:0)(cid:2)BAB2(cid:3) −(cid:2)BA3(cid:3)(cid:1) 2(q6 − 1)(cid:0)(cid:2)CB2(cid:3) −(cid:2)CA2(cid:3)(cid:1) (q + q−1)2 + [BAB] β +(cid:2)BA2(cid:3) α q3(q − q−1) [BAC] γ + − + (122) holds in ∆. In view of Remark 3.3, write each of the left and right sides of (122) as a linear combination of irreducible ∆-words. This yields the same linear combination of the basis vectors (16) of ∆. Apply −ad A, −ad B, −ad B, −ad C to both sides of (106),(108),(109),(110), respectively. Write all Lie monomials in standard form. We get (cid:2)BA2(cid:3) α = f1, (cid:2)CA2(cid:3) β = f2, [CAB] γ = f3, [BAC] γ + k [CAB] γ = f4, (123) (124) (125) (126) for some k ∈ F and some f1, f2, f3, f4 ∈ L. Eliminate (cid:2)BA2(cid:3) α,(cid:2)CA2(cid:3) β, [BAC] γ in (122) using (123) to (126). The result is (118). Apply ρ, ρ2 to both sides of (118) in order to obtain (119),(120), respectively. We now show (121) holds. Add (116) and (117). We get where g1, g2 are the right sides of (116),(117), respectively. Apply −ρ to both sides of (117). We get [CB] α + [CA] β = g1 + g2, [CB] α + [CA] β = g3, (127) (128) for some g3 ∈ L such that [[CAC] , [CB]] appears with nonzero coefficient in g3. Eliminate [CB] α + [CA] β in (127),(128). We get (121) as desired. (cid:4) At this point we have shown that each of (cid:2)BA2BC(cid:3) ,(cid:2)(cid:2)CB2(cid:3) , [CA](cid:3) ,(cid:2)(cid:2)CA2(cid:3) , [CB](cid:3) , [[CAC] , [CB]] , (129) is linearly dependent on standard Lie monomials of length at most 5 that are not in (129). In what follows, we shall show that the standard Lie monomials of length at most 5 except (129) are linearly independent in ∆. Proposition 7.3. The following hold in ∆. (cid:2)CA3B(cid:3) − (q + q−1)(q − q−1)4A2Ω ∈ ∆4, (cid:2)CA2B2(cid:3) − q(q − q−1)4ABΩ ∈ ∆4, (cid:2)CA2BC(cid:3) + q−1(q − q−1)4ACΩ ∈ ∆4, (cid:2)BAB2C(cid:3) + (q + q−1)(q − q−1)4B2Ω ∈ ∆4, (cid:2)BABC2(cid:3) + q(q − q−1)4BCΩ ∈ ∆4, [[CBC] , [CA]] + (q + q−1)(q − q−1)4C2Ω ∈ ∆4. Proof. Use Proposition 5.5 to show (130) to (134) hold. To show (135), set i = 2 in (49). We get (cid:2)CA2(cid:3) − q−2(q − q−1)2A2C ∈ ∆2. 15 (130) (131) (132) (133) (134) (135) (136) Applying −ρ2 to the element in (136) and using the fact that ∆2 is invariant under ρ, we have Set i = 1 in (49). We get [CBC] + q−2(q − q−1)2C2B ∈ ∆2. [CA] + q−1(q − q−1)AC ∈ ∆1. By taking the Lie bracket of the elements in (137) and (138), we have [[CBC] , [CA]] − q−3(q − q−1)3(cid:0)C2BAC − AC3B(cid:1) ∈ ∆3 ⊂ ∆4. Using (14), it is routine to show that C2A − q−4AC2 ∈ ∆2, from which we obtain Combining (139) and (140), we obtain (q − q−1)3(cid:0)qC2ACB − q−3AC3B(cid:1) ∈ ∆4. Using the fact that (25) and (26) are both equal to Ω, we have [[CBC] , [CA]] − (q − q−1)3C2(cid:0)q−3BAC − qACB(cid:1) ∈ ∆4. (q + q−1)(q − q−1)Ω + q−3BAC − qACB ∈ ∆2. Finally, we get (135) from (141) and (142). Proposition 7.4. For nonzero j, k ∈ N, the following hold in ∆. (cid:2)BA3Bj(cid:3) − (−1)jq3(q6 − 1)j(q − q−1)3A3Bj+1 ∈ ∆j+3, (cid:2)CA3C k(cid:3) + q−3(2k+1)(q6 − 1)k(q − q−1)3A3C k+1 ∈ ∆k+3, (cid:2)CB3C k(cid:3) − (−1)kq3(q6 − 1)k(q − q−1)3B3C k+1 ∈ ∆k+3. Proof. Use Proposition 5.3. (137) (138) (139) (140) (141) (142) (cid:4) (143) (144) (145) (cid:4) Lemma 7.5. Assume q is not a sixth root of unity. Fix a nonzero n ∈ N. The following vectors are linearly independent in ∆ for any i, j, k ∈ N such that 1 ≤ i, j, k ≤ n. 1, A, B, C, [CAB] , [BAC] , (cid:2)CA2B(cid:3) , [BABC] , [[CB] , [CA]] , (cid:2)CA3B(cid:3) ,(cid:2)CA2B2(cid:3) ,(cid:2)CA2BC(cid:3) , (cid:2)BAB2C(cid:3) ,(cid:2)BABC2(cid:3) , [[CBC] , [CA]] , (cid:2)BAi(cid:3) ,(cid:2)BABj(cid:3) ,(cid:2)BA2Bj(cid:3) ,(cid:2)BA3Bj(cid:3) , (cid:2)CAi(cid:3) ,(cid:2)CAC k(cid:3) ,(cid:2)CA2C k(cid:3) ,(cid:2)CA3C k(cid:3) , (cid:2)CBj(cid:3) ,(cid:2)CBC k(cid:3) ,(cid:2)CB2C k(cid:3) ,(cid:2)CB3C k(cid:3) . (146) (147) (148) (149) (150) (151) (152) (153) Proof. The proof is similar to that of Lemma 6.2. In order to construct the desired upper triangular transition matrix, we combine the data from (90) to (103) to that in (130) to (135), and (143) to (145). Recall that the transition matrix that can be constructed from the data is upper triangular if the scalar coefficients of the leading terms are nonzero. Those in (90) to (103) are nonzero as shown in the proof of Lemma 6.2. The scalar coefficients in (130) to (135) are nonzero by the manner q is defined. Finally, the scalar coefficients in (143) to (145) are nonzero since q is further assumed to be not a sixth root of unity. (cid:4) Notation 7.6. Let Jn denote the set consisting of all the linearly independent vectors in Lemma 7.5. 16 Lemma 7.7. Assume q is not a sixth root of unity. Fix nonzero m, n ∈ N. The vectors Y αrβsγt are linearly independent in ∆ for any Y ∈ Jn and any r, s, t ∈ N such that r + s + t ≤ m. Proof. The proof is similar to that of Lemma 6.2. For each of the data used in the proof of Lemma 7.5, which are (90) to (103), (130) to (135), and (143) to (145), multiply the element by αrβsγt and add r + s + t to the index of the filtration subspace. Use these new data to construct a similar upper triangular transition matrix. (cid:4) Notation 7.8. Let J m Observe that the vectors n denote the set consisting of all the linearly independent vectors in Lemma 7.7. [CB] γ, [BA] β, [CA] α, [CB] β, [CA] γ, [BA] α, (154) are in J 1 4 . Let J ∗ denote the set obtained from J 1 4 by replacing the vectors in (154) by the following vectors (cid:2)CAB2(cid:3) , (cid:2)BA2C(cid:3) , [CABC] , (cid:2)BAC2(cid:3) , [[CA] , [BA]] , [[CB] , [BA]] . (155) Lemma 7.9. Assume q is not a sixth root of unity. The vectors in J ∗ are linearly independent in ∆. Proof. The proof is similar to that of Lemma 6.7. Proposition 7.10. Assume q is not a sixth root of unity. If V is a subspace of Span J ∗ such that Span J ∗ = V + Span J 0 4 , (direct sum) then a basis for V is J ∗\J 0 4 . Proof. This follows from the fact that J ∗, J 0 4 are both linearly independent sets and that J 0 4 ⊂ J ∗. Lemma 7.11. The following hold in ∆. (cid:4) (cid:4) (156) (157) (158) (159) (160) (161) (162) (163) (q − q−1)2 ∈ Span J 0 4 , (q − q−1)2 ∈ Span J 0 4 , [CAC] α + (cid:2)CABC2(cid:3) (q − q−1)2 + (cid:2)CABC2(cid:3) (q6 − 1)(cid:2)(cid:2)BA2(cid:3) , [BA](cid:3) q3(q − q−1)3 [CBC] β − (cid:2)BAC3(cid:3) (cid:2)BA2(cid:3) γ − (q + q−1)2 [[BAB] , [BA]] [BAB] γ + (cid:2)CA2(cid:3) β − [CAC] β + (cid:2)CB2(cid:3) α − [CBC] α + (q − q−1)2 (q6 − 1)(cid:2)(cid:2)CA2(cid:3) , [CA](cid:3) q3(q − q−1)3 (q + q−1)2 [[CAC] , [CA]] (q − q−1)2 (q6 − 1)(cid:2)(cid:2)CB2(cid:3) , [CB](cid:3) q3(q − q−1)3 (q + q−1)2 [[CBC] , [CB]] (q − q−1)2 ∈ Span J 0 4 , ∈ Span J 0 4 , ∈ Span J 0 4 , ∈ Span J 0 4 , ∈ Span J 0 4 , ∈ Span J 0 4 . Proof. Apply −ad C to both sides of (111),(107) and write all Lie monomials in standard form in order to get (156),(157), respectively. To get (158),(159), we first show that the equation (cid:2)BA2(cid:3) γ 2(q + q−1)2 = (q6 − 1)(cid:2)(cid:2)BA2(cid:3) , [BA](cid:3) 2q3(q + q−1)2(q − q−1)3 − [[CA] , [BA]] +(cid:2)BA2C(cid:3) 2(q − q−1) − (cid:2)BA3B(cid:3) 2(q + q−1)2(q − q−1)2 + [CAC] − [BAB] (164) 17 holds in ∆. In view of Remark 3.3, write each of the left and right sides of (164) as a linear combination of irreducible ∆-words. This yields the same linear combination of the basis vectors (16) of ∆. This proves (164), from which (158) follows. To prove (159), apply σ to both sides of (164). We obtain [BAB] γ 2(q + q−1)2 = − [[BAB] , [BA]] 2(q − q−1)2 − [[CB] , [BA]] + [BABC] + 2(q + q−1)2(q − q−1)2 (cid:2)BA2B2(cid:3) − [CBC] −(cid:2)BA2(cid:3) , 2(q − q−1) from which (159) follows. Finally, to prove (160) to (163), apply ρ, ρ2 to both sides of (164),(165). (cid:4) Notation 7.12. Observe that the vectors (165) (166) (167) (168) (169) are in J ∗. Let K0 denote the set obtained from J ∗ by replacing the vectors in (166),(167) by the vectors [CAC] α, [CBC] β, (cid:2)BA2(cid:3) γ, [BAB] γ, (cid:2)CA2(cid:3) β, [CAC] β, (cid:2)CB2(cid:3) α, [CBC] α, (cid:2)CABC2(cid:3) , (cid:2)BAC3(cid:3) , (cid:2)(cid:2)BA2(cid:3) , [BA](cid:3) , [[BAB] , [BA]] , (cid:2)(cid:2)CA2(cid:3) , [CA](cid:3) , [[CAC] , [CA]] , (cid:2)(cid:2)CB2(cid:3) , [CB](cid:3) , [[CBC] , [CB]] . Lemma 7.13. Assume q is not a sixth root of unity. The vectors in K0 are linearly independent in ∆. Proof. We use Lemma 7.11 in order to obtain a transition matrix from the vectors in J ∗ into the vectors in K0. Denote each of (156) to (163) by Mi + fi ∈ Span J 0 4 , where 1 ≤ i ≤ 8, the vector Mi is an element of J ∗ that is to be replaced in order to form K0, while the standard Lie monomials that appear in fi are the replacements. Using the usual ordering of standard Lie monomials, define M i as the largest standard Lie monomial in fi. Let j, k ∈ N, with 1 ≤ j, k ≤ 8. Observe that if Mj 6= Mk then M j 6= M k. Observe also that Mi ∈ J ∗\J 0 4 for 1 ≤ i ≤ 8. By Proposition 7.10, the coefficient of Mi in fi is −1 for all i. By these observations, it follows that there exists an upper triangular transition matrix from the vectors in J ∗ into the vectors in K0 with nonzero diagonal entries. These diagonal entries are precisely the scalar coefficients of M i for all i. These coefficients are all nonzero since q is assumed to be not a sixth root of unity. Since the vectors in J ∗ are linearly independent, the existence of a transition matrix just described implies that the vectors in K0 are also linearly independent. (cid:4) Theorem 7.14. Assume q is not a sixth root of unity. The standard Lie monomials of L of length at most 5 except the vectors from (129) form a basis for L5. Proof. Let K denote the set obtained from K0 by replacing the vectors by the vectors [BAC] γ, [CAB] γ, [BAB] β, (cid:2)BA2(cid:3) α, (cid:2)BA2(cid:3) β, (cid:2)CA2(cid:3) γ, [CA] β, [BAC] β, [CAC] γ, (cid:2)CA2(cid:3) α, [CAB] β, (cid:2)CB2(cid:3) γ, [CB] α, [BAB] α, [CAB] α, (cid:2)CB2(cid:3) β, [BAC] α, [CBC] γ, [[CAB] , [BA]] , [[BAC] , [BA]] , [[BAB] , [CA]] ,(cid:2)(cid:2)BA2(cid:3) , [CB](cid:3) , (cid:2)BA3C(cid:3) ,(cid:2)(cid:2)CA2(cid:3) , [BA](cid:3) ,(cid:2)(cid:2)BA2(cid:3) , [CA](cid:3) , (cid:2)BA2C2(cid:3) , [[CAC] , [BA]] , [[CAB] , [CA]] , [[BAC] , [CA]] , (cid:2)CAB3(cid:3) ,(cid:2)(cid:2)CB2(cid:3) , [BA](cid:3) , [[BAB] , [CB]] , (cid:2)CAB2C(cid:3) , [[CBC] , [BA]] , [[CAB] , [CB]] , [[BAC] , [CB]] . 18 (170) (171) (172) (173) (174) (175) (176) (177) (178) (179) We claim that K is linearly independent. Observe that all vectors mentioned in the statement of the theorem are in K. By Theorem 7.2, the vectors in (129) are linearly dependent on these vectors. The result follows. We now prove our claim. We construct five more sets in a manner similar to the construction of J ∗ from J 1 4 and to that of K0 from J ∗. The goal is that at each step, we prove that the constructed set is linearly independent. Let K1 denote the set obtained from K0 by replacing the vectors (170) in K0 by the vectors (175). Let K2 denote the set obtained from K1 in a similar manner until K5, which is obtained from K4 by replacing the vectors (174) in K4 by the vectors (179). Observe that K5 = K. We show that each of K1, . . . , K5 is a linearly independent set in ∆. Apply −ad A, −ad B, −ad B, −ad A to both sides of (110),(108),(109),(111), respectively. Write all Lie monomials in standard form. We get [BAC] γ + k [CAB] γ = f1, [CAB] γ = f2, [BAB] β = f3, (cid:2)BA2(cid:3) α = f4, (180) (181) (182) (183) for some k ∈ F and some f1, f2, f3, f4 ∈ L. Eliminate (cid:2)BA2BC(cid:3) in (180),(182),(183) using the relation (118). Solve the resulting system in order to obtain [BAC] γ = g1, [CAB] γ = g2, [BAB] β = g3, where each of g1, . . . , g4 ∈ L, is a linear combination of (cid:2)BA2(cid:3) α = g4, (184) (185) (186) (187) [[CAB] , [BA]] , [[BAC] , [BA]] , [[BAB] , [CA]] ,(cid:2)(cid:2)BA2(cid:3) , [CB](cid:3) , together with the vectors in K0. We use the equations (184) to (187) to construct a transtion matrix T1 from the vectors in K0 into those in K1. Index the columns of T1 such that the last 4 correspond to (170), while index the rows such that the last 4 correspond to (175). We find that T1 has the form 0 L1 (cid:21) T1 = (cid:20) I1 U1 where I1 is an identity matrix, U1 is some matrix with 4 columns, and L1 is a 4 × 4 matrix, with det L1 = − 1 2(q + q−1)2(q − q−1)8 6= 0, which implies that det T1 6= 0. Thus, K1 is linearly independent. For 2 ≤ i ≤ 5, we can also construct a transtion matrix Ti from the vectors in Ki−1 into those in Ki in a similar manner. The equations that can be used to construct Ti are also derived from (106) to (111) with the application of the appropriate map ad X where X ∈ {A, B, C}. Furthermore, we find that Ti can be partitioned into four matrices similar to T1, and that if we denote the bottom right partition as Li, then 1 (q − q−1)7 = − det L2 = det L4 1 (q − q−1)8 = − det L3 = − det L5 6= 0, 6= 0, which imply that Ti is invertible for 2 ≤ i ≤ 5. Therefore, K5 = K is a linearly independent set in ∆. (cid:4) 8 Acknowledgements This work was done while the author is a graduate student in De La Salle University, Manila, Philippines, and in part while he was an Honorary Fellow (Sept. to Dec. 2015) in the University of Wisconsin - Madison. He extends his thanks to his mentors, Prof. Arlene Pascasio and Prof. Paul Terwilliger. This research is supported by the Science Education Institute of the Department of Science and Techonology (DOST-SEI), Republic of the Philippines. 19 References [1] R. Alperin, P SL2(Z) = Z2 ∗ Z3, Amer. Math. Monthly 100 (1993) 385-386. [2] R. Askey, J. Wilson, Some basic hypergeometric polynomials that generalize Jacobi polynomials, Mem. Amer. Math. Soc. 54 (1985) 319. [3] G. Bergman, The diamond lemma for ring theory, Adv. Math. 29 (1978) 178-218. [4] A. Bonfiglioli, E. Lanconelli, F. Uguzzoni, Stratified Lie groups and potential theory for their sub- Laplacians, Springer-Verlag, 2007. [5] M. Hall, A basis for free Lie rings and higher commutators in free groups. Proc. Amer. Math. Soc. 1 (1950) 575-581. [6] C. Reutenauer, Free Lie algebras, Oxford Univ. Press, New York, 1993. [7] P. Terwilliger, The universal Askey-Wilson algebra, SIGMA 7 (2011) 069, 24 pages, arXiv:1104.2813. [8] A. Zhedanov, "Hidden symmetry" of the Askey-Wilson polynomials, Theoret. and Math. Phys. 89 (1991) 1146-1157. 20
1811.01392	1	1811	2018-11-04T16:01:16	On linear representation of $\ast$-regular rings having representable ortholattice of projections	[ "math.RA" ]	This paper aims at the following results: \begin{enumerate} \item The class of all $$-regular rings forms a variety. \item A subdirectly irreducible $$-regular ring $R$ is faithfully representable (i.e. isomorphic to a subring of an endomorphisms ring of vector spaces, where the involution is given by adjunction with respect to a scalar product on the vector space) if so is its ortholattice of projections. \end{enumerate} This is a short version of the second author's PhD thesis.	math.RA	math	On linear representation of ∗-regular rings having representable ortholattice of projections Christian Herrmann and Niklas Niemann Abstract. The origings of regular and ∗-regular rings lie in the works of J. von Neu- mann and F.J. Murray on operator algebras, von Neumann-algebras and projection lattices. They constitute a strong connection between operator theory, ring theory and lattice theory. This paper aims at the following results: 1. The class of all ∗-regular rings forms a variety. 2. A subdirectly irreducible ∗-regular ring R is faithfully representable (i.e. isomorphic to a subring of an endomorphisms ring of vector spaces, where the involution is given by adjunction with respect to a scalar product on the vector space) if so is its ortholattice of projections. This is a short version of the second author's PhD thesis. Contents 1 Introduction 1.1 Rings and Lattices . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1 Orthogonalisation of a (n, k)-Frame via J´onsson . . . . 1.2.2 Rings and Frames . . . . . . . . . . . . . . . . . . . . . 1.2.3 Projectivity of Frames . . . . . . . . . . . . . . . . . . 1.3 Concepts of Representability . . . . . . . . . . . . . . . . . . . 2 3 4 7 8 8 9 2 The Variety of ∗-Regular Rings 11 2.1 Directed Unions and Rings without Unit . . . . . . . . . . . . 12 2.2 Representability and Universal Algebra . . . . . . . . . . . . . 13 3 Representability of ∗-Regular Rings 13 3.1 Convention and Notation . . . . . . . . . . . . . . . . . . . . . 14 3.2 General Framework . . . . . . . . . . . . . . . . . . . . . . . . 15 3.2.1 Decomposition Systems & Abstract Matrix Rings . . . 16 3.2.2 Frames and Induced Structures . . . . . . . . . . . . . 18 1 3.3 Representability of ∗-Regular Rings . . . . . . . . . . . . . . . 20 ∗-Regular Rings with Frames . . . . . . . . . . . . . . 21 3.3.1 3.3.2 Simple ∗-Regular Rings . . . . . . . . . . . . . . . . . . 26 3.3.3 Representations of Ideals . . . . . . . . . . . . . . . . . 27 Subdirectly Irreducible ∗-Regular Rings . . . . . . . . . 29 3.3.4 1 Introduction In the present section, we will fix notational conventions and introduce the different concepts of representability. The term ring is used for rings with or without unit. Rings are denoted by R, S, T, C. We consider rings with an involution ∗ : R → R (an anti- automorphism of order two). We consider the unary map as part of the signature of the ring R. A (von Neumann-)regular ring is a ring such that every element x has at least one quasi-inverse y, i.e., for x ∈ R there exists an y ∈ R such that xyx = x. A ∗-regular ring is a regular involutive ring satisfying the implication xx∗ = 0 ⇒ x = 0. The term idempotent is used for a ring element x satisfying x2 = x, the term projection is used for a ring element x satisfying x2 = x∗ = x. We use the letters p, q for projections and e, f, g for idempotents and projections. The term lattice is used for a partially ordered set with binary operations join and meet. These operations are denoted by + and ·. All lattices con- sidered have a smallest element 0. By a bounded lattice, we mean a lattice with top and bottom. We use the terms interval and section of a lattice in the usual way. Intervals and sections are bounded lattices in their own right, with the inherited operations. We use the notation a ⊕ b or L ai for the join of independent elements a and b or for the join of the independent fam- ily {ai : i ∈ I}. We define the height h(L) of a lattice as usual to be the supremum of all cardinalities C − 1, C a chain in L. In this paper, we deal mainly with modular lattices. Of particular interest are (relatively or sectionally) complemented modular lattices and (sectional) modular ortholattices. We use the abbreviations CML and MOL, respectively. We denote the orthocomplementation on a MOL L by ⊥ : L → L. 2 1.1 Rings and Lattices In this section, we recall well-known results about regular rings and the connections between regular rings and complemented modular lattices. Theorem 1.1. A ring with unit is regular if and only if the set of all its principal right ideals is a complemented modular lattice. If R does not contain a unit, the equivalence holds for complemented re- placed by relatively complemented. For a ring R, we denote the set of all its principal right (left) ideals by L(RR) (by L(RR)). Lemma 1.2. A ring with unit (without unit) is ∗-regular if and only if L(RR) is a (sectional) MOL. ⊲ Proof. Folklore. If R is ∗-regular, every principal right ideal is generated by a projection. The orthogonality on L(RR) is given by aR ⊥ bR ⇔ b∗a = 0 If R contains a unit, then the orthocomplement of eR, e a projection in R, is given by (1 − e)R. ⊳ Proposition 1.3. If R is regular, then the lattices L(RR) and L(RR) of prin- cipal right ideals and principal left ideals respectively, are anti-isomorphic. If R is ∗-regular, L(RR) and L(RR) are isomorphic. ⊲ Proof. See [Mic03], [Skor64] and [Mae58]. ⊳ Lemma 1.4. If R is a simple ∗-regular ring (without unit), then L(RR) is a simple (sectional) MOL. ⊲ Proof. See [HR99, Theorem 2.5]. ⊳ Lemma 1.5. In a simple MOL, each non-trivial interval [0, a] is simple. ⊲ Proof. [J´on60, Lemma 2.2]. ⊳ Lemma 1.6. If R is ∗-regular and e a projection in R, then the set eRe is a ∗-regular subring (with unit e) of R. ⊲ Proof. Since e is a projection, eRe is a subring and the involution on R restricts to an involution on eRe. For regularity, consider x ∈ eRe. Take an quasi-inverse y of x in R and reflect that eye is also an quasi-inverse of x. ⊳ Let R be a ∗-regular ring and e a projection in R. We write Re for the ∗-regular subring eRe. Furthermore, we define the height h(R) by the height h(L(RR)) of its principal ideal lattice L(RR). 3 Lemma 1.7. Let R be a ∗-regular ring and e a projection in R. Then the lattice L((Re)Re) of all principal right ideals in Re = eRe is isomorphic to the section [0, eR] ⊆ L(RR). ⊲ Proof. [J´on60, Lemma 8.2]. ⊳ Lemma 1.8. If R is a simple ∗-regular ring and e a projection in R, then Re is a simple ∗-regular ring with unit e. ⊲ Proof. consider the ideal generated by A in R. ⊳ It is left to show simplicity. For a non-vanishing ideal A in eRe, Lemma 1.9. Let R be a ∗-regular ring. Then for each x ∈ R, there exists a projection ex ∈ R such that exxex = x. ⊲ Proof. Let px be the projection that generates the left ideal generated by x and qx be the projection that generates the right ideal generated by x. Take ex := px ∨ qx to be the supremum in the lattice of all projections of R. ⊳ 1.2 Frames In this section, we recall the notion of perspectivity of elements of a lattice and the concept of a frame. The reader familiar with frames of modular lat- tices might give the following descriptions only a short glance and then jump to Definition 1.16 of a stable orthogonal frame and Corollary 1.22 that a simple MOL of height at least n contains a stable orthogonal (n, k)-frame. Two elements a, b of a lattice L are called perspective (to each other) if they have a common complement c in L. If a and b are perspective, we write a ∼ b. We say that a is subperspective to b or perspective to a part of b if there exists an element d ≤ b such that a ∼ d. We write a . b. An element c establishing a (sub)perspectivity between elements a and b is called an axis of (sub)perspectivity between a and b. If a . b, the part d ≤ b such that a ∼ d is called the image of a under the perspectivity between a and b. There exist different notions of a frame. In [vN60], von Neumann defined a homogeneous basis for a CML L (p. 93) and a (normalised) system of axes of perspectivity for a given homogeneous basis (p. 118). The combined system was called a (normalised) frame for L. Equivalently, G. Bergmann and A. Huhn introduced the notion of a n-frame (originally, a (n − 1)-diamond) in a modular lattice (see the survey articles [Day82], [Day84] or the article of C. Herrmann in memory of A. Day [Herr95]). The notion of a frame was subject to further development and generalisa- tion. See [J´on60] for the introduction of a partial frame, a large partial frame 4 and a global frame. In [J´on60], J´onsson defined a large partial n-frame in a bounded modular lattice B to be a subset of B consisting of independent elements a0, . . . , an−1, d and the entries of a symmetric matrix c = (cij)i,j<n such that the supremum of a0, . . . , an−1 and d equals the unit element 1B, d consists of a sum of finitely many elements each of which is subperspective to a0, and cij is an axis of perspectivity between aj and ai. We adapt the definition of J´onsson in the following way: Decomposing d into k summands, each of which is subperspective to a0, we incorporate these summands and their axes of subperspectivity to a0 into the frame. Further- more, we demand that the spanning elements of the frame are independent. Definition 1.10. Large partial (n, k)-frame A large partial frame of format (n, k) in a bounded modular lattice L is a subset Φ := {ai, a0i : 0 ≤ i < n + k} ⊆ L such that the following conditions are satisfied. 1. Li<n+k ai = 1L 2. a0 + ai = a0 ⊕ a0i = ai ⊕ a0i for i = 1, . . . , n − 1 3. a0(ai + a0i) + ai = ai ⊕ a0i = a0(ai + a0i) ⊕ a0i for i = n, . . . , n + k − 1 That is, Φ contains n + k independent elements ai spanning the lattice L (condition (1)). Conditions (2) and (3) state that a1, . . . , an−1 are perspec- tive to a0 and an, . . . , an+k−1 are subperspective to a0, where the axes of (sub)perspectivity are just the a0i. In particular, we have a0 · a0i = ai · a0i = 0 for all i. The axes of perspectivity between ai, aj for indices i, j < n can be con- structed via the axes a0i, a0j: We have aji = [a0j + a0i] · [aj + ai] and conse- quently, we have aki = [akj + aji] · [ak + ai] for i, j, k < n. Likewise, we can construct the axis of subperspectivity aji between ai and aj for indices i, j such that j < n and n ≤ i < n + k. For short, we call Φ a large partial (n, k)-frame or an (n, k)-frame, dropping the attribute large partial for the ease of notation and to avoid confusion with the notion of a large partial n-frame in the sense of J´onsson. In the following, we state some helpful results and develop the appropriate notion of a frame for a modular ortholattice. Lemma 1.11. Let L be a CML and assume that a0, a, b ∈ L are elements such that a0 ≤ a, a0 · b = 0, and b is subperspective to a0. Then the relative complement d of ab in [0, b] is subperspective to a0 and a ⊕ d = a + b. 5 Lemma 1.12. Let L be a CML. If L contains a large partial n-frame in the sense of J´onsson, then L contains an (n, k)-frame. ⊲ Proof. Construct summands ai, n ≤ i < n + k of d with the desired prop- erties (subperspective to a0 and such that a0, . . . , an+k−1 are independent) inductively. ⊳ Next, we introduce the concept of a stable frame. The main difference is that we incorporate all the axes of (sub)perspectivity (see Definition 1.10) and a set of relative complements. Definition 1.13. Stable (n, k)-frame Let L be a CML. A subset Φ = {ai, aij : 0 ≤ i < n, 0 ≤ j < n + k} ∪ {zij : j < n, n ≤ i < n + k} ⊆ L will be called a stable (n, k)-frame Φ in L if 1. {ai, a0i : 0 ≤ i < n + k} ⊆ L is a (n, k)-frame in L 2. for i, j ∈ I, i < n, aij is the axis of (sub)perspectivity between aj and ai 3. for each pair (i, j) of indices with j < n and n ≤ i < n + k, the element zij is a complement of bji in [0, aj], where bji is the image of ai under the subperspectivity aji between ai and aj. Lemma 1.14. Let L be a CML. If L contains an (n, k)-frame, L contains a stable (n, k)-frame. ⊲ Proof. Choose the necessary relative complements. ⊳ Definition 1.15. Orthogonal (n, k)-frame Let L be a MOL. An (n, k)-frame Φ in L is called an orthogonal (n, k)-frame if the following additional condition is satisfied: ∀i ∈ {0, . . . , n + k − 1}. a⊥ i =Xj6=i aj Definition 1.16. Stable orthogonal (n, k)-frame Let L be a MOL. A stable orthogonal (n, k)-frame is a stable frame such that 1. Φ as a frame satisfies the condition of Definition 1.15, and 2. the relative complements zij are relative orthocomplements. 6 1.2.1 Orthogonalisation of a (n, k)-Frame via J´onsson Now we want to show that the notion of a (stable) orthogonal (n, k)-frame is the appropriate one for a MOL: We will proove that a given frame can be orthogonalised. We will base the proof on arguments and results presented in [J´on60]. In fact, one could choose an alternative approach via ideas of Fred Wehrung, presented in [Weh98], using the notion of a normal equivalence in a modular lattice and the concept of a normal modular lattice. Lemma 1.17. Let L be a MOL and a, b projective elements in L. Then there exist four elements b0, b1, b2, b3 in L such that b is the direct orthogonal sum of b0, b1, b2, b3 and each bi is perspective to a part of a. ⊲ Proof. The lines of argument follow J´onsson's proof of Lemma 1.4 in [J´on60]. The only difference is that we choose the relative complements in J´onsson's proof to be relative orthocomplements in the considered intervals. ⊳ Lemma 1.18. Let L be a MOL and a0, a, b elements in L such that a0 ≤ a, a · b = 0, and b . a0. Then b decomposes into a direct orthogonal sum of five elements b0, b1, b2, b3, b4 such that each bi is subperspective to a0. ⊲ Proof. We choose c := b · a⊥ and d as the relative orthocomplement of c in [0, (a + b) · a⊥]. As part of b, c is subperspective to a0. Furthermore, one can show that d is projective to a part x ≤ a0. Consequently, by Lemma 1.17, we can decompose d into the direct orthogonal sum of 4 elements b1, . . . , b4, each of which is subperspective to a0. Together with b0 := c, we have the desired result. ⊳ Lemma 1.19. Let L be a simple MOL and a, b ∈ L non-trivial independent elements. Then there exist non-trivial elements a0, b0 with a0 ≤ a, b0 ≤ b such that a0 and b0 are perspective to each other. In particular, this holds if b ≤ a⊥. ⊲ Proof. Since L is simple, the neutral ideal generated by a is the whole lattice. Then b is the sum of finitely many elements, each of which is perspective to a part of a. Choose one such non-trivial summand as b0 and the corresponding perspective part of a as a0. ⊳ Lemma 1.20. Let L be a simple MOL with h(L) ≥ n. Then there exists a large partial n-frame Φ (in the sense of J´onsson) such that the first n elements a0, . . . , an−1 of Φ are orthogonal, that is, we have ak ≤ (Mi<k ai)⊥ 7 ⊲ Proof. By induction. ⊳ Lemma 1.21. Let L be a MOL containing a (n, k)-frame Φ such that a0, . . . an−1 are orthogonal, that is, if for all k < n, we have ak ≤ (⊕i<kai)⊥. Then L contains an orthogonal (n, k′)-frame for some k′. ⊲ Proof. By induction over the elements an, . . . , an+k−1 and Lemma 1.17. ⊳ Corollary 1.22. Let L be a simple MOL of height at least n. Then L contains a stable orthogonal (n, k)-frame. 1.2.2 Rings and Frames Combining Corollary 1.22 with Lemma 1.4 and Lemma 1.8, we get the fol- lowing results. Corollary 1.23. Let R be a simple ∗-regular ring with unit and h(R) ≥ n. Then the MOL L(RR) contains a stable orthogonal (n, k)-frame. Corollary 1.24. If R is a simple ∗-regular ring and e a projection in R, then the MOL L(ReRe) of principal right ideals of Re contains a stable orthogonal frame. Remark 1.1. Clearly, the format of the frame depends on the height of Re. 1.2.3 Projectivity of Frames It is well-known that global frames are projective. In this section, we state similar results for the above introduced frames. We begin with large partial (n, k)-frames. Lemma 1.25. Let K, L be CMLs, f : K ։ L a surjective 0-1-lattice homo- morphism and Φ ⊆ L a large partial (n, k)-frame in L. Then there exists a section M ≤ K and a set Ψ ⊆ M such that 1. fM : M → L is a surjective lattice homomorphism, 2. Ψ is a large partial frame in M of the same format as Φ, 3. f [Ψ] = Φ. ⊲ Proof. Inductive process and appropriate choices of preimages. ⊳ Similarly, we have the following. 8 Lemma 1.26. Let K, L be CMLs, f : K ։ L a surjective 0-1-lattice homo- morphism and Φ ⊆ L a stable (n, k)-frame in L. Then there exists a section M ≤ K and a set Ψ ⊆ M such that 1. fM : M → L is a surjective lattice homomorphism, 2. Ψ is a stable frame in M of the same format as Φ, 3. f [Ψ] = Φ. ⊲ Proof. Incorporate the choice of the necessary relative complements in the inductive procedure. To accomplish this, it is enough to show that if a, b are in K such that b ≤ a and f (b) ⊕ c = f (a) for some c ∈ L, then there exists d ∈ K such that b ⊕ d = a and f (d) = c. ⊳ Lemma 1.27. Let K, L be MOLs, f : K ։ L a surjective 0-1-lattice ho- momorphism and Φ ⊆ L a stable orthogonal (n, k)-frame in L. Then there exists a section M ≤ K and a set Ψ ⊆ M such that 1. fM : M → L is a surjective lattice homomorphism, 2. Ψ is a stable orthogonal frame in M of the same format as Φ, 3. f [Ψ] = Φ. ⊲ Proof. Incorporate the orthogonality into the inductive process. ⊳ 1.3 Concepts of Representability Definition 1.28. Linear representation As in [Mic03] and [Nie03], a linear positive representation of a ∗-regular ring R is a tuple σ = (D, VD, φ, ρ) where D is an (involutive) skew field, VD a right vector space over D, φ a scalar product on VD, and ρ : R → End(VD) a ring homomorphism such that ∀r ∈ R. ρ(r∗) = ρ(r)∗φ If the morphism ρ : R → End(VD) is injective, we call σ a faithful repre- sentation. 9 Definition 1.29. Generalised representation Let R be an involutive ring, I an arbitrary non-empty index set and σ a tuple σ = (I, {Di}i∈I, {Vi}i∈I, {φi}i∈I, ρ) consisting of an indexed family of (involutive) skew fields, an indexed family of vector spaces and an indexed family of scalar products, such that for each i ∈ I, Vi is a right vector space over Di with scalar product φi, and a map End(ViDi). ρ : R →Yi∈I πi(ρ(r∗)) =(cid:0)πi(ρ(r))(cid:1)∗φi If ρ is a ∗-ring morphism, i.e., for all r ∈ R and all i ∈ I the condition holds, we call σ a positive generalised representation of R. For short, we speak of a positive g-representation, or just a g-representation. If ρ is injective, we call σ a faithful g-representation. Remark 1.2. Since this paper deals with ∗-regular rings only, we suppress the adjective positive when speaking of a linear or a generalised representa- tion of a ∗-regular ring. We use the term representation for a linear repre- sentation as well as for a generalised representation, if the context leaves no ambiguity or both concepts are considered simultaneously. Remark 1.3. Note that the properties of a structure to be a (faithful linear) representation of a ring can be expressed in first-order logic [Mic03]. Definition 1.30. Representation of a (sectional) MOL A representation of a (sectional) MOL L consists of a tuple with D, VD, h·, ·i, as above and a morphism ς = (D, VD, h·, ·i, ι) ι : L → L(VD, h·, ·i) of (bounded) lattices between L and the subspace lattice of VD such that the (sectional) orthocomplementation on L corresponds to the (sectional) orthocomplementation on VD given by the scalar product, that is, for all x ∈ L, we have ι(x′) = ι(x)⊥ (ι(x′ b) = ι(x)⊥ ∩ ι(b) for a sectional MOL). We call a representation ς faithful if the morphism ι is injective. g-representations are defined, analoguously. A representation of an MOL L in an inner product space (VF , Φ) is an (0, 1)-lattice homomorphism ε : L → Sub(VF , Φ) such that ε(x′) = ε(x)⊥ for all x ∈ L. Observe that, by modularity, ε(x) = ε(x)⊥⊥ for all x ∈ L. A representation ε is faithful, if it is one-to-one. Both ∗-regular rings and MOLs will be called representable if they admit som faithful representation. 10 2 The Variety of ∗-Regular Rings The term variety is used in the usual sense: A variety is a class of algebraic structures of the same type that is closed under products, homomorphic images and substructures. Obviously, an arbitrary product of ∗-regular rings, where the operations are as usual defined componentwise, is itself a ∗-regular ring. For homomorphic images, the following holds. Proposition 2.1. A homomorphic image of a ∗-regular ring is ∗-regular. ◮ Proof. Due to [Good91], Lemma 1.3 and [Mic03], Proposition 1.7, every two-sided ideal of a ∗-regular ring is ∗-regular. ◭ For substructures, we recall the notion of the Rickart relative inverse of an element of a ∗-regular ring. Some preliminary work is needed. Definition 2.2. Left and right projection Let R be a ∗-regular ring. For an element a ∈ R, we call the unique projection e in R that generates the principal right ideal aR the left projection of a and the unique projection f in R that generates the principal left ideal Ra the right projection of a. Remark 2.1. This terminology can be found in [Kap68], p. 27 -- 28 or [Kap55], p. 525. We denote the left and right projection of a by l(a) and r(a), respec- tively. Furthermore, if R has a unit, we have annl R(a) = R(1 − e) and annr R(a) = (1 − f )R. The following result holds. Lemma 2.3. The left and right projection of an element a can be constructed in the following way: For x ∈ R, we set l(x) := x(x∗x)′x∗ and r(x) := x∗(xx∗)′x, where x′ denotes any quasi-inverse of x. ⊲ Proof. See [Mic03], p. 9 -- 10. ⊳ Lemma 2.4. Let R be a ∗-regular ring. Then for each element a ∈ R there exists a unique element q(a) such that the following conditions hold. 1. e := l(a) = aq(a∗a)a∗. 2. f := r(a) = a∗q(aa∗)a. 11 3. f q(a) = q(a). 4. aq(a) = e. Furthermore, q(a) has the properties that q(a)a = f , the left projection of q(a) is f and the right projection of q(a) is e. ⊲ Proof. See [Kap68] or [Kap55]). We have defined a function q : R → R that maps each a ∈ R to the unique element y with the listed properties. ⊳ Remark 2.2. We call q(a) the relative inverse of a. We note that a is the relative inverse of q(a), so q2 = idR. We arrive at the following result. Proposition 2.5. Let R be a ∗-regular ring. The q-subrings of R are exactly the ∗-regular subrings of R. Consequently, we incorporate the unary map q : R → R into the signature of ∗-regular rings, that is, a ∗-regular ring R is an algebra of type (R, +, ·,∗ , q, 0). ◮ Proof. Assume that S is closed under q. For an element a ∈ S, the map q gives a quasi-inverse q(a) of a, so S is regular. Since S is a ∗-subring of the ∗-regular subring R, S is itself ∗-regular. Conversely, assume that S ≤ R is a ∗-regular subring of R. Let x ∈ S. Due to Lemma 2.3, we can construct the left and the right projection of x within S, using the involution on S and any quasi-inverses of x, x∗, xx∗, x∗x in S. By Lemma 2.4, there exists an element y with the desired properties within the ∗-regular ring S. Since y is the unique element with these properties, we have y = q(x). Hence, S is closed under q. ◭ Combining Propositions 2.1 and 2.5, we have proven the first result. Theorem 2.6. The class R of ∗-regular rings forms a variety. 2.1 Directed Unions and Rings without Unit Definition 2.7. Directed union of rings Let R be a ring and S = {Si : i ∈ I} be a directed family of subrings of R. We say that R is a directed union of the family S if for each r ∈ R there exists k ∈ I such that r ∈ Sk. Remark 2.3. Casually, we speak of R being the directed union of the Si, without giving the family of the Si an extra name, and we write R =Si∈I Si, using the usual symbol for an ordinary union. Of course, an arbitrary union of rings is in general not a ring; hence, the lax notion does not lead to the risk of misunderstandings. 12 Lemma 2.8. Let R be a ∗-regular ring and assume that R is the directed union of a family S of ∗-regular subrings Si of R. Then R is a ∗-regular subring of an ultraproduct of the rings Si, i ∈ I. ⊲ Proof. Since the class of all ∗-regular rings forms a variety, this follows from [Gor98], Theorem 1.2.12 (1). ⊳ 2.2 Representability and Universal Algebra We finish the first section with a look on representability of ∗-regular rings under an universal-algebraic perspective. Lemma 2.9. Let R be a ∗-regular ring with a representation σ = (D, VD, h·, ·i, ρ). Then every ∗-regular subring S of R is representable. If the representation of R is faithful, so is the representation of S. ⊲ Proof. Just take the restriction ρS . ⊳ Proposition 2.10. Let {Si : i ∈ I} be a family of ∗-regular rings, I an arbi- trary index set. Assume that each Si admits a linear positive representation. Let U be an ultrafilter on I. Then the ultraproduct R := (Yi∈I Si)/U admits a linear positive representation. If every Si has a faithful linear posi- tive representation, then so does R. ◮ Proof. Consider the class of 2-sorted structures K := {(R, V ) : R a ∗-regular ring, V a vector space such that R has a linear positive representation in V } Since the relation that the ∗-regular ring R has a (faithful) linear positve representation in the vector V can be expressed in first-order logic, an ultra- product of a family of structures (Ri, Vi) ∈ K lies again in K. ◭ 3 Representability of ∗-Regular Rings This section is devided into the following parts: In the first part, we will intro- duce notation and convention. In the second part, we will develop the general framework that is needed to tackle the problem of representability. In the third part, we will present a proof that a ∗-regular ring R is g-representable if and only if so is L(RR). 13 3.1 Convention and Notation We consider right modules over rings, denoted by MS, NT . Submodules will be denoted by Mi, Ni, neglecting the respective underlying ring. If the contrary is not explicitly stated (or obvious from the context), we assume that the underlying ring is a ∗-regular ring (with or without unit). For morphisms between submodules Mi, Mj ≤ M, we write ϕji : Mi → Mj, where the indices should be read from right to left. If Mi, Mj have trivial intersection, we define the graph of a morphism ϕji : Mi → Mj by Γ(ϕji) = {x − ϕji(x) : x ∈ Mi}. Observation 3.1. Let Mi ∩ Mj = {0}. Note that Γ(ϕji) is a relative com- plement of Mj in [0, Mi + Mj] and, conversely, each such relative complement gives rise to a morphism ψji : Mi → Mj. Observation 3.2. Let M be a module with a direct decomposition M =Mi∈I Mi and denote the corresponding projections and embeddings by πi and εj, re- spectively. Consider a morphism ϕji : Mi → Mj. Then the composition of ϕji with the projection πi : M → Mi yields a morphism ϕji ◦ πi : M → Mj defined on all of M, i.e.,, ϕji ◦ πi ∈ Hom(M, Mi). Since Mi ≤ M, we can consider ϕji ◦ πi as an element of End(M), too. For the latter point of view, the formally correct approach would be to consider εj ◦ ϕji ◦ πi. To avoid technical and notational overload, we will treat ϕji ◦ πi itself as an element of End(M). Note that the composition ϕji ◦ πi is nothing else than the extension of the map ϕji : Mi → Mj to the module M, by defining the action of the extension to be trivial on the other summands of M. Very rarely, we write ϕji for this extension: We just use overlined symbols if we want to distinguish between a partial map and its extension. Conversely, consider a morphism ϕ ∈ M. We define ϕi := ϕ ◦ ǫi : Mi → M ϕji := πj ◦ ϕi = πj ◦ ϕ ◦ ǫi : Mi → Mj Then we have a 1-1-correspondence between a morphism ϕ : M → M, and a family {ϕi : i ∈ I}, where ϕi : Mi → M, and a family {ϕji : i, j ∈ I}, where ϕji : Mi → Mj since each ϕ ∈ End(M) can be decomposed in the following ways: ϕ =Mi∈I ϕi =Mi∈I Xj∈I ϕji 14 We agree to write ϕ = Pi,j∈I ϕji, with the convention stated above. We agree to not impose a rigorous notational strictness, but to understand the notation in the natural sense. Similar to the observation above, we note that ϕi ◦ πi is nothing else that the extension of the map ϕi = ϕ ◦ ǫi : Mi → M to all of M. We note that the conventions are compatible with addition and multipli- cation: We can form the sum ϕji + ψji and the composition ϕjk ◦ ψki in the natural sense, and for ϕ, ψ ∈ End(M), we have (ϕ + ψ)ji = ϕji + ψji and (ϕ ◦ ψ)ji =Xk∈I ϕjk ◦ ψki. We agree to write 1 = idM : M → M. Then we have 1ii = idMi : Mi → Mi (that is, the corresponding extension 1ii acts like the identity on Mi and trivially on every other summand Mj) and 1ji = 0ji : Mi → Mj (that is, the extension 1ji coincides with the zero map on M). Observation 3.3. For cyclic modules MS, NS with generators x, y, a mor- phism ϕ : M → N is determined by its action on the generator x of M. If M = xS, we have f (xs) = f (x)s for every xs ∈ M. Conversely, each choice z ∈ yS defines a morphism g : M → N via xs 7→ zs. In particular, let R be a regular ring and consider the module RR. Assume that I = eR, J = f R are principal right ideals in R (that is, cyclic submodules of RR). Since R is regular, the generators e, f can be taken to be idempotent. Let r ∈ R such that re ∈ J, that is, re = f c for some c ∈ R (or, equiv- alently, f (re) = re). Then the left multiplication with r defines a right-R- module-homomorphism br between I and J es 7→ r(es). br : I → J Remark 3.1. From now on, if possible, we denote the action defined by left multiplication with an element r bybr. We will speak of the left multiplication morphism (or left multiplication map or left multiplication) br. 3.2 General Framework In this section, we will develop the necessary machinery for the proof of the desired result. In order to simplify the lines of argument and to clarify the applied technique, we have chosen to separate the ring-theoretical aspects, the lattice- and frame-theoretical aspects and the general module-theoretical mechanisms as far as possible. 15 3.2.1 Decomposition Systems & Abstract Matrix Rings Definition 3.1. Decomposition system of a module Let MS be a right S-module over S and I = {i : 0 ≤ i < n + k} an index set, where n < ω and k ≤ ω. A decomposition system ε of M of format (n, k) consists of 1. a decomposition M =Li∈I Mi of M into a direct sum of submodules, 2. corresponding projections πi : M ։ Mi and embeddings ǫi : Mi ֒→ M, 3. a family {ǫij : i, j ∈ I} of maps ǫij, 4. submodules zij of M for i ∈ I, j < n, and 5. a 1-subring C ≤ End(M0) such that the following conditions are satisfied: 1. For i = j, we have ǫii = idMi. 2. For i, j < n, ǫij, ǫji are mutually inverse morphisms, i.e., ǫij ◦ ǫji = idMi. 3. For i ∈ I, we have ǫi0 ◦ ǫ0i = idMi (in particular, ǫ0i is injective). 4. For distinct indices i, j, k such that k, j < n, we have ǫki = ǫkj ◦ ǫji 5. For j < n, zij is a relative complement of im(ǫji) in [0, Mj]. 6. For i ∈ I, ǫ0i ◦ ǫi0 ∈ C. In other words, for i, j < n, the submodules Mi, Mj are isomorphic, while for i ∈ I, j < n, Mi is isomorphic to a submodule of Mj -- and the morphisms ǫji are the corresponding isomorphisms and embeddings. The relative complements zij are integrated into the notion of a decom- position systems for the following reason: For i, j with j < n, the injective morphism ǫji : Mi ֒→ Mj has a left inverse ǫij :: Mj → Mi, defined only on im(ǫji) ≤ Mj. Taking the relative complement zij ≤ Mj of im(ǫji) in [0, Mj], we can extend the partial morphism ǫij :: Mj → Mi to a morphism ǫij : Mj → Mi by setting ǫij(x) := 0 for all x ∈ zij (i.e., the extension ǫij : Mj → Mi acts trivially on zij). Remark 3.2. For the ease of notation, we stated that a decomposition sys- tem contains a family of maps ǫij for i, j ∈ I. The required conditions should have made clear that only particular maps have to exist. Of course, the maps that do exist are (partial) morphisms satisfying the desired relations. (One 16 might take the view that the other maps are partial maps with trivial do- main.) We write ε = ε(C, M) to indicate the ring C and the module M under consideration. We recall Observation 3.2 for the natural identifications and conventions. Definition 3.2. Morphisms between decomposition systems S ′ be modules over S, S′ and ε, ε′ decomposition systems of M, M ′, Let MS, M ′ respectively.1 A morphism between the two decomposition systems ε, ε′ is a map η : ε → ε′ such that the components of ε get mapped onto the components of ε′. In particular, the following hold. 1. η(Mi) = M ′ i and η(πi) = π′ i, η(ǫi) = ǫ′ i for all i ∈ I. 2. η(zij) = z′ ij for all i, j ∈ I. 3. η(ǫij) = ǫ′ ij for all i, j ∈ I. 4. η : C → C ′ is a morphism of rings with units. A morphism η between decomposition systems will be called injective or an embedding of decomposition systems if η : C → C ′ is injective. Definition 3.3. Abstract matrix ring Let MS be a module and ε a decomposition system of M. The abstract matrix ring with respect to the decomposition system ε of M is R(ε, C, M) := {ϕ ∈ End(MS) : ǫ0j ◦ ϕji ◦ ǫi0 ∈ C for all i, j} ⊆ End(MS) where, as above, ϕji = πj ◦ ϕ ◦ ǫi and πj, ǫi are the natural projections and embeddings belonging to decomposition system ε. The following result justifies this definition. Proposition 3.4. The set R(ε, C, M) is a 1-subring of End(MS). Proposition 3.5. Let MS, M ′ S ′ be two modules with decomposition systems ε, ε′ and η : ε → ε′ a morphism of decomposition systems between ε and ε′. Declaring η(ϕji) := ǫ′ j0 ◦ η(cid:16)ǫ0j ◦ ϕji ◦ ǫi0(cid:17) ◦ ǫ′ 0i for morphisms ϕji : Mi → Mj, the map η can be extended to a map η : R(Φ, C, M) → R(Φ′, C ′, M ′) 1Similarly, we denote the components of the two systems by the same letters, once with prime, once without. 17 in the following way. Since ϕ ∈ End(M) decomposes into ϕ =L ϕi =P ϕji, we can define η(ϕi) :=Pj∈I η(ϕji) for a fixed index i ∈ I and η(ϕji). η(ϕ) :=Xi∈I η(ϕi) = Xi,j∈I With this definition, η : R(Φ, C, M) → R(Φ′, C ′, M ′) is a morphism of rings with unit. If the restriction ηC : C → C ′ is injective, then so is the map η : R(Φ, C, M) → R(Φ′, C ′, M ′). 3.2.2 Frames and Induced Structures In this section, we approach the connection between the general framework and our particular setting. Starting with a frame in L(M), we will develop the notion of the coefficient ring of a frame and the notion of an induced decomposition system. Definition 3.6. Coefficient ring of a frame Let MS be a right module over S and Φ a stable (n, k)-frame in L(MS), contained in the sublattice L ≤ L(MS) with n ≥ 3. The coefficient ring of (Φ, L, M) is C(Φ, L, M) := {ϕ ∈ End(M0) : Γ(ǫ10 ◦ ϕ) ∈ L} ⊆ End(M0) The following result justifies this definition. Proposition 3.7. The set C(Φ, L, M) is a 1-subring of End(M0). Remark 3.3. The lines of argument and the technique of this proof are well- known: It is possible to express the ring operations via lattice terms with constants in Φ. (See the works of von Neumann, J´onsson and Handelman.) These terms are uniform in the frame Φ. In particular, they are independent of the particular module MS. Proposition 3.8. The Decomposition System of a Frame Let MS be a right module over S and Φ a stable (n, k)-frame in L(MS) contained in the sublattice L ≤ L(MS) with n ≥ 3. Then Φ induces a decomposition system in the following way. Since Φ is a frame, we have a decomposition of M into a direct sum M =L Mi, together with corresponding projections and embeddings. As usual, the axes of per- spectivity as well as the axes of subperspectivity are the graphs of morphisms between the summands. Since Φ is stable, we have relative complements zij as required.2 As ring C, we take the coefficient ring C(Φ, L, M). 2That is, zij a relative complement of bji in [0, aj], where bji is the image of ai under aji in [0, aj]. 18 We denote the decomposition system induced by Φ by ξ = ξΦ,L(C, M). ◮ Proof. The only thing left to show is Property 5 in Definition 3.1. Consider ǫ0i, ǫi0. Both graphs Γ(ǫ0i), Γ(ǫi0) and of course Γ(ǫ10) are part of the frame Φ. Since we can express composition of maps by lattice terms with constants in Φ, we have Γ(ǫ10 ◦ ǫ0i ◦ ǫi0) ∈ L. ◭ Definition 3.9. Matrix ring of a frame Let MS be a right module over S and Φ a stable (n, k)-frame in L(MS) (with n ≥ 3), contained in the sublattice L ≤ L(MS), C(Φ, L, M) the coefficient ring as defined in Defintion 3.6 and ξ = ξΦ,L(C, M) the induced decomposi- tion system as defined in Definition 3.8. The ring R(Φ, L, M) := R(ξ, C(Φ, L, M), M) will be called the matrix ring (of Φ, L, M). We consider the following situation: Let M and M ′ be modules over S and S′, L ≤ L(MS) a complemented 0-1-sublattice and Φ a stable frame in L(MS) contained in L of format (n, k) with n ≥ 3. Assume that we are given a morphism ι : L ֒→ L(M ′) of bounded complemented lattices. Observation 3.4. The image Φ′ := ι[Φ] is a stable frame in L(M ′), contained i , ι(πi) = π′ in L′ i, ι(ǫi) = ǫ′ := ι[L] ≤ L(M ′). In particular, we have ι(Mi) = M ′ i and ι(zij) = zij. Proposition 3.10. The morphism ι : L → L′ induces a morphism η between the induced decomposition systems ξ := ξΦ,L(C(Φ, L, M), L, M) and ξ′ := ξ′ Φ′,L′(C(Φ′, L′, M), L′, M ′). If ι : L → L′ is injective, then so is η : ξ → ξ′. ◮ Proof. We want to define η via the lattice morphism ι : L → L′. For the first two properties of a morphism between two decomposition systems (see Definition 3.2), we define η to coincide with ι on the submodules Mi, zij of M and recall Observation 3.4. Now consider the morphisms ǫji given by the frame Φ, i.e., Γ(ǫji) = aji ∈ Φ. Then Setting ι(cid:0)Γ(ǫji)(cid:1) = ι(aji) = a′ ji = Γ(ǫ′ ji) ∈ Φ′ η(ǫji) := ǫ′ ji for i 6= j < n and η(ǫii) := ǫ′ ii for arbitrary i 19 we have guaranteed that η maps the morphism ǫji to ǫ′ ji. For appropriate indices i, j, k, the compatibility ǫki = ǫkj ◦ ǫji is determined by the lattice-theoretical equation [akj + aji] · [ak + ai] = aki of the elements of the frame Φ (and similarly for Φ′). Therefore, we have η(ǫkj ◦ ǫji) = η(ǫki) = ǫ′ ji ki = ǫ′ kj ◦ ǫ′ for appropriate indices i, j, k. Secondly, we consider an element ϕ of the coefficient ring C = C(Φ, L, M), i.e., ϕ : M0 → M0 with Γ(ǫ10 ◦ ϕ) ∈ L. The property that ǫ10 ◦ ϕ is a morphism between M0 and M1 is equivalent to the lattice-theoretical property that Γ(ǫ10 ◦ ϕ) is a relative complement of M1 in [0, M0 + M1]. Since ι : L → L′ is a lattice morphism mapping Φ to Φ′, ι(Γ(ǫ10 ◦ ϕ)) is a relative complement of M ′ 1], i.e., the graph of a morphism ψ : M ′ 1. Composing ψ with ǫ′ 01, we can define 0 → M ′ 1 in [0, M ′ 0 + M ′ η(ϕ) := ǫ′ 01 ◦ ψ : M ′ 0 → M ′ 0 Thirdly, we can capture the ring operations on C(Φ, L, M) via lattice terms with constants in Φ. Hence, the ring operations are transferred via ι : L → L′ to Φ′ and C ′. Accordingly, the map η : C(Φ, L, M) → C(Φ′, L′, M ′) is a morphism of rings. Finally, injectivity of ι implies injectivity of η. ◭ Corollary 3.11. In the given situation, there exists a morphism η : R(Φ, L, C) → R(ι[Φ], ι[L], C ′) of rings with unit. If ι : L → L′ is injective, then so is η : R → R′. In particular, if L embedds into the subspace lattice L(V ) of a vector space V , we have a ring embedding η : R(Φ, L, C M ) ֒→ End(VD) ⊲ Proof. Combine Proposition 3.10 and Proposition 3.5. ⊳ 3.3 Representability of ∗-Regular Rings This section is dedicated to the desired result on representability of ∗-regular rings R such that L(RR) is representable. First, we focus our attention on a ∗-regular ring R with unit such that the MOL L(RR) contains a stable 20 orthogonal frame. With that restriction, we aim at representability of simple ∗-regular rings with unit. Subsequently, we will deal with simple ∗-regular rings without unit and finally, with subdirectly irreducible ∗-regular rings (with and without unit). Due to the first main theorem that the class of all ∗-regular rings is a variety, with Theorem 3.31 , we reach the desired result that a ∗-regular ring R is g-representable if so is L(RR). 3.3.1 ∗-Regular Rings with Frames Remark 3.4. For this subsection, we assume that 1. R is a ∗-regular ring with unit, 2. L := L(RR) is a MOL of height h(L) ≥ 3, 3. Φ is a stable orthogonal (n, k)-frame in L with n = 3, 4. M = RR, if not stated otherwise. Remark 3.5. Moreover, we assume that there exists a faithful representation ι : L ֒→ L(VD, h·, ·i) of the MOL L. Consequently, Corollary 3.11 applies in its full strength. Corollary 3.12. Let ei, ej be projections in R and eiR, ejR the corresponding cyclic modules. Any morphism ϕji : eiR → ejR is a left multiplication by a ring element ejsei ∈ ejRei. Observation 3.5. By Proposition 3.8, the stable orthogonal frame Φ induces a decomposition system ξ = ξΦ,L(C, M) = ξΦ,L(C, RR). More exactly, we have the following correspondences. 1. The summands Mi correspond to principal right ideals eiR generated by left multiplication with ei and coincides with the (extension of the) embedding ǫi = idMi. by a projection ei. Each projection πi corresponds to a map bei given 2. The morphisms ǫji : eiR → ejR are given by beji with eji := ǫji(ei) an element of eiRej. 3. For the coefficient ring of the frame, we have C = C(Φ, L, M) = C(Φ, L, RR) = {br : r ∈ e0Re0}. Remark 3.6. From now on, we will denote the morphisms ǫji given by the decomposition above by ǫji and ceji interchangedly, as it suits the particular situation. 21 Corollary 3.13. We have an isomorphism θ : R → R(Φ, L, RR) of rings with unit. ⊲ Proof. As stated in Remark 3.4 at the beginning of the section, we agree to write M for RR. We recall the definition of the matrix ring of a frame: R(Φ, L, M) = {ϕ ∈ End(M) : ∀i, j ∈ I. ǫ0j ◦ ϕji ◦ ǫi0 ∈ C(Φ, L, M)} Since R contains a unit, an endomorphism ϕ ∈ End(M) is given by left multiplication br for some r ∈ R. We notice that (br)ji = πj ◦br ◦ ǫi = bej ◦br ◦bei = [ejrei =crji, Then we have with rji := ejrei. ǫ0j ◦ (br)ji ◦ ǫi0 = e0j ◦ [ejrei ◦ ei0 = \e0jrei0, The equality since e0jej = e0j, eiei0 = ei0. ǫ0j ◦ (br)ji ◦ ǫi0 = \e0jrei0 and Observation 3.5 lead to ǫ0j ◦ (br)ji ◦ ǫi0 ∈ C(Φ, L, M). Since the indices i, j were arbitrary, we have br ∈ R(Φ, L, M). In particular, for an element r ∈ R, the left multiplication br : M → M br =Xcrji where crji : eiR → ejR, that is, the isomorphism θ : R → R(Φ, L, M) is given by θ : r 7→br. and rji = ejrei ∈ ejRei. decomposes into Of course, we have Γ(ǫ10 ◦ \e0jrei0) = Γ(ce10 ◦ \e0jrei0) = Γ(\e1jrei0) = (e0 − e1jrei0)R ∈ L = L(RR) ⊳ Corollary 3.14. We have an embedding ρ : R ֒→ End(VD) of rings with unit. 22 ⊲ Proof. By Remark 3.5 the MOL L = L(RR) is assumed to be representable in L(VD, h., .i). We combine that with Corollaries 3.11 and 3.13 and define ρ := η ◦ θ to get the desired isomorphism. ⊳ It is left to show that the isomorphism translates the involution on R into adjunction with respect to the scalar product. In the following, assume that in addition to the assumptions of ref..., we have the following. 1. (VD, h., .i) a vector with scalar product, 2. K a MOL represented in L(VD, h., .i), i.e., K is a modular sublattice of L(VD) and the orthocomplementation on K is induced by the scalar product h., .i on VD, 3. Ψ a stable orthogonal frame in the MOL K, 4. Ui, Uj ≤ V with Ui, Uj ∈ Ψ and f : Ui → Uj, g : Uj → Ui linear maps, If we discuss both situations - that is, for the frame K with its frame Ψ or subspaces of V - simultaneously, we use the symbol Υ for the frame, Ni for submodules or subspaces and a, b for morphisms. Definition 3.15. Adjointness on End(VD, h., .i) We call f and g adjoint to each other (with respect to h., .i) if ∀v ∈ Ui, w ∈ Uj. hf v, wi = hv, gwi. Remark 3.7. Notice that Ui, Uj are elements of the orthogonal frame Ψ, in particular, if i 6= j, then Ui and Uj are orthogonal to each other. Lemma 3.16. The following conditions are equivalent: 1. f and g are adjoint to each other in the sense of Definition 3.15. 2. The extensions f , g : V → V are adjoint to each other in the usual sense. If i 6= j, both these conditions are equivalent to Γ(f ) ⊥ Γ(−g).3 ⊲ Proof. The equivalence of the first two conditions is immediate. Now, if i 6= j, we have Γ(f ) = {v − f v : v ∈ Ui} Γ(−g) = {w + gw : w ∈ Uj} 3Notice that Γ(f ), Γ(g), Γ(−g) are contained in the MOL K. 23 and Γ(f ) ⊥ Γ(−g) ⇔ hv − f v, w + gwi = 0 for all v ∈ Ui, w ∈ Uj ⇔ hv, wi + hv, gwi − hf v, wi − hf v, gwi = 0 for all v ∈ Ui, w ∈ Uj ⇔ hv, gwi − hf v, wi = 0 for all v ∈ Ui, w ∈ Uj ⇔ hv, gwi = hf v, wi for all v ∈ Ui, w ∈ Uj ⇔ f and g are adjoint to each other in the sense of Definition 3.15, where the terms hv, wi, hf v, gwi vanish since Ui, Uj are orthogonal to each other. ⊳ Now, we derive a result similar to Lemma 3.16 for a ∗-regular ring R and the relation between the involution on R and the orthogonality on L = L(RR). Lemma 3.17. The involution on R can be captured via the orthogonality on L, more exactly, for aij ∈ eiRej, bji ∈ ejRei with i 6= j, the following conditions are equivalent: 1. aij = b∗ ji 2. Γ(caij) ⊥ Γ(−cbji) ⊲ Proof. Since caij : ejR → eiR and −cbji : eiR → ej, we have Γ(−cbji) = (ei + bjiei)R Γ(caij) = (ej − aijej)R The orthogonality on L is given by pR ⊥ qR :⇔ q∗p = 0. Calculating yields (ei + bjiei)∗ · (ej − aijej) = (ei + eib∗ ji)(ej − aijej) = eiej − eiaijej + eib∗ = ei(−aij + b∗ jiej − eib∗ jiaijej ji)ej = −aij + b∗ ji, since aij, b∗ ji ∈ eiRej. Hence aij = b∗ Corollary 3.18. Uniqueness Let (i, j) be an arbitrary pair of indices. ji iff Γ(caij) ⊥ Γ(−cbji). ⊳ A linear map f : Ui → Uj has at most one adjoint g : Uj → Ui. Due to this uniqueness, it is legitimate to write f ∗ = g if f, g are adjoint to each other. Likewise, a map caij : ejR → eiR gives rise to a map cbji : eiR → ejR, namely cbji = ca∗ ij. If i 6= j, we have cbji = ca∗ Lemma 3.19. For each morphism ǫki : Ni → Nk, there exists an adjoint ǫ∗ ki : Nk → Ni in Υ. ij iff Γ(caij) ⊥ Γ(−cbji). 24 Corollary 3.20. For i, k ∈ I with k < n we have ǫ∗ ki ◦ ǫ∗ ik = idUi. ⊲ Proof. We have ⊳ ǫ∗ ki ◦ ǫ∗ ik = (ǫik ◦ ǫki)∗ = (idNi)∗ = idNi. Remark 3.8. Obviously, Lemma 3.19 and Corollary 3.20 hold for arbitrary indices i, k ∈ I: Recall that for if i = k, we have ǫik = ǫii = ǫi = idNi, which is an hermitian idempotent map. Corollary 3.21. Let a : Ni → Nj and b : Nj → Ni be as before. Then a and b are adjoint to each other iff ǫ∗ i0 ◦ b ◦ ǫ∗ 1j and ǫ1j ◦ a ◦ ǫi0 are adjoint to each other. ⊲ Proof. Assume that ǫ∗ 1j and ǫ1j ◦ a ◦ ǫi0 are adjoint to each other, where adjoint is either understood in the sense of Definition 3.15 or in the sense of Lemma 3.17. Then i0 ◦ b ◦ ǫ∗ ǫ∗ 0i ◦ (ǫ1j ◦ a ◦ ǫi0)∗ ◦ ǫ∗ j1 = (ǫj1ǫ1jaǫi0ǫ0i)∗ = a∗ and ǫ∗ 0i ◦ (ǫ1j ◦ a ◦ ǫi0)∗ ◦ ǫ∗ j1 = ǫ∗ 0i ◦ (ǫ∗ i0 ◦ b ◦ ǫ∗ 1j) ◦ ǫ∗ j1 = (ǫ∗ 0iǫ∗ i0)b(ǫ∗ 1jǫ∗ j1) = b, so a∗ = b. Now, assume that a, b are adjoint to each other. Then (ǫ1j ◦ a ◦ ǫi0)∗ = ǫ∗ i0 ◦ a∗ ◦ ǫ∗ 1j = ǫ∗ i0 ◦ b ◦ ǫ∗ 1j. ⊳ Remark 3.9. Corollary 3.21 holds for arbitrary indices i, j, too. In particu- lar, we can complete Lemma 3.17 and Corollary 3.18 by noting the following: If aii, bii ∈ eiRei, we have a∗ ii = bii ⇔ Γ(ǫ∗ Proposition 3.22. The map i0 ◦caii ◦ ǫ∗ 1j) ⊥ Γ(−ǫ1j ◦ bbii ◦ ǫi0) ρ = η ◦ θ : R → End(V,h., .i) defined in Corollary 3.14 is a ∗-ring-embedding of involutive rings with unit. 25 ◮ Proof. First, we recall that the map η : R(Φ, L, M) ֒→ End(VD) defined in Corollary 3.11 was defined via the lattice embedding ι : L ֒→ L(VD). In this situation, we consider a MOL L and a MOL-embedding of L into L(VD, h., .i). As shown before, for morphisms ǫji : Mi → Mj ∈ Φ, we can define an adjoint operator ǫ∗ ji : Mj → Mi via the orthogonality on L. For morphisms ϕ10 : M0 → M1 and ψ01 : M1 → M0, the relation of adjointness could be captured via orthogonality on graphs. In particular, we have ρ(ǫ∗ and ρ(ϕ∗ ji) =(cid:0)ρ(ǫji)(cid:1)∗ 10) =(cid:0)ρ(ϕ10)(cid:1)∗. Now, for r ∈ R, consider ejrei ∈ ejRei. Then (cid:0)ρ(ejrei)(cid:1)∗ = (cid:0)ρ(ej1(e1jejreiei0)e0i)(cid:1)∗ =(cid:0)ρ(ej1)ρ(e1jejreiei0)ρ(e0i)(cid:1)∗ j1) j1) i0e∗ i r∗e∗ = ρ(e∗ = ρ(e∗ = (cid:0)ρ(e0i)(cid:1)∗(cid:0)ρ(e1jejreiei0)(cid:1)∗(cid:0)ρ(ej1)(cid:1)∗ = ρ(cid:0)e∗ 0i)ρ(cid:0)(e1jejreiei0)∗(cid:1)ρ(ǫ∗ 0i)ρ(cid:0)e∗ 1j(cid:1)ρ(ǫ∗ j1(cid:1) = ρ(e∗ =(cid:16)X ρ(ejrei)(cid:17)∗ i r∗e∗ i r∗e∗ j e∗ j e∗ 1jǫ∗ 0ie∗ i0e∗ (cid:0)ρ(r)(cid:1)∗ = (cid:16)ρ(cid:0)X ejrei(cid:1)(cid:17)∗ Hence, we have j ) = ρ(cid:0)(ejrei)∗(cid:1). =X(cid:0)ρ(ejrei)(cid:1)∗ j ) = ρ(cid:16)X e∗ = X ρ(cid:0)(ejrei)∗(cid:1) =X ρ(e∗ = ρ(cid:16)X(ejrei)∗(cid:17) = ρ(cid:16)(cid:0)X ejrei(cid:1)∗(cid:17) = ρ(r∗). i r∗e∗ i r∗e∗ j(cid:17) ◭ 3.3.2 Simple ∗-Regular Rings Corollary 3.23. Every simple ∗-regular ring S with unit admits a faithful linear representation provided that L(SS) does so. ⊲ Proof. We may assume that S is non-Artinian, hence, we can assume that S has height at least 3. By Corollary 1.23, the MOL L = L(SS) contains a stable orthogonal frame of format (n, k) with n ≥ 3. It follows by Proposition 3.22 that S is faithfully representable. ⊳ Proposition 3.24. Every simple ∗-regular ring R admits a faithful linear representation provided that L(RR) does so. ◮ Proof. Let R be a simple ∗-regular ring R without unit. Consider the set P (R) of all projections in R. Since R is ∗-regular, P (R) is a lattice. In particular, it is a directed set. By Lemma 1.9, we have that R is the directed 26 union of its subrings Re, e ∈ P (R). By Lemma 2.8, R is a ∗-regular subring of an ultraproduct of the Re, e ∈ P (R). By Lemma 1.8, for each projection e, the ring Re is a simple ∗-regular ring with unit e and ¯L(ReRe) ∼= [0, eR] is representable. By Corollary 3.23, each Re is faithfully representable. Hence, we can conclude with Lemma 2.9 and Proposition 2.10 that R has a faithful representation. ◭ 3.3.3 Representations of Ideals Observation 3.6. Let I be a two-sided ideal of the ∗-regular ring R. We can consider I as a ∗-regular ring (without unit, if I is non-trivial) on its own; hence, we can consider representations of I. Proposition 3.25. Let R be a ∗-regular ring and I a two-sided ideal in R with a representation : I → End(VD, h·, ·i). Denote the set of all projections in I by P (I), abbreviate Vp := (p)[V ] for a projection p ∈ P (I) and set ρ(r) := [p∈P (I) (rp)Vp Then ρ is a representation of R an appropriate subspace U of V , where the scalar product on UD is given by restriction. ◮ Proof. First, we have to show that the given definition of ρ indeed defines a map ρ : R → End(VD). Recalling that the set of all projections of a ∗-regular ring is directed, we consider two projections e, f ∈ P (I) with e ≤ f , that is, with e = f e. We have to show that the restrictions coincide on Ve, i.e., (rf )Ve = (re)Ve. Since e = f e, we have (re)Ve = (rf e)Ve = ((rf ) ◦ (e))Ve = (rf )Ve, as desired. Second, we have to show that the map ρ : R → End(VD) is a ∗-ring- homomorphism. For 0 in R, we have If 1 ∈ R, we have ρ(0) = [p∈P (I) ρ(1) = [p∈P (I) (0p)Vp = 0V . (1p)Vp = 1U with U :=Sp∈P (I) Vp, that is, ρ[R] acts on the subspace U of VD. 27 For addition, let r, s ∈ R. We have ρ(r + s) = [p∈P (I) = [p∈P (I) ((r + s)p)Vp = [p∈P (I) (rp)Vp + (sp)Vp = [p∈P (I) (rp + sp)Vp (rp)Vp + [q∈P (I) (sq)Vq = ρ(r) + ρ(s). Therefore, ρ(r + s) = ρ(r) + ρ(s) and ρ(−r) = −ρ(r)for all r, s ∈ R. For multiplication, let r, s be in R. We note that for each p ∈ P (I), there exists a qp ∈ P (I) such that sp = qpsp. We claim that [p∈P (I) ((rqp) ◦ (sp))Vp = [q∈P (I) (rq)Vq ◦ [p∈P (I) (sp)Vp. ⊲ Proof. Take v ∈ U. Then there exists pv ∈ P (I) with v ∈ Vpv, so on the one hand (cid:18) [p∈P (I)(cid:0)(rqp) ◦ (sp)(cid:1)Vp(cid:19)(v) =(cid:0)(rqpv)v ◦ (spv)(cid:1)(v), while on the other hand (cid:18) [q∈P (I) = [q∈P (I) (sp)Vp(cid:19)(v) = [q∈P (I) (rq)Vq ◦ [p∈P (I) (rq)Vq(cid:16)(spv)(v)(cid:17) = (rqpv)((spv)(v)). (rq)Vq(cid:16) [p∈P (I) (sp)Vp(v)(cid:17) ⊳ Thus, we have ρ(rs) = [p∈P (I) = [p∈P (I) ((rs)p)Vp = [p∈P (I) ((rqp) ◦ (sp))Vp = [q∈P (I) (rqpsp)Vp (rq)Vq ◦ [p∈P (I) (sp)Vp = ρ(r) ◦ ρ(s). Now we examine the involution on R. For r ∈ R, consider v, w ∈ V . Then take e ∈ P (I) with v ∈ Ve. There exists f1 ∈ P (I) such that f1re = re and 28 f2 ∈ P (I) such that w ∈ Vf2. Choosing f := f1 ∨ f2, we have hρ(r)v, wi = h [p∈P (I) (rp)Vpv, wi = h(re)v, wi = h(f re)v, wi = hv, (er∗f )wi = hv, (e)(r∗f )wi = h(e)v, (r∗f )wi = hv, (r∗f )wi = hv, ρ(r∗)wi, where we have used that : I → End(VD) is a ∗-ring-homomorphism and v ∈ Ve, w ∈ Vf . ◭ Lemma 3.26. Let R be a ∗-regular ring and I a two-sided ideal in R with a representation : L → End(VD, h., .i). Denote the action of R on the ideal I given by left multiplication by λI, that is λI : R → End(II) λI(r)(x) :=br(x) = rx. If the representation : I → End(VD, h·, ·i) is faithful and λI : R → End(II) is injective, then the representation ρ : R → End(VD, h·, ·i) defined in Propo- sition 3.25 is faithful. ⊲ Proof. Assume that ρ(r) = 0. This is equivalent to (re) = 0 for all e ∈ P (I). As : I → End(VD, h·, ·i) is faithful, this means that re = 0 for all e ∈ P (I). Since I is a ∗-regular ring, for every element x ∈ I there exists e ∈ P (I) such that ex = x. Hence, we have that rx = 0 for all x ∈ I. Since we assumed the action of R on I given by left multiplication to be injective, we have that r = 0. This shows that ρ is injective. ⊳ 3.3.4 Subdirectly Irreducible ∗-Regular Rings In this section, we will show that each subdirectly irreducible ∗-regular ring R has a faithful representation provided that L(RR) does so. Observation 3.7. We may assume that R is non-Artinian: Since every reg- ular ring is semi-prime, a subdirectly irreducible ∗-regular ring which is Ar- tinian is semi-simple, hence representable. Furthermore, the minimal two-sided ideal of R is non-Artinian, too (see [HS], Proposition 2). Proposition 3.27. Let R be a subdirectly irreducible ∗-regular ring and let J be the minimal two-sided ideal of R. Then the action λJ : R → End(JJ) of R defined by λJ (r)(x) = r(x) = rx is injective. 29 ◮ Proof. Consider the left annihilator A := annl R(J) of J in R. While a priori A is only a left ideal, it can be shown that A is indeed closed under left and right multiplication by elements of R. Since A is closed under addition, A is a two-sided ideal in R. Since J does not annihilate itself, we can conclude that A is trivial. Therefore, the action of R on J defined by left multiplication is injective. ◭ Lemma 3.28. The minimal ideal J of a subdirectly irreducible ∗-regular ring R is a simple ∗-regular ring. ⊲ Proof. For a non-vanishing ideal A in J, consider the ideal generated by A in R. ⊳ Remark 3.10. Note that one does not need that R contains a unit. Lemma 3.29. Let R be a ∗-regular ring and I a minimal two-sided ideal in R, in particular simple as a ring. Let e be a projection in I. Then the ring Re = eRe is simple. ⊲ Proof. For a non-vanishing ideal A in Re, consider the ideal generated by A in I. ⊳ Proposition 3.30. Considered as simple ∗-regular ring, the minimal ideal J of a subdirectly irreducible ∗-regular ring R has a faithful representation provided that ¯L(IR) does so. ⊲ Proof. Proposition 3.24. ⊳ Theorem 3.31. Every subdirectly irreducible ∗-regular ring R has a faithful representation provided that L(RR) does so. ⊲ Proof. Lemma 3.26. ⊳ References [Good91] K. R. Goodearl, Von Neumann Regular Rings Krieger Publishing Company, 1991. [Gor98] V. A. Gorbunov, Algebraic Theory of Quasivarieties Siberian School of Logic and Algebra, Consultants Bureau, New York, 1998. [Kap68] I. Kaplansky, Rings of Operators, W. A. Benjamin, Inc., 1968. [Mae58] F. Maeda Kontinuierliche Geometrien Springer, Berlin, 1958. 30 [Skor64] L. A. Skornyakov, Complemented Modular Lattices and Regular Rings Oliver and Boyd Ltd., 1964. [vN60] J. von Neumann, Continous Geometries, Princeton, 1960. [Day82] [Day84] [Herr95] [HR99] [HS] [J´on60] [Kap55] [Mic03] A. Day, Geometrical applications in modular Universal Algebra and Lattice Theory (R. S. Freese ans O. C. Garcia, eds.) Lecture Notes in Mathematics, 1004, 1983, p. 111 -- 141. Springer. A. Day, Applications of coordinatization in modular lattice theory: The legacy of J. von Neumann Order, 1, 1984, p. 295 -- 300. Reidel Publishing Company, Boston, U.S.A. C. Herrmann, Alan Day's work on modular and Arguesian lattices Algebra Universalis, 34, 1995, p. 35 -- 60. Dedicated to the memory of Alan Day. C. Herrmann, M. Roddy, Proatomic modular ortholattices: Represen- tation and equational theory Note di matematica e fisica, 10, 1999, p. 57 -- 88. C. Herrmann, M. Semenova, Existence varieties of regular rings and complemented modular lattices J. Algebra 314, 2007, 235 -- 251 B. J´onsson, Representations of complemented modular lattices Transactions of the AMS, 97, 1960, p. 64 -- 94. I. Kaplansky, Any orthocomplemented complete modular lattice is a continuous geometry Annals of Mathematics, 61, 1955, p. 524 -- 542. F. Micol, On representability of ∗-regular rings and modular ortholat- tices FB 04 TUD, 2003. http://tuprints.ulb.tu-darmstadt.de/303/ [Nie03] N. Niemann, Representations of involutive rings FB 04 TUD, 2003. [Weh98] F. Wehrung, The dimension monoid of a lattice Algebra Universalis, 40 (3), 1998, p. 247 -- 411. Authors address: Fachbereich Mathematik der Technischen Universitat Darmstadt 31
1301.5289	1	1301	2013-01-22T19:22:47	Split strongly abelian p-chief factors and first degree restricted cohomology	[ "math.RA", "math.RT" ]	In this paper we investigate the relation between the multiplicities of split strongly abelian p-chief factors of finite-dimensional restricted Lie algebras and first degree restricted cohomology. As an application we obtain a characterization of solvable restricted Lie algebras in terms of the multiplicities of split strongly abelian p-chief factors. Moreover, we derive some results in the representation theory of restricted Lie algebras related to the principal block and the projective cover of the trivial irreducible module of a finite-dimensional restricted Lie algebra. In particular, we obtain a characterization of finite-dimensional solvable restricted Lie algebras in terms of the second Loewy layer of the projective cover of the trivial irreducible module. The analogues of these results are well known in the modular representation theory of finite groups.	math.RA	math	SPLIT STRONGLY ABELIAN p-CHIEF FACTORS AND FIRST DEGREE RESTRICTED COHOMOLOGY J ORG FELDVOSS, SALVATORE SICILIANO, AND THOMAS WEIGEL Abstract. In this paper we investigate the relation between the multiplici- ties of split strongly abelian p-chief factors of finite-dimensional restricted Lie algebras and first degree restricted cohomology. As an application we obtain a characterization of solvable restricted Lie algebras in terms of the multi- plicities of split strongly abelian p-chief factors. Moreover, we derive some results in the representation theory of restricted Lie algebras related to the principal block and the projective cover of the trivial irreducible module of a finite-dimensional restricted Lie algebra. In particular, we obtain a charac- terization of finite-dimensional solvable restricted Lie algebras in terms of the second Loewy layer of the projective cover of the trivial irreducible module. The analogues of these results are well known in the modular representation theory of finite groups. 1. Introduction Let p be an arbitrary prime number, and let G be a finite group whose order is divisible by p. Moreover, let Fp[G] denote the group algebra of G over the field Fp with p elements, and let S be an irreducible (unital left) Fp[G]-module. Then [G : S]p−split denotes the number of p-elementary abelian chief factors or for short p-chief factors Gj /Gj−1 (1 ≤ j ≤ n) of a given chief series {1} = G0 ⊂ G1 ⊂ · · · ⊂ Gn = G that are isomorphic to S as Fp[G]-modules and for which the exact sequence {1} → Gj/Gj−1 → G/Gj−1 → G/Gj → {1} splits in the category of groups. It is well known that [G : S]p−split is independent of the choice of the chief series of G (see also Theorem 1.2 below). W. Gaschutz proved the "only if"-part of the following result on split (or com- plementable) p-chief factors of finite p-solvable groups (see [9, Theorem VII.15.5]). The converse of Gaschutz' theorem is due to U. Stammbach [13, Corollary 1]), but in an equivalent form it was already proved earlier by W. Willems [15, Theorem 3.9]. p H 1(G, S) = Theorem 1.1. A finite group G is p-solvable if, and only if, dimF dimF p[G](S) · [G : S]p−split holds for every irreducible Fp[G]-module S. p EndF Let CG(M ) := {g ∈ G g · m = m for every m ∈ M } denote the centralizer of an Fp[G]-module M in G. In order to be able to apply his cohomological charac- terization of p-solvable groups (see [12, Theorem A]) in the proof of Theorem 1.1, Stammbach established the following result (see the main result of [13]): Date: January 21, 2013. 2000 Mathematics Subject Classification. 17B05, 17B30, 17B50, 17B55, 17B56. Key words and phrases. Solvable restricted Lie algebra, irreducible module, p-chief factor, strongly abelian p-chief factor, split p-chief factor, multiplicity of a split strongly abelian p-chief factor, restricted cohomology, transgression, principal block, projective indecomposable module, Loewy layer. 1 2 J ORG FELDVOSS, SALVATORE SICILIANO, AND THOMAS WEIGEL Theorem 1.2. Let G be a finite group, and let S be an irreducible Fp[G]-module with centralizer algebra D := EndF p[G](S). Then [G : S]p−split = dimD H 1(G, S) − dimD H 1(G/CG(S), S) holds. In particular, [G : S]p−split is independent of the choice of the chief series of G. The goal of this paper is to investigate whether analogues of Theorem 1.1 and Theorem 1.2 hold in the context of restricted Lie algebras. Recently, the authors have obtained analogues of these results for ordinary Lie algebras (see the equiva- lence (i)⇐⇒(iii) in [6, Theorem 4.3] and [6, Theorem 2.1], respectively). The main result of this paper is a restricted analogue of Theorem 1.2 (see Theorem 2.3) from which all the other results will follow. The proof given here follows the argument used in the proof of [6, Theorem 2.1] very closely. An important tool in the proof is a one-to-one correspondence between the set of equivalence classes of restricted ex- tensions of a strongly abelian restricted Lie algebra M by a restricted Lie algebra L acting on M and the second restricted cohomology space H 2 ∗ (L, M ) that is defined via the transgression in the five-term exact sequence associated to the restricted extension of M by L. As a consequence of Theorem 2.3 and the equivalence (i)⇐⇒(iv) in [6, Theorem 5.5], we obtain the analogue of Theorem 1.1 for split strongly abelian p-chief factors of restricted Lie algebras (see Theorem 2.7). In the final section we apply the results obtained in Section 2 to the second Loewy layer of the projective cover of the trivial irreducible module. The equivalence (i)⇐⇒(ii) in Theorem 3.3 is an analogue of Willems' module-theoretic characterization of p-solvable groups (see [15, Theorem 3.9] and also [13, Corollary 2]) for restricted Lie algebras. Let hXiF denote the F-subspace of a vector space V over a field F spanned by a subset X of V . For more notation and some well-known results from the structure and representation theory of restricted Lie algebras we refer the reader to Chapters 2 and 5 in [14]. 2. Split strongly abelian p-chief factors and restricted cohomology In analogy to group theory we define a p-chief series for a finite-dimensional restricted Lie algebra L to be an ascending chain 0 = L0 ⊂ L1 ⊂ · · · ⊂ Ln = L of p-ideals in L such that Lj/Lj−1 is a minimal (non-zero) p-ideal of L/Lj−1 for every integer j with 1 ≤ j ≤ n. Any Lj/Lj−1 is then called a p-chief factor of L. We say that Lj/Lj−1 is a strongly abelian p-chief factor if it is an abelian Lie algebra with zero p-mapping (see [8, p. 565] for the notion of a strongly abelian restricted Lie algebra). Observe that strongly abelian p-chief factors are irreducible restricted modules but this is not the case for arbitrary p-chief factors. Let S be a simple Lie algebra that is not restrictable, and let L be the minimal p-envelope of S. Then L has no non-zero proper p-ideals (see [5, Proposition 1.4(1)]), and therefore L is a p-chief factor of L which is not irreducible as an L-module, because S is a non-zero proper L-submodule of L (see [5, Proposition 1.1(1)]). Note also that every p-chief factor of a solvable restricted Lie algebra is abelian but not necessarily strongly abelian as any non-zero torus shows. For an irreducible L-module S and a given p-chief series 0 = L0 ⊂ L1 ⊂ · · · ⊂ Ln = L of L we denote by [L : S]p−split the number of strongly abelian p-chief SPLIT STRONGLY ABELIAN p-CHIEF FACTORS AND RESTRICTED COHOMOLOGY 3 factors Lj/Lj−1 that are isomorphic to S as an L-module and for which the exact sequence 0 → Lj/Lj−1 → L/Lj−1 → L/Lj → 0 splits in the category of restricted Lie algebras. Since we will show in Theorem 2.3 that [L : S]p−split is independent of the choice of the p-chief series, we will not indicate the p-chief series in the notation. Let L be a finite-dimensional restricted Lie algebra over a field F of prime characteristic, and let u(L) denote the restricted universal enveloping algebra of L (see [10, p. 192] or [14, p. 90]). Then every restricted L-module is an u(L)- module and vice versa, and so there is a bijection between the irreducible re- stricted L-modules and the irreducible u(L)-modules. In particular, as u(L) is finite-dimensional (see [10, Theorem 12, p. 191] or [14, Theorem 2.5.1(2)]), every irreducible restricted L-module is finite-dimensional. Following Hochschild [8] we define the restricted cohomology of L with coefficients in a restricted L-module M by H n u(L)(F, M ) for every non-negative integer n. ∗ (L, M ) := Extn In [8, Section 3], Hochschild discusses restricted extensions of a strongly abelian restricted Lie algebra M by a restricted Lie algebra L with a fixed action of L on M . In particular, he shows in [8, Theorem 3.3] that the set of equivalence classes of restricted extensions of M by L, which he denotes by ext∗(M, L), is a vector space canonically isomorphic to H 2 ∗ (L, M ). In the following we indicate how to derive part of this result in a different way by using the transgression dE 2 in the five-term exact sequence (1) 0 → H 1 ∗ (L, M ) → H 1 ∗ (E, M ) → HomL(M, M ) dE 2→ H 2 ∗ (L, M ) → H 2 ∗ (E, M ) associated to any restricted extension E : 0 → M → E → L → 0 of M by L. It is clear from the naturality of (1) that dE 2 only depends on the equivalence class [E] of the restricted extension E. Hence one can define a map ∆ : ext∗(M, L) → H 2 ∗ (L, M ) by ∆([E]) := dE 2 (idM ). As for ordinary Lie algebras one has the following result (see [7, Theorem VII.3.3]): Lemma 2.1. Let L be a restricted Lie algebra, and let M be a strongly abelian restricted Lie algebra over a field of prime characteristic p. Furthermore, assume that M is a restricted L-module. Then ∆ : ext∗(M, L) → H 2 ∗ (L, M ) is a bijection. Moreover, the equivalence class of a restricted extension of M by L is mapped to zero if, and only if, its representatives are split. Proof. As we will not need the surjectivity in this paper, we only prove the injec- tivity of ∆ and that restricted extensions are mapped to zero if, and only if, they are split. For more details we refer the reader to the proofs of the analogous results for groups (see [7, Section VI.10]). Let F : 0 → R → F → L → 0 of L be any free presentation of the restricted Lie algebra L, and let E : 0 → M → E → L → 0 be any restricted extension of M by L. Then there exist restricted Lie algebra homomorphisms ϕ : F → E and ρ : R → M such that the diagram (2) commutes. 0 0 / R / F ρ ϕ / M / E / L / L / 0 / 0 Moreover, ρ induces an L-module homomorphism ψ : R → M , where R := R/[R, R] + hR[p]iF. Observe that ψ is surjective if, and only if, ϕ is surjective. / / / / / / / / 4 J ORG FELDVOSS, SALVATORE SICILIANO, AND THOMAS WEIGEL The commutativity of the diagram (2) in conjunction with the naturality of the five-term exact sequence yields that the diagram H 1 ∗ (E, M ) / HomL(M, M ) ϕ∗ ψ∗ H 1 ∗ (F, M ) τ / HomL(R, M ) dE 2 dF 2 / H 2 ∗ (L, M ) / H 2 ∗ (E, M ) / H 2 ∗ (L, M ) / 0 is commutative as well. In particular, we obtain that (3) ∆([E]) = dE 2 (idM ) = dF 2 (ψ∗(idM )) = dF 2 (ψ) . ι2→ E2 We are ready to prove the injectivity of ∆. Let E1 : 0 → M π1→ L → 0 and π2→ L → 0 be two restricted extensions of M by L, and suppose E2 : 0 → M that ∆([E1]) = ∆([E2]). Let F := F (E1 ⊕ E2) denote the free restricted Lie algebra generated by the underlying vector space of E1 ⊕E2. Consider the free presentation F : 0 → R → F → L → 0 of the restricted Lie algebra L. Then the restricted Lie algebra homomorphisms ϕj : F → Ej (j = 1, 2) are surjective, and the L-module homomorphisms ψj : R → M induced by ρj (j = 1, 2) are surjective as well. ι1→ E1 It follows from ∆([E1]) = ∆([E2]) and (3) that ψ1 − ψ2 ∈ Ker(dF 2 ). Thus there exists a restricted derivation D : F → M such that ψ1 − ψ2 = τ (D). Put It is easily seen that ϕ′ Consider the commutative diagram 2 : F → E2 is a restricted Lie algebra homomorphism. ϕ′ 2 := ϕ2 + ι2 ◦ D . 0 0 / R / F ρ′ 2 ϕ′ 2 / M / E2 / L / L / 0 / 0 2 : R → M denotes the map induced by ρ′ If ψ′ 2 = ψ2 + τ (D) = ψ1. In particular, ψ′ 2. Moreover, as ψ1 = ψ′ 2). Consequently, 2) ∼= E2. Consider the map η : E2 → E1 defined by E1 ∼= F/ Ker(ϕ1) = F/ Ker(ϕ′ ϕ′ 2(f ) 7→ ϕ1(f ) for every f ∈ F . Then η is a restricted Lie algebra homomorphism making the following diagram commutative: 2, which implies that Ker(ϕ1) = Ker(ϕ′ 2 is surjective, and, in turn, so is ϕ′ 2 on R, then it is clear that ψ′ 2, we have ρ1 = ρ′ E1 : 0 / M E2 : 0 / M ι2 ι1 / E2 π2 / L η / E1 π1 / L / 0 / 0 We conclude that [E1] = [E2], which finishes the proof of the injectivity of ∆. Finally, let us prove that ∆ maps split restricted extensions to zero. Because of the injectivity of ∆, split restricted extensions are the only ones that are mapped to zero. Without loss of generality we can assume that E is the semi-direct product M ⋊ L of M and L. Recall that the p-mapping on E is defined by (m, x)[p] := (xp−1 · m, x[p]) for every m ∈ M and every x ∈ L (see [8, p. 572]). We have to show that dE 2 (idM ) = 0, or by the exactness of the corresponding five-term sequence, that there exists a restricted derivation D from E to M such that the restriction of D / / / / / / / / / / / / / / / / / / / / / / SPLIT STRONGLY ABELIAN p-CHIEF FACTORS AND RESTRICTED COHOMOLOGY 5 to M is the identity. It is straightforward to check that D(m, x) := m for every m ∈ M and every x ∈ L defines such a restricted derivation. (cid:3) We are ready to prove a restricted analogue of [1, Lemma 2] which will be essential in the proof of our main result (see Theorem 2.3). Lemma 2.2. Let L be a finite-dimensional restricted Lie algebra over a field of prime characteristic p, let I be a minimal p-ideal of L that is strongly abelian, and let E denote the equivalence class of the restricted extension 0 → I → L → L/I → 0. Then the following statements hold: (a) If E splits, then the transgression dE (b) If E does not split, then the transgression dE 2 : HomL(I, I) → H 2 2 : HomL(I, I) → H 2 ∗ (L/I, I) is ∗ (L/I, I) is zero. injective. Proof. (a): It follows from Lemma 2.1 that dE the action of D := HomL(I, I), this implies that dE 2 = 0 2 (idI ) = 0. As dE 2 is compatible with (b): By virtue of the D-linearity of dE 2 , it is enough to show that Ker(dE 2 ) = 0. According to Lemma 2.1, dE 2 (idI ) 6= 0. Then the claim follows from dimD Ker(dE 2 ) + dimD Im(dE 2 ) = dimD HomL(I, I) = 1 . (cid:3) Remark. If one ignores in the proofs of Lemma 2.1 and Lemma 2.2 the compat- ibility of the homomorphisms with the p-mappings, then one obtains conceptual proofs of [1, Lemma 1 and 2], respectively. A restricted Lie algebra L over F is called p-perfect if L = [L, L] + hL[p]iF. By virtue of [3, Proposition 2.7], L is p-perfect if, and only if, H 1 ∗ (L, F) = 0. Our main result is completely analogous to the main result of [13] (see also [6, Theorem 2.1] for the analogue for ordinary Lie algebras): Theorem 2.3. Let L be a finite-dimensional restricted Lie algebra over a field of prime characteristic p, and let S be an irreducible L-module with centralizer algebra D := EndL(S). Then (4) [L : S]p−split = dimD H 1 ∗ (L, S) − dimD H 1 ∗ (L/ AnnL(S), S) holds. In particular, [L : S]p−split is independent of the choice of the p-chief series of L. Proof. We proceed by induction on the dimension of L. If L is one-dimensional, then L is either a torus or strongly abelian. For a torus both sides of (4) vanish and in the strongly abelian case the only irreducible restricted L-module is trivial so that both sides of (4) are also equal. Thus we may assume that the dimension of L is greater than one, and that the claim holds for all restricted Lie algebras of dimension less than dimF L. Let 0 = L0 ⊂ L1 ⊂ · · · ⊂ Ln = L be a p-chief series of L. For the remainder of the proof the multiplicity [L : S]p−split always refers to this fixed p-chief series. If AnnL(S) = 0, then the right-hand side of (4) is zero. But as strongly abelian p-chief factors have non-zero annihilators, the left-hand side also vanishes and the assertion holds. So we may assume that AnnL(S) 6= 0. 6 J ORG FELDVOSS, SALVATORE SICILIANO, AND THOMAS WEIGEL We first assume that L1 ⊆ AnnL(S). Then the five-term exact sequence for restricted cohomology in conjunction with [3, Proposition 2.7] yields the exactness of (5) 0 −→ H 1 ∗ (L/L1, S) −→ H 1 ∗ (L, S) −→ HomL(L1/[L1, L1] + hL[p] 1 iF, S) −→ H 2 ∗ (L/L1, S) . Since S is also an irreducible restricted L/L1-module, one obtains by induction that (6) [L/L1 : S]p−split = dimD H 1 ∗ (L/L1, S) − dimD H 1 ∗ (L/ AnnL(S), S) . As L1 is a minimal p-ideal of L, L1 is either p-perfect or strongly abelian. In the former case, the third term in (5) vanishes, and thus H 1 ∗ (L, S). Since L1 is not strongly abelian, one has [L : S]p−split = [L/L1 : S]p−split. Hence (4) holds in this case. ∗ (L/L1, S) ∼= H 1 If L1 is strongly abelian, one has HomL(L1/[L1, L1]+hL[p] 1 iF, S) = HomL(L1, S). If in addition L1 and S are not isomorphic as L-modules, then HomL(L1, S) = 0, and the assertion follows as before. For L1 ∼= S one has to distinguish two cases depending on the strongly abelian p-chief factor L1 being split, or being not split. In case that L1 is split, one has (7) [L : S]p−split = [L/L1 : S]p−split + 1 = dimD H 1 ∗ (L/L1, S) − dimD H 1 ∗ (L/ AnnL(S), S) + 1 , and Lemma 2.2(a) shows that the transgression HomL(L1, S) → H 2 zero. Thus the exactness of (5) implies that the restriction H 1 is surjective, and therefore ∗ (L/L1, S) is ∗ (L, S) → HomL(L1, S) (8) dimD H 1 ∗ (L, S) = dimD H 1 = dimD H 1 ∗ (L/L1, S) + dimD HomL(L1, S) ∗ (L/L1, S) + 1 . Hence (7) and (8) yield the assertion. Suppose now that L1 is not split. In this case Lemma 2.2(b) implies that the transgression HomL(L1, S) → H 2 ∗ (L/L1, S) is injective. According to (5), the inflation H 1 ∗ (L, S) is bijective. Then one has [L : S]p−split = [L/L1 : S]p−split, and the claim follows from (6). ∗ (L/L1, S) → H 1 Finally, assume that L1 6⊆ AnnL(S), i.e., L1 ∩ AnnL(S) = 0 and SL1 = 0. Suppose that Lj/Lj−1 is strongly abelian and Lj/Lj−1 ∼= S as L-modules for some integer j with 1 ≤ j ≤ n. Then Lj -- and thus L1 -- would be contained in AnnL(S), a contradiction. Hence [L : S]p−split = 0. As SL1 = 0, one concludes from the beginning of the five-term exact sequence 0 −→ H 1 ∗ (L/L1, SL1) −→ H 1 ∗ (L, S) −→ H 1 ∗ (L1, S)L −→ H 2 ∗ (L/L1, SL1) that the vertical mappings in the commutative diagram H 1 ∗ (L/ AnnL(S), S) α H 1 ∗ (L, S) H 1 ∗ (L1 + AnnL(S)/ AnnL(S), S)L β / H 1 ∗ (L1, S)L are isomorphisms. Because β is an isomorphism, α is an isomorphism as well. This shows that in this case the right-hand side of (4) is also zero. Since the right-hand side of (4) does not depend on the choice of the p-chief series, (cid:3) the left-hand side does not either. This completes the proof of the theorem. / / / SPLIT STRONGLY ABELIAN p-CHIEF FACTORS AND RESTRICTED COHOMOLOGY 7 In the extreme case AnnL(S) = L, Theorem 2.3 in conjunction with [3, Propo- sition 2.7] has the following consequence: Corollary 2.4. Let L be a finite-dimensional restricted Lie algebra over a field F of prime characteristic p. Then the trivial irreducible L-module occurs with multi- plicity dimF L/[L, L] + hL[p]iF as a split strongly abelian p-chief factor of L. Moreover, the next result follows from Hochschild's six-term exact sequence re- lating ordinary and restricted cohomology (see [8, p. 575]) in conjunction with Corollary 2.4 and [6, Corollary 2.2]. (Here [L : S]split denotes the multiplicity of S as a split abelian chief factor of the ordinary Lie algebra L.) Corollary 2.5. Let L be a finite-dimensional restricted Lie algebra over a field F of prime characteristic p. If S is an irreducible restricted L-module, then [L : S]p−split = (cid:26) [L : S]split [L : S]split − dimF(hL[p]iF/[L, L] ∩ hL[p]iF) if S 6∼= F if S ∼= F . In particular, [L : S]p−split ≤ [L : S]split. The equality of [L : S]p−split and [L : S]split for non-trivial irreducible restricted L-modules S explains why the results in Section 5 of [6] could be obtained although their ingredients belong to different categories. Recall that the principal block of a restricted Lie algebra is the block that contains the trivial irreducible module. For the convenience of the reader we include a proof of the following result which is completely analogous to the corresponding proof for modular group algebras (see [12, Proposition 1]). Proposition 2.6. Every strongly abelian p-chief factor of a finite-dimensional re- stricted Lie algebra L belongs to the principal block of L. Proof. Let S = I/J be a strongly abelian p-chief factor of L. In particular, S is a trivial I-module. Then the five-term exact sequence for restricted cohomology in conjunction with [3, Proposition 2.7] yields the exactness of 0 −→ H 1 ∗ (L/I, S) −→ H 1 ∗ (L/J, S) −→ HomL(S, S) −→ H 2 ∗ (L/I, S) . Since the third term is non-zero, the second or fourth term must also be non-zero. According to [4, Lemma 1(a)], in either case S belongs to the principal block of a restricted Lie factor algebra of L. Then it follows from [4, Lemma 4] that S also belongs to the principal block of L. (cid:3) The analogue of Theorem 1.1 for restricted Lie algebras is another consequence of Theorem 2.3 in conjunction with the equivalence (i)⇐⇒(iv) in [6, Theorem 5.5]. Theorem 2.7. Let L be a finite-dimensional restricted Lie algebra over a field F of prime characteristic p. Then the following statements are equivalent: (i) L is solvable. (ii) dimF H 1 ∗ (L, S) = dimF EndL(S) · [L : S]p−split holds for every irreducible L-module S. (iii) dimF H 1 ∗ (L, S) = dimF EndL(S) · [L : S]p−split holds for every irreducible L-module S belonging to the principal block of L. 8 J ORG FELDVOSS, SALVATORE SICILIANO, AND THOMAS WEIGEL Proof. The equivalence of (i) and (ii) is a consequence of Theorem 2.3 and the equivalence (i)⇐⇒(iv) in [6, Theorem 5.5], and the equivalence of (ii) and (iii) follows from [4, Lemma 1(a)] in conjunction with Proposition 2.6. (cid:3) Remark. It is an immediate consequence of Theorem 2.3 that dimF H 1 ∗ (L, S) = dimF EndL(S)·[L : S]p−split holds for the trivial irreducible L-module S. Hence one can also obtain Theorem 2.7 immediately from Corollary 2.5 and the equivalence of (i), (vi), and (vii) in [6, Theorem 5.5]. We included the proof given above since it is the precise analogue of the proof of [13, Corollary 1]. 3. Split strongly abelian p-chief factors and the 0-PIM Let A be a finite-dimensional (unital) associative algebra with Jacobson radical Jac(A), and let M be a (unital left) A-module. Then the descending filtration M ⊃ Jac(A)M ⊃ Jac(A)2M ⊃ Jac(A)3M ⊃ · · · ⊃ Jac(A)ℓM ⊃ Jac(A)ℓ+1M = 0 is called the Loewy series of M and the factor module Jac(A)n−1M/ Jac(A)nM is called the nth Loewy layer of M (see [2, Definition 1.2.1] or [9, Definition VII.10.10a)]). Recall that a projective module PA(M ) is a projective cover of M , if there ex- ists an A-module epimorphism πM from PA(M ) onto M such that the kernel of πM is contained in the radical Jac(A)PA(M ) of PA(M ). If projective covers exist, then they are unique up to isomorphism. It is well known that projective covers of finite-dimensional modules over finite-dimensional associative algebras exist and are again finite-dimensional. Moreover, every projective indecomposable A-module is isomorphic to the projective cover of some irreducible A-module. In this way one obtains a bijection between the isomorphism classes of the projective indecompos- able A-modules and the isomorphism classes of the irreducible A-modules. In the sequel we use the notation PL(F) := Pu(L)(F) for the projective cover of the trivial irreducible module of a finite-dimensional restricted Lie algebra L over a field F of prime characteristic. Using [2, Proposition 2.4.3] and Theorem 2.3 we obtain a lower bound for the multiplicity [Jac(u(L))PL(F)/ Jac(u(L))2PL(F) : S] of an irreducible restricted L-module S in the second Loewy layer of PL(F) (see [15, Theorem 3.7] for the analogue in the modular representation theory of finite groups): Theorem 3.1. Let L be a finite-dimensional restricted Lie algebra over a field F of prime characteristic p. Then [Jac(u(L))PL(F)/ Jac(u(L))2PL(F) : S] ≥ [L : S]p−split for every irreducible restricted L-module S. Proof. We obtain from [2, Proposition 2.4.3] and Theorem 2.3 that dimF EndL(S) · [Jac(u(L))PL(F)/ Jac(u(L))2PL(F) : S] u(L)(F, S) = dimF H 1 = dimF Ext1 ≥ dimF H 1 = dimF EndL(S) · [L : S]p−split . ∗ (L, S) ∗ (L, S) − dimF H 1 ∗ (L/ AnnL(S), S) Cancelling dimF EndL(S) yields the desired inequality. (cid:3) SPLIT STRONGLY ABELIAN p-CHIEF FACTORS AND RESTRICTED COHOMOLOGY 9 Remark. If one uses the main result of [13] instead of Theorem 2.3, then the above proof would also work in the case of finite-dimensional modular group algebras. This provides an alternative proof of [15, Theorem 3.7]. The following example shows that equality does not necessarily hold in Theo- rem 3.1. We will see soon that equality holds if, and only if, the restricted Lie algebra is solvable (see the equivalence (i)⇐⇒(ii) in Theorem 3.3). Example. Consider the three-dimensional restricted simple Lie algebra L := sl2(F) over an algebraically closed field F of characteristic p > 2. Take for S the (p − 1)-dimensional irreducible restricted L-module. Then it follows from [11, The- orem 1(ii)] that [Jac(u(L))PL(F)/ Jac(u(L))2PL(F) : S] = 2, but [L : S]p−split = 0. As an immediate consequence of Theorem 3.1, we obtain the following weak analogue of a well-known result for finite modular group algebras: Corollary 3.2. Every split strongly abelian p-chief factor of a finite-dimensional restricted Lie algebra L is a direct summand of the second Loewy layer of the pro- jective cover PL(F) of the trivial irreducible L-module. In particular, every split strongly abelian p-chief factor of a finite-dimensional restricted Lie algebra L is a composition factor of PL(F). Question. In view of Corollary 3.2, it is natural to ask whether every strongly abelian p-chief factor of a finite-dimensional solvable restricted Lie algebra L is a composition factor of PL(F), or even more generally (see Proposition 2.6), whether every irreducible module in the principal block of u(L) is a composition factor of PL(F) (for an affirmative answer to the analogous question in the modular repre- sentation theory of finite p-solvable groups see [9, Theorem VII.15.8]). Finally, we obtain the following characterization of solvable restricted Lie alge- bras which was motivated by [6, Theorem 5.5] but contrary to the latter allows to include the trivial irreducible module in the implications (i)=⇒(ii) and (i)=⇒(iii) (see [15, Theorem 3.9] for the analogue of (i)⇐⇒(ii) in the modular representation theory of finite groups). Theorem 3.3. Let L be a finite-dimensional restricted Lie algebra over a field F of prime characteristic p. Then the following statements are equivalent: (i) L is solvable. (ii) [Jac(u(L))PL(F)/ Jac(u(L))2PL(F) : S] = [L : S]p−split for every irreducible restricted L-module S. (iii) [Jac(u(L))PL(F)/ Jac(u(L))2PL(F) : S] = [L : S]p−split for every irreducible restricted L-module S belonging to the principal block of L. Proof. The equivalence of the three statements is a consequence of Theorem 2.7 in conjunction with dimF EndL(S) · [Jac(u(L))PL(F)/ Jac(u(L))2PL(F) : S] = dimF H 1 ∗ (L, S) . (cid:3) Remark. It follows from the proof of Theorem 3.1 that the equality in statements (ii) and (iii) of Theorem 3.3 holds for the trivial irreducible L-module. Hence one 10 J ORG FELDVOSS, SALVATORE SICILIANO, AND THOMAS WEIGEL can also obtain Theorem 3.3 from Corollary 2.5 and the equivalence of (i), (viii), and (ix) in [6, Theorem 5.5]. Acknowledgements. The first and the second author would like to thank the Dipartimento di Matematica e Applicazioni at the Universit`a degli Studi di Milano- Bicocca for the hospitality during their visit in May 2012 when parts of this paper were written. References [1] D. W. Barnes: First cohomology groups of soluble Lie algebras, J. Algebra 46 (1977), no. 1, 292 -- 297. [2] D. J. Benson: Representations and Cohomology I: Basic Representation Theory of Finite Groups and Associative Algebras, Cambridge Studies in Advanced Mathematics, vol. 30, Cambridge University Press, Cambridge, 1991. [3] J. Feldvoss: On the cohomology of restricted Lie algebras, Comm. Algebra 19 (1991), no. 10, 2865 -- 2906. [4] J. Feldvoss: On the block structure of supersolvable restricted Lie algebras, J. Algebra 183 (1996), no. 2, 396 -- 419. [5] J. Feldvoss: On the cohomology of modular Lie algebras, in: Lie Algebras, Vertex Operator Algebras and Their Applications, Raleigh, NC, 2005 (eds. Y.-Z. Huang and K. C. Misra), Contemp. Math., vol. 442, Amer. Math. Soc., Providence, RI, 2007, pp. 89 -- 113. [6] J. Feldvoss, S. Siciliano, and T. Weigel: Split abelian chief factors and first degree cohomology for Lie algebras, arXiv:1206.3669 (12 pages), accepted for publication in J. Algebra. [7] P. J. Hilton and U. Stammbach: A Course in Homological Algebra (Second edition), Graduate Texts in Mathematics, vol. 4, Springer-Verlag, New York/Berlin/Heidelberg, 1997. [8] G. Hochschild: Cohomology of restricted Lie algebras, Amer. J. Math. 76 (1954), no. 3, 555 -- 580. [9] B. Huppert and N. Blackburn: Finite Groups II , Grundlehren der Mathematischen Wis- senschaften, vol. 242, Springer-Verlag, Berlin/Heidelberg/New York, 1982. [10] N. Jacobson: Lie Algebras, Dover Publications, Inc., New York, 1979 (unabridged and cor- rected republication of the original edition from 1962). [11] R. D. Pollack: Restricted Lie algebras of bounded type, Bull. Amer. Math. Soc. 74 (1968), no. 2, 326 -- 331. [12] U. Stammbach: Cohomological characterisations of finite solvable and nilpotent groups, J. Pure Appl. Algebra 11 (1977/78), no. 1-3, 293 -- 301. [13] U. Stammbach: Split chief factors and cohomology, J. Pure Appl. Algebra 44 (1987), no. 1-3, 349 -- 352. [14] H. Strade and R. Farnsteiner: Modular Lie Algebras and Their Representations, Mono- graphs and Textbooks in Pure and Applied Mathematics, vol. 116, Marcel Dekker, Inc., New York/Basel, 1988. [15] W. Willems: On p-chief factors of finite groups, Comm. Algebra 13 (1985), no. 11, 2433 -- 2447. Department of Mathematics and Statistics, University of South Alabama, Mobile, AL 36688 -- 0002, USA E-mail address: [email protected] Dipartimento di Matematica e Fisica "Ennio De Giorgi", Universit`a del Salento, Via Provinciale Lecce-Arnesano, I-73100 Lecce, Italy E-mail address: [email protected] Dipartimento di Matematica e Applicazioni, Universit`a degli Studi di Milano-Bicocca, Via Roberto Cozzi, No. 53, I-20125 Milano, Italy E-mail address: [email protected]
1101.5697	5	1101	2013-09-02T04:43:31	Recollements and Hochschild theory	[ "math.RA", "math.KT", "math.RT" ]	It is shown that a recollement of derived categories of algebras induces those of tensor product algebras and opposite algebras respectively, which is applied to clarify the relations between recollements of derived categories of algebras and smoothness and Hochschild cohomology of algebras.	math.RA	math	Recollements and Hochschild theory Yang Han KLMM, ISS, AMSS, Chinese Academy of Sciences, Beijing 100190, P.R. China. E-mail: [email protected] Dedicated to the memory of Dieter Happel Abstract It is shown that a recollement of derived categories of algebras in- duces those of tensor product algebras and opposite algebras respec- tively, which is applied to clarify the relations between recollements of derived categories of algebras and smoothness and Hochschild co- homology of algebras. Mathematics Subject Classification (2010) : 16E40, 16E35, 18E30 Keywords : recollement, smoothness, Hochschild cohomology. 1 Introduction Let T1, T and T2 be triangulated categories. A recollement of T relative to T1 and T2 is given by ✛ i∗ T1 i∗ = i! ✲ ✛ i! T ✛ j! j! = j∗ ✲ ✛ j∗ T2 and denoted by 9-tuple (T1, T , T2, i∗, i∗ = i!, i!, j!, j! = j∗, j∗) such that (R1) (i∗, i∗), (i!, i!), (j!, j!) and (j∗, j∗) are adjoint pairs of triangle functors; (R2) i∗, j! and j∗ are full embeddings; (R3) j!i∗ = 0 (and thus also i!j∗ = 0 and i∗j! = 0); (R4) for each X ∈ T , there are triangles 1 j!j!X → X → i∗i∗X → i!i!X → X → j∗j∗X → where the arrows to and from X are the counit and the unit, respectively. Recollements of triangulated categories are "short exact sequences" of tri- angulated categories. They were introduced by Beilinson-Bernstein-Deligne [5] and play an important role in algebraic geometry [5], representation theory [11, 35], etc. Let k be a field and ⊗ := ⊗k. Throughout the paper, all alge- bras are assumed to be associative k-algebras with identity, and all modules are right unitary modules unless stated otherwise. Here, we focus on recolle- ments of derived categories of algebras, i.e., all triangulated categories in the recollements are derived categories of algebras, which are closely related to tilting theory [1, 23, 32], (co)localization theory [31, 29], some important ho- mological invariants of algebras such as global dimension [41, 23, 2], finitistic dimension [18], Hochschild homology and cyclic homology [22], and so on. In this paper, we shall show that a recollement of derived categories of algebras induces those of tensor product algebras (see section 3) and oppo- site algebras (see section 4) respectively. As applications, we shall clarify the relations between recollements of derived categories of algebras and smooth- ness, i.e., finiteness of Hochschild dimension, and Hochschild cohomology of algebras. Note that the relations between recollements of derived categories of algebras and Hochschild homology and cyclic homology have been clar- ified already in [22]. More precisely, we shall show in section 5 that, in a recollement of derived categories of algebras, the middle algebra is smooth if and only if so are the algebras on both sides. As a corollary, a triangular matrix algebra is smooth if and only if so are the algebras on diagonal. In sec- tion 6, we shall obtain three triangles on Hochschild cocomplexes which can induce three long exact sequences on Hochschild cohomologies of algebras. Note that these long exact sequences on Hochschild cohomologies have been widely studied for one-point extensions [17, 15], triangular matrix algebras [9, 30, 16, 10, 7], stratifying ideals [25], homological epimorphisms [36, 39], etc. 2 Standard recollements In this section, we show that every recollement of derived categories of algebras is equivalent to a standard one in which all triangle functors are 2 naturally isomorphic to derived functors. This result is already known for some experts. 2.1 Recollements of derived categories of algebras Let A be an algebra. Denote by ProjA (resp. projA) the category of projective (resp. finitely generated projective) A-modules. Denote by D(A) the unbounded derived category of complexes of A-modules. Let X be an object in D(A). Denote by X ⊥ the full subcategory of D(A) consisting of all objects Y ∈ D(A) such that HomD(A)(X, Y [n]) = 0 for all n ∈ Z. Denote by TriaX the smallest full triangulated subcategory of D(A) which contains X and is closed under small coproducts. We say X is exceptional if HomD(A)(X, X[n]) = 0 for all n ∈ Z\{0}. We say X is compact if the functor HomD(A)(X, −) preserves small coproduct, or equivalently, X is perfect, i.e., isomorphic in D(A) to an object in K b(projA), the homotopy category of bounded complexes of finitely generated projective A-modules. We say X is self-compact if HomD(A)(X, −) preserves small coproducts in TriaX (ref. [20]). A very important criterion for the right bounded derived category of an [34, Theorem 3]). algebra to admit a recollement is provided in [23] (cf. It was extended and modified to suit for the unbounded derived categories of algebras (ref. [20, Theorem 3.3]) and differential graded categories (ref. [33, Corollary 3.4]). [33, Corollary 3.4]), differential graded algebras (ref. Proposition 1. (Konig [23]; Jørgensen [20]; Nicol´as-Saorin [33]) Let A1, A and A2 be algebras. Then D(A) admits a recollement relative to D(A1) and D(A2) if and only if there are objects X1 and X2 in D(A) such that (1) EndD(A)(Xi) ∼= Ai as algebras for i = 1, 2; (2) X2 (resp. X1) is exceptional and compact (resp. self-compact); (3) X1 ∈ X ⊥ 2 ; 1 ∩ X ⊥ (4) X ⊥ 2 = {0}. An important example of recollements of derived categories of algebras is given by stratifying ideals: Example 1. (Cline-Parshall-Scott [12]) Let A be an algebra, e an idem- potent of A, and AeA a stratifying ideal of A, i.e., the multiplication in A induces an isomorphism Ae ⊗eAe eA ∼= AeA and ToreAe n (Ae, eA) = 0 for all n ≥ 1. Then there is a recollement (D(A/AeA), D(A), D(eAe), i∗, i∗ 3 = i!, i!, j!, j! = j∗, j∗) where A A/AeA, i∗ = − ⊗L i∗ = i! = − ⊗L i! = RHomA(A/AeA, −), A/AeA A/AeA, j! = − ⊗L eAe eA, j! = j∗ = − ⊗L j∗ = RHomeAe(Ae, −). A Ae, 2.2 Standard recollements Definition 1. Let A1, A and A2 be algebras. A recollement (D(A1), D(A), D(A2), i∗, i∗ = i!, i!, j!, j! = j∗, j∗) is said to be standard and defined by Y ∈ D(Aop ⊗ A1) and Y2 ∈ D(Aop 2 ⊗ A) if i∗ ∼= − ⊗L A Y and j! ∼= − ⊗L A2 Y2. Proposition 2. Let A1, A and A2 be algebras, and (D(A1), D(A), D(A2), i∗, i∗ = i!, i!, j!, j! = j∗, j∗) a standard recollement defined by Y ∈ D(Aop ⊗ A1) and Y2 ∈ D(Aop 2 ⊗ A). Then A Y, i∗ ∼= − ⊗L i∗ = i! i! ∼= RHomA(RHomA1(Y, A1), −), ∼= RHomA1(Y, −), ∼= − ⊗L A2 Y2, j! j! = j∗ ∼= RHomA(Y2, −), j∗ ∼= RHomA2(RHomA(Y2, A), −). Proof. Since i∗ is a left adjoint of i!, it commutes with small coproduct. A Y ∼= i∗ has a right adjoint i∗ which commutes with small The functor − ⊗L coproduct, thus Y is compact in D(A1). Therefore, RHomA1(Y, −) ∼= − ⊗L A1 [32, Lemma 2.6]). Since the right adjoint is unique RHomA1(Y, A1) (ref. ∼= RHomA1(Y, −) and i! ∼= up to natural isomorphism, we have i∗ = i! RHomA(RHomA1(Y, A1), −). Similar for j! = j∗ and j∗. Proposition 3. Let A, A1 and A2 be algebras. If D(A) admits a recolle- ment relative to D(A1) and D(A2) then D(A) admits a standard recollement relative to D(A1) and D(A2). Proof. Let (D(A1), D(A), D(A2), i∗, i∗ = i!, i!, j!, j! = j∗, j∗) be a recollement. It follows from Proposition 1 that there are objects Xi, i = 1, 2, in D(A) such that they satisfy all conditions in Proposition 1. Clearly, we may assume that X2 is homotopically projective. Since X2 is exceptional, it follows from [24, 8.3.1] that there exists Y2 ∈ D(Aop 2 ⊗ A) such that the derived tensor functor − ⊗L A2 Y2 : D(A2) → D(A) sends A2 to X2. By [32, Theorem 2.8], we have a recollement (X ⊥ ∗) where ! = − ⊗L j′ ∗ = RHomA2(RHomA(Y2, A), −), and i′ ∗ = i′ A2 Y2, j′! = j′∗ = RHomA(Y2, −), j′ ! is the natural embedding. 2 , D(A), D(A2), i′∗, i′ !, j′! = j′∗, j′ !, i′!, j′ ∗ = i′ 4 By [6, Theorem 3] or [33, §4 Theorem], there is a homological epimor- phism of differential graded algebras f : A → C such that the essential image Imf∗ of the induced functor f∗ : D(C) → D(A) and Imi∗ = X ⊥ 2 coincide. In particular, D(A1) and D(C) are equivalent as triangulated categories. It follows from [21] that there is a two-sided tilting differential graded A1-C bimodule Z such that − ⊗L A1 Z : D(A1) → D(C) is a triangle equivalence. f∗→ D(A) is a fully faithful triangle Thus the composition D(A1) functor which is naturally isomorphic to − ⊗L A1 Y1, where Y1 is the image of Z under the functor f∗ : D(Aop 1 ⊗ C) → D(Aop 1 ⊗ A). Since A1Z is compact in D(Aop (Y1, A1) is a differential graded A-A1- bimodule, i.e., a complex of A-A1-bimodules, which is compact in D(A1) so there is a natural isomorphism RHomA1(Y, −) ∼= − ⊗L A1 Y1 : D(A1) → D(A). Consequently, − ⊗L A1 Y1. Thus D(A) admits a stan- dard recollement relative to D(A1) and D(A2) defined by Y ∈ D(Aop ⊗ A1) and Y2 ∈ D(Aop 1 ), so is A1Y1. Thus Y := RHomAop A Y is left adjoint to − ⊗L −⊗L A1 → D(C) 2 ⊗ A). Z 1 1 , T ′, T ′ Remark 1. Two recollements (T1, T , T2, i∗, i∗ = i!, i!, j!, j! = j∗, j∗) and !, j′! = j′∗, j′ (T ′ ∗) are said to be equivalent if (Imi∗, Imj!, Imj∗) = (Imi′ ∗). From the proof of Proposition 3, it is easy to see that the new constructed recollement is equivalent to the original given one. ∗ = i′ ∗, Imj′ !, i′!, j′ !, Imj′ 2 , i′∗, i′ 3 Recollements on tensor product algebras In this section, we show that from a standard recollement of derived categories of algebras we can obtain those of tensor product algebras. Lemma 1. Let A and B be algebras, and X, Y ∈ D(A). Then the canonical homomorphism B ⊗ HomD(A)(X, Y ) → HomD(B⊗A)(B ⊗ X, B ⊗ Y ) is an isomorphism when (1) X is compact, or (2) X is self-compact and Y ∈ TriaX. Proof. Since the functor B ⊗ − : D(A) → D(B ⊗ A) is left adjoint to the forgetful functor, we have HomD(B⊗A)(B ⊗ X, B ⊗ Y ) ∼= HomD(A)(X, B ⊗ Y ), which is further isomorphic to B ⊗ HomD(A)(X, Y ) when X is compact, or X is self-compact and Y ∈ TriaX. Lemma 2. Let A, B and C be algebras, and Y ∈ D(Aop ⊗ C). Then there are natural isomorphisms 5 B⊗A (B ⊗ Y ) ∼= − ⊗L (1) − ⊗L (2) RHomB⊗C(B ⊗ Y, −) ∼= RHomC(Y, −) : D(B ⊗ C) → D(B ⊗ A). A Y : D(B ⊗ A) → D(B ⊗ C), and Proof. (1) holds since − ⊗L A Y . (2) holds since the functor B ⊗ − : D(A) → D(B ⊗ A) is left adjoint to the forgetful functor. B⊗A (B ⊗ Y ) ∼= (− ⊗L A Y ∼= − ⊗L B B) ⊗L Theorem 1. Let A, A1, A2 and B be algebras, and Y ∈ D(Aop ⊗ A1) and Y2 ∈ D(Aop 2 ⊗A) define a standard recollement of D(A) relative to D(A1) and D(A2). Then B ⊗ Y and B ⊗ Y2 define a standard recollement of D(B ⊗ A) relative to D(B ⊗ A1) and D(B ⊗ A2). Moreover, the six triangle functors in this recollement, now denoted with capital letters, are A Y, I ∗ ∼= − ⊗L I∗ = I! I ! ∼= RHomA(RHomA1(Y, A1), −), ∼= RHomA1(Y, −), ∼= − ⊗L A2 Y2, J! J ! = J ∗ ∼= RHomA(Y2, −), J∗ ∼= RHomA2(RHomA(Y2, A), −). Proof. We may assume that Y and Y2 are homotopically projective, and := (Yi)A, i = 1, 2, in D(A) Y1 := RHomA1(Y, A1). Then the objects Xi satisfy all conditions in Proposition 1 (ref. [20, 23]). Let Zi := B ⊗ Xi for i = 1, 2. Now we show that Z1 and Z2 satisfy all conditions in Proposition 1 for tensor product algebras. Step 1. Since X2 is compact in D(A), Z2 is compact in D(B⊗A). Since X2 is compact and exceptional and EndD(A)(X2) ∼= A2 as algebras, by Lemma 1, we have HomD(B⊗A)(Z2, Z2[n]) ∼= B ⊗HomD(A)(X2, X2[n]) ∼= (cid:26) B ⊗ A2, Thus Z2 is exceptional and EndD(B⊗A)(Z2) ∼= B ⊗ A2 as algebras. 0, if n = 0; otherwise. Step 2. The forgetful functor D(B ⊗ A) → D(A) maps the objects in TriaD(B⊗A)(B ⊗ X1) to the objects in TriaD(A)(X1). Indeed, let C be the full triangulated subcategory of D(B ⊗A) consisting of the objects which become objects of TriaD(A)(X1) when applying the forgetful functor. Then C is a full triangulated subcategory of D(B ⊗ A) containing B ⊗ X1 and closed under small coproducts. Thus TriaD(B⊗A)(B ⊗ X1) ⊆ C. Since X1 is self-compact, by Lemma 2, for any index set Λ and Tλ ∈ TriaD(B⊗A)(B ⊗ X1) with λ ∈ Λ, we have HomD(B⊗A)(Z1, ⊕λ∈ΛTλ) ∼= HomD(A)(X1, ⊕λ∈ΛTλ) ∼= ⊕λ∈ΛHomD(A)(X1, Tλ) ∼= ⊕λ∈ΛHomD(B⊗A)(Z1, Tλ). 6 Thus Z1 is self-compact. Since X1 is self-compact and exceptional and EndD(A)(X1) ∼= A1 as alge- bras, by Lemma 1, we have HomD(B⊗A)(Z1, Z1[n]) ∼= B ⊗HomD(A)(X1, X1[n]) ∼= (cid:26) B ⊗ A1, Thus Z1 is exceptional and EndD(B⊗A)(Z1) ∼= B ⊗ A1 as algebras. 0, if n = 0; otherwise. Step 3. Since X2 is compact and X1 ∈ X ⊥ 2 , by Lemma 1, we have HomD(B⊗A)(Z2, Z1[n]) ∼= B ⊗ HomD(A)(X2, X1[n]) = 0 for all n ∈ Z. Thus Z1 ∈ Z ⊥ 2 . Step 4. For any Z ∈ Z ⊥ 1 ∩ Z ⊥ 2 , by Lemma 2, we have HomD(A)(Xi, Z[n]) ∼= HomD(B⊗A)(Zi, Z[n]) = 0 1 ∩ X ⊥ 2 = {0}. Hence Z ⊥ for all n ∈ Z and i = 1, 2. Thus Z ∈ X ⊥ 2 = {0}. Step 5. By the Steps 1 to 4 above, we have shown that Z1 and Z2 satisfy all conditions in Proposition 1 for tensor product algebras. Analogous to Jørgensen's construction (ref. [20, Theorem 3.3 and Remark 3.4]), we can obtain a recollement (D(B ⊗A1), D(B ⊗A), D(B ⊗A2), I ∗, I∗ = I!, I !, J!, J ! = J ∗, J∗) such that 1 ∩ Z ⊥ I∗ = I! = − ⊗L I ! = RHomB⊗A(B ⊗ Y1, −), B⊗A1 (B ⊗ Y1), J ! = J ∗ = RHomB⊗A(B ⊗ Y2, −), J∗ = RHomB⊗A2(RHomB⊗A(B ⊗ Y2, B ⊗ A), −). J! = − ⊗L B⊗A2 (B ⊗ Y2), By Lemma 2, we have I∗ = I! A2 Y2, I ! ∼= RHomA(Y1, −) and J ! = J ∗ ∼= RHomA(Y2, −). Since Y2 is compact in D(A), by Lemma 2 and Lemma 1, we have ∼= − ⊗L ∼= − ⊗L A1 Y1, J! J∗ = RHomB⊗A2(RHomB⊗A(B ⊗ Y2, B ⊗ A), −) ∼= RHomB⊗A2(B ⊗ RHomA(Y2, A), −) ∼= RHomA2(RHomA(Y2, A), −). ∼= − ⊗L Clearly, I∗ B⊗A (B ⊗ Y ). Thus we have I ∗ ∼= − ⊗L A1 Y1 − ⊗L ∼= RHomA1(Y, −) has a left adjoint − ⊗L A Y ∼= B⊗A (B ⊗ Y ) ∼= − ⊗L A Y . 7 4 Recollements on opposite algebras In this section, we show that from a standard recollement of derived categories of algebras we can obtain that of opposite algebras. Let A and B be algebras. Denote by rep(B, A) the full subcategory of D(Bop ⊗ A) consisting of all complexes of B-A-bimodules which are perfect when restricted to complexes of A-modules. The following result is folklore: Lemma 3. Let A and B be algebras. Then the derived Hom functor RHomA(−, A) : D(Bop ⊗ A) → D(Aop ⊗ B) induces a duality from rep(B, A) to rep(Bop, Aop). Theorem 2. Let A, A1 and A2 be algebras, and Y ∈ D(Aop ⊗ A1) and Y2 ∈ D(Aop 2 ⊗ A) define a standard recollement of D(A) relative to D(A1) and D(A2). Then Y ⋆ := RHomA1(Y, A1) and Y ∗ 2 := RHomA(Y2, A) define a standard recollement of D(Aop) relative to D(Aop 1 ) and D(Aop 2 ). Proof. Step 1. Since (Y2)A is compact and exceptional and EndD(A)(Y2) ∼= A2, we have HomD(Aop)(Y ∗ 2 , Y ∗ 2 [n]) ∼= HomD(A)(Y2, Y2[n]) = (cid:26) A2, 0, if n = 0; otherwise. Therefore, A(Y ∗ 2 ) is compact and exceptional and EndD(Aop)(Y ∗ 2 ) ∼= Aop 2 . Step 2. It follows from Theorem 1 by taking B = Aop A Y ∼= A1 A1 − is a full embedding. Hence AY is self-compact 1 that Y ⋆ ⊗L in D(Aop in D(A) (ref. [20, Lemma 1.7]). 1 ⊗A1). Thus Y ⊗L By Lemma 3, we have Y ∼= Y ⋆⋆ in D(Aop ⊗ A1). Since A1Y ⋆ is compact, we have HomD(Aop)(Y, Y [n]) ∼= HomD(Aop)(Y, Y ⋆⋆[n]) (Y ⋆, A1[n])) A Y, A1[n]) 1 1 )(Y ⋆ ⊗L 1 )(A1, A1[n]) ∼= HomD(Aop)(Y, RHomAop ∼= HomD(Aop ∼= HomD(Aop ∼= H 0(A1[n]) ∼= (cid:26) A1, if n = 0; otherwise. 0, Thus AY is exceptional and EndD(Aop)(Y ) ∼= Aop 1 as algebras where the isomorphism is given by sending an element a1 ∈ A1 to the right multiplica- tion of Y by a1. 8 Step 3. Since (Y2)A is compact and Y ⋆ ∈ Y ⊥ 2 in D(A), we have HomD(Aop)(Y ∗ 2 , Y ) ∼= HomD(Aop)(Y ∗ ∼= HomD(Aop ∼= HomD(Aop (Y ⋆, A1)) 2 , RHomAop 2 , A1) 1 A Y ∗ 1 )(Y ⋆ ⊗L 1 )(RHomA(Y2, Y ⋆), A1) = 0. Thus Y ∈ Y ∗⊥ in D(Aop). 2 A X ∼= Y ∗∗ 2 ⊗L 2 )⊥, we have Y2 ⊗L 2 )⊥ ⊆ D(Aop), since A(Y ∗ Step 4. For any X ∈ Y ⊥ ∩ (Y ∗ A X ∼= RHomAop(Y ∗ 2 ) is compact and X ∈ (Y ∗ 2 , X) = 0. It follows from Theorem 1 by taking B = Aop that there exists a triangle Y ∗ A2 Y2⊗L 2 ⊗L A X → X → Y ⊗L A X in A X, X) ∼= D(Aop). Furthermore, HomD(Aop)(X, X) ∼= HomD(Aop)(Y ⊗L A X, RHomAop(Y, X)) = 0, since X ∈ Y ⊥. Hence, Y ⊥ ∩ HomD(Aop (Y ∗ A1 Y ⋆ → in D(Aop⊗A), further a triangle Y ∗ A X → in D(Aop). Therefore, X ∼= Y ⊗L A2 Y2 → A → Y ⊗L A1 Y ⋆ ⊗L 2 )⊥ = {0}. Step 5. By the Steps 1 to 4 above, we have shown that AY and AY ∗ 2 satisfy all conditions in Proposition 1 for opposite algebras. Analogous to Jørgensen's construction (ref. [20, Theorem 3.3 and Remark 3.4]), we can obtain a recollement (D(Aop 2 ), i∗, i∗ = i!, i!, j!, j! = j∗, j∗) such that 1 ), D(Aop), D(Aop 2 ⊗L A1 Y ⋆ ⊗L 1 )(Y ⋆ ⊗L A1 Y ⋆ ⊗L i∗ = i! = Y ⊗L i! = RHomAop(Y, −), A1 −, A2 −, 2 ⊗L j! = Y ∗ j! = j∗ = RHomAop(Y ∗ j∗ = RHomAop (Y2, −). 2 2 , −), A − is a left adjoint of Y ⊗L A1 − ∼= RHomAop 1 2 define a standard recollement of D(Aop) relative to D(Aop Clearly, Y ⋆ ⊗L Y ⋆ and Y ∗ D(Aop 2 ). (Y ⋆, −). Thus 1 ) and Remark 2. One referee showed me an elegant proof of Theorem 2 by using the correspondence between the smashing subcategories of D(A) and the idempotent ideals of the category Dc(A) of compact objects of D(A) (ref. [28, 33]). Here, we just provide a relatively elementary proof. 5 Recollements and smoothness In this section, we shall apply the results obtained in sections 3 and 4 to study the relation between recollements of derived categories of algebras and smoothness of algebras. For this, we need to know the relation between rec- ollements of derived categories of algebras and global dimensions of algebras. Recently, the following result is proved: 9 Proposition 4. (Angeleri Hugel-Konig-Liu-Yang [2]) Let A1, A and A2 be algebras, and D(A) admit a recollement relative to D(A1) and D(A2). Then A is of finite global dimension if and only if so are A1 and A2. Let A be an algebra and Ae := Aop ⊗ A its enveloping algebra. The Hochschild dimension dimA of A is the projective dimension of A as a left or right Ae-module. The Hochschild dimensions of algebras were studied very early [8]. An algebra A is of Hochschild dimension 0 if and only if Ae is semisimple [8, Theorem 7.9]. In case A is finitely generated, A is of Hochschild dimension 0 if and only if A is separable [8, Theorem 7.10]. The algebras of Hochschild dimension ≤ 1 are called quasi-free or formally smooth [13, 26]. An algebra A is said to be smooth if it has finite Hochschild dimen- sion, i.e., the projective dimension of A as Ae-module is finite (ref. [40]), or equivalently, A is isomorphic to an object in K b(ProjAe), the homotopy cat- egory of bounded complexes of projective Ae-modules. It is well-known that A is smooth if and only if gl.dimAe < ∞ where gl.dimAe denotes the global dimension of the algebra Ae. Indeed, this follows from the lemma below (ref. [8, Chap. IX, Proposition 7.5, 7.6] and [14, Proposition 2]). Lemma 4. Let A and B be algebras. Then the following assertions hold true: (1) gl.dimA ≤ dimA ≤ gl.dimAe, (2) gl.dimA ⊗ B ≤ gl.dimA + dimB. Remark 3. (1) Sometimes gl.dimA < ∞ ⇔ gl.dimAe < ∞: Let A be either a com- mutative Noetherian algebra over a perfect field k, or a finite-dimensional k-algebra such that the factor algebra A/J of A modulo its Jacobson radical J is separable. Then gl.dimA < ∞ if and only if gl.dimAe < ∞ (ref. [19, Theorem 2.1] and [3, Theorem 16]). (2) In general gl.dimA < ∞ ; gl.dimAe < ∞: Let A be a finite insep- arable field extension of an imperfect field k. Then gl.dimA = 0. However, gl.dimAe = ∞, since A ⊗k A is not semisimple (ref. [4, Page 65, Remark]). Let the algebras A and B be derived equivalent. Then by [37, Proposition 9.1] and [38, Theorem 2.1] we know Ae and Be are derived equivalent. Thus gl.dimAe < ∞ if and only if gl.dimBe < ∞. Hence, A is smooth if and only if so is B, i.e., the smoothness of algebras is invariant under derived equivalences. More general, we have the following result: 10 Theorem 3. Let A1, A and A2 be algebras, and D(A) admit a recollement relative to D(A1) and D(A2). Then A is smooth if and only if so are A1 and A2. Proof. By Proposition 3 and Theorem 1, we have a recollement of D(Aop⊗A) relative to D(Aop ⊗ A1) and D(Aop ⊗ A2). It follows from Proposition 4 that gl.dimAe < ∞ if and only if gl.dimAop ⊗ Ai < ∞ for all i = 1, 2. By Proposition 3, Theorem 2 and Theorem 1, we have a recollement of D(Aop ⊗ Ai) relative to D(Aop 2 ⊗ Ai). It follows from Proposition 4 that for i = 1, 2, gl.dimAop ⊗ Ai < ∞ if and only if gl.dimAop j ⊗ Ai < ∞ for all j = 1, 2. Therefore, gl.dimAe < ∞ if and only if gl.dimAe i < ∞ for all i = 1, 2, by Lemma 4. 1 ⊗ Ai) and D(Aop Theorem 3 can be applied to judge the smoothness of some algebras or construct some smooth algebras. For instance, when applied to triangular matrix algebras, we have the following result: Corollary 1. Let A1 and A2 be algebras, M an A2-A1-bimodule, and A = 0 M A2 (cid:3). Then A is smooth if and only if so are A1 and A2. (cid:2) A1 Proof. Note that X1 := (cid:2) 1A1 0 (cid:3)A and X2 := (cid:2) 0 ditions in Proposition 1. Thus there is a recollement of D(A) relative to D(A1) and D(A2) (ref. [23, Corollary 15]). Now the corollary follows from Theorem 3. 0 1A2 (cid:3)A satisfy all con- 0 0 0 0 V [27]). Let A be the infinite Kronecker algebra (cid:20) k Remark 4. An algebra A is said to be homologically smooth if A is com- pact in D(Ae), i.e., A is isomorphic in D(Ae) to an object in K b(projAe) (ref. k (cid:21), where V is an infinite-dimensional k-vector space. Choose X2 to be the simple projective A-module and X1 the other simple A-module. Then X1 and X2 satisfy all conditions in Proposition 1. Thus D(A) admits a recollement relative to D(k) and D(k) (ref. [23, Example 9]). By Corollary 1, we know the infi- nite Kronecker algebra is smooth. However, it is not homologically smooth, because the finitely generated projective Ae-module resolution of A would induce a finitely generated projective A-module resolution of the nonprojec- tive simple A-module. Hence, Corollary 1 and Theorem 3 are not correct for homological smoothness. 6 Recollements and Hochschild cohomology In this section, we shall apply the results obtained in sections 3 and 4 to observe the relations between recollements of derived categories of algebras 11 and Hochschild cohomology of algebras. Note that the relation between recollements of derived categories of algebras and Hochschild homology of algebras had been clarified by Keller in [22]. Recall that the n-th Hochschild homology of an algebra A is HHn(A) := TorAe Ae A). In D(k) the complex A ⊗L Ae A is isomorphic to the Hochschild complex of A. The following result is due to Keller, which is a corollary of [22, Theorem 3.1] (ref. [22, Remarks 3.2 (a)]) and can be also proved by using Theorem 1. n (A, A) ∼= H −n(A ⊗L Proposition 5. (Keller [22]) Let A, A1 and A2 be algebras, and D(A) admit a recollement relative to D(A1) and D(A2). Then there is a triangle in D(k): A2 ⊗L Ae 2 A2 → A ⊗L Ae A → A1 ⊗L Ae 1 A1 → . From the triangle in Proposition 5, by taking cohomologies, we can obtain a long exact sequence on the Hochschild homologies of the algebras: Corollary 2. (Keller [22]) Let A, A1 and A2 be algebras, and D(A) admit a recollement relative to D(A1) and D(A2). Then there is a long exact sequence on the Hochschild homologies of these algebras · · · → HHn+1(A1) → HHn(A2) → HHn(A) → HHn(A1) → · · · . Now we consider Hochschild cohomology. Recall that the n-th Hochschild Ae(A, A) ∼= H n(RHomAe(A, A)). cohomology of an algebra A is HH n(A) := Extn Note that in D(k) the complex RHomAe(A, A) is isomorphic to the Hochschild cochain complex or Hochschild cocomplex of A. From a recollement of de- rived categories of algebras, we shall obtain three triangles on Hochschild cocomplexes of these algebras, which can induce three long exact sequences on their Hochschild cohomologies. The following lemma is essentially due to Konig and Nagase (cf. [25, Lemma 2.1]). Lemma 5. Let A be an algebra and X u→ Y v→ Z → a triangle in D(A) such that RHomA(X, Z) = 0 in D(k). Then there are three triangles in D(k): (1) RHomA(Y, X) → RHomA(Y, Y ) (2) RHomA(Z, Y ) → RHomA(Y, Y ) (3) RHomA(Z, X) → RHomA(Y, Y ) φ → RHomA(Z, Z) → , ψ → RHomA(X, X) → , ϕ → RHomA(X, X) ⊕ RHomA(Z, Z) → . Moreover, φ (resp. ψ, ϕ) induces a homomorphism of graded rings ¯φ (resp. ¯ψ, ¯ϕ) between the corresponding cohomology rings. 12 Proof. Applying the bifunctor RHomA(−, −) to the triangle X u→ Y v→ Z →, we have the following commutative diagram: RHomA(X[1], Z[−1]) → RHomA(X[1], X) → RHomA(X[1], Y ) → RHomA(X[1], Z) ↓ ↓ ↓ ↓ RHomA(Z, Z[−1]) → RHomA(Z, X) → RHomA(Z, Y ) → RHomA(Z, Z) ↓ ↓ ↓ ↓ RHomA(Y, Z[−1]) → RHomA(Y, X) → RHomA(Y, Y ) → RHomA(Y, Z) ↓ ↓ ↓ ↓ RHomA(X, Z[−1]) → RHomA(X, X) → RHomA(X, Y ) → RHomA(X, Z) in which the four corners are zero by the assumption RHomA(X, Z) = 0 in D(k). It follows two triangles (1) and (2). By Octahedral axiom, we have the following commutative diagram: RHomA(Z, Z[−1]) = RHomA(Z, Z[−1]) ↓ ↓ RHomA(Z, X) → RHomA(Y, X) → RHomA(X, X) → RHomA(Z[−1], X) k ↓ ↓ k RHomA(Z, X) → RHomA(Y, Y ) → RHomA(X, X) ⊕ RHomA(Z, Z) → RHomA(Z[−1], X) ↓ ↓ RHomA(Z, Z) = RHomA(Z, Z) where the morphism RHomA(Z, Z[−1]) → RHomA(X, X) is zero. It follows the triangle (3). For the last statement, it is enough to note that φ induces a map ¯φ : ⊕n∈ZHomD(A)(Y, Y [n]) → ⊕n∈ZHomD(A)(Z, Z[n]) sending fn ∈ HomD(A)(Y, Y [n]) to the unique morphism ¯φ(fn) ∈ HomD(A)(Z, Z[n]) satisfying ¯φ(fn) ◦ v = v[n] ◦ fn, i.e., the following diagram in D(A) is commutative: Y fn ↓ Y [n] v→ Z ↓ ¯φ(fn) v[n] → Z[n], which is clearly a homomorphism of graded rings. Similar for ¯ψ and ¯ϕ. The main result in this section is the following: Theorem 4. Let A1, A and A2 be algebras, and (D(A1), D(A), D(A2), i∗, i∗ = i!, i!, j!, j! = j∗, j∗) a standard recollement given by Y ∈ D(Aop ⊗ A1) and 13 Y2 ∈ D(Aop 2 ⊗ A). Then there are three triangles in D(k): (1) RHomAe(A, RHomA(Y2, A) ⊗L φ → RHomAe → RHomAe(A, A) A2 Y2) 1(A1, A1) → , (2) RHomAe(RHomA1(Y, Y ), A) → RHomAe(A, A) (A2, A2) → , (3) RHomAe(RHomA1(Y, Y ), RHomA(Y2, A) ⊗L 2 ψ → RHomAe A2 Y2) → RHomAe(A, A) ϕ → RHomAe (A1, A1) ⊕ RHomAe 2 (A2, A2) → . 1 Moreover, φ (resp. ψ, ϕ) induces a homomorphism of graded rings ¯φ (resp. ¯ψ, ¯ϕ) between the corresponding Hochschild cohomology rings. Proof. By Theorem 1, we have a recollement (D(Aop ⊗ A1), D(Aop ⊗ A), D(Aop ⊗ A2), I ∗, I∗ = I!, I !, J!, J ! = J ∗, J∗) such that A Y, I ∗ ∼= − ⊗L I∗ = I! I ! ∼= RHomA(RHomA1(Y, A1), −), ∼= RHomA1(Y, −), ∼= − ⊗L A2 Y2, J! J ! = J ∗ ∼= RHomA(Y2, −), J∗ ∼= RHomA2(RHomA(Y2, A), −). Thus we obtain a triangle J!J !A → A → I∗I ∗A → in D(Aop ⊗ A). Note that RHomAe(J!J !A, I∗I ∗A) = 0 due to the recollement. By Lemma 5, we have three triangles in D(k): (1) RHomAe(A, J!J !A) → RHomAe(A, A) → RHomAe(I∗I ∗A, I∗I ∗A) → , (2) RHomAe(I∗I ∗A, A) → RHomAe(A, A) → RHomAe(J!J !A, J!J !A) → , (3) RHomAe(I∗I ∗A, J!J !A) → RHomAe(A, A) → RHomAe(J!J !A, J!J !A)⊕ RHomAe(I∗I ∗A, I∗I ∗A) → . By Lemma 3 and Theorem 1, we have RHomAe(I∗I ∗A, I∗I ∗A) ∼= RHomAop⊗A1(I ∗A, I ∗A) ∼= RHomAop⊗A1(Y, Y ) ∼= RHomAop ∼= RHomAe 1 1 ⊗A(RHomA1(Y, A1), RHomA1(Y, A1)) (A1, A1) and RHomAe(J!J !A, J!J !A) ∼= RHomAop⊗A2(J !A, J !A) ∼= RHomAop⊗A2(RHomA(Y2, A), RHomA(Y2, A)) ∼= RHomAop ∼= RHomAe 2 ⊗A(Y2, Y2) 2(A2, A2), 14 where the last steps follow from the full embeddings I∗ and J! in Theorem 1 by taking B = Aop respectively. Thus there are three triangles in D(k) as required. 1 and Aop 2 The last statement follows from the isomorphisms above and Lemma 5. From the three triangles in Theorem 4, by taking cohomologies, we can obtain three long exact sequences on the Hochschild cohomologies of the algebras: Corollary 3. Let A1, A and A2 be algebras, and (D(A1), D(A), D(A2), i∗, i∗ = i!, i!, j!, j! = j∗, j∗) a standard recollement given by Y ∈ D(Aop ⊗ A1) and Y2 ∈ D(Aop 2 ⊗ A). Then there are three long exact sequences: (1) (2) (3) → HH n(A) · · · → HomD(Ae)(A, RHomA(Y2, A) ⊗L φn→ HH n(A1) → · · · , · · · → HomD(Ae)(RHomA1(Y, Y ), A[n]) ψn→ HH n(A2) → · · · , → HH n(A) A2 Y2[n]) · · · → HomD(Ae)(RHomA1(Y, Y ), RHomA(Y2, A) ⊗L A2 Y2[n]) → HH n(A) ϕn→ HH n(A1) ⊕ HH n(A2) → · · · . Moreover, ⊕n∈Nφn (resp. ⊕n∈Nψn, ⊕n∈Nϕn) is a homomorphism of graded rings between the corresponding Hochschild cohomology rings. Applying Corollary 3 to Example 1 by taking A1 = A/AeA, A2 = eAe, Y2 = eA and Y = A/AeA, we can obtain the following result due to Konig and Nagase: Corollary 4. (Konig-Nagase [25]) Let A be an algebra, e an idempotent of A and AeA a stratifying ideal of A. Then there are three long exact sequences: (1) (2) (3) · · · → Extn · · · → Extn · · · → Extn Ae(A, AeA) → HH n(A) Ae(A/AeA, A) → HH n(A) Ae(A/AeA, AeA) → HH n(A) ψn→ HH n(A/AeA) → · · · , φn→ HH n(eAe) → · · · , ϕn→ HH n(A/AeA) ⊕ HH n(eAe) → · · · . Moreover, ⊕n∈Nφn (resp. ⊕n∈Nψn, ⊕n∈Nϕn) is a homomorphism of graded rings between the corresponding Hochschild cohomology rings. ACKNOWLEDGMENT. I thank Dong Yang for pointing out an error in the original version and some helpful discussions. I am indebted to Steffen Konig and Dong Yang for 15 letting me know their recent work [2]. I am grateful to both referees for their numerous very valuable suggestions which improve some original results especially the present Theorem 2 and Theorem 3 and make the paper much readable. I am sponsored by Project 10731070 and 11171325 NSFC. References [1] L. Angeleri Hugel, S. Konig and Q.H. Liu, Recollements and tilting objects, J. Pure Appl. Algebra 215 (2011), 420 -- 438. [2] L. Angeleri Hugel, S. Konig, Q.H. Liu and D. Yang, Derived simple algebras and restrictions of recollements of derived module categories, Preprint. [3] M. Auslander, On the dimension of modules and algebras, III : Global dimension, Nagoya Math. J. 9 (1955), 67 -- 77. [4] M. Auslander, On the dimension of modules and algebras, VI : Comparison of global and algebra dimension, Nagoya Math. J. 11 (1957), 61 -- 65. [5] A.A. Beilinson, J. Bernstein and P. Deligne, Faisceaux pervers, Ast´erique 100 (1982). [6] K. Bruning and B. Huber, Realising smashing localisations as morphisms of DG algebras, Appl. Categor. Struct. 16 (2008), 669 -- 687. [7] B. Bendiffalah and D. Guin, Cohomologie de l'alg`ebre triangulaire et applications, J. Algebra 282 (2004), 513 -- 537. [8] H. Cartan and S. Eilenberg, Homological algebra, Princeton University Press, 1956. [9] C. Cibils, Tensor Hochschild homology and cohomology, in : Interactions between ring theory and representations of algebras, Lecture Notes in Applied Mathematics, Vol. 210, Dekker, New York, 2000, pp. 35 -- 51. [10] C. Cibils, E. Marcos, M.J. Redondo and A. Solotar, Cohomology of split algebras and trivial extensions, Glasgow Math. J. 45 (2003), 21 -- 40. [11] E. Cline, B. Parshall and L. Scott, Finite-dimensional algebras and highest weight categories, J. Reine Angew. Math. 391 (1988), 85 -- 99. [12] E. Cline, B. Parshall and L. Scott, Stratifying endomorphism algebras, Mem. Amer. Math. Soc. 591 (1996), 1 -- 119. [13] J. Cuntz and D. Quillen, Algebra extensions and nonsingularity, J. Amer. Math. Soc. 8 (1995), 251 -- 289. [14] S. Eilenberg, A. Rosenberg and D. Zelinsky, On the dimension of modules and alge- bras, VIII : Dimension of tensor products, Nagoya Math. J. 12 (1957), 71 -- 93. [15] E.L. Green, E.N. Marcos and N. Snashall, The Hochschild cohomology ring of a one point extension, Comm. Algebra 31 (2003), 357 -- 379. [16] E.L. Green and Ø. Solberg, Hochschild cohomology rings and triangular rings, in : Representations of algebras, Vol. II, Beijing Normal University Press, Beijing, 2002, pp. 192 -- 200. 16 [17] D. Happel, Hochschild cohomology of finite-dimensional algebras, in : S´eminaire d'Alg`ebre Paul Dubreil et Marie-Paul Malliavin, 39`eme Ann´ee (Paris, 1987/1988), Lecture Notes in Math., Vol. 1404, Springer, Berlin, 1989, pp. 108 -- 126. [18] D. Happel, Reduction techniques for homological conjectures, Tsukuba J. Math. 17 (1993), 115 -- 130. [19] G. Hochschild, B. Kostant and A. Rosenberg, Differential forms on regular affine algebras, Trans. Amer. Math. Soc. 102 (1962), 383 -- 408. [20] P. Jørgensen, Recollement for differential graded algebras, J. Algebra 299 (2006), 589 -- 601. [21] B. Keller, Deriving DG categories, Ann. Sci. ´Ecole Norm. Sup. 27 (1994), 63 -- 102. [22] B. Keller, Invariance and localization for cyclic homology of DG algebras, J. Pure Appl. Algebra 123 (1998), 223 -- 273. [23] S. Konig, Tilting complexes, perpendicular categories and recollements of derived module categories of rings, J. Pure Appl. Algebra 73 (1991), 211 -- 232. [24] S. Konig and A. Zimmermann, Derived equivalences for group rings, Lecture Notes in Math. 1685, Springer-Verlag, 1998. [25] S. Konig and H. Nagase, Hochschild cohomology and stratifying ideals, J. Pure Appl. Algebra 213 (2009), 886 -- 891. [26] M. Kontsevich and A. Rosenberg, Noncommutative smooth spaces, The Gelfand Mathematical Seminars, 1996 -- 1999, 85 -- 108, Birkhauser Boston, 2000. [27] M. Kontsevich and Y. Soibelman, Notes on A∞-algebras, A∞-categories and non- commutative geometry. I, arXiv:math.RA/0606241. [28] H. Krause, Smashing subcategories and the telescope conjecture -- an algebraic ap- proach, Invent. Math. 139 (2000), 99 -- 133. [29] H. Krause, Localization theory for triangulated categories, arXiv:0806.1324 [math.CT]. [30] S. Michelena and M.I. Platzeck, Hochschild cohomology of triangular matrix algebras, J. Algebra 233 (2000), 502 -- 525. [31] J. Miyachi, Localization of triangulated categories and derived categories, J. Algebra 141 (1991), 463 -- 483. [32] J. Miyachi, Recollement and tilting complexes, J. Pure Appl. Algebra 183 (2003), 245 -- 273. [33] P. Nicol´as and M. Saorin, Parametrizing recollement data for triangulated categories, J. Algebra 322 (2009), 1220 -- 1250. [34] P. Nicol´as and M. Saorin, Lifting and restricting recollement data, Appl. Categor. Struct. 19 (2011), 557 -- 596. [35] B. Parshall and L. Scott, Derived categories, quasi-hereditary algebras and algebraic groups, Carlton Univ. Math. Notes 3 (1988), 1 -- 104. 17 [36] J.A. de la Pena and C.C. Xi, Hochschild cohomology of algebras with homological ideals, Tsukuba J. Math. 30 (2006), 61 -- 79. [37] J. Rickard, Morita theory for derived categories, J. London Math. Soc. 39 (1989), 436 -- 456. [38] J. Rickard, Derived equivalences as derived functors, J. London Math. Soc. 43 (1991), 37 -- 48. [39] M. Su´arez-Alvarez, Applications of the change-of -rings spectral sequence to the com- putation of Hochschild cohomology, arXiv:0707.3210. [40] M. Van den Bergh, A relation between Hochschild homology and cohomology for Gorenstein rings, Proc. Amer. Math. Soc. 126 (1998), 1345 -- 1348. Erratum, Proc. Amer. Math. Soc. 130 (2002), 2809 -- 2810. [41] A. Wiedemann, On stratifications of derived module categories, Canad. Math. Bull. 34 (1991), 275 -- 280. 18
1009.0830	1	1009	2010-09-04T12:07:43	More sublattices of the lattice of local clones	[ "math.RA", "math.LO" ]	We investigate the complexity of the lattice of local clones over a countably infinite base set. In particular, we prove that this lattice contains all algebraic lattices with at most countably many compact elements as complete sublattices, but that the class of lattices embeddable into the local clone lattice is strictly larger than that: For example, the lattice $M_{2^\omega}$ is a sublattice of the local clone lattice.	math.RA	math	MORE SUBLATTICES OF THE LATTICE OF LOCAL CLONES MICHAEL PINSKER Abstract. We investigate the complexity of the lattice of local clones over a countably infinite base set. In particular, we prove that this lattice contains all algebraic lattices with at most countably many compact elements as complete sublattices, but that the class of lattices embeddable into the local clone lattice is strictly larger than that: For example, the lattice M2ω is a sublattice of the local clone lattice. 1. Local clones 1.1. Defining local clones. Fix a countably infinite base set X, and denote for all n ≥ 1 the set X X n = {f : X n → X} of n-ary operations on X by O (n). Then O (n) is the set of all finitary operations on X. A clone C is the union O := Sn≥1 a subset of O satisfying the following two properties: • C contains all projections, i.e., for all 1 ≤ k ≤ n the operation πn k ∈ O (n) defined by πn k (x1, . . . , xn) = xk, and • C is closed under composition, i.e., whenever f ∈ C is n-ary and g1, . . . , gn ∈ C are m-ary, then the operation f (g1, . . . , gn) ∈ O (m) defined by (x1, . . . , xm) 7→ f (g1(x1, . . . , xm), . . . , gn(x1, . . . , xm)) also is an element of C . Since arbitrary intersections of clones are again clones, the set of all clones on X, equipped with the order of inclusion, forms a complete lattice Cl(X). In this paper, we are not interested in all clones of Cl(X), but only in clones which satisfy an additional topological closure property: Equip X with the discrete topology, and O (n) = X X n with the corresponding product topology (Tychonoff topology), for every n ≥ 1. A clone C is called locally closed or just local iff each of its n-ary fragments C ∩ O (n) is a closed subset of O (n). Equivalently, a clone C is local iff it satisfies the following interpolation property: For all n ≥ 1 and all g ∈ O (n), if for all finite A ⊆ X n there exists an n-ary f ∈ C which agrees with g on A, then g ∈ C . Again, taking the set of all local clones on X, and ordering them according to set-theoretical inclusion, one obtains a complete lattice, which we denote by Clloc(X): This is because intersections of clones are clones, and because arbitrary intersections of closed sets are closed. We are interested in the structure of Clloc(X), in particular in how complicated it is as a lattice. 2000 Mathematics Subject Classification. Primary 08A40; secondary 08A05. Key words and phrases. clone; local closure; algebraic lattice; embedding. The author is grateful for support through Erwin Schrodinger Fellowship J2742-N18 of the Austrian Science Fund (FWF). 1 2 MICHAEL PINSKER Before we start our investigations, we give an alternative description of local clones which will be useful. Let f ∈ O (n) and let ρ ⊆ X m be a relation. We say that f preserves ρ iff f (r1, . . . , rn) ∈ ρ whenever r1, . . . , rn ∈ ρ; here, the m-tuple f (r1, . . . , rn) is calculated componentwise, i.e., it is the m-tuple whose i-th component is obtained by applying f to the n-tuple consisting of the i-th components of the tuples r1, . . . , rn. For a set of relations R, we write Pol(R) for the set of those operations in O which preserve all ρ ∈ R. The operations in Pol(R) are called polymorphisms of R. The following is due to [Rom77], see also the textbook [Sze86]. Theorem 1. Pol(R) is a local clone for all sets of relations R. Moreover, every local clone is of this form. Similarly, for an operation f ∈ O (n) and a relation ρ ⊆ X m, we say that ρ is invariant under f iff f preserves ρ. Given a set of operations F ⊆ O, we write Inv(F ) for the set of all relations which are invariant under all f ∈ F . Since arbitrary intersections of local clones are local clones again, the mapping on the power set of O which assigns to every set of operations F ⊆ O the smallest local clone hF iloc containing F is a closure operator, the closed elements of which are exactly the local clones. Using the operators Pol and Inv which connect operations and relations, one obtains the following well-known alternative for describing this operator (confer [Rom77] or [Sze86]). Theorem 2. Let F ⊆ O. Then hF iloc = Pol Inv(F ). As already mentioned, the aim of this paper is to investigate the structure of the local clone lattice. So far, this lattice has been studied only sporadically, e.g. in [RS82], [RS84]. There, the emphasis was put on finding local completeness criteria for sets of operations F ⊆ O, i.e., on how to decide whether or not hF iloc = O. Only very recently has the importance of the local clone lattice to problems from model theory and theoretical computer science been revealed: 1.2. The use of local clones. Let Γ = (X, R) be a countably infinite structure; that is, X is a countably infinite base set and R is a set of finitary relations on X. Consider the expansion Γ′ of Γ by all relations which are first-order definable from Γ. More precisely, Γ′ has X as its base set and its relations R ′ consist of all finitary relations which can be defined from relations in R using first-order formulas. A reduct of Γ′ is a structure ∆ = (X, D), where D ⊆ R ′. We also call ∆ a reduct of Γ, which essentially amounts to saying that we expect our structure Γ to be closed under first-order definitions. Clearly, the set of reducts of Γ is in one-to-one correspondence with the power set of R ′, and therefore not of much interest as a partial order. However, it might be more reasonable to consider such reducts up to, say, first-order interdefinability. That is, we may consider two reducts ∆1 = (X, D1) and ∆2 = (X, D2) the same iff their first-order expansions coincide, or equivalently iff all relations of ∆1 are first-order definable in ∆2 and vice-versa. In 1976, P. J. Cameron [Cam76] showed that there are exactly five reducts of (Q, <) up to first-order interdefinability. Recently, M. Junker and M. Ziegler gave a new proof of this fact, and established that (Q, <, a), the expansion of (Q, <) by a constant a, has 116 reducts [JZ08]. S. Thomas proved that the first-order theory of the random graph also has exactly five reducts, up to first-order interdefinabil- ity [Tho91]. MORE SUBLATTICES OF THE LATTICE OF LOCAL CLONES 3 These examples have in common that the structures under consideration are ω-categorical, i.e., their first-order theories determine their countable models up to isomorphism. This is no coincidence: For, given an ω-categorical structure Γ, its reducts up to first-order interdefinability are in one-to-one correspondence with the locally closed permutation groups which contain the automorphism group of Γ, providing a tool for describing such reducts (confer [Cam90]). A natural variant of these concepts is to consider reducts up to primitive positive interdefinability. That is, we consider two reducts ∆1, ∆2 of Γ the same iff their expansions by all relations which are definable from each of the structures by prim- itive positive formulas coincide. (A first-order formula is called primitive positive iff it is of the form ∃x(φ1 ∧ · · · ∧ φl) for atomic formulas φ1, . . . , φl.) It turns out that for ω-categorical structures Γ, the local clones containing all automorphisms of Γ are in one-to-one correspondence with those reducts of the first-order expansion of Γ which are closed under primitive positive definitions. This recent connection, which relies on a theorem from [BN06], has already been exploited in [BCP], where the reducts of (X, =), the structure whose only relation is the equality, have been classified using this method. It turns out that despite the simplicity of this struc- ture, the lattice of its reducts is quite complex, and in particular has the size of the continuum. We mention in passing that distinguishing relational structures up to primitive positive interdefinability, and therefore understanding the structure of Clloc(X), has recently gained significant importance in theoretical computer science, more precisely for what is known as the Constraint Satisfaction Problem; see [BKJ05], [Bod04], or also the introduction in [BCP]. 1.3. Main results of this paper. In this paper, which is the journal version of a shorter article which appeared in the conference proceedings of the ROGICS'08 conference [Pin08], we are concerned with Problem V of the survey paper [GP08], which asks which sublattices Clloc(X) contains. We prove that every algebraic lat- tice with countably many compact elements is a complete sublattice of Clloc(X) (Theorem 5). We also show that Clloc(X) is, with respect to size, not too far from such lattices as it as a join-preserving embedding into an algebraic lattice with countably many compacts (Theorem 8). All this is done in Section 2. In Section 3, we prove that the lattice M2ω embeds into Clloc(X) (Theorem 14), thereby pro- viding the first example of a sublattice of Clloc(X) which does not embed into any algebraic lattice with countably many compacts. We also pose a series of open problems, one in Section 2 (Problem 11), and three more in an own open problems section, Section 4 (Problems 20, 21, 22). 1.4. Acknowledgement. I am grateful to Marina Semenova for her critical re- marks which forced me to find correct proofs of the theorems. 2. Algebraic lattices and the local clone lattice We start our investigations by observing that whereas the lattice Cl(X) of all (not necessarily local) clones over X is algebraic, it has been discovered recently in [GP08] that the local clone lattice Clloc(X) is far from being so; in fact, the following has been shown. Proposition 3. The only compact element in the lattice Clloc(X) is the clone of projections. 4 MICHAEL PINSKER We remark that it follows from the proof of the preceding proposition given in [GP08] that Clloc(X) is not even upper continuous. We now turn to sublattices of Clloc(X). The following is a first easy observation which tells us that there is practically no hope that Clloc(X) can ever be fully described, since it is believed that already the clone lattice over a three-element set is too complex to be completely understood. Proposition 4. Let Cl(A) be the lattice of all clones over a finite set A. Then Cl(A) is an isomorphic copy of an interval of Clloc(X). Proof. Assume without loss of generality that A ⊆ X. Assign to every operation f (x1, . . . , xn) on A a set of n-ary operations Sf ⊆ O (n) on X as follows: An operation g ∈ O (n) is an element of Sf iff g agrees with f on An. Let σ map every clone C on the base set A to the set S{Sf : f ∈ C }. Then the following hold: (1) For every clone C on A, σ(C ) is a local clone on X. (2) σ maps the clone of all operations on A to Pol({A}). (3) All local clones (in fact: all clones) which contain σ({f : f is a projection on A}) (i.e., which contain the local clone on X which, via σ, corresponds to the clone of projections on A) and which are contained in Pol({A}) are of the form σ(C ) for some clone C on A. (4) σ is one-to-one, and both σ and its inverse are order preserving. (1) and (2) are easy verifications and left to the reader. To see (3), let D be any clone in the mentioned interval, and denote by C the set of all restrictions of operations in D to appropriate powers of A. Since D ⊆ Pol({A}), all such restrictions are operations on A, and since D is closed under composition and contains all projections, so does C . Thus, C is a clone on A. We claim D = σ(C ). By the definitions of C and σ, we have that σ(C ) clearly contains D. To see the less obvious inclusion, let f ∈ σ(C ) be arbitrary, say of arity m. The restriction of f to Am is an element of C , hence there exists an m-ary f ′ ∈ D which has the same restriction to Am as f . Define s(x1, . . . , xm, y) ∈ O (m+1) by s(x1, . . . , xm, y) =(y f (x1, . . . , xm) , if(x1, . . . , xm) ∈ Am , otherwise. Since s behaves on Am+1 like the projection onto the last coordinate, and since D contains σ({f : f is a projection on A}), we infer s ∈ D. But f (x1, . . . , xm) = s(x1, . . . , xm, f ′(x1, . . . , xm)), proving f ∈ D. (4) is an immediate consequence of the definitions. (cid:3) It is known that all countable products of finite lattices embed into the clone lattice over a four-element set [Bul94], so by the preceding proposition they also embed into Clloc(X). However, there are quite simple countable lattices which do not embed into the clone lattice over any finite set: The lattice Mω consisting of a countably infinite antichain plus a smallest and a greatest element is an exam- ple [Bul93]. We shall see now that the class of lattices embeddable into Clloc(X) properly contains the class of lattices embeddable into the clone lattice over a finite set. In fact, the structure of Clloc(X) is at least as complicated as the structure of any algebraic lattice with ℵ0 compact elements. Before we prove this, observe that similarly to the local clones, the set of locally closed (that is: topologically closed in the space of all permutations on X) permutation groups on X forms a complete MORE SUBLATTICES OF THE LATTICE OF LOCAL CLONES 5 lattice with respect to inclusion; denote this lattice by Grloc(X). Moreover, the set of locally closed (that is: closed in the space X X , where X is discrete) trans- formation monoids on X forms a complete lattice with respect to inclusion, which we denote by Monloc(X). Note that the elements of Monloc(X) are precisely the objects of the form C ∩ O (1), where C ∈ Clloc(X). Theorem 5. Every algebraic lattice with a countable number of compact elements is a complete sublattice of Clloc(X). In fact, every such lattice is a complete sublattice of Grloc(X) and of Monloc(X). It is clear that the lattice Monloc(X) of locally closed transformation monoids embeds completely into Clloc(X) via the assignment which sends every local monoid to the local clone it generates. Local clones arising in this way will contain only operations which depend on at most one variable: In fact, the operations of such a local clone are exactly the functions of the monoid, with (possibly) some ficti- tious variables added. Therefore, the statement of the theorem about the lattice Monloc(X) implies the statement about Clloc(X). To prove Theorem 5, we cite the following deep theorem from [Tum89]. Theorem 6. Every algebraic lattice with a countable number of compact elements is isomorphic to an interval in the subgroup lattice of a countable group. Proof of Theorem 5. We prove the statement about Grloc(X). Let L be the alge- braic lattice to be embedded into Grloc(X). Let X = (X, +, −, 0) be the group pro- vided by Theorem 6. Let [G1, G2] be the interval in the subgroup lattice of X that L is isomorphic to. We will assign to every group in the interval its Cayley representa- tion as a group of permutations on X: That is, for every a ∈ X, define a unary op- eration fa ∈ O (1) by fa(x) = a+x. Clearly, we have fa(fb(x)) = a+b+x = fa+b(x) for all a, b ∈ X. Define a mapping µ : [G1, G2] → Grloc(X) sending every group H = (H, +, −, 0) in the interval to CH := {fa : a ∈ H}. It is easy to see (and folklore) that the CH are permutation groups; we only have to check that they are locally closed. To see this, let f ∈ O (1) be an element of the topological (local) closure of CH in the full symmetric group on X. We claim f ∈ CH . Indeed, ob- serve that f agrees with some fa ∈ CH on the finite set {0} ⊆ X. Suppose that there is b ∈ X such that f (b) 6= fa(b) = a + b. Then take any fc ∈ CH such that f and fc agree on {0, b}. But then c = fc(0) = f (0) = fa(0) = a, and thus f (b) = fc(b) = fa(b) 6= f (b), an obvious contradiction. Hence, f = fa ∈ CH and we are done. With the explicit description of the CH and given that they are indeed closed permutation groups, a straightforward check shows that µ preserves arbitrary meets and joins. The proof for the embedding into Monloc(X) is identical. By the discussion (cid:3) above the statement about Clloc(X) follows. Since in particular, Clloc(X) contains Mω as a sublattice, and since according to [Bul93], Mω is not a sublattice of the clone lattice over any finite set, we have the following corollary to Theorem 5. Corollary 7. Clloc(X) does not embed into the clone lattice over any finite set. Observe also that Theorem 5 is a strengthening of Proposition 4 in so far as the clone lattice over a finite set is an example of an algebraic lattice with countably 6 MICHAEL PINSKER many compact elements. However, in that proposition we obtain an embedding as an interval, not just as a complete sublattice. What about other lattices, i.e., lattices which are more complicated or larger than algebraic lattices with countably many compact elements? We will now find a restriction on which lattices can be sublattices of Clloc(X). A partial clone of finite operations on X is a set of partial operations of finite domain on X which contains all restrictions of the projections to finite domains and which is closed under composition. A straightforward verification shows that the set of partial clones of finite operations on X forms a complete algebraic lattice Clfpart(X), the compact elements of which are precisely the finitely generated partial clones; in particular, the number of compact elements of Clfpart(X) is countable. Theorem 8. The mapping σ from Clloc(X) into Clfpart(X) which sends every C ∈ Clloc(X) to the partial clone of all restrictions of its operations to finite domains is one-to-one and preserves arbitrary joins. Proof. It is obvious that σ(C ) is a partial clone of finite operations, for all local (in fact: also non-local) clones C . Let C , D ∈ Clloc(X) be distinct. Say without loss of generality that there is an n-ary f ∈ C \ D; then since D is locally closed, there exists some finite set A ⊆ X n such that there is no g ∈ D which agrees with f on A. The restriction of f to A then witnesses that σ(C ) 6= σ(D). We show that σ(C ) ∨ σ(D) = σ(C ∨ D); the proof for arbitrary joins works the same way. It follows directly from the definition of σ that it is order-preserving. Thus, σ(C ∨ D) contains both σ(C ) and σ(D) and hence also their join. Now let f ∈ σ(C ) ∨ σ(D). This means that it is a composition of partial operations in σ(C ) ∪ σ(D). All partial operations used in this composition have extensions to operations in C or D, and if we compose these extensions in the same way as the partial operations, we obtain an operation in C ∨ D which agrees with f on the domain of the latter. Whence, f ∈ σ(C ∨ D). (cid:3) Corollary 9. Clloc(X) embeds as a suborder into the power set of ω. In particular, the size of Clloc(X) is 2ℵ0, and Clloc(X) does not contain any uncountable ascending or descending chains. Proof. The number of partial operations with finite domain on X is countable; therefore, partial clones of finite operations can be considered as subsets of ω, which proves the first statement. In particular, Clloc(X) cannot have more than 2ℵ0 elements, and the fact that all algebraic lattices with at most ℵ0 compact elements embed into Clloc(X) shows that it must contain at least 2ℵ0 elements. The third statement is an immediate consequence of the first one. (cid:3) It also follows from Theorem 8 that for all sets of local clones S ⊆ Clloc(X), there exists a countable S′ ⊆ S such that W S =W S′. It might be interesting to observe at this point that Clfpart(X) is universal for the class of algebraic lattices with countably many compact elements: Denote by Monfpart(X) the lattice of all sets of finite partial unary operations on X which are closed under composition, and which contain all restrictions of the identity. In other words, the elements of Monfpart(X) are precisely the unary fragments of the elements of Clfpart(X). As before with local monoids and local clones, Monfpart(X) MORE SUBLATTICES OF THE LATTICE OF LOCAL CLONES 7 embeds naturally into Clfpart(X) by adding fictitious variables to the partial oper- ations of an element of Monfpart(X). We have: Proposition 10. Every algebraic lattice with a countable number of compact ele- ments is a complete sublattice of Clfpart(X). In fact, every such lattice is a complete sublattice of Monfpart(X). Proof. Let L be the lattice to be embedded, and embed it into Monloc(X) as in the proof of Theorem 5 via the mapping µ of that proof. Then apply the mapping σ from Theorem 8 to the local monoids thus obtained. This obviously gives us a join-preserving mapping from L into Monfpart(X). We claim that σ, restricted to monoids in the image of µ, preserves arbitrary meets: Let CH , CK be two such monoids. Clearly, σ(CH ∩ CK) is contained in σ(CH ) ∩ σ(CK ). Now let p ∈ σ(CH ) ∩ σ(CK ). Then, using the notation from the proof of Theorem 5, there exist a finite set A ⊆ X, a ∈ H and b ∈ K such that p = fa ↾A= fb ↾A. We may assume that A is non-void; pick c ∈ A. We have p(c) = fa(c) = a + c = fb(c) = b + c. Hence, a = b, so fa = fb ∈ CH ∩ CK, and p ∈ σ(CH ∩ CK ). Larger meets work the same way, and thus we have a complete lattice embedding of L into Monfpart(X). (cid:3) Until today, no other restriction to embeddings into Clloc(X) except for Theo- rem 8 is known, and we ask: Problem 11. Does every lattice which has a complete join-embedding into an alge- braic lattice with countably many compacts have a lattice embedding into Clloc(X)? It seems, however, difficult to embed even the simplest lattices which are not covered by Theorem 5 into Clloc(X). In the conference version of this paper, [Pin08], the following problem was posed: Does the lattice M2ω , which consists of an antichain of length 2ω plus a smallest and a largest element, embed into Clloc(X)? We will give an affirmative answer to this problem in Section 3. This is the first example of a lattice which embeds into Clloc(X) but not into any algebraic lattice with countably many compacts (for the latter statement, see the proof of Corollary 13). It is much easier to see the following: Proposition 12. There exist a join-preserving embedding as well as a meet-preserving embedding of M2ω into Clloc(X). Proof. Denote by 0 and 1 the smallest and the largest element of M2ω , respectively, and enumerate the elements of its antichain by (ai)i∈2ω . We first construct a join-preserving embedding. Enumerate the non-empty proper subsets of X by (Ai)i∈2ω . Consider the mapping σ which sends 0 to the clone of projections, 1 to O, and every ai to Pol({Ai}). Now it is well-known (see [RS84]) that for any non-empty proper subset A of X, Pol({A}) is covered by O, i.e. there exist no local (in fact even no global) clones between Pol({A}) and O. Hence, we have that σ(ai)∨σ(aj ) = hPol({Ai})∪Pol({Aj})iloc = O = σ(1) for all i 6= j. Since clearly σ(ai) contains σ(0) for all i ∈ 2ω, the mapping σ indeed preserves joins. To construct a meet-preserving embedding, fix any distinct a, b ∈ X and define for every non-empty subset A of X \ {a, b} an operation fA ∈ O (1) by fA(x) =(a, b, if x ∈ A otherwise. 8 MICHAEL PINSKER Enumerate the non-empty subsets of X \{a, b} by (Bi : i ∈ 2ω). Denote the constant unary operation with value b by cb. Let the embedding σ map 0 to h{cb}iloc, for all i ∈ 2ω map ai to h{fBi}iloc, and let it map 1 to O. One readily checks that σ(ai) = h{fBi}iloc contains only projections and, up to fictitious variables, the operations fBi and cb. Therefore, for i 6= j we have σ(ai)∧σ(aj ) = h{cb}iloc = σ(0). Since clearly σ(ai) ⊆ σ(1) = O for all i ∈ 2ω, we conclude that σ does indeed preserve meets. (cid:3) Simple as the preceding proposition is, it still shows us as a consequence that Theorem 5 is not optimal. Corollary 13. Clloc(X) is not embeddable into any algebraic lattice with countably many compact elements. Proof. It is well-known and easy to check (confer also [CD73]) that any algebraic lattice L with countably many compact elements can be represented as the subal- gebra lattice of an algebra over the base set ω. The meet in the subalgebra lattice L is just the set-theoretical intersection. Now there is certainly no uncountable family of subsets of ω with the property that any two distinct members of this family have the same intersection D; for the union of such a family would have to be uncountable. Consequently, L cannot have M2ω as a meet-subsemilattice. But Clloc(X) has, hence L cannot have Clloc(X) as a sublattice. (cid:3) Observe that this corollary is a strengthening of Corollary 7, since the clone lat- tice over a finite set is an algebraic lattice with countably many compact elements. We conclude this section by remarking that the lattice Cl(X) of all (not necessar- ily local) clones on X is infinitely more complicated than Clloc(X): It contains all algebraic lattices with at most 2ℵ0 compact elements, and in particular all lattices of size continuum (including Clloc(X)), as complete sublattices [Pin07]. 3. How to embed M2ω into Clloc(X) This section is devoted to the proof of the following theorem. Theorem 14. M2ω is isomorphic to a sublattice of Clloc(X). Write X = A ∪B ∪{∞}, with both A and B infinite. Without loss of generality assume B = 2<ω, the set of all finite 0-1-sequences. For every infinite 0-1-sequence α ∈ 2ω, write Bα := {α ↾n: n < ω}. Clearly, the Bα form an almost disjoint family of infinite subsets of B, that is, Bα ∩ Bβ is finite whenever α, β ∈ 2ω are distinct. For all α ∈ 2ω, set Gα ⊆ O (1) to consist of all functions f ∈ O (1) satisfying the following two properties: • f maps {∞} ∪ B to ∞, and • f maps A injectively into Bα. For c, d ∈ B, we write c ⊥ d iff none of the two elements is an initial segment of the other, i.e. iff there is no α such that c, d ∈ Bα. If C, D ⊆ B, we write C ⊥ D iff c ⊥ d for all (c, d) ∈ C × D. For a function σ, write dom(σ) and ran(σ) for the domain and the range of σ, respectively. Now for all functions σ with the properties • dom(σ) = C × D for some finite C, D ⊆ B with C ⊥ D, and • ran(σ) ⊆ B, and • σ is injective, MORE SUBLATTICES OF THE LATTICE OF LOCAL CLONES 9 define an operation mσ ∈ O (2) by mσ(x, y) :=  ∞ y σ(x, y) x , x = ∞ ∨ y = ∞ , x ∈ A ∧ y ∈ B , (x, y) ∈ dom(σ) , otherwise. In words, mσ does the following: If one of its arguments equals ∞, then it returns ∞. If one of the arguments is in A and the other one in B, then it returns the one in B. If both arguments are in B and σ is defined for (x, y), then mσ(x, y) = σ(x, y). Otherwise, mσ returns the first argument. Denote by M the set of all such operations mσ, together with the constant function with value ∞, which we denote by ∞ as well. For all α ∈ 2ω, set Dα := hM ∪ Gαi, which is to denote the normal clone closure of M ∪ Gα, i.e., the set of all operations which can be written as a term of the operations in M ∪ Gα and the projections. Set moreover Cα := hM ∪ Gαiloc (the topological closure of Dα), for all α ∈ 2ω. Finally, set C := hSα∈2ω Cαiloc. α, β ∈ 2ω, which clearly implies our theorem. We now aim at proving hCα ∪ Cβiloc = C and Cα ∩ Cβ = hM iloc, for all distinct Lemma 15. Let m ∈ M , and f, g ∈ Gα for some α ∈ 2ω. Then the following equations hold: m(x, x) = x and m(f (x), x) = m(x, f (x)) = m(f (x), g(x)) = f (x). Proof. We leave the straightforward verification to the reader. (cid:3) Denote the identity operation on X by id. α . Then f ∈ Gα, or f = ∞, or f = id. Lemma 16. Let f ∈ C (1) Proof. We first prove the statement for all f ∈ D (1) α , using induction over terms. The beginning is trivial, so let f = g(t), with g ∈ Gα ∪ {id} and t ∈ D (1) α satisfying the induction hypothesis. There is nothing to show if t is the identity or ∞. If t ∈ Gα, then also f ∈ Gα if g is the identity; otherwise, g ∈ Gα and f = ∞. Now assume f = m(s, t), with m ∈ M and s, t satisfying the induction hypothesis. Then the preceding lemma immediately implies our assertion. Now with this description of D (1) α it is easy to check that D (1) Hence C (1) α = D (1) α , proving the lemma. α is locally closed. (cid:3) Lemma 17. Let t(x1, . . . , xn) ∈ Dα, and assume that t has a representation as a term over Gα ∪ M (without projections!) which uses at least one symbol from Gα. Then either t = ∞, or t(x, . . . , x) ∈ Gα. Proof. We use induction over the complexity of t. The beginning is trivial, so write t = g(s), where g ∈ Gα and s satisfies the induction hypothesis. Assume first that s does not contain any symbol from Gα; then s(x, . . . , x) = x by Lemma 15 and a standard induction, implying t(x, . . . , x) = g(s(x, . . . , x)) = g(x) ∈ Gα. If on the other hand s does contain a symbol from Gα, then the range of s is contained in B∪{∞}, as is easily verified by a straightforward induction using Lemma 15. Hence, g(s) = ∞. Now write t = m(r, s), with m ∈ M and r, s satisfying the induction hypothesis. Then our assertion follows from Lemma 15 and the definition of the operations in M . (cid:3) 10 MICHAEL PINSKER Lemma 18. Let α, β ∈ 2ω be distinct. Then Cα ∩ Cβ = hM iloc. Proof. It suffices to show Cα ∩ Cβ ⊆ hM iloc. To see this, let t(x1, . . . , xn) ∈ Cα ∩ Cβ, and suppose there is a finite set F ⊆ X such that no operation in hM i (normal clone closure) agrees with t on F n. By expanding F , we may assume that F ∩ A > Bα ∩ Bβ (since the latter set is finite). Let s be a term over Gα ∪ M which interpolates t on F n. By our assumption, there appears some function from Gα in s. Then by Lemma 17, s(x, . . . , x) ∈ Gα. Therefore, also t(x, . . . , x) behaves like an operation from Gα on F , and in particular on F ∩ A. Hence, it maps F ∩ A injectively into Bα. Now the same argument shows us that t(x, . . . , x) maps F ∩ A injectively into Bβ, hence it maps F ∩ A injectively into Bα ∩ Bβ, in contradiction with the size of these sets. (cid:3) Lemma 19. Let α, β ∈ 2ω be distinct. Then hCα ∪ Cβiloc = C . Proof. Let γ ∈ 2ω, γ /∈ {α, β} be arbitrary, and consider any h ∈ Gγ. It suffices to show that h ∈ hCα ∪ Cβ iloc. So let F ⊆ X be finite. We have to find an operation in hCα ∪ Cβi which agrees with h on F . Assume for the moment that F ⊆ A. Pick any C ⊆ Bα, D ⊆ Bβ with C = D = F and such that C ⊥ D. Pick f ∈ Gα mapping F onto C, and g ∈ Gβ mapping F onto D. Then (f, g) : X → X 2 maps F injectively into C × D. Thus there exists σ : C × D → Bγ such that σ(f (x), g(x)) = h(x) for all x ∈ F . Now mσ(f (x), g(x)) = σ(f (x), g(x)) = h(x) for all x ∈ F , so mσ(f (x), g(x)) ∈ hCα ∪ Cβi agrees with h on F . For the case where F is not a subset of A, one constructs the interpolation for F ′ := F ∩ A. Then observe that the operation mσ(f (x), g(x)) constructed sends all b ∈ B ∪ {∞} to ∞, thus behaving like h outside A anyway. (cid:3) Lemmas 18 and 19 clearly prove Theorem 14. 4. Open problems Suprisingly, the method of the previous section seems to be hard to generalize: For example, we do not know: Problem 20. Does every lattice of finite height which has cardinality 2ℵ0 embed into Clloc(X)? When proving that all algebraic lattices with countably many compacts embed into Clloc(X), we in fact embedded them into Monloc(X) (Theorem 5). The same could be true with M2ω (which we do not believe, though): Problem 21. Does M2ω embed into Monloc(X)? Does it embed into Grloc(X)? In theory, even the following could hold: Problem 22. Does Clloc(X) embed into Monloc(X)? Does it embed into Grloc(X)? References [BCP] M. Bodirsky, H. Chen, and M. Pinsker. The reducts of equality up to primitive pos- itive interdefinability. Journal of Symbolic Logic. To appear. Preprint available from http://arxiv.org/abs/0810.2270. [BKJ05] A. Bulatov, A. Krokhin, and P. G. Jeavons. Classifying the complexity of constraints using finite algebras. SIAM Journal on Computing, 34:720 -- 742, 2005. [BN06] M. Bodirsky and J. Nesetril. Constraint satisfaction with countable homogeneous tem- plates. Journal of Logic and Computation, 16(3):359 -- 373, 2006. MORE SUBLATTICES OF THE LATTICE OF LOCAL CLONES 11 [Bod04] M. Bodirsky. Constraint satisfaction with infinite domains. Dissertation, Humboldt- Universitat zu Berlin, 2004. [Bul93] A. Bulatov. Identities in lattices of closed classes. Discrete Mathematics and Applica- tions, 3(6):601 -- 609, 1993. [Bul94] A. Bulatov. Finite sublattices in the lattice of clones. Algebra and Logic, 33(5):287 -- 306, 1994. [Cam76] P. J. Cameron. Transitivity of permutation groups on unordered sets. Mathematische Zeitschrift, 148:127 -- 139, 1976. [Cam90] P. J. Cameron. Oligomorphic permutation groups, volume 152 of London Mathematical Society Lecture Note Series. Cambridge University Press, Cambridge, 1990. [CD73] P. Crawley and R. P. Dilworth. Algebraic theory of lattices. Prentice-Hall, 1973. [GP08] M. Goldstern and M. Pinsker. A survey of clones on infinite sets. Algebra Universalis, 59:365 -- 403, 2008. [JZ08] M. Junker and M. Ziegler. The 116 reducts of (Q, <, a). Journal of Symbolic Logic, 73(3):861 -- 884, 2008. [Pin07] M. Pinsker. Algebraic lattices are complete sublattices of the clone lattice on an infinite set. Fundamenta Mathematicae, 195(1):1 -- 10, 2007. [Pin08] M. Pinsker. Sublattices of the lattice of local clones. In Proceedings of the ROGICS'08 conference, pages 80 -- 87, 2008. [Rom77] B. A. Romov. Galois correspondence between iterative post algebras and relations on [RS82] infinite sets. Cybernetics, 3:377 -- 379, 1977. I. G. Rosenberg and D. Schweigert. Locally maximal clones. Elektron. Informationsver- arb. Kybernet., 18(7-8):389 -- 401, 1982. in German, Russian summary. I. G. Rosenberg and L. Szab´o. Local completeness. I. Algebra Universalis, 18(3):308 -- 326, 1984. ´A. Szendrei. Clones in universal algebra. Les Presses de L'Universit´e de Montr´eal, 1986. [Sze86] [Tho91] S. Thomas. Reducts of the random graph. Journal of Symbolic Logic, 56(1):176 -- 181, [RS84] 1991. [Tum89] J. Tuma. Intervals in subgroup lattices of infinite groups. Journal of Algebra, 125(2):367 -- 399, 1989. E-mail address: [email protected] URL: http://dmg.tuwien.ac.at/pinsker/ Laboratoire de Math´ematiques Nicolas Oresme, CNRS UMR 6139, Universit´e de Caen, 14032 Caen Cedex, France
1606.01566	4	1606	2017-12-04T16:57:57	Finite Gr\"obner basis algebra with unsolvable nilpotency problem and zero divisors problem	[ "math.RA" ]	This work presents a sample constructions of two algebras both with the ideal of relations defined by a finite Gr\"obner basis. For the first algebra the question whether a given element is nilpotent is algorithmically unsolvable, for the second one the question whether a given element is a zero divisor is algorithmically unsolvable. This gives a negative answer to questions raised by Latyshev.	math.RA	math	FINITE GR OBNER BASIS ALGEBRAS WITH UNSOLVABLE NILPOTENCY PROBLEM AND ZERO DIVISORS PROBLEM ILYA IVANOV-POGODAEV, SERGEY MALEV Abstract. This work presents a sample constructions of two algebras both with the ideal of relations defined by a finite Grobner basis. For the first algebra the question whether a given element is nilpotent is algorithmically unsolvable, for the second one the question whether a given element is a zero divisor is algorithmically unsolvable. This gives a negative answer to questions raised by Latyshev. for which the question whether a given element is nilpotent is algorithmically unsolvable. This gives a negative answer to a question raised by Latyshev. 1. Introduction The word equality problem in finitely presented semigroups (and in algebras) cannot be algorithmically solved. This was proved in 1947 by Markov ([Ma]) and independently by Post([Po]). In 1952 Novikov constructed the first example of the group with unsolvable problem of word equality (see [N1] and [N2]). In 1962 Shirshov proved solvability of the equality problem for Lie algebras with one relation and raised a question about finitely defined Lie algebras (see [Sh]). In 1972 Bokut settled this problem. In particular, he showed the existence of a finitely defined Lie algebra over an arbitrary field with algorithmically unsolvable identity problem ([Bo]). A detailed overview of algorithmically unsolvable problems can be found in [BK]. Otherwise, some problems become decidable if a finite Grobner basis defines a relations ideal. In this case it is easy to determine whether two elements of the algebra are equal or not (see [Be]). Grobner bases for various structures are investigated by the Bokut school in Guangzhou ([BC]). In his work, Piontkovsky extended the concept of obstruction, introduced by Latyshev (see [Pi1], [Pi2], [Pi3], [Pi4]). Latyshev raised the question concerning the existence of an algorithm that can find out if a given element is either a zero divisor or a nilpotent element when the ideal of relations in the algebra is defined by a finite Grobner basis. Similar questions for monomial automaton algebras can be solved. In this case the existence of an algorithm for nilpotent element or a zero divisor was proved by Kanel- Belov, Borisenko and Latyshev [KBBL]. Note that these algebras are not Noetherian and not weak Noetherian. Iyudu showed that the element property of being one-sided zero divisor is recognizable in the class of algebras with a one-sided limited processing (see [I1], [I2]). It also follows from a solvability of a linear recurrence relations system on a tree (see [KB1]). We would like to thank Agata Smoktunowicz and Alexei Kanel-Belov for interesting and fruitful discussions regarding this paper. This research was supported by ERC Advanced grant Coimbra 320974. This research was supported by Young Russian Mathematics award. This research was supported by Russian Science Foundation (grant no. 17-11-01377). 1 2 ILYA IVANOV-POGODAEV, SERGEY MALEV An example of an algebra with a finite Grobner basis and algorithmically unsolvable problem of zero divisor is constructed in [IP]. A notion of Grobner basis (better to say Grobner-Shirshov basis) first appeared in the context of noncommutative (and not Noetherian) algebra. Note also that Poincar´e- Birkhoff-Witt theorem can be canonically proved using Grobner bases. More detailed discussions of these questions see in [Bo], [U], [KBBL]. In the present paper we construct an algebra with a finite Grobner basis and algorith- mically unsolvable problem of nilpotency. We also provide a shorter construction for the zero divisors question. For these constructions we simulate a universal Turing machine, each step of which corresponds to a multiplication from the left by a chosen letter. Thus, to determine whether an element is a zero divisor or is a nilpotent, it is not enough for an algebra to have a finite Grobner basis. 2. The plan of construction Let A be an algebra over a field K. Fix a finite alphabet of generators {a1, . . . , aN }. A word in the alphabet of generators is called a word in algebra. The set of all words in the alphabet is a semigroup. The main idea of the construction is a realization of a universal Turing machine in the semigroup. We use the universal Turing machine constructed by Marvin Minsky in [Mi]. This machine has 7 states and 4-color tape. The machine can be completely defined by 28 instructions. Note that 27 of them have a form (i, j) → (L, q(i, j), p(i, j)) or (i, j) → (R, q(i, j), p(i, j)), where 0 ≤ i ≤ 6 is the current machine state, 0 ≤ j ≤ 3 is the current cell color, L or R (left or right) is the direction of a head moving after execution of the current instruction, q(i, j) is the state after current instruction, p(i, j) is the new color of the current cell. Thus, the instruction (2, 3) → (L, 3, 1) means the following: "If the color of the current cell is 3 and the state is 2, then the cell changes the color to 1, the head moves one cell to the left, the machine changes the state to 3. The last instruction is (4, 3) → STOP. Hence, if the machine is in state 4 and the current cell has color 3, then the machine halts. Letters. By Qi, 0 ≤ i ≤ 6 denote the current state of the machine. By Pj , 0 ≤ j ≤ 3 denote the color of the current cell. The action of the machine depends on the current state Qi and current cell color Pj. Thus every pair Qi and Pj corresponds to one instruction of the machine. The instructions moving the head to the left (right) are called left (right) ones. There- fore there are left pairs (i, j) for the left instructions, right pairs for the right ones and instruction STOP for the pair (4, 3). All cells with nonzero color are said to be non-empty cells. We shall use letters a1, a2, a3 for nonzero colors and letter a0 for color zero. Also, we use R for edges of colored area. Hence, the word Rau1 au2 . . . auk QiPjav1 av2 . . . avl R presents a full state of Turing machine. We model head moving and cell painting using computations with powers of ai (cells) and Pi and Qi (current cell and state of the machine's head). 3. Universal Turing machine We use the universal Turing machine constructed by Minsky. This machine is defined by the following instructions: (0, 0) → (L, 4, 1) (0, 1) → (L, 1, 3) (0, 2) → (R, 0, 0) (0, 3) → (R, 0, 1) (1, 0) → (L, 1, 2) (1, 1) → (L, 1, 3) (1, 2) → (R, 0, 0) (1, 3) → (L, 1, 3) FINITE GR OBNER BASIS ALGEBRAS WITH UNSOLVABLE PROBLEMS 3 (2, 0) → (R, 2, 2) (2, 1) → (R, 2, 1) (2, 2) → (R, 2, 0) (2, 3) → (L, 4, 1) (3, 0) → (R, 3, 2) (3, 1) → (R, 3, 1) (3, 2) → (R, 3, 0) (3, 3) → (L, 4, 0) (4, 0) → (L, 5, 2) (4, 1) → (L, 4, 1) (4, 2) → (L, 4, 0) (4, 3) → STOP (5, 0) → (L, 5, 2) (5, 1) → (L, 5, 1) (5, 2) → (L, 6, 2) (5, 3) → (R, 2, 1) (6, 0) → (R, 0, 3) (6, 1) → (R, 6, 3) (6, 2) → (R, 6, 2) (6, 3) → (R, 3, 1) We use the following alphabet: {t, a0, . . . a3, Q0, . . . Q6, P0 . . . P3, R} For every pair except (4, 3) the following functions are defined: q(i, j) is a new state, p(i, j) is a new color of the current cell (the head leaves it). 4. Defining relations for the nilpotency question Consider the following defining relations: tRal = Rtal; 0 ≤ l ≤ 3 talR = alRt; 0 ≤ l ≤ 3 takaj = aktaj; 0 ≤ k, j ≤ 3 takQiPj = Qq(i,j)Pktap(i,j); for left pairs (i, j) and 0 ≤ k ≤ 3 tRQiPj = RQq(i,j)P0tap(i,j); for left pairs (i, j) and 0 ≤ k ≤ 3 talQiPjakan = alap(i,j)Qq(i,j)Pktan; for right pairs (i, j) and 0 ≤ k ≤ 3 talQiPj akR = alap(i,j)Qq(i,j)PkRt; for right pairs (i, j) and 0 ≤ k ≤ 3 tRQiPjakan = Rap(i,j)Qq(i,j)Pktan; for right pairs (i, j) and 0 ≤ k ≤ 3 tRQiPj akR = Rap(i,j)Qq(i,j)PkRt; for right pairs (i, j) and 0 ≤ k ≤ 3 talQiPjR = alap(i,j)Qq(i,j)P0Rt; for right pairs (i, j) and 0 ≤ l ≤ 3 tRQiPj R = Rap(i,j)Qq(i,j)P0Rt; for right pairs (i, j) Q4P3 = 0. h4.1i h4.2i h4.3i h4.4i h4.5i h4.6i h4.7i h4.8i h4.9i h4.10i h4.11i h4.12i The relations h4.1i and h4.3i are used to move t from the left edge to the last letter al standing before QiPj which represent the head of the machine. The relations h4.4i -- h4.11i represent the computation process. The relation h4.2i is used to move t through the finishing letter R. Finally, the relation h4.12i halts the machine. 5. Nilpotency of the fixed word and machine halt Let us call the word tRau1 au2 . . . auk QiPjav1 av2 . . . avl R the main word. The main goal is to prove the following theorem: Theorem 5.1. The machine halts if and only if the main word is nilpotent in the algebra presented by the defining relations h4.1i -- h4.12i. First, we prove some propositions. Remark. We use sign ≡ for lexicographical equality and sign = for equality in algebra. Consider a full state of our Turing machine represented by the word Rau1 au2 . . . auk QiPjav1 av2 . . . avl R. Suppose that U ≡ au1 au2 . . . auk and V ≡ av1 av2 . . . avl . Therefore U and V repre- sent the colors of all cells on the Turing machine tape. We denote the full state of this machine as M (i, j, U, V ). Suppose that M (i′, j ′, U ′, V ′) is the next state (M (i, j, U, V ) → M (i′, j ′, U ′, V ′)). 4 ILYA IVANOV-POGODAEV, SERGEY MALEV Consider a semigroup G presented by the defining relations h4.1i -- h4.12i. Suppose that W (i, j, U, V ) is a word in G corresponding to machine state M (i, j, U, V ). (Actually W (i, j, U, V ) ≡ Rau1au2 . . . auk QiPj av1 av2 . . . avl R.) Proposition 5.1. Let us move all the words from the relations h4.1i -- h4.12i to the left- hand side. There exists a reduction order on the free monoid generated by alphabet Φ = {t, a0, . . . , a3, Q0, . . . , Q6, P0, . . . , P3, R}, such that the left-hand sides of the ob- tained equalities comprise a Grobner basis in the ideal generated by them. Proof. Recall, that by reduction order on the free monoid Φ∗ we mean a well order such that the empty word is the minimal one, and for any a, b, s1, s2 ∈ Φ∗, if s1 ≺ s2 then as1b ≺ as2b. Any word w from Φ∗ can be uniquely written as X0tX1t · · · tXn, where Xi ∈ Φ∗ are free from the letter t. Each Xi can be empty, even all of them (if the word is tn). By height of this word we call h(w) = n X i=0 2i deg Xi. We define the following order. Given two words w1 and w2, we compare them with respect to the degree of t. If degt(w1) < degt(w2) then w1 ≺ w2. If degt(w1) = degt(w2) then we compare them with respect to the height. If h(w1) < h(w2) then w1 ≺ w2. If their heights are also equal then we use a deglex order to compare them. We need to prove that this order is a reduction order. Note that an empty word is the minimal (it has a zero degree of t, a zero height and a zero degree). Assume a, b, s1, s2 ∈ Φ∗ and s1 ≺ s2. If degt(s1) < degt(s2) , then degt(as1b) < degt(as2b), therefore as1b ≺ as2b. Assume degt(s1) = degt(s2) = n and h(s1) < h(s2). In this case we will show that multiplication of inequality by one symbol does not change it. In other words, we will show for any symbol x ∈ Φ that xs1 ≺ xs2 and s1x ≺ s2x. First assume that x 6= t. Then multiplication by x from the left increases a height by 1 of both sides, thus an inequality remains. Note that multiplication by x from the right increases a height by 2n of both sides, and an inequality remains also. The multiplication by t from the left multiplies both heghts by 2 and multiplication by t from the right does not change it. Now assume that degt(s1) = degt(s2) and h(s1) = h(s2). Hence degt(as1b) = In this case we compare both pairs (s1, s2) and degt(as2b) and h(as1b) = h(as2b). (as1b, as2b) by deglex order which is a reduction order. Note that every left-hand side contains a leading monomial. There is no such word that (cid:3) begins some leading monomial in the basis and ends some other leading monomial. Lemma 5.1. For any nonempty word U ≡ ai1 · · · ail we have tU R = U Rt. Proof. We can use the relation h4.3i (l−1) times and transform tU R to ai1 · · · ail−1 tail R. After that we use the relation h4.2i. (cid:3) Proposition 5.2. If i = 4 and j = 3 then tW (i, j, U, V ) = 0. Otherwise, the following condition holds: tW (i, j, U, V ) = W (i′, j ′, U ′, V ′)t. Proof. Consider the word tW (i, j, U, V ) = tRU QiPjV R. If i = 4 and j = 3 then we can apply relation h4.12i. Otherwise, suppose that (i, j) is a left pair. If U is not empty word, then we can write U = U ak for some 0 ≤ k ≤ 3. In this case we have the word tR U akQiPj V R. We can use the relation h4.1i to transform it to Rt UakQiPj V R. Now we use the relation h4.3i the degree of U times: our words transforms to R U takQiPj V R. After that we use the relation h4.4i and our word transforms to R U Qi′ Pj ′ tap(i,j)V R. Now we use Lemma 5.1. FINITE GR OBNER BASIS ALGEBRAS WITH UNSOLVABLE PROBLEMS 5 If U is empty, then tW (i, j, U, V ) ≡ tRQiPj V R. In this case we will start our chain with using relation h4.5i: tRQiPjV R = RQi′ P0tap(i,j)V R. After that we use Lemma 5.1. Suppose that (i, j) is a right pair. In this situation we will have six cases: Case 1 U and V are empty words. In this case our word is tRQiPj R and we use the relation h4.11i. Case 2 U is empty and V = ak is a word of degree 1. In this case our word is tRQiPjakR, and we can use a relation h4.9i. Case 3 U is empty and V = ak V is a word of degree greater than 1. In this case our word is tRQiPj ak V R. We can use a relation h4.8i to transform it to Rap(i,j)Qi′ Pj ′ t V R, where V is not empty. Thus we can use Lemma 5.1 to complete the chain. Case 4 U = U al is not empty and V is empty. In this case our word is tR U alQiPjR. We use a relation h4.1i and transform it to Rt U alQiPj R. Using relation h4.3i the degree of U times will transform our word to R U talQiPj R. A relation h4.10i completes a chain. Case 5 U = U al is not empty and V = ak is a word of degree 1. In this case our word is tR U alQiPjakR. Similar to Case 4 we can transform our word to R U talQiPj akR. A relation h4.7i completes a chain. Case 6 U = U al is not empty and V = ak V is a word of degree greater than 1. In this case our word is tR UalQiPj ak V R. Similar to Case 4 we can transform our word to R U talQiPj ak V R. A relation h4.6i transforms it to R U alap(i,j)Qy ′ Pj ′ t V R. Now we use lemma 5.1 to complete our chain. (cid:3) Proposition 5.3. The following statements are equivalent: (i) The Turing machine described above begins with the state M (i, j, U, V ) and halts in several steps. (ii) There exists a positive integer N such that tN RU QiPj V R = 0. Proof. First, prove that second statement is a consequence of the first one. Suppose that M (i, j, U, V ) transforms to M (4, 3, U ′, V ′) in one step. According to Proposition 5.2 tW (i, j, U, V ) = W (4, 3, U ′, V ′)t. Then we can apply Q4P3 = 0 by h4.12i and obtain zero. Suppose that the statement is true for m (and fewer) steps. Let the machine begin with state M (i, j, k, n) and halt after m + 1 step. Consider the first step in the chain. Let it be the step from M (i, j, U, V ) to M (i′, j ′, U ′, V ′). Apply Proposition 5.2 for this step. Hence tRU QiPj V R = RU ′Qi′ Pj ′ V ′Rt. The machine started in the state M (i′, j ′, U ′, V ′) halts in m steps. Using induction we complete the proof. Now let us prove that the first statement is a consequence of the second one. If tN RU QiPjV R = 0, then there exists a chain of equivalent words, starting with tN RU QiPjV R and finishing with 0. The only way to obtain 0 is to use a relation Q4P3 = 0. Therefore the word before 0 in the chain contains Q4P3. By structure of the word W , S(W ) let us denote the word W , where all letters t will be deleted. Each word in the chain will have a structure RUkQik Pjk VkR because the only relation that breaks this structure is Q4P3 = 0, and it will be used only one time, in the end of the chain. Note that each structure corresponds to the Turing machine. The only way to obtain 0 in this chain is to change indices of Q and P in the structure. This can be done by moving t. According to the Proposition 5.2, moving t from the left to the right corresponds to the Turing Machine's one step to the future, and moving t from the right to the left corresponds to the Turing Machine's one step to the past (note that this is not always possible). There is a Grobner basis of relations in our algebra, thus we can assume that in 6 ILYA IVANOV-POGODAEV, SERGEY MALEV our chain words decrease (each word is lower than the previous with respect to the order on the free monoid Φ∗). Therefore letters t move only from the left to the right. Hence there exists k ≤ N such that tN RU QiPjV R = tN −kR U Q4P3 V Rtk. Therefore the machine halts after k steps. (cid:3) Now we are ready to prove the theorem above. Theorem 5.2. Consider an algebra A presented by the defining relations h4.1i -- h4.12i. The word tRU QiPj V R is nilpotent in A if and only if machine M (i, j, U, V ) halts. Proof. Suppose that (tRU QiPjV R)n = 0. The structure of this word corresponds to a row of n separate machines. Using relations we can transform some machine to the next state (note that we have a Grobner basis in the algebra, therefore we can assume that words in the chain will decrease). Thus if we obtain Q4P3 for some machine, we can conclude that this machine halts after several steps. Therefore M (i, j, U, V ) halts. Suppose that M (i, j, U, V ) halts. Then tnRU QiPj V R = 0 for some minimal n. We can obtain (tRU QiPj V R)n = AtnRU QiPj V R (for some word A) by using Proposition 5.2 several times. Therefore (tRU QiPj V R)n = 0. (cid:3) Since the halting problem cannot be algorithmically solved, the nilpotency problem in algebra A is algorithmically unsolvable. 6. Defining relations for a zero divisors question We use the following alphabet: Ψ = {t, s, a0, . . . a3, Q0, . . . Q6, P0 . . . P3, L, R}. For every pair except (4, 3) the following functions are defined: q(i, j) is a new state, p(i, j) is a new color of the current cell (the head leaves it). Consider the following defining relations: tLak = Ltak; 0 ≤ k ≤ 3 takal = aktal; 0 ≤ k, l ≤ 3 sR = Rs; sak = aks; 0 ≤ k ≤ 3 takQiPj = Qq(i,j)Pkap(i,j)s; for left pairs (i, j) and 0 ≤ k ≤ 3 tLQiPj = LQq(i,j)P0ap(i,j)s; for left pairs (i, j) talQiPjak = alap(i,j)Qq(i,j)Pks; for right pairs (i, j) and 0 ≤ k, l ≤ 3 tLQiPj ak = Lap(i,j)Qq(i,j)Pks; for right pairs (i, j) and 0 ≤ k ≤ 3 talQiPjR = alap(i,j)Qq(i,j)P0Rs; for right pairs (i, j) and 0 ≤ l ≤ 3 tLQiPjR = Lap(i,j)Qq(i,j)P0Rs; for right pairs (i, j) Q4P3 = 0; h6.1i h6.2i h6.3i h6.4i h6.5i h6.6i h6.7i h6.8i h6.9i h6.10i h6.11i The relations h6.1i -- h6.2i are used to move t from the left edge to the letters Qi, Pj which present the head of the machine. The relations h6.3i -- h6.4i are used to move s from the letter Qi, Pj to the right edge. The relations h6.5i -- h6.9i represent the computation process. Here we use relations of the form tU = V s. Finally, the relation h6.11i halts the machine. FINITE GR OBNER BASIS ALGEBRAS WITH UNSOLVABLE PROBLEMS 7 7. Zero divisors and machine halt Let us call the word Lau1 au2 . . . auk QiPj av1 av2 . . . avl R the main word. The main goal is to prove the following theorem: Theorem 7.1. The machine halts if and only if the main word is a zero divisor in the algebra presented by the defining relations h6.1i -- h6.11i. Consider a full state of our Turing machine represented by the word Lau1 au2 . . . auk QiPj av1 av2 . . . avl R. Suppose that U = au1 au2 . . . auk and V = av1 av2 . . . avl . Therefore U and V repre- sent the colors of all cells on the Turing machine tape. We denote the full state of this machine as T (i, j, U, V ). Suppose that T (i′, j ′, U ′, V ′) is the next state (T (i, j, U, V ) → T (i′, j ′, U ′, V ′)). Consider a semigroup S presented by the defining relations h6.1i -- h6.11i. Suppose that F (i, j, U, V ) is a word in S corresponding to machine state T (i, j, U, V ). Proposition 7.1. Let us move all the words from relations h6.1i -- h6.11i to the left-hand side. Consider the semi-DEGLEX order: {t, s, a0, . . . a3, Q0, . . . , Q6, P0, . . . , P3, L, R}. The left-hand sides of the obtained equalities comprise a Grobner basis in the ideal gener- ated by them. Proof. We will use a weighted degree instead of the usual: each letter from the alpha- bet (except for t) will have degree 1, however the degree of t equals 2. (For example, deg(tRL) = 4) This order is a reduction order. Note that every left-hand side contains a leading monomial. There is no such word that (cid:3) begins some leading monomial in the basis and ends some other leading monomial. Proposition 7.2. If i = 4 and j = 3 then tF (i, j, U, V ) = 0. Otherwise, the following condition holds: tF (i, j, U, V ) = F (i′, j ′, U ′, V ′)s. Proof. Consider the word tF (i, j, U, V ) = tLU QiPjV R. If i = 4 and j = 3 then we can apply relation h6.11i. Otherwise, suppose that (i, j) is a left pair. If U is an empty word then tF (i, j, U, V ) = tLQiPjV R. Hence we can apply relation h6.6i to obtain tLQiPjV R = LQq(i,j)P0aj sV R. Using h6.3i and h6.4i we finally have tLQiPjV R = LQq(i,j)P0aj sV R = LQq(i,j)P0aj V Rs. According to the definition of q(i, j) and p(i, j), the word LQq(i,j)P0ajV R corresponds to the next state of the machine. If U is not an empty word, we can write U = U1ak for some k. We use the relations h6.1i and h6.2i and obtain that tLU QiPj V R = LU1takQiPj V R. Further, we use relation h6.5i: LU1takQiPj V R = LU1Qq(i,j)Pkap(i,j)V Rs. The word LU1Qq(i,j)Pkap(i,j)V R corresponds to the next state of the machine. Assume that (i, j) is a right pair. If U and V are empty words, than we use relation h6.10i. If U is empty, and V = ak V is not, then we use the relation h6.8i and obtain tLQiPjak V R = Lap(i,j)Qq(i,j)Pks V R. After that we use relations h6.4i and h6.3i and move s to the right. Assume U = U ak is not empty. In this case we use the relation h6.2i the length of U times and obtain t U ak = U tak. If V is empty we can use the relation h6.9i. If V = al V is not empty then we can use the relation h6.7i, after that we will use relations h6.4i and h6.3i and move s to the right. (cid:3) 8 ILYA IVANOV-POGODAEV, SERGEY MALEV Proposition 7.3. The following statements are equivalent: (i) The Turing machine described above begins with the state T (i, j, U, V ) and halts in several steps. (ii) There exists a positive integer N such that tN LU QiPjV R = 0. Proof. First, prove that second statement is a consequence of the first one. Suppose that T (i, j, U, V ) transforms to T (4, 3, U ′, V ′) in one step. According to Propo- sition 7.2 tF (i, j, U, V ) = F (4, 3, U ′, V ′)s. Then we can apply Q4P3 = 0 by h6.11i and obtain zero. Suppose that the statement is true for m (and fewer) steps. Let the machine begin with state T (i, j, k, n) and halt after m + 1 step. Consider the first step in the chain. Let it be the step from T (i, j, U, V ) to T (i′, j ′, U ′, V ′). Apply Proposition 7.2 for this step. Hence tLU QiPj V R = LU ′Qi′ Pj ′ V ′Rs. The machine started in the state T (i′, j ′, U ′, V ′) halts in m steps. Using induction we complete the proof. Now let us prove that the first statement is a consequence of the second one. If tN LU QiPjV R = 0, then there exists a chain of equivalent words, starting with tN LU QiPjV R and finishing with 0. The only way to obtain 0 is to use a relation Q4P3 = 0. Therefore the word before 0 in the chain contains Q4P3. By structure of the word W , S(W ) let us denote the word W , where all letters t and s will be deleted. Each word in the chain will have a structure LUkQik Pjk VkR because the only relation that breaks this structure is Q4P3 = 0, and it will be used only one time, in the end of the chain. Note that each structure corresponds to the Turing machine. The only way to obtain 0 in this chain is to change indices of Q and P in the structure. This can be done by moving t. According to the Proposition 7.2, moving t from the left to the right, and transforming it to s corresponds to the Turing Machine's one step. Note that there is a Grobner basis on our algebra, thus we can assume that words in the chain decrease. In particular, moving s from the right to the left, transforming it to t is impossible. Therefore we can obtain Q4P3 only by moving t from the left to s on the right, and there exists k ≤ N such that tN LU QiPj V R = tN −kL U Q4P3 V Rtk. Therefore the machine halts after k steps. (cid:3) Proposition 7.4. If Xt = 0 in S, then X = 0. If sX = 0 in S, then X = 0. Proof. Suppose that we apply some relations and transform Xt to zero. We say that the letter t is almost last if the word has the form Y1tY2, and Y2 contains ak and L letters only. Note that if an almost last t-letter occurs in some relation then this relation is h6.1i or h6.2i. Therefore that t-letter is always almost last. It is clear that an almost last t-letter always exists in every word which is equivalent to Xt. Since an almost last t-letter never participates in relations h6.3i -- h6.11i, we can situate it on the right edge of we word Xt while we use our relations. We did not use the t-letter, and therefore we can do the same with the word X. Similarly we can prove that if sX = 0 then X = 0. Proposition 7.5. If Xtn = 0 in S, then X = 0. If snX = 0 in S, then X = 0. Proof. We can prove this by induction. Now we are ready to prove the theorem above. (cid:3) (cid:3) Theorem 7.2. Consider an algebra H presented by the defining relations h6.1i -- h6.11i. The word LU QiPj V R is a zero divisor in the algebra H if and only if machine T (i, j, U, V ) halts. FINITE GR OBNER BASIS ALGEBRAS WITH UNSOLVABLE PROBLEMS 9 Proof. Suppose that machine T (i, j, U, V ) halts. Using Proposition 7.3 we have tN LU QiPjV R = 0 for some positive integer N . Thus, the word LU QiPj V R is a zero divisor. Let XLU QiPj V RY = 0 for some algebra elements X, Y 6= 0. Suppose that X, Y are some words. Note that L and R letters cannot disappear from the word. Hence we can divide our word into three parts: to the left of L, to the right of R, and between L and R. There is only one relation which can turn the word XLU QiPjV RY to zero: Q4P3 = 0. Thus this subword Q4P3 can appear in three possible parts of the word. Note that only t letters can pass through L and only s letters can pass through R. Every relation can change nothing in the area to the left side of L and to the right side of R, except t and s-letters occurrences. Therefore if Q4P3 appears to the left of L, then Xsn = 0. Using Proposition 7.5 we obtain a contradiction: X = 0. Similarly if Q4P3 appears to the right of R, then Y = 0. Thus Q4P3 appears between L and R. Consider the structure of the word LU QiPj V R. For any structure of a word equivalent to LU QiPj V R there exists a corresponding state of the machine. Since only t letters can pass through L and only s letters can pass through R, we can change the structure of the word LU QiPjV R by turn to the next or the previous machine state. If Q4P3 appears between L and R then we can obtain a STOP state. Thus the machine T (i, j, U, V ) halts. Now let us consider the general case: X, Y are some algebra elements. Suppose that X = c1X1 + . . . cnXn, Y = d1Y1 + . . . dmYm, where Xk, Yl are words, and ck and dl are elements of the field. Without loss of generality we may assume that n is the minimal possible, and for this n m is the minimal possible. We also may assume that Xk, Yl are written in the reduced form. We assume that either n > 1, or m > 1. Consider the function h : Ψ∗ → N0: for any word w h(w) = degt(w) + degs(w). Note that relations h6.1i-h6.10i do not change value of h, therefore it is invariant under the word reduction. Assume that h(Xk1 ) 6= h(Xk2 ). In this case we will take a subset Sx ⊆ {1, . . . , n} such that h takes a maximal value on Xk for k ∈ Sx. We also take a subset Sy ⊆ {1, . . . , m} such that h takes a maximal value on Yl for l ∈ Sy. We know that (c1X1 + . . . cnXn)W (d1Y1 + . . . dmYm) = Pk,l ckdlXkLU QiPjV RYl = 0. Therefore one can reduce this element to zero. However none of the elements XkLU QiPjV RYl can be reduced to zero. Thus, all elements XkLU QiPjV RYl can be separated to several sets of similar words. Note that all words XkLU QiPjV RYl (where k ∈ Sx and l ∈ Sy) can be similar only to a word Xk′ LU QiPj V RYl′ where k′ ∈ Sx and l′ ∈ Sy. Hence, dlYl) = 0. A contradiction (n was taken as a minimal ( P k∈Sx possible). Therefore h(Xk) does not depend on k and h(Yl) does not depend on l. ckXk) · LU QiPjV R · ( P l∈Sy We have XLU QiPj V RY = Pk,l ckdlXkLU QiPj V RYl. We can consider our defin- ing relations as reductions and use them to find the Grobner basis of every term ktqk . These XkLU QiPjV RYl. Let us fix the t's at the end of the Xk words: Xk = X ′ are lexicographical equalities and qk > 0. ktqk LU QiPjV RY ′ l = 0, Since Pk,l ck,lX ′ this sum (in the reduced form) can be separated into several sets of similar monomials. Consider one of these sets: P X ′ u must be also sim- ilar. Recall that h(Xk) does not depend on k, therefore all xu must be the same. If these monomials are similar then all X ′ utxu LU QiPj V RY ′ u. Hence, n = 1, and m > 1 and we have a situation XLU QiPj V R(Pm l=1 dlYl) = 0, where X ∈ Ψ∗ is a word, and m is minimal. Therefore, all words XLU QiPjV RYl should be equal in the algebra, however Yl should be pairwise different. If we will reduce word W RYl (for W = XLU QiPj V ), only letter s can pass through R, therefore the only case to reduce it is to pass letters s from W to Yl. The number of these letters s depend on W therefore it will be similar. Therefore we will have an equality skY1 = skY2 = · · · = skYm 10 ILYA IVANOV-POGODAEV, SERGEY MALEV (for some non negative number k) in the algebra. Note that relations with letter s do not change a structure of the word, and one can see that for any two different words Y and Z, words sY and sZ must be also different. Therefore skY1 6= skY2 in the algebra, and m cannot be larger than 1. This contradiction completes the proof. (cid:3) Since the halting problem cannot be algorithmically solved, the zero divisors problem in algebra H is algorithmically unsolvable. Remark. We can consider two semigroups corresponding to our algebras: in both algebras each relation is written as an equality of two monomials. Therefore the same alphabets together with the same sets of relations define semigroups. In both semigroups the equality problem is algorithmically solvable, since it is solvable in algebras. However in the first semigroup a nilpotency problem is algorithmically unsolvable, and in the second semigroup a zero divisor problem is algorithmically unsolvable. References [Be] Bergman, G. The diamond lemma for ring theory. Adv. Math., (1978), 29, 2, 178 -- 218. [Bo] Bokut, L Unsolvability of the equality problem and subalgebras of finitely presented lie algebras. Izvestiya Akad. Nauk SSSR. 36:6 (1972), 1173 -- 1219 [BC] Bokut, L., Chen Y. Grobner-Shirshov bases and PBW theorems J. Sib.Fed. Univ. Math. Phys., 2013, Volume 6, Issue 4, 417 -- 427 [BK] Bokut, L., Kukin G. Undecidable algorithmic problems for semigroups, groups and rings. (Rus- sian) Translated in J. Soviet Math. 45 (1989), no. 1, 871 -- 911. Itogi Nauki i Tekhniki, Algebra. Topology. Geometry, Vol. 25 (Russian), 3 -- 66, Akad. Nauk SSSR, Vsesoyuz. Inst. Nauchn. i Tekhn. Inform., Moscow, (1987). [GIL] Gateva-Ivanova, T., Latyshev, V. On the recognizable properties of associative algebras. J.Symb.Comp., 6 (1989) 371 -- 398, Elsevier. [I1] Iyudu, N. Algorithmical solvability of zero divisors problem in one class of algebras. Pure and Applied Math., (1995), 2, 1, 541 -- 544. [I2] Iyudu, N. Standart bases and property solvability in the algebras defined by relations. Dissertation -- Moscow, (1996), 73. [IP] Ivanov-Pogodaev, I. An algebra with a finite Grobner basis and an unsolvable problem of zero divisors. J. Math. Sci. (N. Y.) 152 (2008), no. 2, 191 -- 202 [KB1] Kanel-Belov, A. Linear recurrence relations on tree. Math. zametki, 78, N5, 643 -- 651. [KB2] Kanel-Belov, A. Classification of weakly Noetherian monomial algebras. (Russian. English sum- mary) Fundam. Prikl. Mat. 1, (1995), no.4, 1085 -- 1089. [KBBL] Kanel-Belov, A.; Borisenko V.; Latysev V. Monomial Algebras. NY. Plenum (1997) [L] Latyshev, V. Combinatorial ring theory, standard bases. (Russian) Moskva: Izdatelstvo Moskovskogo Gosudarstvennogo Universiteta. 68 pp. R. 0.15 (1988). [Ma] Markov, A. The impossibility of certain algorithms in the theory of associative systems. (Rus- sian) Doklady Akad. Nauk SSSR, 55 N7, 1947, 587 -- 590 [Mi] Minsky, M. Computation: Finite and Infinite Machines (1967) [N1] Novikov, P., On algorithmic unsolvability of the problem of identity. (Russian) Doklady Akad. Nauk SSSR 85 N4, (1952). 709 -- 712. [N2] Novikov, P., On the algorithmic unsolvability of the word problem in group theory. Trudy Mat. Inst. im. Steklov. no. 44. Izdat. Akad. Nauk SSSR, Moscow, (1955), 3 -- 143 [Po] Post E., Recursive unsolvability of a problem of Thue. J. Symb. Logic, 12, N1, 1947, 1 -- 11 [Pi1] Piontkovsky, D. Grobner base and coherence of monomial associative algebra. Pure and Applied Math., (1996), 2, 2, 501 -- 509. [Pi2] Piontkovsky, D. Noncommutative Grobner bases, coherence of monomial algebras and divisibil- ity in semigroups Pure and Applied Math.,(2001), 7, 2, 495 -- 513. [Pi3] Piontkovsky, D. On the Kurosh problem in varieties of algebras. J. Math. Sci., New York 163, No. 6, 743 -- 750 (2009); translation from Fundam. Prikl. Mat. 14, No. 5, 171 -- 184 (2008). [Pi4] Piontkovsky, D. Graded algebras and their differential graded extensions. J. Math. Sci., New York 142, No. 4, 2267 -- 2301 (2007); translation from Sovrem. Mat. Prilozh. 30, 65 -- 100 (2005). [Sh] Shirshov, A. Some algorithmic problems for Lie algebras Sib. mat. journal, (1962), vol 3 N2, 292 -- 296 [U] Ufnarovsky, V. Combinatorial and asymototic methods in algebra. Itogi nauki i tehniki, Modern problems of pure math. .: VINITI, (1990), 57, 5 -- 177. Moscow Institute of Physics and Technology, Moscow, Russia E-mail address: [email protected] School of mathematics, University of Edinburgh, Edinburgh, UK E-mail address: [email protected]
1510.04381	1	1510	2015-10-15T02:48:00	Notes on Gr\"obner Bases and Free Resolutions of Modules over Solvable Polynomial Algebras	[ "math.RA" ]	This text consists of five relatively systematic notes on Gr\"obner bases and free resolutions of modules over solvable polynomial algebras.	math.RA	math	Notes on Grobner Bases and Free Resolutions of Modules over Solvable Polynomial Algebras 5 1 0 2 t c O 5 1 ] . A R h t a m [ 1 v 1 8 3 4 0 . 0 1 5 1 : v i X r a Huishi Li Department of Applied Mathematics College of Information Science & Technology Hainan University, Haikou 570228, China Introduction Polynomials and power series, May they forever rule the world. -- Shreeram S. Abhyankar Since the late 1980s, the Grobner basis theory for commutative polyno- mial algebras and their modules (cf. [Bu1, 2], [Sch], [BW], [AL2], [Frob], [KR1, 2]) has been successfully generalized to (noncommutative) solvable polynomial algebras and their modules (cf. [AL1], [Gal], [K-RW], [Kr2], [LW], [Li1], [Lev]). In particular, there have been a noncommutative version of Buchberger's criterion and a noncommutative version of the Buchberger algorithm for checking and computing (1) Grobner bases of (one-sided) ideals of solvable polynomial algebras, and (2) Grobner bases of submodules of free modules over solvable polyno- mial algebras, while the noncommutative version of Buchberger algorithm has been im- plemented in some well-developed computer algebra systems, such as Modula-2 [KP] and Singular [DGPS]. By the definition (see Section 1 of [K-RW], or Definition 1.1.3 given in our Chapter 1), a solvable polyno- mial algebra is a polynomial-like but generally noncommutative algebra. Nowadays it is well known that the class of solvable polynomial algebras covers numerous significant noncommutative algebras such as enveloping algebras of finite dimensional Lie algebras, Weyl algebras (including al- gebras of partial differential operators with polynomial coefficients over a i ii Introduction field of characteristic 0), more generally a lot of operator algebras, iterated Ore extensions, and quantum (quantized) algebras. So, comparing with the commutative case, the theory of Grobner bases for solvable polyno- mial algebras and their modules has created great possibility of studying certain noncommutative algebras and their modules in a "solvable set- ting" (see [Wik1] for an introduction to "Decision problem" or "Solvable problem", which may help us to better understand why a "solvable poly- nomial algebra" deserves its name). Along the lines in the literature for developing a computational (one- sided) ideal theory and more generally, a computational module theory over solvable polynomial algebras, it seems that a rapid but relatively sys- tematic and concrete introduction to the subject is worthwhile. Thereby, we wrote these lecture notes so as to provide graduates and researchers (who are interested in noncommutative computational algebra) with an accessible reference on • an concise introduction to solvable polynomial algebras and the theory of Grobner bases for submodules of free nodules over solvable polyno- mial algebras, and • some details concerning applications of Grobner bases in construct- ing finite free resolutions over an arbitrary solvable polynomial alge- bras, minimal finite N-graded free resolutions over an N-graded solv- able polynomial algebra with the degree-0 homogeneous part being the ground field K, and minimal finite N-filtered free resolutions over an N-filtered solvable polynomial algebra (where the N-filtration is deter- mined by a positive-degree function). The first three chapters of these notes grew out of a course of lectures given to graduate students at Hainan University, and the last two chap- ters are adapted from the author's recent research work [Li7] (or [Li5] arXiv:1401.5206v2 [math.RA], [Li6] arXiv:1401.5464 [math.RA]). Also, at the level of module theory, these notes may be viewed as supplements of the sections concerning solvable polynomial algebras and their modules in ([Li1], [Li2]). Throughout the text, K denotes a field, K ∗ = K − {0}; N denotes the additive monoid of all nonnegative integers, and Z denotes the additive group of all integers; all algebras are associative K-algebras with the multiplicative identity 1, and modules over an algebra are meant left unitary modules. Contents Introduction 1. Solvable Polynomial Algebras 1.1 Definition and Basic Properties 1.2 Left Grobner Bases of Left Ideals 1.3 The Noetherianess 1.4 A Constructive Characterization 2. Left Grobner Bases for Modules 2.1 Left Monomial Orderings on Free Modules 2.2 Left Grobner Bases of Submodules 2.3 The Noncommutative Buchberger Algorithm 3. Finite Free Resolutions 3.1 Computation of Syzygies 3.2 Computation of Finite Free Resolutions 3.3 Global Dimension and Stability 3.4 Calculation of p.dimAM 4. Minimal Finite Graded Free Resolutions 4.1 N-graded Solvable Polynomial Algebras 4.2 N-Graded Free Modules 4.3 Computation of Minimal Homogeneous Generating Sets iii i 1 33 51 69 iv Contents 4.4 Computation of Minimal Finite Graded Free Resolutions 5. Minimal Finite Filtered Free Resolutions 89 5.1 N-Filtered Solvable Polynomial Algebras 5.2 N-Filtered Free Modules 5.3 Filtered-Graded Transfer of Grobner Bases for Modules 5.4 F-Bases and Standard Bases with Respect to Good Filtration 5.5 Computation of Minimal F-Bases and Minimal Standard Bases 5.6 Minimal Filtered Free Resolutions and Their Uniqueness 5.7 Computation of Minimal Finite Filtered Free Resolutions 6. Some Examples of Applications 125 1. Solvable Polynomial Algebras In this chapter we give a concise but easily accessible introduction to solvable polynomial algebras and the theory of left Grobner bases for left ideals of such algebras. In the first section we introduce the definition of solvable polynomial algebras and some typical examples; also from the definition we highlight some often used properties of solvable polynomial algebras. In Section 2, we introduce left Grobner bases for left ideals of solvable polynomial algebras via a left division algorithm, and we dis- cuss some basic properties of left Grobner bases. In Section 3, on the basis of Dickson's lemma we show that every left ideal of a solvable poly- nomial algebra has a finite Grobner basis, thereby solvable polynomial algebras are left Noetherian. Since every solvable polynomial algebra has a (two-sided) monomial ordering, a right division algorithm and a theory of right Grobner bases for right ideals hold true as well, it follows that every solvable polynomial algebra is also right Noetherian. Concerning the algorithmic approach to computing a finite left Grobner basis, it will be a job of Chapter 2 for modules. In the final Section 4, by employing Grobner bases of two-sided ideals in free algebras we give a characteriza- tion of solvable polynomial algebras, so that such algebras are completely recognizable and constructible in a computational way. The main references of this chapter are [AL1], [Gal], [K-RW], [Kr2], [LW], [Li1], [Li4], [DGPS], [Ber2], [Mor], [Gr], [Uf]. 1 2 Solvable Polynomial Algebras 1.1. Definition and Basic Properties as a finite sum of the form a = Pi λiaα1 Let K be a field and let A = K[a1, . . . , an] be a finitely generated K- algebra with the set of generators {a1, . . . , an}, that is, A is an associative ring with the multiplicative identity 1, every element a ∈ A is expressed with λi ∈ K, aij ∈ {a1, . . . , an}, αj ∈ N, t ≥ 1, and A is also a K-vector space with respect to its additive operation, such that λ(ab) = (λa)b = a(λb) holds for all λ ∈ K and a, b ∈ A. Moreover, we assume that any proper subset of the given generating set {a1, . . . , an} of A cannot generate A as a K-algebra, i.e., the given set of generators is minimal. · · · aαt it i1 · · · aαn If, for some permutation τ = i1i2 · · · in of 1, 2, . . . , n, the set B = {aα = aα1 in α = (α1, . . . , αn) ∈ Nn}, forms a K-basis of A, then B i1 is referred to as a PBW K-basis of A (where the phrase "PBW K-basis" is abbreviated from the well-known Poincar´e-Birkhoff-Witt Theorem con- cerning the standard K-basis of the enveloping algebra of a Lie algebra, e.g., see [Hu], P. 92). It is clear that if A has a PBW K-basis, then we can always assume that i1 = 1, . . . , in = n. Thus, we make the following convention once for all. Convention From now on in this paper, if we say that an algebra A has the PBW K-basis B, then it means that B = {aα = aα1 1 · · · aαn n α = (α1, . . . , αn) ∈ Nn}. Moreover, adopting the commonly used terminology in computational algebra, elements of B are referred to as monomials of A. Monomial ordering and admissible system Suppose that the K-algebra A = K[a1, . . . , an] has the PBW K-basis B as presented above and that ≺ is a total ordering on B. Then every nonzero element f ∈ A has a unique expression f = λ1aα(1) + λ2aα(2) + · · · + λmaα(m), · · · aαnj n ∈ B, 1 ≤ j ≤ m. where λj ∈ K ∗, aα(j) = aα1j 1 aα2j 2 If aα(1) ≺ aα(2) ≺ · · · ≺ aα(m) in the above representation, then the leading monomial of f is defined as LM(f ) = aα(m), the leading coefficient of F is defined as LC(f ) = λm, and the leading term of f is defined as LT(f ) = λmaα(m). Solvable Polynomial Algebras 3 1.1.1. Definition Suppose that the K-algebra A = K[a1, . . . , an] has the PBW K-basis B. If ≺ is a total ordering on B that satisfies the following three conditions: (1) ≺ is a well-ordering (i.e. every nonempty subset of B has a min- imal element); (2) For aγ, aα, aβ, aη ∈ B, if aγ 6= 1, aβ 6= aγ, and aγ = LM(aαaβaη), then aβ ≺ aγ (thereby 1 ≺ aγ for all aγ 6= 1), (3) For aγ, aα, aβ, aη ∈ B, if aα ≺ aβ, LM(aγaαaη) 6= 0, and LM(aγaβaη) 6∈ {0, 1}, then LM(aγaαaη) ≺ LM(aγaβaη); then ≺ is called a monomial ordering on B (or a monomial ordering on A). If ≺ is a monomial ordering on B, then the data (B, ≺) is referred to as an admissible system of A. (i) Definition 1.1.1 above is borrowed from the theory of Remark. Grobner bases for general finitely generated K-algebras, in which the algebras considered may be noncommutative, may have divisors of zero, and the K-bases used may not be a PBW basis, but with a (one-sided, two-sided) monomial ordering that, theoretically, enables such algebras have a (one-sided, two-sided) Grobner basis theory. For more details on this topic, one may referrer to ([Li2], Section 1 of Chapter 3 and Section 3 of Chapter 8). Also, to see the essential difference between Definition 1.2.1 and the classical definition of a monomial ordering in the commu- tative case, one may refer to (Definition 1.4.1 and the proof of Theorem 1.4.6 given in [AL2]). (ii) The conditions (2) and (3) listed in Definition 1.1.1 mean that ≺ is two-sided compatible with the multiplication operation of the algebra A. As one will see soon, that the use of a (two-sided) monomial ordering ≺ on a solvable polynomial algebra A first guarantees that A is a domain, and furthermore guarantees an effective (left, right, two-sided) finite Grobner basis theory for A. Note that if a K-algebra A = K[a1, . . . , an] has the PBW K-basis B = {aα = aα1 n α = (α1, . . . , αn) ∈ Nn}, then for any given n- tuple (m1, . . . , mn) ∈ Nn, a weighted degree function d( ) is well defined on nonzero elements of A, namely, for each aα = aα1 n ∈ B, d(aα) = i=1 λiaα(i) ∈ A with m1α1 + · · · + mnαn, and for each nonzero f = Ps 1 · · · aαn 1 · · · aαn 4 Solvable Polynomial Algebras λi ∈ K ∗ and aα(i) ∈ B, d(f ) = max{d(aα(i)) 1 ≤ i ≤ s}. If d(ai) = mi > 0 for 1 ≤ i ≤ n, then d( ) is referred to as a positive-degree function on A. 1.1.2. Definition Let d( ) be a positive-degree function on A. If ≺ is a monomial ordering on B such that for all aα, aβ ∈ B, aα ≺ aβ implies d(aα) ≤ d(aβ), then we call ≺ a graded monomial ordering with respect to d( ). Convention Unless otherwise stated, from now on in this book we always use ≺gr to denote a graded monomial ordering with respect to a positive- degree function on A. As one may see from the literature (or loc. cit) that in both the com- mutative and noncommutative computational algebra, the most popularly used graded monomial orderings on an algebra A = K[a1, . . . , an] with the PBW K-basis B are those graded (reverse) lexicographic orderings with respect to the degree function d( ) such that d(ai) = 1, 1 ≤ i ≤ n. Definition of solvable polynomial algebras and examples Originally, a (noncommutative) solvable polynomial algebra (or an alge- bra of solvable type) R′ was defined in [K-RW] by first fixing a monomial ordering ≺ on the standard K-basis B = {X α1 αi ∈ N} of 1 the commutative polynomial algebra R = K[X1, . . . , Xn] in n variables X1, . . . , Xn over a field K, and then introducing a new multiplication ∗ on R, such that certain axioms ([K-RW], AXIOMS 1.2) are satisfied. In [LW], a solvable polynomial algebra was redefined in the formal language of associative K-algebras, as follows. · · · X αn n 1.1.3. Definition If a K-algebra A = K[a1, . . . , an] satisfies the follow- ing two conditions: (S1) A has the PBW K-basis B; (S2) There is a monomial ordering ≺ on B, i.e., (B, ≺) is an admissible Solvable Polynomial Algebras 5 system of A, such that for all aα = aα1 1 · · · aαn n , aβ = aβ1 1 · · · aβn n ∈ B, aαaβ = λα,βaα+β + fα,β, where λα,β ∈ K ∗, aα+β = aα1+β1 fα,β ∈ K-spanB with LM(fα,β) ≺ aα+β if fα,β 6= 0, · · · aαn+βn , and n 1 or alternatively, such that for all generators a1, . . . , an of A and i < j, ajai = λjiaiaj + fji, where λji ∈ K ∗, and fji ∈ K-spanB with LM(fji) ≺ aiaj if fji 6= 0, then A is said to be a solvable polynomial algebra. Remark As one will see later, that the pure-ring-theoretical definition of a solvable polynomial algebra above may at least provide us with the following two advantages. (i) Solvable polynomial algebras can be characterized in a constructive way (see subsequent Section 1.4), so that more noncommutative algebras (and hence their modules) can be studied in a "solvable setting" (see [Wik1] for an introduction on "Decision problem" (or "Solvable prob- lem")). (ii) It is quite helpful in determining whether a solvable polynomial al- gebra A is an N-graded algebra as specified in (Section 4.1 of Chapter 4), or an N-filtered algebra as specified in (Section 5.1 of Chapter 5) respectively. Example (1) By Definition 1.1.3, every commutative polynomial algebra K[x1, . . . , xn] in variables x1, . . . , xn over a field K is trivially a solvable polynomial algebra. From [KR-W] and [Li1] we recall several most well-known noncom- mutative solvable polynomial algebras as follows. Example (2) The nth Weyl algebra The nth Weyl algebra An(k) over a field K is defined to be the K-algebra generated by 2n generators x1, ..., xn, y1, ..., yn subject to the relations: xixj = xjxi, yiyj = yjyi, yjxi = xiyj + δij the Kronecker delta, 1 ≤ i, j ≤ n. 1 ≤ i < j ≤ n, Historically, the Weyl algebra is the first "quantum algebra" ([Dir] 1926, It is a well-known fact that if charK = 0, then [Wey] 1928). 6 Solvable Polynomial Algebras An(k) coincides with the algebra of linear partial differential operators K[x1, . . . , xn, ∂1, . . . , ∂n] of the polynomial ring K[x1, ..., xn] (or the ring of polynomial functions of the affine n-space An , 1 ≤ i ≤ n. k ), where ∂i = ∂ ∂xi Example (3) The additive analogue of the Weyl algebra This algebra was introduced in Quantum Physics in ([Kur] 1980) and studied in ([JBS] 1981), that is, the algebra An(q1, ..., qn) generated over a field K by x1, ..., xn, y1, ..., yn subject to the relations: xixj = xjxi, yiyj = yjyi, yixi = qixiyi + 1, xjyi = yixj, 1 ≤ i < j ≤ n, 1 ≤ i ≤ n, i 6= j, where qi ∈ K ∗. If qi = q 6= 0, i = 1, ..., n, then this algebra becomes the algebra of q-differential operators. Example (4) The multiplicative analogue of the Weyl algebra This is the algebra stemming from ([Jat] 1984 and [MP] 1988, where one may see why this algebra deserves its title), that is, the algebra On(λji) generated over a field K by x1, ..., xn subject to the relations: xjxi = λjixixj, 1 ≤ i < j ≤ n, where λji ∈ K ∗. If n = 2, then O2(λ21) is the quantum plane in the sense of Manin [Man]. If λji = q−2 6= 0 for some q ∈ K ∗ and all i < j, then On(λji) becomes the coordinate ring of the so called quantum affine n-space (see [Sm]). Example (5) The enveloping algebra of a finite dimensional Lie algebra Let g be a finite dimensional vector space over the field k with basis {x1, ..., xn} where n = dimkg. If there is a binary operation on g, called the bracket product and denoted [ , ], which is bilinear, i.e., for a, b, c ∈ g, λ ∈ k, [a + b, c] = [a, c] + [b, c] [a, b + c] = [a, b] + [a, c] λ[a, b] = [λa, b] = [a, λb], and satisfies: [a, b] = −[b, a], [[a, b], c] + [[c, a], b] + [[b, c], a] = 0 Jacobi a, b ∈ g, identity, a, b, c ∈ g, Solvable Polynomial Algebras 7 then g is called a finite dimensional Lie algebra over k. Note that [ , ] need not satify the associative low. If [a, b] = [b, a] for every a, b in a Lie algebra g, g is called abelian. The enveloping algebra of g, denoted U (g), is defined to be the associative k-algebra generated by x1, ..., xn subject to the relations: xjxi − xixj = [xj, xi], 1 ≤ i < j ≤ n. For instance, the Heisenberg Lie algebra h has the k-basis {xi, yj, z i, j = 1, ..., n} and the bracket product is given by [xi, yi] = z, [xi, xj] = [xi, yj] = [yi, yj] = 0, [z, xi] = [z, yi] = 0, 1 ≤ i ≤ n, i 6= j, 1 ≤ i ≤ n. Example (6) The q-Heisenberg algebra This is the algebra stemming from ([Ber] 1992, [Ros] 1995) which has its root in q-calculus (e.g., [Wal] 1985), that is, the algebra hn(q) generated over a field K by the set of elements {xi, yi, zi i = 1, ..., n} subject to the relations: xixj = xjxi, yiyj = yjyi, zjzi = zizj, 1 ≤ i < j ≤ n, xizi = qzixi, ziyi = qyizi, xiyi = q−1yixi + zi, xiyj = yjxi, xizj = zjxi, yizj = zjyi, 1 ≤ i ≤ n, 1 ≤ i ≤ n, 1 ≤ i ≤ n, i 6= j, where q ∈ K ∗. Example (7) The algebra of 2 × 2 quantum matrices This is the algebra Mq(2, K) introduced in ([Man] 1988), which is gener- ated over a field K by a, b, c, d subject to the relations: ba = qab, db = qbd, ca = qac, dc = qcd, cb = bc, da − ad = (q − q−1)bc, where q ∈ K ∗. Example (8) The algebra U in constructing Hayashi algebra In order to get bosonic representations for the types of An and Cn of the Drinfield-Jimbo quantum algebras, Hayashi introduced in ([Hay] 1990) the q-Weyl algebra A− q , which is constructed as follows (see [Ber1]). Let 8 Solvable Polynomial Algebras U be the algebra generated over the field K = C by the set of elements {xi, yi, zi i = 1, ..., n} subject to the relations: xjxi = xixj, yjyi = yiyj, zjzi = zizj, 1 ≤ i < j ≤ n, xjyi = q−δij yixj, zjxi = q−δij xizj, zjyi = yizj, ziyi − q2yizi = −q2x2 i , 1 ≤ i, j ≤ n, i 6= j, 1 ≤ i ≤ n, where q ∈ K ∗. Then A− q = S−1U, One may also use the technique given in the subsequent Section 1.4 to directly verify that the algebras listed above are solvable polynomial algebras. Basic properties By Definition 1.1.1 and Definition 1.1.3, the properties listed in the next two propositions are straightforward. 1.1.4. Proposition Let A = K[a1, . . . , an] be a solvable polynomial algebra with admissible system (B, ≺). The following statements hold. (i) If f, g ∈ A with LM(f ) = aα, LM(g) = aβ, then LM(f g) = LM(LM(f )LM(g)) = LM(aαaβ) = aα+β. (ii) A is a domain, that is, A has no (left and right) divisors of zero. Proof Exercise. (cid:3) 1.1.5. Proposition Let A1 = K[a1, . . . , an] and A2 = K[b1, . . . , bm] be solvable polynomial algebras with admissible systems (B1, ≺1) and (B2, ≺2) respectively. Then A = A1⊗KA2 is a solvable polynomial algebra with the admissible system (B, ≺), where B = {aα ⊗ bβ aα ∈ B1, bβ ∈ B2}, while ≺ is defined on B subject to the rule: for aα ⊗ bβ, aγ ⊗ bη ∈ B, aα ⊗ bβ ≺ aγ ⊗ bη ⇔ Proof Exercise. aα ≺1 aγ; or aα = aγ and bβ ≺2 bη. Solvable Polynomial Algebras 9 1.2. Left Grobner Bases of Left Ideals Let A = K[a1, . . . , an] be a solvable polynomial algebra with admissible system (B, ≺). In this section we introduce left Grobner bases for left ideals of A via a left division algorithm in A, and we record some basic facts determined by left Grobner bases. Moreover, minimal left Grobner bases and reduced left Grobner bases are discussed. Left division algorithm Let aα, aβ ∈ B. We say that aα divides aβ from the left side, denoted aαLaβ, if there exists aγ ∈ B such that aβ = LM(aγaα). It follows from Proposition 1.1.4(i) that the division defined above is implementable. Let F be a nonempty subset of A. Then the division of monomials defined above yields a subset of B: N (F ) = {aα ∈ B LM(f ) 6 L aα, f ∈ F }. If F = {g} consists of a single element g, then we simply write N (g) in place of N (F ). Also we write K-spanN (F ) for the K-subspace of A spanned by N (F ). 1.2.1. Definition Elements of N (F ) are referred to as normal monomi- als (mod F ). Elements of K-spanN (F ) are referred to as normal elements (mod F ). In view of Proposition 1.1.4(i), the left division we defined for mono- mials in B can naturally be used to define a left division procedure for let f, g ∈ A with LC(f ) = µ 6= 0, elements in A. More precisely, LC(g) = λ 6= 0. If LM(g)L LM(f ), i.e., there exists aα ∈ B such that LM(f ) = LM(aαLM(g)), then put f1 = f − λ−1µaαg; otherwise, put f1 = f − LT(f ). Note that in both cases we have f1 = 0, or f1 6= 0 and LM(f1) ≺ LM(f ). At this stage, let us refer to such a procedure of canceling the leading term of f as the left division procedure by g. With f := f1 6= 0, we can repeat the left division procedure by g and so on. 10 Solvable Polynomial Algebras This returns successively a descending sequence LM(f ) ≻ LM(f1) ≻ LM(f2) ≻ · · · . Since ≺ is a well-ordering, it follows that such a division procedure termi- nates after a finite number of repetitions, and consequently f is expressed as f = qg + r, where q, r ∈ A with r ∈ K-spanN (g), i.e., r is normal (mod g), such that either LM(f ) = LM(qg) or LM(f ) = LM(r). Furthermore, the left division procedure demonstrated above can be extended to a left division procedure by a finite subset G = {g1, . . . , gs} in A, which yields the following division algorithm: Algorithm-LDIV INPUT: f, G = {g1, . . . , gs} with gi 6= 0 (1 ≤ i ≤ s) OUTPUT: q1, . . . , qs, r such that r ∈ K-spanN (G), f =Ps i=1 qigi + r, LM(qigi) (cid:22) LM(f ) for qi 6= 0, LM(r) (cid:22) LM(f ) if r 6= 0 · · · , qs := 0; r := 0; h := f INITIALIZATION: q1 := 0, q2 := 0, BEGIN WHILE h 6= 0 DO IF there exist i and aα(i) ∈ B such that LM(h) = LM(aα(i)LM(gi)) THEN choose i least such that LM(h) = LM(aα(i)LM(gi)) qi := qi + LC(gi)−1LC(h)aα(i) h := h − LC(gi)−1LC(h)aα(i)gi ELSE r := r + LT(h) h := h − LT(h) END END END 1.2.2. Definition The element r obtained in Algorithm-LDIV is called a remainder of f on left division by G, and is denoted by f = r. If f = 0, then we say that f is reduced to 0 (mod G). , i.e., f G G G Solvable Polynomial Algebras 11 Remark Note that the element r obtained in Algorithm-LDIV de- pends on how the order of elements in G is arranged. This implies that r may be different if a different order for elements of G is given (acturally as in the commutative case, e.g. [AL2], P.31). That is why we use the phrase "a remainder" in the above definition. Summing up, we have reached the following 1.2.3. Theorem Given a set of nonzero elements G = {g1, . . . , gs} and f in A, the Algorithm-LDIV produces elements q1, . . . , qs, r ∈ A with i=1 qigi + r and LM(qigi) (cid:22) LM(f ) (cid:3) r ∈ K-spanN (G), such that f =Ps whenever qi 6= 0, LM(r) (cid:22) LM(f ) if r 6= 0. Left Grobner bases Theoretically the left division procedure by using a finite subset G of nonzero elements in A can be extended to a left division procedure by means of an arbitrary proper subset G of nonzero elements, for, it is a true statement that if f ∈ A with LM(f ) 6= 0, then either there exists g ∈ G such that LM(g)L LM(f ) or such g does not exist. This leads to the following 1.2.4. Proposition Let N be a left ideal of A and G a proper subset of nonzero elements in N . The following two statements are equivalent: (i) If f ∈ N and f 6= 0, then there exists g ∈ G such that LM(g)L LM(f ). j=1 qigij with qi ∈ A (cid:3) (ii) Every nonzero f ∈ N has a representation f =Ps and gij ∈ G, such that LM(qigij ) (cid:22) LM(f ) whenever qi 6= 0. 1.2.5. Definition Let N be a left ideal of A. With respect to a given monomial ordering ≺ on B, a proper subset G of nonzero elements in N is said to be a left Grobner basis of N if G satisfies one of the equivalent conditions of Proposition 1.2.4. If G is a left Grobner basis of N , then the expression f =Ps j=1 qigij appeared in Proposition 1.2.4(ii) is called a left Grobner representation of f . Clearly, if N is a left ideal of A and N has a left Grobner basis G, then G is certainly a generating set of N , i.e., N = Pg∈G Ag. But the 12 Solvable Polynomial Algebras converse is not necessarily true. For instance, in the solvable polynomial algebra A = C[a1, a2, a3] generated by {a1, a2, a3} subject to the relations a2a1 = 3a1a2, a3a1 = a1a3, a3a2 = 5a2a3, let g1 = a2 1a2 − a3, g2 = a2, and N = Ag1 + Ag2. Then since a3 ∈ N , G = {g1, g2} is not a left Grobner basis with respect to any given monomial ordering ≺ on B such that LM(g1) = a2 1a2, LM(g2) = a2. The existence of left Grobner bases Noticing 1 ∈ B, if we define on B the ordering: aα ≺′ aβ ⇔ aαLaβ, aα, aβ in B, then, by Definition 1.1.1, Proposition 1.1.4(i) and the divisibility defined for monomials, it is an easy exercise to check that ≺′ is reflexive, anti- symmetric, transitive, and moreover, aα ≺′ aβ implies aα ≺ aβ. Since the given monomial ordering ≺ is a well-ordering on B, it follows that every nonempty subset of B has a minimal element with respect to the ordering ≺′ on B. Now, let N 6= {0} be a left ideal of A, LM(N ) = {LM(f ) f ∈ N }, and Ω = {aα aα is minimal in LM(N ) w.r.t. ≺′}. 1.2.6. Theorem With the notation as above, the following statements hold. (i) Ω 6= ∅ and Ω is a proper subset of LM(N ). (ii) Let G = {g ∈ N LM(g) ∈ Ω}. Then G is a left Grobner basis of N . Proof (i) That Ω 6= ∅ follows from N 6= {0} and the remark about ≺′ we made above. Since N is a left ideal and A is a domain (Proposition 1.1.4), if f ∈ N and f 6= 0, then hf ∈ N for all nonzero h ∈ A. It follows from Proposition 1.1.4(i) that LM(hf ) ∈ LM(N ) and LM(f )L LM(hf ), i.e., LM(f ) ≺′ LM(hf ) in LM(N ). It is clear that if LM(h) 6= 1, then LM(hf ) 6= LM(f ). This shows that Ω is a proper subset of LM(N ). (ii) By the definition of ≺′ and the remark we made above the theorem, if f ∈ N with LM(f ) 6= 0 and LM(f ) 6∈ Ω, then there is some g ∈ G Solvable Polynomial Algebras 13 such that LM(g)L LM(f ). Hence the selected G is a left Grobner basis for N . (cid:3) In Section 3 we will show that every nonzero left ideal N of A has a finite left Grobner basis. Basic facts determined by left Grobner bases By referring to Proposition 1.1.4, Theorem 1.2.3 and Proposition 1.2.4, the foregoing discussion allows us to summarize some basic facts deter- mined by left Grobner bases, of which the detailed proof is left as an exercise. 1.2.7. Proposition Let N be a left ideal of A, and let G be a left Grobner basis of N . Then the following statements hold. (i) If f ∈ N and f 6= 0, then LM(f ) = LM(qg) for some q ∈ A and g ∈ G. (ii) If f ∈ A and f 6= 0, then f has a unique remainder f G. (iii) If f ∈ A and f 6= 0, then f ∈ N if and only if f = 0. Hence the membership problem for left ideals of A can be solved by using left Grobner bases. (iv) As a K-vector space, A has the decomposition on division by G G A = N ⊕ K-spanN (G). (v) As a K-vector space, A/N has the K-basis N (G) = {aα aα ∈ N (G)}, where aα denotes the coset represented by aα in A/N . (vi) N (G) is a finite set, or equivalently, DimK A/N < ∞, if and only if for each i = 1, . . . , n, there exists gji ∈ G such that LM(gji) = ami , where i mi ∈ N. (cid:3) Minimal and reduced left Grobner bases 1.2.8. Definition Let N 6= {0} be a left ideal of A and let G be a left grobner basis of N . If any proper subset of G cannot be a left Grobner basis of N , then G is called a minimal left Grobner basis of N . 14 Solvable Polynomial Algebras 1.2.9. Proposition Let N 6= {0} be a left ideal of A. A left Grobner basis G of N is minimal if and only if LM(gi)6 L LM(gj ) for all gi, gj ∈ G with gi 6= gj. Proof Suppose that G is minimal. If there were gi 6= gj in G such that LM(gi)L LM(gj), then since the left division is transitive, the proper subset G′ = G − {gj} of G would form a left Grobner basis of N . This contradicts the minimality of G. Conversely, if the condition LM(gi)6 L LM(gj ) holds for all gi, gj ∈ G with gi 6= gj, then the definition of a left Grobner basis entails that any proper subset of G cannot be a left Grobner basis of N . This shows that G is minimal. (cid:3) 1.2.10. Corollary Let N 6= {0} be a left ideal of A, and let the notation be as in Theorem 1.2.6. (i) The left Grobner basis G obtained there is indeed a minimal left Grobner basis for N . (ii) If G is any left Grobner basis of N , then Ω = LM(G) = {LM(g) g ∈ G} ⊆ LM(G) = {LM(g′) g′ ∈ G}. Therefore, any two minimal left Grobner bases of N have the same set of leading monomials Ω. Proof This follows immediately from the definition of a left Grobner basis, the construction of Ω and Proposition 1.2.9. (cid:3) We next introduce the notion of a reduced left Grobner basis. 1.2.11. Definition Let N 6= {0} be a left ideal of A and let G be a left grobner basis of N . If G satisfies the following conditions: (1) G is a minimal left Grobner basis; (2) LC(g) = 1 for all g ∈ G; (3) For every g ∈ G, h = g − LM(g) is a normal element (mod G), i.e., h ∈ K-spanN (G), then G is called a reduced left Grobner basis of N . 1.2.12. Proposition Every nonzero left ideal N of A has a unique reduced left Grobner basis. Therefore, two left ideals N1, N2 have the Solvable Polynomial Algebras 15 same reduced left Grobner basis if and only if N1 = N2. Proof Note that if G and G′ are reduced left Grobner bases for I, then LM(G) = LM(G′) by Corollary 1.2.10. If LM(gi) = LM(g′ j) then gi − j ∈ N ∩ K-spanN (G) = N ∩ K-spanN (G′). It follows from Proposition g′ 1.2.7(iv) that gi = g′ j. Hence G = G′. By the uniqueness, the second assertion is clear. (cid:3) Since we will see in Section 3 that every left ideal N of A has a finite left Grobner basis, a minimal left Grobner basis, thereby the reduced left Grobner basis for N , can be obtained in an algorithmic way. More precisely, the next proposition holds true. 1.2.13. Proposition Let N 6= {0} be a left ideal of A, and let G = {g1, . . . , gm} be a finite left Grobner basis of N . (i) The subset G0 = {gi ∈ G LM(gi) is minimal in LM(G) w.r.t. ≺′} of G forms a minimal left Grobner basis of N (see the definition of ≺′ given before Theorem 1.2.6). An algorithm written in pseudo-code is omitted here. (ii) With G0 as in (i) above, we may assume, without loss of gener- ality, that G0 = {g1, . . . , gs} with LC(gi) = 1 for 1 ≤ i ≤ s. Put G1. Then LM(h1) = LM(g1). Put G1 = {g2, ..., gs} and h1 = g1 G2. Then LM(h2) = LM(g2). Put G2 = {h1, g3, ..., gs} and h2 = g2 G3, and so on. The last obtained G3 = {h1, h2, g4, ..., gs} and h3 = g3 Gs+1 = {h1 . . . , hs−1} ∪ {hs} is then the reduced left Grobner basis. An algorithm written in pseudo-code is omitted here. Proof This can be verified directly, so we leave it as an exercise. (cid:3) 1.3. The Noetherianess Let A = K[a1, . . . , an] be a solvable polynomial algebra with admissible system (B, ≺). In this section we show that A is a (left and right) Noethe- rian K-algebra by showing that every (left, right) ideal of A has a finite (left, right) Grobner basis. The following lemma, which will be essential not only in prov- ing our next theorem but also in proving the existence of finite left Grobner bases and the termination property of Buchberger's algorithm 16 Solvable Polynomial Algebras for modules over solvable polynomial algebras (Section 3 of Chapter 2), is usually attributed to the American algebraist L. E. Dickson (http://en.wikipedia.org/wiki/Dickson's− lemma). For a detailed argu- ment on Dickson's lemma in commutative computational algebra, we refer to ([BW], 4.4). 1.3.1. Lemma (Dickson's Lemma) Every subset S of Nn has a finite subset B such that for each (α1, . . . , αn) ∈ S, there exists (γ1, . . . , γn) ∈ B with γi ≤ αi for 1 ≤ i ≤ n. (cid:3) Let aα, aβ ∈ B with α = (α1, . . . , αn), β = (β1, . . . , βn) ∈ Nn. Recall that aα ≺′ aβ if and only if aαLaβ, while the latter is defined subject to the property that aβ = LM(aγaα) for some aγ ∈ B. Thus, by Proposition 1.1.4(i), aα ≺′ aβ is actually equivalent to the property that αi ≤ βi for 1 ≤ i ≤ n. Also note that the correspondence α = (α1, . . . , αn) ←→ aα gives the bijection between Nn and B. Combining this observation with Dickson's lemma, the following result is obtained. 1.3.2. Theorem With notations as in Theorem 1.2.6, let N 6= {0} be a left ideal of A, LM(N ) = {LM(f ) f ∈ N }, Ω = {aα ∈ LM(N ) aα is minimal w.r.t. ≺′}, and G = {g ∈ N LM(g) ∈ Ω}. Then Ω is a finite subset of LM(N ), thereby G is a finite left Grobner basis of N . (cid:3) Since every solvable polynomial algebra has a (two-sided) monomial ordering, a right division algorithm and a theory of right Grobner bases for right ideals hold true as well, we then have the following 1.3.3. Corollary Every solvable polynomial algebra A is (left and right) Noetherian. (cid:3) In the next chapter we will see that the Buchberger algorithm, which computes a finite Grobner basis for a finitely generated commutative poly- nomial ideal, has a complete noncommutative version (Algorithm-LGB presented in Section 3 of Chapter 2), that computes a finite left Grobner basis for a finitely generated submodule of a free module over a solvable polynomial algebra. Solvable Polynomial Algebras 17 1.4. A Constructive Characterization From Definition 1.1.3 we see that the two conditions (S1) and (S2), which determine a solvable polynomial algebra A = K[a1, . . . , an], are mutually independent factors. In this section we give a characterization of solvable polynomial algebras by employing Grobner bases of ideals in free alge- bras, so that solvable polynomial algebras are completely recognizable and constructible in a computational way. To make the text self-contained, we start by recalling some ba- sics on Grobner bases of two-sided ideals in a free K-algebra KhXi = KhX1, . . . , Xni on X = {X1, . . . , Xn}. Division algorithm in KhXi Let B = {1, Xi1 · · · Xis Xij ∈ X, s ≥ 1} be the standard K-basis of KhXi. For convenience, elements of B are also referred to as monomials, and we use capital letters U, V, W, S, . . . to denote monomials in B. Recall that a monomial ordering ≺X on B (or equivalently on KhXi) is a well- ordering such that for any W, U, V, S ∈ B, (1) U ≺X V implies W U ≺X W V, U S ≺X V S, or equivalently, W U S ≺X W V S; (2) if U 6= V , then V = W U S implies U ≺X V (thereby 1 ≺X W for all 1 6= W ∈ B). If ≺X is a monomial ordering on B, then the data (B, ≺X) is referred to as an admissible system of KhXi. Note that for any given n-tuple (m1, . . . , mn) ∈ Nn, a weighted degree function d( ) is well defined on nonzero elements of KhXi, namely, by assigning each Xi the degree d(Xi) = mi, 1 ≤ i ≤ n, we may define for each W = Xi1 · · · Xis ∈ B the degree d(W ) = mi1 + · · · + mis, and for i=1 λiWi ∈ KhXi with λi ∈ K ∗ and Wi ∈ B, the each nonzero f = Ps degree of f is then defined as d(f ) = max{d(Wi) 1 ≤ i ≤ s}. If d(Xi) = mi > 0 for 1 ≤ i ≤ n, then d( ) is referred to as a positive-degree function on KhXi. Let d( ) be a positive-degree function on KhXi. If ≺X is a monomial 18 Solvable Polynomial Algebras ordering on B such that for all U, V ∈ B, U ≺X V implies d(U ) ≤ d(V ), then, as in Section 1, we call ≺X a graded monomial ordering with respect to d( ). For instance, with respect to any given positive-degree function d( ), the lexicographic graded ordering ≺grlex on B can be defined as follows. For U, V ∈ B, U ≺grlex V ⇔ d(U ) < d(V ); or d(U ) = d(V ) and U ≺lex V, where the lexicographic ordering ≺lex on B may be defined by ordering X1, . . . , Xn arbitrarily, say Xi1 ≺lex Xi2 ≺lex · · · ≺lex Xin, i.e., if U = Xℓ1 · · · Xℓs, V = Xt1 · · · Xtm ∈ B and s ≤ m, then s < m, Xℓk = Xtk , 1 ≤ k ≤ s; or s = m, there exists p ≤ s such that Xℓk = Xtk for k < p and Xℓp ≺lex Xtp. U ≺lex V ⇔ At this point we should point out that though the ordering ≺lex is a total ordering on B, it is not a monomial orderings on B. For instance, considering the free algebra KhX1, X2i with X1 ≺lex X2, we have X2 ≻lex X1X2 ≻lex X1X1X2 ≻lex X1X1X1X2 ≻lex · · · . This shows that ≺lex is not a well-ordering on KhXi. One may refer to loc. cit. (e.g. [Gr]) for more monomial orderings on KhXi. f =Pm Let ≺X be a monomial ordering on B. If f ∈ KhXi is such that i=1 λiWi with λi ∈ K ∗, Wi ∈ B, and W1 ≺X W2 ≺X · · · ≺X Wm, then we write LM(f ) = Wm for the leading monomial of f , LC(f ) = λm for the leading coefficient of f , and we write LT(f ) = λWm for the leading term of f . Note that B forms a multiplicative monoid with the identity 1 and the multiplication in B is just simply the concatenation of monomials, i.e., if U, V ∈ B with U = Xi1 · · · Xip, V = Xj1 · · · Xiq , then U V = Solvable Polynomial Algebras 19 Xi1 · · · XipXj1 · · · Xjq ∈ B. Thereby if there exist W, S ∈ B such that V = W U S, then we say that U divides V , denoted U V . Let F be a nonempty subset of KhXi. Then the division of monomials defined above yields a subset of B: N (F ) = {W ∈ B LM(f )6 W , f ∈ F }. If F = {g} consists of a single element g, then we simply write N (g) in place of N (F ). Also we write K-spanN (F ) for the K-subspace of KhXi spanned by N (F ). 1.4.1. Definition Elements of N (F ) are referred to as normal monomi- als (mod F ). Elements of K-spanN (F ) are referred to as normal elements (mod F ). Given a monomial ordering ≺X on B, if f ∈ KhXi and U, V ∈ B, then the definition of ≺X entails that LM(U f V ) = U LM(f )V . So, the division we defined for monomials in B can naturally be used to define a division procedure for elements in KhXi. More precisely, let f, g ∈ KhXi with LC(f ) = µ 6= 0, LC(g) = λ 6= 0. If LM(g)LM(f ), i.e., there exists U, V ∈ B such that LM(f ) = U LM(g)V , then put f1 = f − λ−1µU gV ; otherwise, put f1 = f − LT(f ). Note that in both cases we have f1 = 0, or f1 6= 0 and LM(f1) ≺X LM(f ). At this stage, let us refer to such a procedure of canceling the leading term of f as the division procedure by g. With f := f1 6= 0, we can repeat the division procedure by g and so on. This, in turn, gives rise to LM(fi+1) ≺X LM(fi) ≺X · · · ≺X LM(f1) ≺X LM(f ), i ≥ 2. Since ≺X is a well-ordering, it follows that such a division procedure terminates after a finite number of repetitions, and consequently f is expressed as f = λiUigV i + r, tXi where λi ∈ K ∗, Ui, Vi ∈ B, and r ∈ K-spanN (g), i.e., r is normal (mod g), such that either LM(f ) = max{LM(UigVi) 1 ≤ i ≤ t} or LM(f ) = LM(r). 20 Solvable Polynomial Algebras Furthermore, the division procedure demonstrated above can be ex- tended to a division procedure canceling the leading term of f by a finite subset G = {g1, . . . , gs}, which therefore gives rise to an effective division algorithm in KhXi, that is, we have reached the following 1.4.2. Theorem Given a set of nonzero elements G = {g1, . . . , gs} and f in A, the Algorithm-DIV given below produces finitely many λij ∈ K ∗, Uij, Vij ∈ B, and an r ∈ K-spanN (G), such that f =Pi,j λijUijgjVij + r and LM(UijgjVij) (cid:22) LM(f ), LM(r) (cid:22) LM(f ) if r 6= 0. Algorithm-DIV INPUT: f, G = {g1, . . . , gs} with gi 6= 0 (1 ≤ i ≤ s) OUTPUT: λij ∈ K ∗, Uij, Vij ∈ B, r ∈ K-spanN (G) INITIALIZATION: i = 0; r := 0; h := f Λ1 := ∅, · · · , Λs = ∅; Q1 := ∅, · · · , Qs := ∅ Solvable Polynomial Algebras 21 BEGIN WHILE h 6= 0 DO IF there exist j and U, V ∈ B such that LM(h) = U LM(gj)V THEN choose j least such that LM(h) = U LM(gj )V i := i + 1, λij := LC(gj )−1LC(h), Uij := U, Vij := V Λj := Λj ∪ {λij}, Qj := Qj ∪ {Uij, Vij} h := h − LC(gj )−1LC(h)UijgjVij ELSE r := r + LT(h) h := h − LT(h) END END END 1.4.3. Definition The element r obtained in Algorithm-DIV is called a remainder of f on division by G, and is denoted by f = r. If f = 0, then we say that f is reduced to 0 (mod G). G G G , i.e., f Remark Actually as in the case of (Section 2, Algorithm-LDIV), the element r obtained in Algorithm-DIV depends on how the order of elements in G is arranged. This implies that the r may not be unique. That is why the phrase "a remainder" is used in the above definition. Grobner bases for two-sided ideals of KhXi For the remainder of this section, ideals of KhXi are always meant two- sided ideals; if M ⊂ KhXi is a nonempty subset, then we write LM(M ) = {LM(f ) f ∈ M } for the set of leading monomials of M , and we write I = hM i for the two-sided ideal I of KhXi generated by M ; moreover, we fix an admissible system (B, ≺X) of KhXi. As with a solvable polynomial algebra in Section 2, in principle the division procedure by using a finite subset G of nonzero elements in KhXi can be extended to a left division procedure by means of an arbitrary proper subset G of nonzero elements, for, it is a true statement that if f ∈ KhXi with LM(f ) 6= 0, then either there exists g ∈ G such that LM(g)LM(f ) or such g does not exist. This leads to the following 22 Solvable Polynomial Algebras 1.4.4. Proposition Let I be an ideal of KhXi and G a proper subset of nonzero elements in I. The following statements are equivalent: (i) If f ∈ I and f 6= 0, then there exists g ∈ G such that LM(g)LM(f ). (ii) Every nonzero f ∈ I has a finite representation f =Pi,j λijUijgjVij with λij ∈ K ∗, Uij, Vij ∈ B and gj ∈ G, such that LM(UijgjVij) (cid:22)X LM(f ). (iii) hLM(I)i = hLM(G)i. (cid:3) 1.4.5. Definition Let I be an ideal of KhXi. With respect to a given monomial ordering ≺X on B, a proper subset G of nonzero elements in I is said to be a Grobner basis of I if G satisfies one of the equivalent conditions of Proposition 1.4.4. If G is a Grobner basis of I, then the expression f =Pi,j λijUijgjVij appeared in Proposition 1.4.4(ii) is called a Grobner representation of f . Clearly, if I is an ideal of KhXi and I has a Grobner basis G, then G is certainly a generating set of I, i.e., I = hGi. But the converse is not necessarily true. For instance, let g1 = X 2 1 X2 − X3, g2 = X2, and I = hg1, g2i. Then since X3 ∈ I, G = {g1, g2} is not a Grobner basis with respect to any given monomial ordering ≺X on B such that LM(g1) = X 2 1 X2, LM(g2) = X2. The existence of Grobner bases Noticing 1 ∈ B, if we define on B the ordering: U ≺ V ⇔ U V, U, V in B, then, by the divisibility defined for monomials, it is easy to see that ≺ is reflexive, anti-symmetric, transitive, and moreover, U ≺ V implies U ≺X V. Since the given monomial ordering ≺X is a well-ordering on B, it follows that every nonempty subset of B has a minimal element with respect to the ordering ≺ on B. Now, let I be a nonzero ideal of KhXi, LM(I) = {LM(f ) f ∈ I}, and Ω = {U ∈ LM(I) U is minimal w.r.t. ≺}. Solvable Polynomial Algebras 23 1.4.6. Theorem With the notation as above, the following statements hold. (i) Ω 6= ∅ and Ω is a proper subset of LM(I). (ii) Let G = {g ∈ I LM(g) ∈ Ω}. Then G is a Grobner basis of I. Proof (i) That Ω 6= ∅ follows from I 6= {0} and the remark about ≺ we made above. Since I is a two-sided ideal and KhXi is a domain, if f ∈ I then hf g ∈ I, thereby LM(hf g) ∈ LM(I) for all h, g ∈ KhXi. Also note that the leading monomials of elements in KhXi are taken with respect to the given monomial ordering ≺X. It follows that LM(hf g) = LM(h)LM(f )LM(g) which implies LM(f )LM(hf g), thereby LM(f ) ≺ LM(hf g) in LM(I). It is clear that if LM(h) 6= 1 or LM(g) 6= 1, then LM(hf g) 6= LM(f ). This shows that Ω is a proper subset of LM(I). (ii) By the definition of ≺ and the remark we made above the theorem, if f ∈ I with LM(f ) 6= 0 and LM(f ) 6∈ Ω, then there is some g ∈ G such that LM(g)LM(f ). Hence the selected G is a Grobner basis for I. (cid:3) Basic facts determined by Grobner bases By referring to Proposition 1.4.4, Definition 1.4.5 and the foregoing dis- cussion, the next proposition summarizes some basic facts determined by Grobner bases of ideals in KhXi, of which the detailed proof is left as an exercise. 1.4.7. Proposition Let I be an ideal of KhXi, and let G be a Grobner basis of I. Then the following statements hold. (i) If f ∈ I and f 6= 0, then LM(f ) = LM(U gV ) for some U, V ∈ B and g ∈ G. (ii) If f ∈ KhXi and f 6= 0, then f has a unique remainder f by G. (iii) If f ∈ KhXi and f 6= 0, then f ∈ I if and only if f = 0. Hence the membership problem for ideals of KhXi can be solved by using Grobner bases. (iv) As a K-vector space, KhXi has the decomposition on division G G KhXi = I ⊕ K-spanN (G). (v) As a K-vector space, KhXi/I has the K-basis N (G) = {U U ∈ N (G)}, 24 Solvable Polynomial Algebras where U denotes the coset represented by U in KhXi/I. (cid:3) Minimal and reduced Grobner bases 1.4.8. Definition Let I 6= {0} be an ideal of KhXi and let G be a grobner basis of I. If any proper subset of G cannot be a Grobner basis of I, then G is called a minimal Grobner basis of I. 1.4.9. Proposition Let I 6= {0} be an ideal of KhXi. A Grobner basis G of I is minimal if and only if LM(gi)6 LM(gj) for all gi, gj ∈ G with gi 6= gj. Proof Suppose that G is minimal. If there were gi 6= gj in G such that LM(gi)LM(gj), then since the division of monomials is transitive, the proper subset G′ = G − {gj } of G would form a Grobner basis of I. This contradicts the minimality of G. Conversely, if the condition LM(gi)6 LM(gj) holds for all gi, gj ∈ G with gi 6= gj, then the definition of a Grobner basis entails that any proper subset of G cannot be a Grobner basis of I. This shows that G is minimal. (cid:3) 1.4.10. Corollary Let I 6= 0 be an ideal I of KhXi. With the notation as in Theorem 1.4.6, the Grobner basis G obtained there is indeed a minimal Grobner basis for I. Moreover, if G is any Grobner basis of I, then Ω = LM(G) = {LM(g) g ∈ G} ⊆ LM(G) = {LM(g′) g′ ∈ G}. Therefore, any two minimal Grobner bases of I have the same set of leading monomials Ω. Proof This follows immediately from the definition of a Grobner basis, the construction of Ω and Proposition 1.4.9. (cid:3) We next introduce the notion of a reduced Grobner basis. 1.4.11. Definition Let I 6= {0} be an ideal of KhXi and let G be a grobner basis of I. If G satisfies the following conditions: (1) G is a minimal Grobner basis; (2) LC(g) = 1 for all g ∈ G; Solvable Polynomial Algebras 25 (3) For every g ∈ G, h = g − LM(g) is a normal element (mod G), i.e., h ∈ K-spanN (G), then G is called a reduced Grobner basis of N . 1.4.12. Proposition Every nonzero ideal I of KhXi has a unique re- duced Grobner basis. Therefore,two ideals I and J have the same reduced Grobner basis if and only if I = J. Proof Note that if G and G′ are reduced Grobner bases for I, then LM(G) = LM(G′) by Corollary 1.4.10. j) then gi − g′ j = I ∩ K-spanN (G) = I ∩ K-spanN (G′). It follows from Proposi- tion 1.4.7(iv) that gi = g′ j. Hence G = G′. By the uniqueness, the second assertion is clear. (cid:3) If LM(gi) = LM(g′ In the case that an ideal I has a finite Grobner basis, a minimal Grobner basis, thereby the reduced Grobner basis for I, can be obtained in an algorithmic way. More precisely, the next proposition holds true. 1.4.13. Proposition Let I 6= {0} be an ideal of KhXi, and let G = {g1, . . . , gm} be a finite Grobner basis of I. (i) The subset G0 = {gi ∈ G LM(gi) is minamal in LM(G) w.r.t. ≺} of G forms a minimal Grobner basis of I (see the definition of ≺ given before Theorem 1.4.6). An algorithm written in pseud-code is omitted here. (ii) With G0 as in (i) above, we may assume, without loss of gener- ality, that G0 = {g1, . . . , gs} with LC(gi) = 1 for 1 ≤ i ≤ s. Put G1. Then LM(h1) = LM(g1). Put G1 = {g2, ..., gs} and h1 = g1 G2. Then LM(h2) = LM(g2). Put G2 = {h1, g3, ..., gs} and h2 = g2 G3, and so on. The last obtained Gs is G3 = {h1, h2, g4, ..., gs} and h3 = g3 then the reduced Grobner basis of I. An algorithm written in pseud-code is omitted here. Proof This can be verified directly, so we leave it as an exercise. (cid:3) Construction of Grobner bases Let the admissible system (B, ≺X) of KhXi be as fixed before. A subset G of nonzero elements in KhXi is said to be LM-reduced if LM(gi)6 LM(gj) for all gi 6= gj in G. For f, g ∈ G, if there are monomials u, v ∈ B such that 26 Solvable Polynomial Algebras (1) LM(f )u = vLM(g), and (2) LM(f )6 v and LM(g) 6 u, then the element o(f, u; v, g) = 1 LC(f ) (f · u) − 1 LC(g) (v · g) is referred to as an overlap element of f and g. Obviously o(f, u; v, g) is generally not unique and it includes also the 2 and case f = g. For instance, look at the cases where LM(f ) = X1X 2 LM(g) = X 2 2 X1; LM(f ) = X 4 3 = LM(g). Let I = hGi be an ideal of KhXi generated by a finite subset G = {g1, . . . , gt} of nonzero elements. Then, G may be reduced to an LM- reduced subset G′ in an algorithmic way as in Proposition 1.4.13, such that I = hG′i. Thus, we may always assume that I is generated by a reduced finite subset G. The next theorem and the following algorithm are known as the im- plementation of Bergman's diamond lemma ([Ber2], [Mor], [Gr]), and are practically used to check whether G is a Grobner basis of I or not; if not, the algorithm produces a (possibly infinite) Grobner basis for I. For the detailed proof of the theorem we refer the reader to loc. cit. 1.4.14. Theorem Let G = {g1, . . . , gt} be an LM-reduced subset of nonzero elements in KhXi. Then G is a Grobner basis for the ideal I = hGi if and only if for each pair gi, gj ∈ G, including gi = gj, every overlap = 0, element o(gi, u; v, gj ) of gi and gj has the property o(gi, u; v, gj ) that is, o(gi, u; v, gj) is reduced to 0 (mod G). G If G is not a Grobner basis of I, then the algorithm presented below returns a (finite or countably infinite) Grobner basis G for I. Algorithm-GB INPUT: G0 = {g1, ..., gt} OUTPUT: G = {g1, ..., gm, ...}, a Grobner basis for I INITIALIZATION: G := G0, O := {o(gi, gj) gi, gj ∈ G0} − {0} BEGIN WHILE O 6= ∅ DO Choose any o(gi, gj ) ∈ O O := O − {o(gi, gj )} Solvable Polynomial Algebras G = r 6= 0 THEN IF o(gi, gj) O := {o(g, r), o(r, g), o(r, r) g ∈ G} − {0} G := G ∪ {r} END ... 27 (cid:3) Since generally the WHILE loop in algorithm Algorithm-GB does not terminate after a finite number of executions, the algorithm is written without the ending statement for the WHILE loop. Getting PBW K-bases via Grobner bases Concerning the first condition (S1) that a solvable polynomial algebra should satisfy, we recall from [Li2] the following result, which is a gener- alization of ([Gr], Proposition 2,14; [Li1], CH.III, Theorem 1.5). 1.4.15. Proposition ([Li2], Ch 4, Theorem 3.1) Let I 6= {0} be an ideal of the free K-algebra KhXi = KhX1, . . . , Xni, and A = KhXi/I. Suppose that I contains a subset of n(n−1) elements 2 G = {gji = XjXi − Fji Fji ∈ KhXi, 1 ≤ i < j ≤ n} such that with respect to some monomial ordering ≺X on B, LM(gji) = XjXi holds for all the gji. The following two statements are equivalent: αn n αj ∈ N} where each (i) A has the PBW K-basis B = {X X i denotes the coset of I represented by Xi in A. (ii) Any subset G of I containing G is a Grobner basis for I with respect to ≺X . (cid:3) α2 2 · · · X α1 1 X Remark (i) Obviously, Proposition 1.4.15 holds true if we use any per- mutation {Xk1, . . . , Xkn} of {X1, . . . , Xn} (see an example given in the end of this section). So, in what follows we conventionally use only {X1, . . . , Xn}. (ii) Let the ideal I be as in Proposition 1.4.15. If G = {gji = XjXi − Fji Fji ∈ KhXi, 1 ≤ i < j ≤ n} is a Grobner basis of I such that LM(gji) = XjXi for all the gji, then it is not difficult to see that the 28 Solvable Polynomial Algebras reduced Grobner basis of I is of the form G =( gji = XjXi −Xq q X α1q µji 1 X α2q 2 · · · X αnq n where µji q ∈ K and (α1q, α2q, . . . , αnq) ∈ Nn. LM(gji) = XjXi, 1 ≤ i < j ≤ n ) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) A characterization of solvable polynomial algebras Bearing in mind Definition 1.1.3 and the remark made above, we are now able to present the main result of this section. 1.4.16. Theorem ([Li4], Theorem?) Let A = K[a1, . . . , an] be a finitely generated algebra over the field K, and let KhXi = KhX1, . . . , Xni be the free K-algebras with the standard K-basis B = {1, Xi1 · · · Xis Xij ∈ X, s ≥ 1}. With notations as before, the following two statements are equivalent: (i) A is a solvable polynomial algebra in the sense of Definition 1.1.3. (ii) A ∼= A = KhXi/I via the K-algebra epimorphism π1: KhXi → A with π1(Xi) = ai, 1 ≤ i ≤ n, I = Kerπ1, satisfying (a) with respect to some monomial ordering ≺X on B, the ideal I has a finite Grobner basis G and the reduced Grobner basis of I is of the form G =(cid:26)gji = XjXi − λjiXiXj − Fji (cid:12)(cid:12)(cid:12)(cid:12) LM(gji) = XjXi, 1 ≤ i < j ≤ n (cid:27) · · · X αnq where λji ∈ K ∗, µji n αn with (α1q, α2q, . . . , αnq) ∈ Nn, thereby B = {X n αj ∈ N} forms a PBW K-basis for A, where each X i denotes the coset of I represented by Xi in A; and q ∈ K, and Fji = Pq µji q X α1q 1 X α2q α1 1 X 2 α2 2 · · · X (b) there is a monomial ordering ≺ on B such that LM(F ji) ≺ X iX j αni n , 1 ≤ i < whenever F ji 6= 0, where F ji =Pq µji α1i 1 X j ≤ n. · · · X q X α2i 2 Proof (i) ⇒ (ii) Let B = {aα = aα1 n α = (α1, . . . , αn) ∈ Nn} be the PBW K-basis of the solvable polynomial algebra A and ≺ a mono- mial ordering on B. By Definition 1.1.3, the generators of A satisfy the relations: 1 · · · aαn ajai = λjiaiaj + fji, 1 ≤ i < j ≤ n, (∗) Solvable Polynomial Algebras 29 where λji ∈ K ∗ and fji =Pq µji Consider in the free K-algebra KhXi = KhX1, . . . , Xni the subset q aα(q) ∈ K-spanB with LM(fji) ≺ aiaj. G = {gji = XjXi − λjiXiXj − Fji 1 ≤ i < j ≤ n}, 2 n q aα1q q X α1q · · · aαnq 1 aα2q 1 X α2q then Fji =Pq µji where if fji =Pq µji · · · X αnq for 1 ≤ i < j ≤ n. We write J = hGi for the ideal of KhXi generated by G and put A = KhXi/J. Let π1: KhXi → A be the K-algebra epi- morphism with π1(Xi) = ai, 1 ≤ i ≤ n, and let π2: KhXi → A be the canonical algebra epimorphism. It follows from the universal property of the canonical homomorphism that there is an algebra epimorphism ϕ: A → A defined by ϕ(X i) = ai, 1 ≤ i ≤ n, such that the following diagram of algebra homomorphisms is commutative: 2 n π2−→ A ւϕ ϕ ◦ π2 = π1 KhXi π1y A H ∈ A may be written as H =Pj µjX On the other hand, by the definition of each gji we see that every element βnj n with µj ∈ K and (β1j, . . . , βnj) ∈ Nn, where each X i is the coset of J represented by Xi in A. Noticing the relations presented in (∗), it is straightforward to check that the correspondence β1j 1 X · · · X β2j 2 ψ : A λiaα1i 1 Xi · · · aαni n −→ 7→ Xi λiX A α1i 1 · · · X αni n is an algebra homomorphism such that ϕ ◦ ψ = 1A and ψ ◦ ϕ = 1A, where 1A and 1A denote the identity maps of A and A respectively. This shows that A ∼= A, thereby Kerπ1 = I = J; moreover, B = αj ∈ N} forms a PBW K-basis for A, and ≺ is a {X monomial ordering on B. α2 2 · · · X α1 1 X αn n We next show that G forms the reduced Grobner basis for I as de- scribed in (a). To this end, we first show that the monomial ordering ≺ on B induces a monomial ordering ≺X on the standard K-basis B of KhXi. For convenience, recall that we have used capital letters U, V, W, S, . . . to denote elements (monomials) in B. Also let us fix a graded lexicographic 30 Solvable Polynomial Algebras ordering ≺grlex on B with respect to a given positive-degree function d( ) on KhXi (see the definition given before Definition 1.4.1), such that X1 ≺lex X2 ≺lex · · · ≺lex Xn. Then, for U, V ∈ B we define U ≺X V if  LM(π1(U )) ≺ LM(π1(V )), or LM(π1(U )) = LM(π1(V )) and U ≺grlex V. Since A is a domain (Proposition 1.1.4(ii)) and π1 is an algebra homomor- phism with π1(Xi) = ai for 1 ≤ i ≤ n, it follows that LM(π1(W )) 6= 0 for all W ∈ B. We also note from Proposition 1.1.4(i) that if f, g ∈ A are nonzero elements, then LM(f g) = LM(LM(f )LM(g)). Thus, if U, V, W ∈ B and U ≺X V subject to LM(π1(U )) ≺ LM(π1(V )), then LM(π1(W U )) = LM(LM(π1(W ))LM(π1(U ))) ≺ LM(LM(π1(W ))LM(π1(V ))) = LM(π1(W V )) implies W U ≺X W V ; if U ≺X V subject to LM(π1(U )) = LM(π1(V )) and U ≺grlex V , then LM(π1(W U )) = LM(LM(π1(W ))LM(π1(U ))) = LM(LM(π1(W ))LM(π1(V ))) = LM(π1(W V )) and W U ≺grlex W V implies W U ≺X W V . Similarly, if U ≺X V then U S ≺X V S for all S ∈ B. Moreover, if W, U, V, S ∈ B, W 6= V , such that W = U V S, then LM(π1(W )) = LM(π1(U V S)) and clearly V ≺grlex W , thereby V ≺X W . Since ≺ is a well-ordering on B and ≺grlex is a well- ordering on B, the above argument shows that ≺X is a monomial ordering on B. With this monomial ordering ≺X in hand, by the definition of Fji we see that LM(Fji) ≺X XiXj. Furthermore, since LM(π1(XjXi)) = aiaj = LM(π1(XiXj)) and XiXj ≺grlex XjXi, we see that XiXj ≺X XjXi. It follows that LM(gji) = XjXi for 1 ≤ i < j ≤ n. Now, by Proposition 1.4.14 we conclude that G forms a Grobner basis for I with respect to ≺X . Finally, by the definition of G, it is clear that G is the reduced Grobner basis of I, as desired. Solvable Polynomial Algebras 31 (ii) ⇒ (i) Note that (a) + (b) tells us that the generators of A satisfy the relations X jX i = λjiX iX j + F ji, 1 ≤ i < j ≤ n, and that if F ji 6= 0 then LM(F ji) ≺ X iX j with respect to the given monomial ordering ≺ on B. It follows that A and hence A is a solvable polynomial algebra in the sense of Definition 1.1.3. (cid:3) Remark The monomial ordering ≺X we defined in the proof of Theorem 1.4.16 is a modification of the lexicographic extension defined in [EPS]. But our definition of ≺X involves a graded monomial ordering ≺grlex on the standard K-basis B of the free K-algebra KhXi = KhX1, . . . , Xni. The reason is that the monomial ordering ≺X on B must be compatible with the usual rule of division, namely, W, U, V, S ∈ B, W 6= V , and W = U V S implies V ≺X W . While it is clear that if we use any lexicographic ordering ≺lex in the definition of ≺X , then this rule will not work in general. We end this section by an example illustrating Theorem 1.4.16, in particular, illustrating that the monomial ordering ≺X used in the condi- tion (a) and the monomial ordering ≺ used in the condition (b) may be mutually independent, namely ≺ may not necessarily be the restriction of ≺X on B, and the choice of ≺ is indeed quite flexible. Example (1) Considering the positive-degree function d( ) on the free K-algebra KhXi = KhX1, X2, X3i such that d(X1) = 2, d(X2) = 1, and d(X3) = 4, let I be the ideal of KhXi generated by the elements g1 = X1X2 − X2X1, g2 = X3X1 − λX1X3 − µX3X 2 g3 = X3X2 − X2X3, 2 − f (X2), where λ ∈ K ∗, µ ∈ K, f (X2) is a polynomial in X2 which has degree ≤ 6, or f (X2) = 0. The following properties hold. (1) If we use the graded lexicographic ordering X2 ≺grlex X1 ≺grlex X3 on KhXi, then the three generators have the leading monomials LM(g1) = X1X2, LM(g2) = X3X1, and LM(g3) = X3X2. It is an exercise to verify that G = {g1, g2, g3} forms a Grobner basis for I with respect to ≺grlex. (2) With respect to the fixed ≺grlex in (1), the reduced Grobner basis G′ 32 of I consists of Solvable Polynomial Algebras g1 = X1X2 − X2X1, g2 = X3X1 − λX1X3 − µX 2 g3 = X3X2 − X2X3, 2 X3 − f (X2), 1 aα3 3 (3) Writing A = K[a1, a2, a3] for the quotient algebra KhXi/I, where a1, a2 and a3 denote the cosets X1 + I, X2 + I and X3 + I in KhXi/I respectively, it follows that A has the PBW basis B = {aα = aα2 2 aα1 α = (α2, α1, α3) ∈ N3}. Noticing that a2a1 = a1a2, it is clear that B′ = {aα = aα1 α = (α1, α2, α3) ∈ N3} is also a PBW basis for A. Since a3a1 = λa1a3 + µa2 2a3 + f (a2), where f (a2) ∈ K- span{1, a2, a2 2}, we see that A has the monomial ordering ≺lex on B′ such that a3 ≺lex a2 ≺lex a1 and LM(µa2 2a3+f (a2)) ≺lex a1a3, thereby A is turned into a solvable polynomial algebra with respect to ≺lex. 2, . . . , a6 1 aα2 2 aα3 3 Moreover, it is also an exercise to check that if we use the positive- degree function d( ) on B′ such that d(a1) = 2, d(a2) = 1, and d(a3) = 4, then A has another monomial ordering on B′, namely the graded lexicographic ordering ≺grlex such that a3 ≺grlex a2 ≺grlex a1 and LM(µa2 2a3 + f (a2)) ≺grlex a1a3, thereby A is turned into a solvable polynomial algebra with respect to ≺grlex. 2. Left Grobner Bases for Modules Based on the theory of left Grobner bases for left ideals of solvable poly- nomial algebras presented in Chapter 1, in this chapter we introduce left Grobner bases for submodules of free left modules over solvable polynomial algebras. More precisely, let A = K[a1, . . . , an] be a solvable polynomial algebra with admissible system (B, ≺) as described in Chapter 1. In the first section, we define left monomial orderings, especially the graded left monomial orderings and Schreyer orderings, on free A-modules. In Sec- tion 2, we introduce left Grobner bases, minimal left Grobner bases and reduced left Grobner bases for submodules of free left A-modules via a left division algorithm, and we discuss some basic properties of left Grobner bases, in particular, we show that every submodule of a free left A-module L = ⊕s I=1Aei has a finite left Grobner basis. In Section 3, we define left S-polynomials for pairs of elements in free A-modules, and we show that a noncommutative version of Buchberger's criterion holds true for sub- modules of free A-modules, and that a noncommutative version of the Buchberger algorithm works effectively for computing finite left Grobner bases of submodules in free A-modules. The main references of this chapter are [AL2], [Eis], [KR1], [K-RW], [Li1], [Lev], [DGPS]. Throughout this chapter, modules are meant left A-modules, and all notions and notations used in Chapter 1 are maintained. 33 34 Grobner Bases for Modules 2.1. Left Monomial Orderings on Free Modules Let A = K[a1, . . . , an] be a solvable polynomial algebra with admissi- ble system (B, ≺) in the sense of Definition 1.1.3, where B = {aα = aα1 n α = (α1, . . . , αn) ∈ Nn} is the PBW K-basis of A and ≺ is a 1 · · · aαn monomial ordering on B, and let L = ⊕s i=1Aei be a free A-module with the A-basis {e1, . . . , es}. Then L has the K-basis B(e) = {aαei aα ∈ B, 1 ≤ i ≤ s}. For convenience, elements of B(e) are also referred to as monomials in L. If ≺e is a total ordering on B(e), and if ξ = Pm where λj ∈ K ∗ and α(j) = (αj1, . . . , αjn) ∈ Nn, such that j=1 λjaα(j)eij ∈ L, aα(1)ei1 ≺e aα(2)ei2 ≺e · · · ≺e aα(m)eim, then by LM(ξ) we denote the leading monomial aα(m)eim of ξ, by LC(ξ) we denote the leading coefficient λm of ξ, and by LT(ξ) we denote the leading term λmaα(m)eim of f . 2.1.1. Definition With respect to the given monomial ordering ≺ on B, a total ordering ≺e on B(e) is called a left monomial ordering if the following two conditions are satisfied: (1) aαei ≺e aβej implies LM(aγaαei) ≺e LM(aγaβej) for all aαei, aβej ∈ B(e), aγ ∈ B; (2) aβ ≺ aβ implies aαei ≺e aβei for all aα, aβ ∈ B and 1 ≤ i ≤ s. If ≺e is a left monomial ordering on B(e), then we also say that ≺e is a left monomial ordering on the free module L, and the data (B(e), ≺e) is referred to as an left admissible system of L. By referring to Proposition 1.1.4, we record two easy but useful facts on a left monomial ordering ≺e on B(e), as follows. 2.1.2. Lemma (i) Every left monomial ordering ≺e on B(e) is a well- ordering, i.e., every nonempty subset of B(e) has a minimal element. (ii) If f ∈ A with LM(f ) = aγ and ξ ∈ L with LM(ξ) = aαei, then LM(f ξ) = LM(LM(f )LM(ξ)) = LM(aγaαei) = aγ+αei. (cid:3) Grobner Bases for Modules 35 Actually as in the commutative case ([AL2], [Eis], [KR1]), any mono- mial ordering ≺ on B may induce two left monomial orderings on B(e): (TOP ordering) aαei ≺e aβej ⇔ aα ≺ aβ, or aα = aβ and i < j; (POT ordering) aαei ≺e aβej ⇔ i < j, or i = j and aα ≺ aβ. Let d( ) be a positive-degree function on A (see Section 1 of Chapter 1) such that d(ai) = mi > 0, 1 ≤ i ≤ n, and let (b1, . . . bs) ∈ Nn be any fixed s-tuple. Then, by assigning ej the degree bj, 1 ≤ j ≤ s, every monomial aαej in the K-basis B(e) of L is endowed with the degree d(aα) + bj. Similar to Definition 1.1.2, if a left monomial ordering ≺e on B(e) satisfies aαei ≺e aβej implies d(aα) + bi ≤ d(aβ) + bj, then we call it a graded left monomial ordering on B(e). We refer the reader to Chapter 4 and Chapter 5 for examples of graded left monomial orderings, also for the reason that a graded left monomial ordering is defined in such a way. Let ≺e be a left monomial ordering on B(e), and let L1 = ⊕m i=1Aεi be another free A-module with the A-basis {ε1, . . . , εm}. Then, as in the commutative case ([AL2], [Eis], [KR1]), for any given finite subset G = {g1, . . . , gm} ⊂ L, an ordering on the K-basis B(ε) = {aαεi aα ∈ B, 1 ≤ i ≤ m} of L1 can be defined subject to the rule: for aαεi, aβεj ∈ B(ε), aαεi ≺s-ε aβεj ⇔ LM(aαgi) ≺e LM(aβgj), or LM(aαgi) = LM(aβgj) and i < j, It is an exercise to check that this ordering is a left monomial ordering on B(ε). ≺s-ε is usually referred to as the Schreyer ordering induced by G with respect to ≺e. 2.2. Left Grobner Bases of Submodules Let A = K[a1, . . . , an] be a solvable polynomial algebra with admissible system (B, ≺), and let L = ⊕s i=1Aei be a free A-module with left admissi- ble system (B(e), ≺e). In this section we introduce left Grobner bases for 36 Grobner Bases for Modules submodules of L via a left division algorithm in L, and we record some basic facts determined by left Grobner bases. Moreover, minimal left Grobner bases and reduced left Grobner bases are discussed. Finally, we show that every nonzero submodule of L has a finite left Grobner basis. Left division algorithm Let aαei, aβej ∈ B(e), where α = (α1, . . . , αn), β = (β1, . . . , βn) ∈ Nn. Then, the left division of monomials in B we introduced in (Section 2 of Chapter 1) gives rise to a left division for monomials in B(e), that is, we say that aαei divides aβej from left side, denoted aαeiLaβej, if i = j and there is some aγ ∈ B such that aβei = LM(aγaαei). It follows from Lemma 2.1.2(ii) that the division defined above is imple- mentable. Let Ξ be a nonempty subset of L. Then the division of monomials defined above yields a subset of B(e): N (Ξ) = {aαei ∈ B(e) LM(ξ) 6 L aαei, ξ ∈ Ξ}. If Ξ = {ξ} consists of a single element ξ, then we simply write N (ξ) in place of N (Ξ). Also we write K-spanN (Ξ) for the K-subspace of L spanned by N (Ξ). 2.2.1. Definition Elements of N (Ξ) are referred to as normal monomials (mod Ξ). Elements of K-spanN Ξ) are referred to as normal elements (mod Ξ). In view of Lemma 2.1.2(ii), the left division we defined for mono- mials in B(e) can naturally be used to define a left division procedure for elements in L. More precisely, let ξ, ζ ∈ L with LC(ξ) = µ 6= 0, LC(ζ) = λ 6= 0. If LM(ζ)L LM(ξ), i.e., there exists aα ∈ B such that LM(ξ) = LM(aαLM(ζ)), then put ξ1 = ξ − λ−1µaαζ; otherwise, put ξ1 = ξ − LT(ξ). Note that in both cases we have ξ1 = 0, or ξ1 6= 0 and LM(ξ1) ≺e LM(ξ). At this stage, let us refer to such a procedure of canceling the leading term of ξ as the left division procedure by ζ. With Grobner Bases for Modules 37 ξ := ξ1 6= 0, we can repeat the left division procedure by ζ and so on. This returns successively a descending sequence LM(ξi+1) ≺e LM(ξi) ≺e · · · ≺e LM(ξ1) ≺e LM(ξ), i ≥ 2. Since ≺e is a well-ordering, it follows that such a division procedure termi- nates after a finite number of repetitions, and consequently ξ is expressed as ξ = qζ + η, where q ∈ A and η ∈ K-spanN (ζ), i.e., η is normal (mod ζ), such that either LM(ξ) = LM(qζ) or LM(ξ) = LM(η). Actually as in (Section 2 of Chapter 1), the left division procedure demonstrated above can be extended to a left division procedure by a finite subset Ξ = {ξ1, . . . , ξs} in L, which yields the following left division algorithm: Algorithm-DIV-L INPUT: ξ, Ξ = {ξ1, . . . , ξs} with ξi 6= 0 (1 ≤ i ≤ s) OUTPUT: q1, . . . , qs ∈ A, η ∈ K-spanN (Ξ), such that ξ =Ps i=1 qiξi + η, LM(qiξi) (cid:22)e LM(ξ) for qi 6= 0, LM(η) (cid:22)e LM(ξ) if η 6= 0 INITIALIZATION: q1 := 0, q2 := 0, BEGIN · · · , qs := 0; η := 0; ω := ξ WHILE ω 6= 0 DO IF there exist i and aα(i) ∈ B such that LM(ω) = LM(aα(i)LM(ξi)) THEN choose i least such that LM(ω) = LM(aα(i)LM(ξi)) qi := qi + LC(ξi)−1LC(ω)aα(i) ω := ω − LC(ξi)−1LC(ω)aα(i)ξi ELSE η := η + LT(ω) ω := ω − LT(ω) END END END 38 Grobner Bases for Modules 2.2.2. Definition The element η obtained in Algorithm-DIV-L is called a remainder of ξ on left division by Ξ, and is denoted by ξ , i.e., Ξ ξ = 0, then we say that ξ is reduced to 0 (mod Ξ). Ξ = η. If ξ Ξ Remark For the reason that the element η obtained in Algorithm- DIV-L depends on how an order of elements in Ξ is arranged, we used the phrase "a remainder" in the above definition. In other words, the element η may be different if a different order for elements in Ξ is given. Summing up, we have reached the following 2.2.3. Theorem Given a set of nonzero elements Ξ = {ξ1, . . . , ξs} and ξ in L, the Algorithm-DIV-L produces elements q1, . . . , qs ∈ A and i=1 qiξi + η and LM(qiξi) (cid:22)e LM(ξ) (cid:3) η ∈ K-spanN (Ξ), such that ξ =Ps whenever qi 6= 0, LM(η) (cid:22)e LM(ξ) if η 6= 0. Left Grobner bases It is theoretically correct that the left division procedure by using a finite subset Ξ of nonzero elements in L can be extended to a left division procedure by means of an arbitrary proper subset G of nonzero elements, for, it is a true statement that if ξ ∈ L with LM(ξ) 6= 0, then either there exists g ∈ G such that LM(g)L LM(ξ) or such g does not exist. This leads to the following 2.2.4. Proposition Let N be a submodule of L and G a proper subset of nonzero elements in N . The following two statements are equivalent: (i) If ξ ∈ N and ξ 6= 0, then there exists g ∈ G such that LM(g)L LM(ξ). j=1 qigij with qi ∈ A (cid:3) (ii) Every nonzero ξ ∈ N has a representation ξ =Ps and gij ∈ G, such that LM(qigij ) (cid:22)e LM(ξ) whenever qi 6= 0. 2.2.5. Definition Let N be a submodule of L. With respect to a given left monomial ordering ≺e on B(e), a proper subset G of nonzero elements in N is said to be a left Grobner basis of N if G satisfies one of the equivalent conditions of Proposition 2.2.4. If G is a left Grobner basis of N , then the expression ξ =Ps j=1 qigij appeared in Proposition 2.2.4(ii) is called a left Grobner representation of ξ. Grobner Bases for Modules 39 Clearly, if N is a submodule L and N has a left Grobner basis G, then G is certainly a generating set of N , i.e., N =Pg∈G Ag. But the converse is not necessarily true. For instance, consider the solvable polynomial algebra A = C[a1, a2, a3] generated by {a1, a2, a3} subject to the relations a2a1 = 3a1a2, a3a1 = a1a3, a3a2 = 5a2a3, and let L = Ae1 ⊕ Ae2 ⊕ Ae3 be the free A-module with the A-basis {e1, e2, e3}. Then, with g1 = a2 1a2e1−a3e3, g2 = a2e1, and N = Ag1+Ag2, we have a3ee ∈ N . Hence the set G = {ξ1, g2} is not a left Grobner basis with respect to any given monomial ordering ≺e on B(e) such that LM(g1) = a2 1a2e1, LM(g2) = a2e1. The existence of left Grobner bases Noticing 1 ∈ B, if we define on B(e) the ordering: aαei ≺′ e aβej ⇔ aαeiLaβej, aαei, aβej in B(e), then, by Definition 2.1.1, Lemma 2.1.2(ii) and the divisibility defined for monomials in B(e), it is an easy exercise to check that ≺′ is reflexive, anti-symmetric, transitive, and moreover, aαei ≺′ e aβej implies aαei ≺e aβej. Since the given left monomial ordering ≺e is a well-ordering on B(e), it follows that every nonempty subset of B(e) has a minimal element with respect to the ordering ≺′ e on B(e). Now, let N 6= {0} be a submodule of L, LM(N ) = {LM(ξ) ξ ∈ N }, and Ω = {aαei ∈ LM(N ) aαei is minimal in LM(N ) w.r.t. ≺′ e}. 2.2.6. Theorem With the notation as above, the following statements hold. (i) Ω 6= ∅ and Ω is a proper subset of LM(N ). (ii) Let G = {g ∈ N LM(g) ∈ Ω}. Then G is a left Grobner basis of N . Proof (i) That Ω 6= ∅ follows from N 6= {0} and the remark about ≺′ e we made above. Since N is a submodule of L and A is a domain (Proposition 1.1.4), if ξ ∈ N and ξ 6= 0, then hξ ∈ N for all nonzero h ∈ A. It follows from Lemma 2.1.2(ii) that 0 6= LM(hξ) ∈ LM(N ) and LM(ξ)L LM(hξ), 40 Grobner Bases for Modules i.e., LM(ξ) ≺′ LM(hξ) 6= LM(ξ). This shows that Ω is a proper subset of LM(N ). e LM(hξ) in LM(N ). It is clear that if LM(h) 6= 1, then (ii) By the definition of ≺′ e and the remark we made above the theo- rem, if ξ ∈ N with LM(ξ) 6= 0 and LM(ξ) 6∈ Ω, then there is some g ∈ G such that LM(g)L LM(ξ). Hence the selected G is a left Grobner basis for N . (cid:3) Since every solvable polynomial algebra A is (left and right) Noethe- rian by Corollary 1.3.3, the free module L = ⊕s i=1Aei is a Noetherian left A-module, thereby every submodule N of L is finitely generated. In the next section we will show that if a finite generating set Ξ = {ξ1, . . . , ξt} of N is given, then a finite left Grobner basis G for N can be produced by means of a noncommutative Buchberger algorithm. Basic facts determined by left Grobner bases By referring to Lemma 2.1.2 Theorem 2.2.3 and Proposition 2.2.4, the foregoing discussion allows us to summarize some basic facts determined by left Grobner bases, of which the detailed proof is left as an exercise. 2.2.7. Proposition Let N be a submodule of the free A-module L = ⊕s i=1Aei, and let G be a left Grobner basis of N . Then the following statements hold. (i) If ξ ∈ N and ξ 6= 0, then LM(ξ) = LM(qg) for some q ∈ A and g ∈ G. (ii) If ξ ∈ L and ξ 6= 0, then ξ has a unique remainder ξ on division by G. (iii) If ξ ∈ L and ξ 6= 0, then ξ ∈ N if and only if ξ = 0. Hence the membership problem for submodules of L can be solved by using left Grobner bases. (iv) As a K-vector space, L has the decomposition G G L = N ⊕ K-spanN (G). (v) As a K-vector space, L/N has the K-basis N (G) = {aαei aαei ∈ N (G)}, where aαei denotes the coset represented by aαei in L/N . (vi) N (G) is a finite set, or equivalently, DimK A/N < ∞, if and only if for j = 1, . . . , n and each i = 1, . . . , s, there exists gji ∈ G such that LM(gji) = amj (cid:3) j ei, where mj ∈ N. Grobner Bases for Modules 41 Minimal and reduced left Grobner bases 2.2.8. Definition Let N 6= {0} be a submodule of L and let G be a left grobner basis of N . If any proper subset of G cannot be a left Grobner basis of N , then G is called a minimal left Grobner basis of N . 2.2.9. Proposition Let N 6= {0} be a submodule of L. A left Grobner basis G of N is minimal if and only if LM(gi)6 L LM(gj) for all gi, gj ∈ G with gi 6= gj. Proof Suppose that G is minimal. If there were gi 6= gj in G such that LM(gi)L LM(gj ), then since the left division is transitive, the proper subset G′ = G − {gj} of G would form a left Grobner basis of N . This contradicts the minimality of G. Conversely, if the condition LM(gi)6 L LM(gj) holds for all gi, gj ∈ G with gi 6= gj, then the definition of a left Grobner basis entails that any proper subset of G cannot be a left Grobner basis of N . This shows that G is minimal. (cid:3) 2.2.10. Corollary Let N 6= {0} be a submodule of L, and let the notation be as in Theorem 2.2.6. (i) The left Grobner basis G obtained there is indeed a minimal left Grobner basis for N . (ii) If G is any left Grobner basis of N , then Ω = LM(G) = {LM(g) g ∈ G} ⊆ LM(G) = {LM(g′) g′ ∈ G}. Therefore, any two minimal left Grobner bases of N have the same set of leading monomials Ω. Proof This follows immediately from the definition of a left Grobner basis, the construction of Ω and Proposition 2.2.9. (cid:3) We next introduce the notion of a reduced left Grobner basis. 2.2.11. Definition Let N 6= {0} be a submodule of L and let G be a left grobner basis of N . If G satisfies the following conditions: (1) G is a minimal left Grobner basis; (2) LC(g) = 1 for all g ∈ G; (3) For every g ∈ G, ξ = g − LM(g) is a normal element (mod G), i.e., ξ ∈ K-spanN (G), 42 Grobner Bases for Modules then G is called a reduced left Grobner basis of N . 2.2.12. Proposition Every nonzero submodule N of L has a unique reduced Grobner basis. Proof Note that if G and G′ are reduced left Grobner bases for N , then LM(G) = LM(G′) by Corollary 2.2.10. If LM(gi) = LM(g′ j) then gi − g′ j ∈ N ∩ K-spanN (G) = N ∩ K-spanN (G′). It follows from Proposition 2.2.7(iv) that gi = g′ (cid:3) j. Hence G = G′. The next proposition shows that if a finite left Grobner basis G of N is given, then a minimal left Grobner basis, thereby the reduced left Grobner basis for N can be obtained in an algorithmic way. 2.2.13. Proposition Let N 6= {0} be a submodule of L, and let G = {g1, . . . , gm} be a finite left Grobner basis of N . (i) The subset G0 = {gi ∈ G LM(gi) is minimal in LM(G) w.r.t. ≺′ e} of G forms a minimal left Grobner basis of N (see the definition of ≺′ e given before Theorem 2.2.6). An algorithm written in pseudo-code is omitted here. (ii) With G0 as in (i) above, we may assume, without loss of gener- ality, that G0 = {g1, . . . , gs} with LC(gi) = 1 for 1 ≤ i ≤ s. Put G1. Then LM(ξ1) = LM(g1). Put G1 = {g2, ..., gs} and ξ1 = g1 G2. Then LM(ξ2) = LM(g2). Put G2 = {ξ1, g3, ..., gs} and ξ2 = g2 G3, and so on. The last obtained G3 = {ξ1, ξ2, g4, ..., gs} and ξ3 = g3 Gs+1 = {ξ1, . . . , ξs−1} ∪ {ξs} is then the reduced left Grobner basis. An algorithm written in pseud0-code is omitted here. Proof This can be verified directly, so we leave it as an exercise. (cid:3) The existence of finite left Grobner bases Finally we show that every nonzero submodule N of the free A-module L = ⊕s i=1Aei has a finite left Grobner basis G with respect to a given left monomial ordering ≺e. Let aαei, aβej ∈ B(e) with α = (α1, . . . , αn), β = (β1, . . . , βn) ∈ Nn. Recall that aαei ≺′ e aβ if and only if i = j and aαL aβ, while the latter is defined subject to the property that aβ = LM(aγaα) for some aγ ∈ B. Grobner Bases for Modules 43 Thus, by Lemma 2.1.2(ii), aαei ≺′ e aβej is actually equivalent to i = j and αi ≤ βi for 1 ≤ i ≤ n. Also note that the correspondence α = (α1, . . . , αn) ←→ aα gives the bijection between Nn and B. Combining this observation with Dickson's lemma (Lemma 1.3.1), we are able to reach the following result. 2.2.14. Theorem With notations as in Theorem 2.2.6, let N 6= {0} be a submodule of L, LM(N ) = {LM(ξ) ξ ∈ N }, Ω = {aαei ∈ LM(N ) aαei e}, and G = {g ∈ N LM(g) ∈ Ω}. Then Ω is a finite subset of LM(N ), thereby G is a finite left Grobner basis of N . is minimal in LM(N ) w.r.t. ≺′ Proof For 1 ≤ i ≤ s, put B(e)i = {aαei aα ∈ B}, Ωi = Ω ∩ B(e)i. Then Ω = ∪s i=1Ωi. Applying Dickson's lemma (Lemma 1.3.1) to Ωi, it turns out that each Ωi has a finite subset Ω′ i = {aα(1i)ei, . . . , aα(ti )ei} such that if aβei ∈ Ωi, then aα(ji)eiLaβei for some aα(ji) ∈ Ω′ i. But this implies aα(ji)ei ≺′ i=1Ω′ i is a finite subset of LM(N ), and consequently G is a finite (minimal) left Grobner basis of N . (cid:3) e aβei in Ω. It follows from the definition of Ω that Ω = ∪s 2.3. The Noncommutative Buchberger Algorithm Recall from the theory of Grobner bases for commutative polynomial algebras ([Bu1], [Bu2], [AL2], [BW], [Frob], [KR1], [KR2]) that the cel- ebrated Buchberger algorithm depends on Buchberger's criterion which establishes the strategy for computing Grobner bases of polynomial ideals. Let A = K[a1, . . . , an] be a solvable polynomial algebra with admissible system (B, ≺), and let L = ⊕s i=1Aei be a free A-module with left admissi- ble system (B(e), ≺e). Then by Theorem 2.2.14, every nonzero submodule N has a finite left Grobner basis G. In this section, by introducing left S-polynomials for elements (ξ, ζ) ∈ L × L, we show that a noncommu- tative version of Buchberger's criterion holds true for submodules of L, and that a noncommutative version of the Buchberger algorithm works effectively for computing finite left Grobner bases of submodules in L. Since the noncommutative version of Buchberger's criterion and the noncommutative version of Buchberger's algorithm for modules over a 44 Grobner Bases for Modules solvable polynomial algebra A (Theorem 2.3.3 and Algorithm-LGB pre- sented below) look as if working the same way as in the commutative case by reducing the S-polynomials, at this stage one is asked to pay more at- tention to compare the argumentation concerning Buchberger's criterion in the commutative case (e.g. [AL2], P.40-42) and that in the noncom- mutative case, given in [K-RW] and we are going to give respectively, so as to see how the barrier made by the noncommutativity of A (i.e., the product of two monomials in A is no longer necessarily a monomial of A) can be broken down. Also, at this point we remind that in the proof of ([K-RW], Theorem 3.11), the argumentation was given in the language of abstract rewriting (see [Wik2] for an introduction of this topic), namely, it was shown that the left reduction of left S-polynomials by G is locally confluent; while in the argumentation we are going to give below, Lemma 2.3.1 will play the key role as the breakthrough point, though our pre- sentation looks quite similar to the most popularly known presentation in the commutative case (e.g. [AL2]). Let ξ, ζ be nonzero elements of L with LM(ξ) = aαei, LM(ζ) = aβej, where α = (α1, . . . , αn), β = (β1, . . . , βn). Put γ = (γ1, . . . , γn) with γk = max{αk, βk). The left S-polynomial of ξ and ζ is defined as the element Sℓ(ξ, ζ) = 1 LC(aγ−αξ) 0, aγ−αξ − 1 LC(aγ−βζ) aγ−βζ, if i = j if i 6= j. Observation If Sℓ(ξ, ζ) 6= 0 then, with respect to the given ≺e on B(e) we have LM(Sℓ(ξ, ζ)) ≺e aγei. 2.3.1. Lemma Let ξ, ζ be as above such that Sℓ(ξ, ζ) 6= 0, and put λ = LC(aγ−αξ), µ = LC(aγ−βζ). If there exist aθ(1), aθ(2) ∈ B such that LC(aθ(1)ξ) = λ1, LC(aθ(2)ζ) = µ1, LM(aθ(1)ξ) = aρei = LM(aθ(2)ζ), where ρ = (ρ1, . . . , ρn), then there exists aδ ∈ B such that Sℓ(aθ(1)ξ, aθ(2)ζ) = b(cid:16)aδSℓ(ξ, ζ) − daθ(2)ζ − f1ξ − f2ζ(cid:17) , where b = λ µ , LM(aδSℓ(ξ, ζ)) ≺e aρei, LM(f1ξ) ≺e aρei, and LM(f2ζ) ≺e aρei, λρ,α, µρ,β ∈ K ∗, f1, f2 ∈ A, which are given , d = λρ,αλ1 λµ1 λρ,αλ1 − µρ,β Grobner Bases for Modules 45 in terms of aρ−γaγ−α = λρ,αaθ(1) + f1 with f1 ∈ A and LM(f1) ≺ aθ(1); aρ−γaγ−β = µρ,βaθ(2) + f2 with f2 ∈ A and LM(f2) ≺ aθ(2). Proof By the assumption and Lemma 2.1.2, θ(1)+α = ρ = θ(2)+β. Since γ = (γ1, . . . , γn) with γk = max{αk, βk), we may put δ = (δ1, . . . , δn) with δi = ρi − γi, i.e., δ = ρ − γ. Now we see that aδSℓ(ξ, ζ) = 1 = 1 λ aδaγ−αξ − 1 λ aρ−γaγ−αξ − 1 µ aδaγ−βζ µ aρ−γaγ−βζ λ λ1 = λρ,αλ1 µ aθ(2)ζ + f2ζ(cid:17) , aθ(1)ξ − 1 − µρ,β λ aθ(1)ξ + f1ξ(cid:17) −(cid:16) µρ,β = (cid:16) λρ,α (cid:16) 1 µ aθ(2)ζ(cid:17) +(cid:16) λρ,αλ1 +(cid:16) λρ,αλ1 µ (cid:17) aθ(2)ζ + f1ξ + f2ζ µ (cid:17) aθ(2)ζ + f1ξ + f2ζ, λ Sℓ(aθ(1)ξ, aθ(2)ζ) = λρ,αλ1 − µρ,β λµ1 λµ1 in which LM(aδSℓ(ξ, ζ)) ≺e aρei, LM(f1ξ) ≺e aρei, and LM(f2ζ) ≺e aρei. (cid:3) 2.3.2. Lemma Let ξ1, ..., ξt ∈ L be such that LM(ξi) = aρeq for all i = 1, ..., t. i=1 ciξi satisfies LM(ξ) ≺e aρeq, then ξ can be expressed as a K-linear combination of the form If, for c1, . . . , ct ∈ K ∗, the element ξ = Pt ξ = d12Sℓ(ξ1, ξ2) + d23Sℓ(ξ2, ξ3) + · · · + dt−1, t(ξt−1, ξt), where dij ∈ K. Proof If we rewrite each ξi as ξi = λiaρeq+ lower terms, where λi = LC(ξi), then the assumption yields a cancelation of leading terms which i=1 ciλi = 0. Since LM(ξi) = aρeq = LM(ξj), it follows that Sℓ(ξi, ξj) = 1 λi gives rise to Ps ξi − 1 λj ξj. Thus ξ = c1ξ1 + · · · + csξt = c1λ1(cid:18) 1 = c1λ1(cid:18) 1 λ1 λ1 ξ1 − ξ1(cid:19) + · · · + ctλt(cid:18) 1 λt ξt(cid:19) 1 λ2 ξ2(cid:19) + (c1λ1 + c2λ2)(cid:18) 1 λ2 ξ2 − 1 λ3 ξ3(cid:19) 46 Grobner Bases for Modules + · · · + (c1λ1 + · · · + ct−1λt−1)(cid:18) 1 λt−1 ξt−1 − 1 λt ξt(cid:19) +(c1λ1 + · · · + ctλt) ξt 1 λt = c1λ1Sℓ(ξ1, ξ2) + (c1λ1 + c2λ2)Sℓ(ξ2, ξ3) + · · · +(c1λ1 + · · · + ct−1λt−1)Sℓ(ξt−1, ξt), {ξ1, . . . , ξm} ⊂ L, it is clear that Sℓ(ξi, ξj) ∈ N for 1 ≤ i < j ≤ m. i=1 Aξi of L generated by Ξ = i=1 ciλi = 0. This completes the proof. (cid:3) becausePt Considering the submodule N = Pm Let N =Pm 2.3.3. Theorem (Noncommutative version of Buchberger's criterion) i=1 Aξi be a submodule of L generated by the set of nonzero elements Ξ = {ξ1, . . . , ξs}. Then, with respect to the given left monomial ordering ≺e on B(e), Ξ is a left Grobner basis of N if and only if every Ξ nonzero Sℓ(ξi, ξj) is reduced to 0 (mod Ξ), i.e., Sℓ(ξi, ξj) = 0. Proof Note that Sℓ(ξi, ξj) ∈ N . If Ξ is a left Grobner basis of N , then Ξ Sℓ(ξi, ξj) = 0 whenever Sℓ(ξi, ξj) 6= 0. Ξ Conversely, suppose that Sℓ(ξi, ξj) = 0 for every nonzero Sℓ(ξi, ξj). We will show that Ξ satisfies Proposition 2.2.4(ii), or in other words, that every nonzero element of N has a left Grobner representation by Ξ. For this purpose, we argue by contradiction. Suppose that there were a i=1 fiξi ∈ N with nonzero fi ∈ A, such that LM(ξ) ≺e LM(fjξj) for some j ≤ s. Let LM(fj) = aθ(j) and LM(ξj) = aα(j)eij . Comparing i=1 fiξi in terms of base elements in B(e), we may assume, without loss of generality, that for some 2 ≤ t ≤ s, ξ =Ps the linear expressions of both sides of ξ = Ps aρeq = LM(f1ξ1) = LM(f2ξ2) = · · · = LM(ftξt) = max{LM(f1ξ1), . . . , LM(fsξs)}, (1) and moreover, we assume that the representation ξ = Ps where ci = LC(fi), and let η0 := Pt i=1 fiξi is cho- sen so that aρeq is the least one. We now proceed to produce a new representation of ξ by elements of Ξ in which the maximal monomial is strictly less than aρeq. To this end, we write fi = ciaθ(i)+ lower terms, i=1 ciaθ(i)ξi, ξ∗ = ξ − η0. Then LM(η0) ≺e aρeq, LM(ξ∗) ≺e aρeq. With λi = LC(aθ(i)ξi), 1 ≤ i ≤ t, we Grobner Bases for Modules may apply Lemma 2.3.2 to η0 so that η0 = d12Sℓ(aθ(1)ξ1, aθ(2)ξ2) + d23Sℓ(aθ(2)ξ2, aθ(3)ξ3) + · · · +dtt−1,tSℓ(aθ(t−1)ξt−1, aθ(t)ξt), 47 (2) where dij ∈ K and Sℓ(aθ(i)ξi, aθ(j)ξj) = 1 λi apply Lemma 2.3.1 to each Sℓ(aθ(i)ξi, aθ(j)ξj) so that aθ(i)ξi − 1 λj aθ(j)ξj. We next Sℓ(aθ(i)ξi, aθ(j)ξj) = bij(cid:16)aδ(ij)Sℓ(ξi, ξj) − zijaθ(j)ξj − hiξi − hjξj(cid:17) , (3) where bij, zij ∈ K, hi, hj ∈ A, LM(aδ(ij)Sℓ(ξi, ξj)) ≺e aρeq, LM(hiξi) ≺e aρeq, and LM(hjξj) ≺e aρeq, thereby LM(Sℓ(aθ(i)ξi, aθ(j)ξj)) = LM(aθ(j)ξj) = aρeq if bij, zij 6= 0. (4) So, if we take the sub-sum η1 :=Pt result of (3) which gives rise to a sub-sum η2 :=Pt Note that j = i + 1 > i in (3). After substituting (3) into (2), and η0 into ξ, we see that aθ(1)ξ1 does not appear in the new representation of ξ. j=2 dijbijzijaθ(j)ξj of ξ, then η1 clearly satisfies LM(η1) ≺e aρeq = LM(aθ(j)ξj), 2 ≤ j ≤ t. Hence we may apply Lemma 2.3.1 and Lemma 2.3.2 to η1 so that we obtain a similar k=3 ukaθ(k)ξk of ξ with uk ∈ K and LM(η2) ≺e aρeq = LM(aθ(k)ξk), 3 ≤ k ≤ t. Repeating such a procedure for at most t − 1 times and at each time, applying the assumption Sℓ(ξi, ξj) = 0 to (3), we eventually obtain a representation of ξ by elements of Ξ which yields the desired contradiction and finishes the proof. (cid:3) Ξ rithm) Let N =Pm 2.3.4. Theorem (The Noncommutative version of Buchberger algo- i=1 Aξi be a submodule of L generated by a finite set of nonzero elements Ξ = {ξ1, . . . , ξs}. Then, with respect to a given left monomial ordering ≺e on B(e), the algorithm presented below returns a finite left Grobner basis G for N . Algorithm-LGB OUTPUT: G = {g1, ..., gt}, a left Grobner basis for N =Pm INITIALIZATION: m′ := m, G := {g1 = ξ1, . . . , gm′ = ξm}, S := {Sℓ(gi, gj) gi, gj ∈ G, i < j } − {0} i=1 Aξi INPUT: Ξ = {ξ1, ..., ξm} BEGIN 48 Grobner Bases for Modules WHILE S 6= ∅ DO Choose any Sℓ(gi, gj ) from S S := S − {Sℓ(gi, gj)} = η 6= 0 with LM(η) = aρek THEN G IF Sℓ(gi, gj ) m′ := m′ + 1, gm′ := η S := S ∪ {Sℓ(gj, gm′ ) gj ∈ G, LM(gj) = aν ek} − {0} G := G ∪ {gm′ }, END END END Proof We first prove that the algorithm terminates after a finite number of executing the WHILE loop. To this end, let Gn+1 denote the new set obtained after the n-th turn of executing the WHILE loop, that is, Gn+1 = Gn ∪nη = Sℓ(gi, gj) Gn 6= 0 for some pair gi, gj ∈ Gno , n ∈ N, where G0 = Ξ , and let G = [n∈N Gn, LM(G) = {LM(g) g ∈ G}. By Dickson's lemma (Lemma 1.3.1), LM(G) has a finite subset U = {LM(g1), . . . , LM(gs)}, such that if g ∈ G then LM(gi)LM(g) for some LM(gi) ∈ U . Since U is finite, we may assume that U ⊂ LM(Gk) for Gk = 0 for all gi, gj ∈ Gk, thereby the some k. This shows that Sℓ(gi, gj) algorithm terminates after the k-th turn of executing the WHILE loop. Now that the algorithm terminates in a finite number of executions, we may assume that G = {g1, . . . , gt}. Then, since G0 = Ξ ⊂ G, the algorithm itself tells us that G generates the submodule N and Sℓ(gi, gj) = 0 for all gi, gj ∈ G. It follows from Theorem 2.3.3 that G is a left Grobner basis for N . (cid:3) G One is referred to the up-to-date computer algebra systems Singu- lar [DGPS] for the implementation of Algorithm-LGB. Also, nowa- days there have been optimized algorithms, such as the signature-based algorithm for computing Grobner bases in solvable polynomial algebras [SWMZ], which is based on the celebrated F5 algorithm [Fau], may be used to speed-up the computation of left Grobner bases for modules. Grobner Bases for Modules 49 2.3.5. Corollary Let N =Pm i=1 Aξi be a submodule of L generated by a finite set of nonzero elements Ξ = {ξ1, . . . , ξm}, and let G = {g1, ..., gt} be a left Grobner basis of N produced by running Algorithm-LGB. Then the following statements hold true. (i) All the properties listed in Proposition 2.2.7 can be recognized in a computational way. (ii) There is a t × m matrix which is obtained after G is computed by running Algorithm-LGB, such that Vt×m = G = g1 ... gt h11 · · · ht1 · · · h1m · · · · · · · · · htm  = Vt×m ξ1 ... ξm hij ∈ A  ,  , 1 ≤ j ≤ s. (Note that for convenience, we formally used the matrix expression to demonstrate an obvious meaning.) (iii) If ξ ∈ N and ξ 6= 0, then a representation of ξ by Ξ = {ξ1, . . . , ξm}, say ξ =Pm i=1 fiξi, can be computed. Proof (i) Now that a left Grobner basis G of N has been computed, all the properties listed in Proposition 2.2.7 can be recognized by means of the division by G. (ii) Recall from Algorithm-LGB that for each gj ∈ G, either gj ∈ Ξ = {ξ1, ..., ξm} or gj is a newly added member of G obtained after a certain pass through the WHILE loop. So, starting with the expressions obtained from the first pass through the WHILE loop: Sℓ(ξi, ξj) =Xk Ξ hkξk + Sℓ(ξi, ξj) , hk ∈ A, 1 ≤ i < j ≤ m, and keeping track of the linear combinations that give rise to the new elements of G, the algorithm eventually yields the desired matrix Vt×m. (iii) By Proposition 2.2.7(iii), after dividing by G the element ξ has a j=1 sjgj with sj ∈ A and gj ∈ G. Now the conclusion (ii) yields the desired representation of ξ by Ξ. representation ξ =Pt 3. Finite Free Resolutions Since every solvable polynomial algebra A is (left and right) Noetherian (Section 3 of Chapter 1), and every nonzero submodule of a free left A- module has a finite left Grobner basis which can be produced by running Algorithm-LGB (Section 3 of Chapter 2), starting from this chapter we shall successively present some details concerning applications of Grobner bases in constructing finite free resolutions over an arbitrary solvable polynomial algebra A, minimal finite N-graded free resolutions over an N- graded solvable polynomial algebra A with the degree-0 homogenous part being the ground field K, and minimal finite N-filtered free resolutions over an N-filtered solvable polynomial algebra A (where the N-filtration of A is determined by a positive-degree function). Let A be a solvable polynomial algebra as before. The current chap- ter consists of four sections. In Section 1 we demonstrate, for a finitely generated submodule N of a free left A-module L, how to compute a generating set of the syzygy module Syz(N ) of N via computing a left Grobner basis G of N . In Section 2 we show, in a constructive way, that a noncommutative version of Hilbert's syzygy theorem holds true for A, and consequently, that a finite free resolution can be algorithmically con- structed for every finitely generated A-module. In Section 3, the non- commutative version of Hilbert's syzygy theorem is applied to highlight two homological properties of A, that is, A has finite global homologi- cal dimension, and every finitely generated projective A-module is stably free. Based on Section 1 -- Section 3, the final Section 4 is devoted to the 51 52 Finite Free Resolutions calculation of projective dimension of a finitely generated A-module M , and meanwhile, the proposed algorithmic procedure verifies whether M is a projective module or not. The main references of this chapter are [AL2], [AF], [Rot], [Li1], [GV], [Lev], [DGPS]. Throughout this chapter, modules are meant left modules over solv- able polynomial algebras, and all notions and notations used in previous chapters are maintained. 3.1. Computation of Syzygies Let A = K[a1, . . . , an] be a solvable polynomial algebra with admissi- ble system (B, ≺) in the sense of Definition 1.1.3, where B = {aα = aα1 n α = (α1, . . . , αn) ∈ Nn} is the PBW K-basis of A and ≺ is 1 · · · aαn a monomial ordering on B. Let L0 = ⊕s i=1Aei be a free left A-module with A-basis {e1, . . . , es}, and ≺e a left monomial ordering on the K-basis B(e) = {aαei aα ∈ B, 1 ≤ i ≤ s} of L0. As in (Section 3 of Chapter 2) we write Sℓ(ξ, ζ) for the left S-polynomial of two elements ξ, ζ ∈ L0, that is, if LM(ξ) = aαei with α = (α1, . . . , αn), LM(ζ) = aβej with β = (β1, . . . , βn), then 1 LC(aγ−αξ) 0, aγ−αξ − 1 LC(aγ−βζ) aγ−βζ, if i = j if i 6= j where γ = (γ1, . . . , γn) with γk = max{αk, βk); moreover, if Sℓ(ξ, ζ) 6= 0, then with respect to the given ≺e on B(e) we have LM(Sℓ(ξ, ζ)) ≺e aγei. i=1 Aξi be a submodule of L0 generated by the set of nonzero elements U = {ξ1, . . . , ξm}, and let L1 = ⊕m i=1Aωi be the free A-module with A-basis {ω1, . . . , ωm}. Then the syzygy module of U (or equivalently the syzygy module of N ), denoted Syz(U ), is the submodule of L1 defined by Sℓ(ξ, ζ) = Let N = Pm Syz(U ) =( mXi=1 hiωi ∈ L1 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12) hiξi = 0) . mXi=1 Our aim of this section is to show that a generating set of the syzygy module Syz(U ) can be computed by means of a left Grobner basis of N . To this end, let G = {g1, . . . , gt} be a left Grobner basis of N with respect Finite Free Resolutions 53 representation Sℓ(gi, gj) = Pt to ≺e, then every nonzero left S-polynomial Sℓ(gi, gj) has a left Grobner i=1 figi with LM(figi) (cid:22)e LM(Sℓ(gi, gj)) whenever fi 6= 0 (note that such a representation is obtained by using the division by G during executing the WHILE loop of Algorithm-LGB presented in (Chapter 2, Theorem 2.3.4)). Considering the syzygy module Syz(G) of G in the free A-module L2 = ⊕t i=1Aεi with A-basis {ε1, . . . , εt}, if we put sij = f1ε1 + · · · +(cid:16)fi − aγ−α(i) +(cid:16)fj + LC(aγ−α(i)ξi)(cid:17) εi + · · · LC(aγ−α(j)ξj )(cid:17) εj + · · · + ftεt, aγ−α(j) S = {sij 1 ≤ i < j ≤ t}, [AL2], then it can be shown, actually as in the commutative case (cf. Theorem 3.7.3), that S generates Szy(G) in L2. However, by employing the Schreyer ordering ≺s-ε on the K-basis B(ε) = {aαεi aα ∈ B, 1 ≤ i ≤ m} of L2 induced by G with respect to ≺e (see Section 1 of Chapter 2), which is defined subject to the rule: for aαεi, aβεj ∈ B(ε), aαεi ≺s-ε aβεj ⇔ LM(aαgi) ≺e LM(aβgj), or LM(aαgi) = LM(aβgj) and i < j, there is indeed a much stronger result, namely the noncommutative ana- logue of Schreyer theorem [Sch] (cf. Theorem 3.7.13 in [AL2] for free modules over commutative polynomial algebras; Theorem 4.8 in [Lev] for free modules over solvable polynomial algebras): 3.1.1. Theorem (2.3.1) With respect to the left monomial ordering ≺s-ε on B(ε) as defined above, the following statements hold. (i) Let sij be determined by Sℓ(gi, gj), where i < j, LM(gi) = aα(i)es with α(i) = (αi1, . . . , αin ), and LM(gj) = aα(j)es with α(j) = (αj1, . . . , αjn). Then LM(sij) = aγ−α(j)εj, where γ = (γ1, . . . , γn) with each γk = max{αik , αjk}. (ii) S is a left Grobner basis of Syz(G), thereby S generates Syz(G). Proof (i) By the definition of Sℓ(gi, gj) we know that LM(aγ−α(i)gi) = aγes = LM(aγ−α(j)gj). Since i < j, by the definition of ≺s-ε we have 54 Finite Free Resolutions the definition of ≺s-ε that LM(sij) = aγ−α(j)εj. aγ−α(i)εi ≺s-ε aγ−α(j)εj. Consequently, it follows from LM(Sℓ(gi, gj)) ≺e i=1 figi, and LM(aγ−α(j)gn), the Grobner representation Sℓ(gi, gj) = Pt sion algorithm by S, ξ has an expression ξ = Psij ∈S fijsij + η, where (ii) To show that S forms a left Grobner basis for Syz(G) with re- spect to ≺s-ε, take any nonzero ξ ∈Syz(G). After executing the divi- LM(fijsij) (cid:22)s-ε LM(ξ) for each term fijsij, and η = ξ is the re- mainder of ξ on division by S. We claim that η = 0, thereby S is a left Grobner basis for Syz(G). Otherwise, assuming the contrary that η =Pi,j λijaθ(ij )εj 6= 0, where λij ∈ K ∗ and θ(ij) = (θij1, . . . , θijn) ∈ Nn, then LM(η) = aθ(lk)εk for some pair (l, k) and S LM(sij) 6 aθ(lk)εk for all sij ∈ S. (1) Let sij be determined by Sℓ(gi, gj), where i < j, LM(gi) = aα(i)es and LM(gj) = aα(j)es. Then by (i) we have LM(sij) = aγ−α(j)εj. If j = k then by (1), aγ−α(k)6 aθ(lk). (2) Since LM(η) = aθ(lk)εk, by the definition of ≺s-ε we have LM(aθ(ij )gj) (cid:22)e LM(aθ(lk)gk), and for (i, j) 6= (l, k), LM(aθ(ij )gj) = LM(aθ(lk)gk) implies j < k. (3) Noticing η = ξ −Psij∈S fijsij ∈ Syz(G), we have Pi,j λijaθ(ij)gj = 0. Since A is a domain, if LM(aθ(ij )gj) 6= LM(aθ(lk)gk) for all (i, j) 6= (l, k), then we would have η = 0 (note that all gi 6= 0), which is a contradiction. So, by (3) we may assume that LM(aθ(ij )gj) = LM(aθ(lk)gk) for some j < k. Let LM(gj) = aα(j)es and LM(gk) = aα(k)es. Then LM(aθ(ij )gj) = LM(aθ(ij )LM(gj)) = LM(aθ(ij )aα(j)es) = aθ(ij )+α(j)es, LM(aθ(lk)gk) = LM(aθ(lk)LM(gk)) = LM(aθ(lk)aα(k)es) = aθ(lk)+α(k)es, thereby θ(ij) + α(j) = θ(lk) + α(k). Now, taking γ = (γ1, . . . , γn) in which γℓ = max{αjℓ, αkℓ} with respect to α(j) = (αj1, . . . , αjn) and α(k) = (αk1, . . . , αkn), it follows that θ(lk) + α(k) = ρ + γ for some ρ = (ρ1, . . . , ρn). Hence θ(lk) = ρ + (γ − α(k)), and this gives rise to aθ(lk) = LM(aρaγ−α(k)), i.e., aγ−α(k)aθ(lk), contradicting (2). Therefore, we must have η = 0, as claimed. This complets the proof. (cid:3) Finite Free Resolutions 55 To go further, again let G = {g1, . . . , gt} be the left Grobner basis of N produced by running Algorithm-LGB presented in (Chapter 2, Theorem 2.3.4) with the initial input data U = {ξ1, . . . , ξm}. Using the usual matrix notation for convenience, we have  ξ1 ... ξm  = Um×t g1 ... gt  ,  g1 ... gt  = Vt×m ξ1 ... ξm  , i=1 ASi with S1, . . . , Sr ∈ L2 = ⊕t we may write Syz(G) = Pr and if Si = Pt where the m × t matrix Um×t (with entries in A) is obtained by the di- vision by G, and the t × m matrix Vt×m (with entries in A) is obtained by keeping track of the reductions during executing the WHILE loop of Algorithm-LGB (Chapter 2, Corollary 2.3.5). By Theorem 3.1.1, i=1Aεi; j=1 fijεj, then we write Si as a 1 × t row matrix, i.e., Si = (fi1 . . . fit), whenever matrix notation is convenient in the accord- ing discussion. At this point, we note also that all the Si may be written down one by one during executing the WHILE loop of Algorithm-LGB successively. Furthermore, we write D(1), . . . , D(m) for the rows of the matrix Dm×m = Um×tVt×m − Em×m where Em×m is the m × m iden- tity matrix. The following proposition is a noncommutative analogue of ([AL2], Theorem 3.7.6). 3.1.2. Theorem With notation fixed above, the syzygy module Syz(U ) of U = {ξ1, . . . , ξm} is generated by {S1Vt×m, . . . , SrVt×m, D(1), . . . , D(m)}, where each 1 × m row matrix represents an element of the free A-module L1 = ⊕m i=1Aωi. Proof Since 0 = Si g1 ... gt  = (fi1 . . . fit) g1 ... gt  = (fi1 . . . fit)Vt×m ξ1 ... ξm  , 56 Finite Free Resolutions we have SiVt×m ∈ Syz(U ), 1 ≤ i ≤ r. Moreover, since ξ1 ... ξm ξ1 ... ξm ξ1 ... ξm ξ1 ... ξm ξ1 ... ξm Dm×m ξ1 ... ξm    = (Um×tVt×m − Em×m)  − = Um×tVt×m = Um×t  −   = 0,  − =   = 0, then 0 = i=1Aωi such that H . This means HUm×t ∈ Syz(G). Hence, HUm×t = . . . hm) represents the ele- if H = (h1 i=1 hiωi ∈ ⊕m ment Pm HUm×t Pr i=1 fiSi with fi ∈ A, and it follows that HUm×tVt×m =Pr Therefore, g1 ... gt we have D(1), . . . , D(r) ∈ Syz(U ). On the other hand, i=1 fiSiVt×m. g1 ... gt ξ1 ... ξm ξ1 ... ξm H = H + HUm×tVt×m − HUm×tVt×m = H(Em − Um×tVt×m) +Pr = −HDm×m +Pr i=1 fi(SiVt×m). i=1 fiSiVt×m This shows that every element of Syz(U ) is generated by {S1Vt×m, . . . , SrVt×m, D(1), . . . , D(m)}, as desired. Finite Free Resolutions 57 3.2. Computation of Finite Free Resolutions Let A = K[a1, . . . , an] be a solvable polynomial algebra with admissi- ble system (B, ≺) in the sense of Definition 1.1.3, where B = {aα = aα1 α = (α1, . . . , αn) ∈ Nn} is the PBW K-basis of A and ≺ 1 · · · aαn n is a monomial ordering on B. Let M be a left A-module. Then a free resolution of M is an exact sequence by free A-modules Li and A-module homomorphisms ϕi: L• · · · ϕi+1−→ Li ϕi−→ · · · ϕ2−→ L1 ϕ1−→ L0 ϕ0−→ M −→ 0 that is, in the sequence L•, ϕ0 is surjective, each Li is a free A-module and Kerϕi = Imϕi+1 = ϕi+1(Li+1) for all i ≥ 0. By classical homological algebra (e.g. [Rot]), theoretically such a free resolution for M exists. If M is a finitely generated A-module, each Li in L• is finitely generated, and Lq+1 = 0 for some q, then L• is called a finite free resolution of M . In this section, we show, in a constructive way, that a noncommutative version of Hilbert's syzygy theorem holds true for A, i.e., every finitely generated A-module M has a finite free resolution, and consequently, an algorithm for computing a finite free resolution of M is obtained. Our argumentation concerning a noncommutative version of Hilbert's syzygy theorem below is adapted from ([Eis], Corollary 15.11) and ([Lev], Section 4.4). Let L = ⊕s i=1Aei be a free A-module with left monomial ordering ≺e on its K-basis B(e), and let G = {g1, . . . , gt} be a left Grobner basis of i=1 Agi ⊂ L. Then, after relabeling the members the submodule N =Pt of G (if necessary), we may always assume that G satisfies (∗) if i < j and LM(gi) = aα(i)ek with α(i) = (αi1, . . . , αin), LM(gj) = aα(j)ek with α(j) = (αj1, . . . , αjn), then aα(i) (cid:22)lex aα(j) under the lexicographic ordering with respect to an ≺lex an−1 ≺lex · · · ≺lex a1, basis of the submodule N =Pt 3.2.1. Lemma Given a free A-module L = ⊕s i=1Aei with left monomial ordering ≺e on its K-basis B(e), let G = {g1, . . . , gt} be a left Grobner i=1 Agi ⊂ L. With notation as in Theorem 3.1.1, let ≺s-ε be the Schreyer ordering on the K-basis B(ε) of the free A- module L1 = ⊕t i=1Aεi induced by G with respect to ≺e. Assume that for some r ≤ n, the generators a1, . . . , ar of A do not appear in every LM(gℓ), 58 Finite Free Resolutions then the generators a1, . . . , ar, ar+1 of A do not appear in LM(sij) for every sij ∈ S. Proof If sij ∈ S with sij 6= 0, then we have i < j and LM(gi) = aα(i)ek with α(i) = (αi1 , . . . , αin), LM(gj) = aα(j)ek with α(j) = (αj1, . . . , αjn), and it follows from the property (∗) mentioned before the lemma that αiℓ = 0 = αjℓ, 1 ≤ ℓ ≤ r, αir+1 ≤ αjr+1. By Theorem 3.1.1, LM(sij) = aγ−α(j)εj where γ = (γ1, . . . , γn) with γk = max{αik , αjk}, in particular, γr+1 = αjr+1. This shows that the generators a1, . . . , ar, ar+1 of A do not appear in LM(sij). (cid:3) Let M = Ps i=1 Avi be a finitely generated left A-module with gen- erating set {v1, . . . , vs}. Then since A is Noetherian, if we consider the i=1Aei and the presentation M ∼= L0/N0 of M by free A-module L0 = ⊕s a submodule N0 of L0, then N0 is a finitely generated submodule of L0. 3.2.2. Theorem (noncommutative version of Hilbert's syzygy theorem) Let A = K[a1, . . . , an] be a solvable polynomial algebra with admissible system (B, ≺). Then every finitely generated left A-module M has a finite free resolution 0 −→ Lq ϕq−→ Lq−1 ϕq−1−→ · · · ϕ1−→ L0 ϕ0−→ M −→ 0 such that q ≤ n. Proof As remarked above we may assume M = L0/N0, where L0 = ⊕s i=1Aei is a free A-module with A-basis {e1, . . . , es}, and N0 is a finitely generated submodule of L0. Then ϕ0 is given by the canonical A-module ϕ0−→ M . Fixing a left monomial ordering ≺e on the homomorphism L0 K-basis B(e) of L0, let G0 = {g1, . . . , gs1} be a left Grobner basis of N0 as specified before Lemma 3.2.1, such that for some r ≤ n, the generators a1, . . . , ar of A do not appear in every LM(gℓ), 1 ≤ ℓ ≤ s1. Let ≺s-ε be the Schreyer ordering on the K-basis B(ε) of the free A-module L1 = ⊕s1 i=1Aεi induced by G0 with respect to ≺e. Then by Lemma 3.2.1, the generators a1, . . . , ar, ar+1 do not appear in LM(sij) for every sij ∈ S, where S is the left Grobner basis of N1 = Syz(G0) ⊂ L1 obtained in Theorem 3.1.1. Defining ϕ1: L1 → L0 by ϕ1(εi) = gi, 1 ≤ i ≤ s1, and working with N1 in place of N0 and so on, we then reach an exact sequence 0 → Nn−r −→ Ln−r ϕn−r−→ Ln−r−1 ϕn−r−1−→ · · · ϕ2−→ L1 ϕ1−→ L0 ϕ0−→ M → 0 Finite Free Resolutions 59 1, . . . , g′ i=1Aωi, this shows that LM(g′ where every Li is a free A-module of finite A-basis and Nn−r =Kerϕn−r which has a left Grobner basis Gn−r = {g′ d} in Ln−r, such that all the generators a1, . . . , an of A do not appear in every LM(g′ j), 1 ≤ j ≤ d. If Ln−r = ⊕t j) = ωkj , 1 ≤ j ≤ d. Without loss of generality, we may assume that all the ωkj are distinct (for instance, Gn−r is minimal), and that LM(g′ j) = ωj, 1 ≤ j ≤ d. Thus, since Gn−r is a left Grobner basis, it follows from (Chapter 2, Proposition 2.2.7(iv)) that in this case Kerϕn−r−1 ∼= Ln−r/Nn−r ∼= ⊕t i=d+1Aωi. Now, with Ln−r replaced by Kerϕn−r−1, the desired (length ≤ n) free resolution of M is obtained. (cid:3) Combining the results of the last section, we are ready to have an algorithmic procedure for constructing a finite free resolution. 3.2.3. Corollary Let M = L0/N0 be a finitely generated A-module as in Theorem 3.2.2, and let G0 = {g1, . . . , gt} be a left Grobner basis of the i=1 Agi with respect to a left monomial ordering ≺e on L0. Then, starting with the exact sequence submodule N0 =Pt 0 → N0 ι−→ L0 ϕ0−→ M → 0, where ϕ0 is the canonical homomorphism, and ι is the inclusion map, the following algorithm returns a finite free resolution of M , which is of length q ≤ n. Algorithm-FRES INPUT L0 = ⊕s OUTPUT L• i=1Aei, ≺e, G0 = {g1, . . . , gt}, L0 ϕ2−→ L1 0 → Lq a finite free resolution of M ϕq−→ Lq−1 ϕq−1−→ · · · ϕ0−→ M → 0 ϕ1−→ L0 ϕ0−→ M → 0 INITIALIZATION i := 0, ≺ := ≺e LOOP IF all the generators a1, . . . , an of A do not appear in LM(gj) for every gj ∈ Gi, THEN by the proof of Theorem 3.2.2, there is some d such that j=1 Agj ∼= ⊕d ι−→ Li ϕi−→ · · · i=1Aei ϕ2−→ L1 ϕ1−→ L0 ϕ0−→ M → 0) Kerϕi ∼= Li/Pt L• := (0 → Kerϕi ELSE i := i + 1, Li := ⊕t j=1Aej, ϕi := (Li → Li−1 with ϕ(ej ) = gj, 1 ≤ j ≤ t) 60 Finite Free Resolutions run Algorithm-DIV-L (with respect to ≺) to compute a left Grobner representation of each Sℓ(gk, gℓ) 6= 0 by Gi−1 in Li−1, 1 ≤ k < ℓ ≤ t, so that a left Grobner basis S = {S1, . . . , Sr} of Syz(G) ⊂ Li is obtained under the Schreyer ordering ≺s-e on Li induced by Gi−1 with respect to ≺ (Theorem 3.1.1) ≺ := ≺s-e Gi := {g1, . . . , gt} with gj = Sj 6= 0 for 1 ≤ j ≤ t ≤ r END UNTIL all the generators a1, . . . , an of A do not appear in LM(gj) for every gj ∈ Gi END Proof By the definition of a free resolution, the sequence L• returned by the algorithm is clearly the desired one for M . The fact that the Algorithm-FRES terminates and returns a sequence L• of finite length q ≤ n is due to Theorem 3.2.2 (or more precisely its proof) 3.3. Global Dimension and Stability In this section, the foregoing Theorem 3.2.2 is applied to highlight that every solvable polynomial algebra A has finite global homological dimen- sion, and that every finitely generated projective A-module is stably free. Let A = K[a1, . . . , an] be a solvable polynomial algebra with admis- sible system (B, ≺) in the sense of Definition 1.1.3, where B = {aα = aα1 n α = (α1, . . . , αn) ∈ Nn} is the PBW K-basis of A and ≺ is a 1 · · · aαn monomial ordering on B. Recall from classical homological algebra (e.g. −→ N → 0 [Rot]) that a left A-module P is said to be projective if M is an exact sequence of A-modules and P α−→ N is an A-module homo- −→ M such morphism, then there exists an A-module homomorphism P that β ◦ γ = α, or in other words, the following diagram commutes: β γ P γւ αy β M −→ N → 0 β ◦ γ = α Finite Free Resolutions 61 Thus, by the universal property of a free module, any free A-module is projective and, an A-module P is projective if and only if there is a free A-module L such that P is isomorphic to a direct summand of L, i.e., L ∼= P ⊕ L1. Let M be a left A-module. Then a projective resolution of M is an ex- act sequence by projective A-modules Pi and A-module homomorphisms ϕi: P• · · · ϕi+1−→ Pi ϕi−→ · · · ϕ2−→ P1 ϕ1−→ P0 ϕ0−→ M −→ 0 that is, in the sequence P•, ϕ0 is surjective, each Pi is a projective A- module and Kerϕi = Imϕi+1 = ϕi+1(Pi+1) for all i ≥ 0. Clearly, any free resolution of M is a projective resolution of M . The projective dimension of M , denoted p.dimAM , is defined to be the shortest length q of a projective resolution or ∞ if no finite projective resolution exists. The long Schanuel lemma (cf. [Rot]) shows that any projective resolution of M can be terminated at this length. The left global dimension of A is then defined to be sup{p.dimAM M any left A-module}. In terms of right A-modules, the right global dimension of A is defined in a similar way. It follows from classical homological algebra (e.g. [Rot]) that the left (right) global dimension is determined by the projective dimension of cyclic modules; moreover, for a (left and right) Noetherian ring A, the left global dimension of A is equal to the right global dimension of A, and this common number is called the global dimension of A, denoted gl.dimA. 3.3.1. Theorem Let A = K[a1, . . . , an] be a solvable polynomial algebra. Then any A-module M has projective dimension p.dimAM ≤ n, thereby A has global dimension gl.dimA ≤ n. Proof Noticing the classical results on global dimension reviewed above, this follows from the fact that A is (left and right) Noetherian (Corol- lary 1.3.3) and the noncommutative version of Hilbert's syzygy theorem (Theorem 3.2.2). (cid:3) Furthermore, since A is Noetherian, A has IBN (invariant basis num- ber), i.e., for every free A-module L, every two A-bases of L have the same cardinal (cf. [Rot], Chapter 3). In this case, the rank of L, denoted 62 Finite Free Resolutions rankAL, is well defined as the cardinal of some A-basis of L, thereby if L1, L2 are free A-modules, then L1 ∼= L2 if and only if rankAL1 = rankAL2. With this well-defined rank for free modules, the stably free modules are then well defined, that is, an A-module P is said to be stably free of rank t if P ⊕ L1 ∼= L2, where L1 is a free module of rankAL1 = s and L2 is a free module of rankAL2 = s + t. Obviously, a stably free module is nec- essarily finitely generated and projective. It follows from the literature (e.g., [Rot], Chapter 4; [MR], Chapter 11) that stably free A-modules can be characterized by finite free resolutions. 3.3.2. Proposition A finitely generated projective A-module P is stably free if and only if P has a finite free resolution. Furthermore, if L• 0 → Lq ϕq−→ Lq−1 ϕq−1−→ · · · ϕ2−→ L1 ϕ1−→ L0 ϕ0−→ P → 0 is a finite free resolution of P , then rankAP =Pq i=0(−1)irankLi. (cid:3) 3.3.3. Theorem Let A = K[a1, . . . , an] be a solvable polynomial algebra. Then every finitely generated projective A-module P is stably free, and moreover, rankAP is computable via constructing a finite free resolution by running Algorithm-FRES given in Corollary 3.2.3. Proof This follows from the noncommutative version of Hilbert's syzygy theorem (Theorem 3.2.2), Theorem 3.2.3, and Proposition 3.3.2 above. 3.4. Calculation of p.dimAM Let A = K[a1, . . . , an] be a solvable polynomial algebra with admissi- ble system (B, ≺) in the sense of Definition 1.1.3, where B = {aα = aα1 α = (α1, . . . , αn) ∈ Nn} is the PBW K-basis of A and ≺ 1 · · · aαn n is a monomial ordering on B. Equipped with the previously developed theory and techniques, in this section we establish an algorithmic pro- cedure which calculates the projective dimension of a finitely generated A-module M and, at the same time, verifies whether M is projective or not. The strategy used in our text was proposed by Gago-Vargas in [GV], though the algebras considered in [GV] are restricted to Weyl algebras. Let L ϕ −→ M → 0 be an A-module epimorphism, and K = Kerϕ. Con- Finite Free Resolutions 63 ι−→ L ϕ ϕ sider the short exact sequence 0 → K −→ M → 0, where ι is the inclusion map. Suppose that there exists an A-module homomorphism −→ L such that ϕ ◦ ϕ = 1M , where 1M is the identity map of M to M . M Then, every ξ ∈ L has a representation as ξ = (ξ − ϕ(ϕ(ξ))) + ϕ(ϕ(ξ)) with ϕ(ϕ(ξ)) ∈ Imϕ and ξ − ϕ(ϕ(ξ)) ∈ K. Since it is clear that K∩ Imϕ = {0} and ϕ ◦ ϕ = 1M implies that ϕ is a monomorphism, it turns out that L = K ⊕ Imϕ = K ⊕ ϕ(M ) ∼= K ⊕ M. (1) This preliminary enables us to prove the following ϕ0−→ M → 0 be an exact sequence 3.4.1. Proposition Let 0 → L1 of A-modules in which L0, L1 are free A-modules. Then M is projective ϕ1−→ L1 such if and only if there exists an A-module homomorphism L0 that ϕ1 ◦ ϕ1 = 1L1, where 1L1 is the identity map of L1 to L1. ϕ1−→ L0 Proof Suppose that M is projective. Then there exists an A-module ϕ0−→ L0 such that ϕ ◦ ϕ0 = 1M , where 1M is the homomorphism M identity map of M to M . Hence, by the formula (1) above we have L0 = ϕ1(L1) ⊕ ϕ0(M ). It follows that if we define L0 ϕ1−→ L1 by ϕ1(ϕ1(ξ1) + ϕ0(m)) = ξ1, ξ1 ∈ L1, m ∈ M, then since ϕ1 is injective, it is easy to see that ϕ1 is an A-module homo- morphism satisfying ϕ1 ◦ ϕ1 = 1L1. Conversely, suppose that there exists an A-module homomorphism ϕ1−→ L1 such that ϕ1 ◦ ϕ1 = 1L1. Then the sequence L0 0 → K = Kerϕ1 ι−→ L0 ϕ1−→ L1 → 0 is exact. Since L1 is free (hence projective), it follows from the formula (1) above that L0 = K ⊕ ϕ1(L1), thereby K ∼= L0/ϕ1(L1) = L0/Kerϕ0 ∼= M . Note that as a direct summand of the free module L0, K is projective. Hence M is projective, as desired. (cid:3) and t respectively, and let L1 j=1Aej, L1 = ⊕t i=1Aεi be free left A-modules of rank s ϕ1−→ L0 be an A-module homomorphism j=1 fijej, 1 ≤ i ≤ t. Then ϕ1 is uniquely determined by Let L0 = ⊕s with ϕ1(εi) =Ps 64 the t × s matrix Finite Free Resolutions i=1 fiεi ∈ L1, then, ϕ1(ξ) is given by left multiplication that is, if ξ =Pt by matrices: ϕ1(ξ) = , · · · f1s · · · f2s ... · · · · · · fts f11 f12 f21 f22 ... ... ft2 ft1  Qϕ1 = fiϕ1(εi) = (f1, . . . , ft)Qϕ1 tXi=1 e1 ... es  . The t × s matrix Qϕ1 is usually referred to as the matrix of ϕ1. Thus, in the language of matrices, Proposition 3.4.1 can be restated as follows. ϕ0−→ M → 3.4.2. Proposition Let the short exact sequence 0 → L1 0 be as in Proposition 3.4.1, and let Qϕ1 be the matrix of ϕ1. Then M is projective if and only if the t × s matrix Qϕ1 is right invertible, i.e., there is an s × t matrix Qϕ1 with entries in A, such that Qϕ1 · Qϕ1 = Et×t, where the latter is the t × t identity matrix with the (i, i)-entry 1 ∈ A. ϕ1−→ L0 (cid:3) Furthermore, let the t × s matrix Qϕ1 of ϕ1 be as above, and write Qj ϕ1 for the j-th column of Qϕ1, 1 ≤ j ≤ s. Considering the free right A-module L1 = ⊕t j=1 ξjA ⊆ L1 be the A- submodule generated by ξj = (ε1, . . . , εt)Qj i=1 εifij, 1 ≤ j ≤ s. Then Proposition 3.4.2 tells us that the left A-module M is projective if and only if N = L1. Since A has also a right Grobner basis theory for right modules, it follows that the following proposition holds true. i=1εiA of rank t, let N = Ps ϕ1 = Pt basis of the right A-submodule N =Ps 3.4.3. Proposition With notation as above, let G be a right Grobner i=1εiA. Then, the left A-module M is projective if and only if εi ∈ G for 1 ≤ i ≤ t. If it is the case, then the right inverse Qϕ1 of Qϕ1 is given by the s × t matrix Vs×t such that j=1 ξjA ⊆ L1 = ⊕t (ε1, . . . , εt) = (ξ1, . . . , ξs)Vs×t, which can be computed via using (Chapter 2, Corollary 2.3.5(ii)). Finite Free Resolutions 65 if and only if all the εi ∈ G, 1 ≤ i ≤ t. Since N =Ps Proof Note that N = L1 if and only if all the εi ∈ N , 1 ≤ i ≤ t. If G is a right Grobner basis of N , then by (Chapter 2, Proposition 2.2.7), N = L1 j=1 ξjA, it follows from (Chapter 2, Corollary 2.3.5(ii)) that the specified s × t matrix Vs×t can be computed. Also since (ξ1, . . . , ξs) = (ε1, . . . , εt)Qϕ1, it turns out that (ε1, . . . , εt) = (ξ1, . . . , ξs)Vs×t = (ε1, . . . , εt)Qϕ1Vs×t. This shows that Vs×t = Qϕ1, as desired. (cid:3) Now, let M be a finitely generated A-module and L• 0 → Lq ϕq−→ Lq−1 ϕq−1−→ · · · ϕ2−→ L1 ϕ1−→ L0 ϕ0−→ M → 0 a finite free resolution of M by free modules of finite rank. Suppose that Imϕi is projective for some i ≥ 0. Then, by the foregoing discussion, it is not difficult to derive inductively that Imϕi+k is projective for every k = 1, 2, . . . , q − i. In particular, Imϕq−1 is projective. By the definition of p.dimAM and the basic property we recalled before Theorem 3.3.1, The next proposition is clear. 3.4.4. Proposition With notation as above, p.dimAM = q if and only if Imϕq−1 is not projective if and only if the matrix Qϕq of ϕq is not right invertible, where the invertibility of Qϕq can be recognized in a computational way via using Proposition 3.4.3. (cid:3) Suppose that Imϕq−1 is projective. It follows from Proposition 3.4.1 (or its proof) that Lq−1 = ϕq(Lq) ⊕ Ker ϕq ψ −→ ∼= Lq ⊕ Im ϕq−1, (2) and the latter isomorphism ψ can be computed, where ϕq is the A-module ϕq−→ Lq such that ϕq ◦ ϕq = 1Lq . The formula (2) homomorphism Lq−1 above enables us to construct another finite free resolution of M 0 → Lq ϕ′ q−→ Lq⊕Lq−1 ϕ′ q−1−→ Lq⊕Lq−2 ϕ′ q−2−→ Lq−3 ϕq−3−→ · · · ϕ1−→ L0 ϕ0−→ M → 0 66 in which Finite Free Resolutions ϕ′ q(ξq) = ϕq(ξq) for all ξq ∈ Lq, ϕ′ q−1(ξq + ξq−1) = ξq + ϕq−1(ξq−1) for all ξq + ξq−1 ∈ Lq ⊕ Lq−1, ϕ′ q−2(ξq + ξq−2) = ϕq−2(ξq−2) for all ξq + ξq−2 ∈ Lq ⊕ Lq−2. By the exactness of free resolution and the formula (2) above, we then have Lq ⊕ ϕq−1(Lq−1) = Imϕ′ q−1 = Kerϕ′ q−2, Lq−1 ψ −→ ∼= thereby M has the following finite free resolution 0 → Lq−1 ψ −→ Lq ⊕ Lq−2 ϕ′ q−2−→ Lq−3 ϕq−3−→ · · · ϕ1−→ L0 ϕ0−→ M → 0 in which the homomorphism ψ can be computed. By Proposition 3.4.4, after the projectiveness of Imϕq−2 is checked we can either have p.dimAM = q − 1, or repeat the above procedure again in order to get a finite free resolution of M which is of length q − 2. It is clear that after a finite number of repetitions of the same procedure we will eventually have p.dimAM = m with m ≤ q. Obviously, if m = 0, then M is projective. Summing up, we have reached the following 3.4.5. Theorem Let M be a finitely generated A-module, and let L• 0 → Lq ϕq−→ Lq−1 ϕq−1−→ · · · ϕ2−→ L1 ϕ1−→ L0 ϕ0−→ M → 0 be a finite free resolution of M computed by running Algorithm-FRES. Then the next algorithm computes p.dimAM and, meanwhile, checks whether M is projective or not. Algorithm-p.dim INPUT L• the given finite free resolution ; q the length of L• OUTPUT p.dimAM INITIALIZATION i := q IF i = 0 THEN p.dimAM := 0 ELSE Finite Free Resolutions 67 LOOP use Proposition 3.3.6 to check the invertibility of the matrix Qϕi of ϕi IF Qϕi is not right invertible THEN p.dimAM := i ELSE i := i − 1 IF i = 0 THEN p.dimAM := 0 ELSE L• := (0 → Li−1 ϕ′ i−2−→ Li−3 ϕi−3−→ · · · ϕ1−→ L0 i−2 ϕ′ i−1−→ L′ i−2 = Li ⊕ Li−2, the homomorphism ϕ′ ϕ0−→ M → 0 in which L′ i−1 = ψ is computed by using Proposition 3.3.4 (or its proof), and ϕ′ i−2 is defined before the theorem.) END END END END 4. Minimal Finite Graded Free Resolutions In this chapter we demonstrate how the methods and algorithms, devel- oped in ([CDNR], [KR2]) for computing minimal homogeneous generating sets of graded submodules and graded quotient modules of free modules over commutative polynomial algebras, can be adapted for computing minimal homogeneous generating sets of graded submodules and graded quotient modules of free modules over an N-graded solvable polynomial K-algebras A with the degree-0 homogeneous part A0 = K, and how the algorithmic procedures of computing minimal graded free resolutions for finitely generated modules over A can be achieved. In the literature, a finitely generated N-graded K-algebra A = ⊕p∈NAp with the degree-0 homogeneous part A0 = K is usually referred to as a connected N-graded K-algebra. Concerning introductions to minimal resolutions of graded modules over a (commutative or noncommutative) connected N-graded K-algebra (or more generally an N-graded local K- algebra) and relevant results, one may refer to ([Eis], Chapter 19), ([Kr1], Chapter 3), and [Li3]. All notions, notations and conventions introduced before are main- tained. 69 70 Minimal Graded Free Resolutions 4.1. N-graded Solvable Polynomial Algebras In this section we specify, by means of positive-degree functions (as de- fined in Section 1.1 of Chapter 1), the structure of N-graded solvable polynomial K-algebras with the degree-0 homogeneous part equal to K. For convenience, we first recall that the condition (S2) given in (Def- inition 1.1.3 of Chapter 1) is equivalent to (S2′) There is a monomial ordering ≺ on B, i.e., (B, ≺) is an admis- sible system of A, such that for all generators ai, aj of A with 1 ≤ i < j ≤ n, ajai = λjiaiaj + fji where λji ∈ K ∗, fji =P µkaα(k) ∈ K-spanB with LM(fji) ≺ aiaj if fji 6= 0. Now, 1 · · · aαn n let A = K[a1, . . . , an] be a solvable polynomial K-algebra with admissible system (B, ≺), where B = {aα = aα1 α = (α1, . . . , αn) ∈ Nn} is the PBW K-basis of A and ≺ is a monomial order- ing on B. Suppose that A is an N-graded algebra with the degree-0 ho- mogeneous part equal to K, namely A = ⊕p∈NAp, where the degree-p ho- mogeneous part Ap is a K-subspace of A, A0 = K, and Ap1Ap2 ⊆ Ap1+p2 for all p1, p2 ∈ N. Then, since conventionally any generator ai of A is not contained in the ground field K, writing dgr(f ) = p for the graded-degree (abbreviated to gr-degree) of a nonzero homogeneous element f ∈ Ap, we have dgr(ai) = mi, 1 ≤ i ≤ n, for some positive integers mi. It turns out that if aα = aα1 n ∈ B, i=1 αimi. This shows that dgr( ) gives rise to a positive- degree function on A as defined in (Section 1.1 of Chapter 1), such that then dgr(aα) =Pn 1 · · · aαn (1) Ap = K-span{aα ∈ B dgr(aα) = p}, p ∈ N; (2) for 1 ≤ i < j ≤ n, all the relations ajai = λjiaiaj + fji with fji = P µkaα(k) presented in (S2′) above, satisfy dgr(aα(k)) = dgr(aiaj) whenever µk 6= 0. Conversely, given a positive-degree function d( ) on A (as defined in Sec- tion 1.1 of Chapter 1) such that d(ai) = mi > 0, 1 ≤ i ≤ n, then we know that A has an N-graded K-module structure, i.e., A = ⊕p∈NAp with Minimal Graded Free Resolutions 71 Ap = K-span{aα ∈ B d(aα) = p}, in particular, A0 = K. It is straight- forward to verify that if furthermore for 1 ≤ i < j ≤ n, all the relations ajai = λjiaiaj + fji with fji =P µkaα(k) presented in (S2′) above, sat- isfy dgr(aα(k)) = dgr(aiaj) whenever µk 6= 0, then Ap1Ap2 ⊆ Ap1+p2 holds for all p1, p2 ∈ N, i.e., A is turned into an N-graded solvable polynomial algebra with the degree-0 homogeneous part A0 = K. Summing up, we have reached the following 4.1.1. Proposition A solvable polynomial algebra A = K[a1, . . . , an] is an N-graded algebra with the degree-0 homogeneous part A0 = K if and only if there is a positive-degree function d( ) on A (as defined in Section 1.1 of Chapter 1) such that for 1 ≤ i < j ≤ n, all the relations ajai = λjiaiaj + fji with fji =P µkaα(k) presented in (S2′) above, satisfy dgr(aα(k)) = dgr(aiaj) whenever µk 6= 0. (cid:3) To make the compatibility with the structure of N-filtered solvable polynomial algebras specified in Chapter 4, it is necessary to emphasize the role played by a positive-degree function in the structure of N-graded solvable polynomial algebras we specified in this section, that is, from now on in the rest of this paper we keep using the following Convention An N-graded solvable polynomial K-algebra A = ⊕p∈NAp with A0 = K is always referred to as an N-graded solvable polynomial algebra with respect to a positive-degree function d( ). Remark Let A = K[a1, . . . , an] = ⊕p∈NAp be an N-graded solvable poly- nomial algebra with respect to a positive-degree function d( ). (i) We emphasize that every aα ∈ B is a homogeneous elements of A and d(aα) = dgr(aα), where dgr( ), as we defined above, is the gr-degree function on nonzero homogeneous elements of A. (ii) Since A is a domain (Theorem 1.2.3), the gr-degree function dgr( ) has the property that for all nonzero homogeneous elements h1, h2 ∈ A, (P1) dgr(h1h2) = dgr(h1) + dgr(h2). From now on we shall freely use this property without additional indica- tion. 72 Minimal Graded Free Resolutions Typical noncommutative N-graded solvable polynomial algebras are provided by the multiplicative analogues On(λji) of the Weyl algebra (see Example (4) given in Section 1.1 of Chapter 1), where the positive-degree function on On(λji) can be defined by setting d(xi) = mi for any fixed tuple (m1, . . . , mn) of positive integers. Another family of noncommutative N-graded solvable polynomial al- gebras are provided by the algebras Mq(2, K) of 2 × 2 quantum matrices (see Example (7) given in Section 1.1 of Chapter 1), where each generator is assigned the degree 1. More generally, let Λ = (λij) be a multiplica- tively antisymmetric n × n matrix over K, and let λ ∈ K ∗ with λ 6= −1. Considering the multiparameter coordinate ring of quantum n × n matri- ces over K (see [Good]), namely the K-algebra Oλ,Λ(Mn(K)) generated by n2 elements aij (1 ≤ i, j ≤ n) subject to the relations aℓmaij = λℓiλjmaijaℓm + (λ − 1)λℓiaimaℓj λλℓiλjmaijaℓm λjmaijaℓm (ℓ > i, m > j) (ℓ > i, m ≤ j) (ℓ = i, m > j) Then Oλ,Λ(Mn(K)) is an N-graded solvable polynomial algebra, where each generator has degree 1. Moreover, by ([LW], [Li1], or the later Chapter 4), the associated graded algebra and the Rees algebra of every N-filtered solvable polyno- mial algebra with a graded monomial ordering ≺gr are N-graded solvable polynomial algebras of the type we specified in this section. Finally, we point out that if a solvable polynomial algebra A = K[a1, . . . , an] has a graded monomial ordering ≺gr with respect to some given positive-degree function d( ) (see Section 1.1 of Chapter 1), then, by Proposition 4.1.1, it is easy to check whether A is an N-graded algebra with respect to d( ) or not. 4.2. N-Graded Free Modules Let A = K[a1, . . . , an] be an N-graded solvable polynomial algebra with respect to a positive-degree function d( ), and let (B, ≺) be an admissible system of A. In this section we demonstrate how to construct N-graded free left modules over A and, furthermore, we highlight that if the input data is a finite set of nonzero homogeneous elements, then Algorithm- LGB (presented in Theorem 2.3.4 of Chapter 2) produces a homogeneous Minimal Graded Free Resolutions 73 left Grobner bases for the graded submodule generated by the given ho- mogeneous elements. We start by a little generality of N-graded modules. Let M be a left A-module. If M = ⊕q∈NMq with each Mq a K-subspace of M , such that ApMq ⊆ Mp+q for all p, q ∈ N, then M is called an N-graded A-module. For each q ∈ N, nonzero elements in Mq are called homogeneous elements of degree q, and accordingly Mq is called the degree-q homogeneous part of M . If ξ ∈ Mq and ξ 6= 0, then we write dgr(ξ) for the graded-degree (abbreviated to gr-degree) of ξ as a homogeneous element of M , i.e., dgr(ξ) = q. Let M = ⊕q∈NMq be a nonzero N-graded A-module, and T a subset of T is called a homogeneous generating set of M . Clearly, if T = {ξi i ∈ I} is a homogeneous generating set of M with dgr(ξi) = bi for ξi ∈ T , then homogeneous elements of M . If T generates M , i.e., M =Pξ∈T Aξ, then Mq =Ppi+bi=q Apiξi for all q ∈ N. If a submodule N of the N-graded A-module M = ⊕q∈NMq is gen- erated by homogeneous elements, i.e., N has a homogeneous generating set, then N is called a graded submodule of M . A graded submodule N has the N-graded structure N = ⊕q∈NNq with Nq = N ∩ Mq, such that ApNq ⊆ Np+q for all p, q ∈ N. With the graded submodule N of M as described above, the quo- tient module M/N is an N-graded A-module with the N-graded structure M/N = ⊕q∈N(M/N )q, where for each q ∈ N, (M/N )q = (Mq + N )/N . Indeed, a submodule N of M is a graded submodule if and only if the quo- tient module M/N is an N-graded A-module with the N-graded structure M/N = ⊕q∈N(Mq + N )/N . let L = ⊕s Now, i=1Aei be a free left A-module with the A-basis {e1, . . . , es}. Then L has the K-basis B(e) = {aαei aα ∈ B, 1 ≤ i ≤ n} and, for an arbitrarily fixed {b1, . . . , bs} ⊂ N, one checks that L can be turned into an N-graded free A-module L = ⊕q∈NLq by setting Lq = {0} if q < min{b1, . . . , bs}; otherwise Lq = Xpi+bi=q Apiei, q ∈ N, or alternatively, for q ≥ min{b1, . . . , bs}, Lq = K-span{aαei ∈ B(e) d(aα) + bi = q}, q ∈ N, 74 Minimal Graded Free Resolutions such that dgr(ei) = bi, 1 ≤ i ≤ s. As with the gr-degree of homogeneous elements in A, noticing that dgr(aαei) = d(aα)+bi for all aαei ∈ B(e) and that A is a domain, from now on we shall freely use the following property without additional indication: for all nonzero homogeneous elements h ∈ A and all nonzero homogeneous elements ξ ∈ L, (P2) dgr(hξ) = dgr(h) + dgr(ξ). Remark Although we have remarked that d(aα) = dgr(aα) for all aα ∈ B, d(aα) is used in constructing Lq just for highlighting the role of d( ). Convention Unless otherwise stated, from now on throughout the subse- quent texts if we say that L is an N-graded free module over an N-graded solvable polynomial algebra A with respect to a positive-degree function d( ), then it always means that L has an N-gradation as constructed above. Let L = ⊕q∈NLq be an N-graded free A-module, and N a graded submodule of L. A left Grobner basis G of N is called a homogeneous left Grobner basis if G consists of homogeneous elements. Note that monomials in B are homogeneous elements of A, thereby left S-polynomials of homogeneous elements are homogeneous elements, and remainders of homogeneous elements on division by homogeneous remain homogeneous elements. Thus, the following assertion is clear now. N = Pm 4.2.1. Theorem With notation as above, if a graded submodule i=1 Aξi of L is generated by the set of nonzero homogeneous elements {ξ1, . . . , ξm}, then, with the initial input data {ξ1, . . . , ξm}, Algorithm-LGB (presented in Theorem 2.3.4 of Chapter 2) produces a finite homogeneous left Grobner basis G for N with respect to any given monomial ordering ≺e on B(e). 4.3. Computation of Minimal Homogeneous Gen- erating Sets In this section, A = K[a1, . . . , an] denotes an N-graded solvable polyno- mial algebra with respect to a positive-degree function d( ), (B, ≺) denotes Minimal Graded Free Resolutions 75 i=1Aei denotes an N-graded free a fixed admissible system of A, L = ⊕s A-module such that dgr(ei) = bi, 1 ≤ i ≤ s, and ≺e denotes a fixed left monomial ordering on the K-basis B(e) of L. Moreover, as before we write Sℓ(ξi, ξj) for the left S-polynomial of two elements ξi, ξj ∈ L. Let N be a finitely generated graded submodule N of L. With the preparation made in the previous two sections, our aim of the current section is to provide an algorithmic way of computing (1) a minimal homogeneous generating set of N , and (2) a minimal homogeneous generating set of the graded quotient module M = L/N . The argumentation we are going to present below is similar to the commutative case (cf. [KR], Section 4.5, Section 4.7). To better under- stand why the similar argumentation can go through the noncommutative case, let us again remind that although monomials from the PBW K-basis B of A can no longer behave as well as monomials in a commutative poly- nomial algebra (namely the product of two monomials is not necessarily a monomial), every monomial from B is a homogeneous element in the N- graded structure of A (as we remarked in section 4.1), thereby the product of two monomials is a homogeneous element. Bearing in mind this fact, one will see that the rule of division, Proposition 1.1.4(i) (Section 1 of Chapter 1), Lemma 2.1.2(ii) (Section 1 of Chapter 2), and the properties (P1), (P2) mentioned in previous Section 1 and Section 2 respectively, all together make the work done. We start by a detailed discussion on computing n-truncated left Grobner bases for graded submodules of L. Except for helping us to compute a minimal homogeneous generating set, from the definition and the characterization given below one may see clearly that having an n- truncated left Grobner basis will be very useful in dealing with certain problems involving only degree-n homogeneous elements. neous elements of L, N =Pt 4.3.1. Definition Let G = {g1, . . . , gt} be a subset of nonzero homoge- i=1 Agi the graded submodule generated by G, and let n ∈ N, G≤n = {gj ∈ G dgr(gj) ≤ n}. If, for each nonzero homogeneous element ξ ∈ N with dgr(ξ) ≤ n, there is some gi ∈ G≤n such that LM(gi)LM(ξ) with respect to ≺e, then we call G≤n an n-truncated left Grobner basis of N with respect to (B(e), ≺e). 76 Minimal Graded Free Resolutions By the definition above, the lemma below is straightforward. 4.3.2. Lemma Let G = {g1, . . . , gt} be a homogeneous left Grobner basis i=1 Agi of L with respect to (B(e), ≺e). For each n ∈ N, put G≤n = {gj ∈ G dgr(gj) ≤ n}, N≤n = ∪n q=0Nq where each Nq is the degree-q homogeneous part of N , and let N (n) = Aξ be the graded submodule generated by N≤n. The following for the graded submodule N =Pt Pξ∈N≤n statements hold. (i) G≤n is an n-truncated left Grobner basis of N . Thus, if ξ ∈ L is a G≤n = 0, homogeneous element with dgr(ξ) ≤ n, then ξ ∈ N if and only if ξ i.e., ξ is reduced to zero on division by G≤n,. Agj, and G≤n is an n-truncated left Grobner basis (ii) N (n) =Pgj ∈G≤n of N (n). Proof Exercise. (cid:3) In light of Theorem 2.3.3 and Theorem 2.3.4 (presented in Section 3 of Chapter 2), an n-truncated left Grobner basis is characterized as follows. 4.3.3. Proposition Let N =Ps i=0 Agi be the graded submodule of L generated by a set of homogeneous elements G = {g1, . . . , gm}. For each n ∈ N, put G≤n = {gj ∈ G dgr(gj) ≤ n}. The following statements are equivalent with respect to the given (B(e), ≺e). (i) G≤n is an n-truncated left Grobner basis of N . (ii) Every nonzero left S-polynomial Sℓ(gi, gj ) of dgr(Sℓ(gi, gj )) ≤ n is reduced to zero on division by G≤n, i.e, Sℓ(gi, gj) G≤n = 0. Proof Recall that if gi, gj ∈ G, LT(gi) = λiaαet with α = (α1, . . . , αn), LT(gj) = λjaβet with β = (β1, . . . , βn), and γ = (γ1, . . . γn) with γi = max{αi, βi}, 1 ≤ i ≤ n, then Sℓ(gi, gj ) = 1 LC(aγ−αgi) aγ−αgi − 1 LC(aγ−βgj) aγ−βgj is a homogeneous element in N with dgr(Sℓ(gi, gj)) = d(aγ) + bt by the foregoing property (P4). If dgr(Sℓ(gi, gj )) ≤ n, then it follows from (i) that (ii) holds. Conversely, suppose that (ii) holds. To see that G≤n is an n-truncated left Grobner basis of N , let us run Algorithm-LGB (presented in Theorem 2.3.4 of Chapter 2) with the initial input data G. Without Minimal Graded Free Resolutions 77 optimizing Algorithm-LGB we may certainly assume that G ⊆ G, thereby G≤n ⊆ G≤n where G is the new input set returned by each pass through the WHILE loop. On the other hand, by the construction of Sℓ(gi, gj ) and the property (P2) given in the lat section, we know that if dgr(Sℓ(gi, gj)) ≤ n, then dgr(gi) ≤ n, dgr(gj) ≤ n. Hence, the assumption (ii) implies that Algorithm-LGB does not append any new element of degree ≤ n to G. Therefore, G≤n = G≤n. By Lemma 4.3.2 we conclude that G≤n is an n-truncated left Grobner basis of N . (cid:3) 4.3.4. Corollary Let N = Pm i=1 Agi be the graded submodule of L generated by a set of homogeneous elements G = {g1, . . . , gm}. Suppose that G≤n = {gj ∈ G dgr(gj) ≤ n} is an n-truncated left Grobner basis of N with respect to (B(e), ≺e). (i) If ξ ∈ L is a nonzero homogeneous element of dgr(ξ) = n such that LM(gi)6 LM(ξ) for all gi ∈ G≤n, then G′ = G≤n ∪ {ξ} is an n-truncated left Grobner basis for both the graded submodules N ′ = N + Aξ and Proof If ξ ∈ L is a nonzero homogeneous element of dgr(ξ) = n1 ≥ n and LM(ξi)6 LM(ξ) for all ξi ∈ G≤n, then noticing the property mentioned in (Lemma 2.1.2(ii) of Chapter 2) and the property (P2) mentioned in previous Section 2, we see that every nonzero left S-polynomial Sℓ(ξ, ξi) with ξi ∈ G has dgr(Sℓ(ξ, ξi)) > n. Hence both (i) and (ii) hold by Proposition 4.3.3. (cid:3) Based on the discussion above, the next proposition tells us that Algorithm-LGB (presented in Theorem 2.3.4 of Chapter 2) can be mod- ified for computing n-truncated left Grobner bases. 4.3.5. Proposition (Compare with ([KR2], Proposition 4.5.10)) Given a finite set of nonzero homogeneous elements U = {ξ1, . . . , ξm} ⊂ L with dgr(ξ1) ≤ dgr(ξ2) ≤ · · · dgr(ξm), and a positive integer n0 ≥ dgr(ξ1), the following algorithm computes an n0-truncated left Grobner basis G = i=1 Aξi such that dgr(g1) ≤ {g1, ..., gt} for the graded submodule N =Pm (ii) If n ≤ n1 and ξ ∈ L is a nonzero homogeneous element of dgr(ξ) = n1 such that LM(gi)6 LM(ξ) for all gi ∈ G≤n, then G′ = G≤n ∪ {ξ} is an n1-truncated left Grobner basis for the graded submodule N ′ = Agj + Aξ of L. N ′′ =Pgj∈G≤n Pgj∈G≤n Agj + Aξ of L. 78 Minimal Graded Free Resolutions dgr(g2) ≤ · · · dgr(gt). Algorithm-TRUNC INPUT: U = {ξ1, ..., ξm} with dgr(ξ1) ≤ dgr(ξ2) ≤ · · · dgr(ξm) n0, where n0 ≥ dgr(ξ1) OUTPUT: G = {g1, ..., gt} an n0-truncated left Grobner basis of N INITIALIZATION: S≤n0 := ∅, W := U, G := ∅, t′ := 0 LOOP n := min{dgr(ξi), dgr(Sℓ(gi, gj)) ξi ∈ W, Sℓ(gi, gj) ∈ S≤n0} Sn := {Sℓ(gi, gj) ∈ S≤n0 dgr(Sℓ(gi, gj)) = n} Wn := {ξj ∈ W dgr(ξj) = n} S≤n0 := S≤n0 − Sn, W := W − Wn WHILE Sn 6= ∅ DO Choose any Sℓ(gi, gj ) ∈ Sn Sn := Sn − {Sℓ(gi, gj )} = η 6= 0 with LM(η) = aρekTHEN G IF Sℓ(gi, gj ) t′ := t′ + 1, gt′ := η S≤n0 := S≤n0S(cid:26)Sℓ(gi, gt′ )(cid:12)(cid:12)(cid:12)(cid:12) G := G ∪ {gt′ } END gi ∈ G, 1 ≤ i < t′, LM(gi) = aτ ek, 0 < dgr(Sℓ(gi, gt′ )) ≤ n0 (cid:27) (cid:27) END WHILE Wn 6= ∅ DO Choose any ξj ∈ Wn Wn := Wn − {ξj} G IF ξj t′ := t′ + 1, gt′ := η S≤n0 := S≤n0 ∪(cid:26)Sℓ(gi, gt′ )(cid:12)(cid:12)(cid:12)(cid:12) G := G ∪ {gt′ } END = η 6= 0 with LM(η) = aρek THEN gi ∈ G, 1 ≤ i < t′, LM(gi) = aτ ek, 0 < dgr(Sℓ(gi, gt′ )) ≤ n0 END UNTIL S≤n0 = ∅ END Proof First note that both the WHILE loops append new elements to G by taking the nonzero normal remainders on division by G. Thus, with a fixed n, by the definition of a left S-polynomial and the normality of Minimal Graded Free Resolutions 79 G gt′ (mod G), it is straightforward to check that in both the WHILE loops every new appended Sℓ(gi, gt′ ) has dgr(Sℓ(gi, gt′ )) > n. To proceed, let us write N (n) for the submodule generated by G which is obtained after Wn is exhausted in the second WHILE loop. If n1 is the first number after 6= 0 n such that Sn1 6= ∅, and for some Sℓ(gi, gj ) ∈ Sn1, η = Sℓ(gi, gj) in a certain pass through the first WHILE loop, then we note that this η is still contained in N (n). Hence, after Sn1 is exhausted in the first WHILE loop, the obtained G generates N (n) and G is an n1-truncated left Grobner basis of N (n). Noticing that the algorithm starts with S = ∅ and G = ∅, inductively it follows from Proposition 4.3.3 and Corollary 4.3.4 that after Wn1 is exhausted in the second WHILE loop, the obtained G is an n1-truncated left Grobner basis of N (n1). Since n0 is finite and all the generators of N with dgr(ξj) ≤ n0 are processed through the second WHILE loop, the algorithm terminates and the eventually obtained G is an n0-truncated left Grobner basis of N . Finally, the fact that the degrees of elements in G are non-decreasingly ordered follows from the choice of the next n in the algorithm. (cid:3) Let the data (A, B, ≺) and (L, B(e)), ≺e) be as fixed before. Com- bining the foregoing results, we now proceed to show that the algorithm given in ([KR], Theorem 4.6.3)) can be adapted for computing minimal homogeneous generating sets of graded submodules in free modules over A. Let N be a graded submodule of the N-graded free A-module L fixed above. We say that a homogeneous generating set U of N is a mini- mal homogeneous generating set if any proper subset of U cannot be a generating set of N . As preparatory result, we first show that the non- commutative analogue of ([KR], Proposition 4.6.1, Corollary 4.6.2) holds true for N . i=1 Aξi be the graded submodule of L generated by a set of homogeneous elements U = {ξ1, . . . , ξm}, where j=1 Aξj, 4.3.6. Proposition Let N = Pm dgr(ξ1) ≤ dgr(ξ2) ≤ · · · ≤ dgr(ξm). Put N1 = {0}, Ni = Pi−1 2 ≤ i ≤ m. The following statements hold. (i) U is a minimal homogeneous generating set of N if and only if ξi 6∈ Ni, 1 ≤ i ≤ m. (ii) The set U = {ξk ξk ∈ U, ξk 6∈ Nk} is a minimal homogeneous generating set of N . 80 Minimal Graded Free Resolutions Proof (i) If U is a minimal homogeneous generating set of N , then clearly ξi 6∈ Ni, 1 ≤ i ≤ m. 6∈ Ni, 1 ≤ i ≤ m. Conversely, suppose ξi If U is not a minimal homogeneous generating set of N , then, there is some i such that N is generated by {ξ1, . . . , ξi−1, ξi+1, . . . , ξm}, thereby ξi =Pj6=i hjξj for some for all j 6= i, then ξi ∈Pi−1 K ∗. Putting i′ = max{i, j fj ∈ K ∗}, we then have ξi′ ∈ Pi′−1 nonzero homogeneous elements hj ∈ A such that dgr(ξi) = dgr(hjgj) = dgr(hj ) + dgr(ξj), where the second equality follows from the foregoing property (P4). Thus dgr(ξj) ≤ dgr(ξi) for all j 6= i. If dgr(ξj) < dgr(ξi) j=1 Aξj, which contradicts the assumption. If dgr(ξi) = dgr(ξj) for some j 6= i, then since hj 6= 0 we have hj ∈ A0−{0} = j=1 Aξj, which again contradicts the assumption. Hence, under the assumption we conclude that U is a minimal homogeneous generating set of N . generating set of N . (ii) In view of (i), it is sufficient to show that U is a homogeneous j=1 Aξj. By Indeed, if ξi ∈ U − U , then ξi ∈ Pi−1 checking ξi−1 and so on, it follows that ξi ∈Pξk∈U Aξk, as desired. (cid:3) 4.3.7. Corollary Let U = {ξ1, . . . , ξm} be a minimal homogeneous generating set of a graded submodule N of L, where dgr(ξ1) ≤ dgr(ξ2) ≤ · · · ≤ dgr(ξm), and let ξ ∈ L−N be a homogeneous element with dgr(ξm) ≤ dgr(ξ). Then bU = U ∪ {ξ} is a minimal homogeneous generating set of the graded submodule bN = N + Aξ. (cid:3) We are ready now to reach the following 4.3.8. Theorem (Compare with ([KR2], Theorem 4.6.3)) Let U = {ξ1, . . . , ξm} ⊂ L be a finite set of nonzero homogeneous elements of L with dgr(ξ1) ≤ dgr(ξ2) ≤ · · · ≤ dgr(ξm). Then the algorithm pre- sented below returns a minimal homogeneous generating set Umin = i=1 Aξi; and mean- while it returns a homogeneous left Grobner basis G = {g1, ..., gt} for N such that dgr(g1) ≤ dgr(g2) ≤ · · · dgr(gt). {ξj1, . . . , ξjr } ⊂ U for the graded submodule N =Pm Minimal Graded Free Resolutions 81 Algorithm-MINHGS INPUT: U = {ξ1, ..., ξm} with dgr(ξ1) ≤ dgr(ξ2) ≤ · · · ≤ dgr(ξm) OUTPUT: Umin = {ξj1, . . . , ξjr } ⊂ U a minimal homogeneous generating set of N ; G = {g1, ..., gt} a homogeneous left Grobner basis of N INITIALIZATION: S := ∅, W := U, G := ∅, t′ := 0, Umin := ∅ LOOP n := min{dgr(ξi), dgr(Sℓ(gi, gj)) ξi ∈ W, Sℓ(gi, gj) ∈ S} Sn := {Sℓ(gi, gj ) ∈ S dgr(Sℓ(gi, gj )) = n}, Wn := {ξj ∈ W dgr(ξj) = n} S := S − Sn, W := W − Wn WHILE Sn 6= ∅ DO Choose any Sℓ(gi, gj) ∈ Sn Sn := Sn − {Sℓ(gi, gj)} G = η 6= 0 with LM(η) = aρek THEN IF Sℓ(gi, gj) t′ := t′ + 1, gt′ := η S := S ∪ {Sℓ(gi, gt′ ) 6= 0 gi ∈ G, 1 ≤ i < t′, LM(gi) = aτ ek} G := G ∪ {gt′ } END END WHILE Wn 6= ∅ DO Choose any ξj ∈ Wn Wn := Wn − {ξj} G = η 6= 0 with LM(η) = aρek THEN IF ξj Umin := Umin ∪ {ξj} t′ := t′ + 1, gt′ := η S := S ∪ {Sℓ(gi, gt′ ) 6= 0 gi ∈ G, 1 ≤ i < t′, LM(gi) = aτ ek} G := G ∪ {gt′ } END END UNTIL S = ∅ END Proof Since this algorithm is clearly a variant of Algorithm-LGB and Algorithm-TRUNC with a minimization procedure which works with the finite set U , it terminates after a certain integer n is executed, and the eventually obtained G is a homogeneous left Grobner basis for N in which the degrees of elements are ordered non-decreasingly. It remains 82 Minimal Graded Free Resolutions to prove that the eventually obtained Umin is a minimal homogeneous generating set of N . As in the proof of Proposition 4.3.5, let us first bear in mind that for each n, in both the WHILE loops every new appended Sℓ(gi, gt′ ) has dgr(Sℓ(gi, gt′ )) > n. Moreover, for convenience, let us write G(n) for the G obtained after Sn is exhausted in the first WHILE loop, and write Umin[n], G[n] respectively for the Umin, G obtained after Wn is exhausted in the second WHILE loop. Since the algorithm starts with O = ∅ and G = ∅, if, for a fixed n, we check carefully how the elements of Umin are chosen during executing the second WHILE loop, and how the new elements are appended to G after each pass through the first or the second WHILE loop, then it follows from Proposition 4.3.3 and Corollary 4.3.4 that after Wn is exhausted, the obtained Umin[n] and G[n] generate the same module, denoted N (n), such that G[n] is an n-truncated left Grobner basis of N (n). We now use induction to show that the eventually obtained Umin is a minimal homogeneous generating set for N . If Umin = ∅, then it is a minimal generating set of the zero module. To proceed, we assume that Umin[n] is a minimal homogeneous generating set for N (n) after Wn is exhausted in the second WHILE loop. Suppose that n1 is the first number after n such that Sn1 6= ∅. We complete the induction proof below by showing that Umin[n1] is a minimal homogeneous generating set of N (n1). G If in a certain pass through the first WHILE loop, Sℓ(gi, gj) = η 6= 0 for some Sℓ(gi, gj ) ∈ Sn1, then we note that η ∈ N (n). It follows that after Sn1 is exhausted in the first WHILE loop, G(n1) generates N (n) and G(n1) is an n1-truncated left Grobner basis of N (n). Next, assume that Wn1 = {ξj1, . . . , ξjs} 6= ∅ and that the elements of Wn1 are processed in the given order during executing the second WHILE loop. Since G(n1) is an n1-truncated left Grobner basis of N (n), if ξj1 ∈ Wn1 is = η1 6= 0, then ξj1, η1 ∈ L − N (n). By Corollary 4.3.4, such that ξj1 we conclude that G(n1) ∪ {η1} is an n1-truncated Grobner basis for the module N (n) + Aη1; and by Corollary 4.3.7, we conclude that Umin[n] ∪ {ξj1} is a minimal homogeneous generating set of N (n) + Aη1. Repeating this procedure, if ξj2 ∈ Wn1 is such that fj2 = η2 6= 0, then ξj2, η2 ∈ L − (N (n) + Aη1). By Corollary 4.3.4, we conclude that G(n1) ∪ {η1, η2} is an n1-truncated left Grobner basis for the module N (n) + Aη1 + Aη2; and by Corollary 4.3.7, we conclude that Umin[n] ∪ {ξj1, ξj2} is G(n1) G(n1)∪{η1} Minimal Graded Free Resolutions 83 a minimal homogeneous generating set of N (n) + Aη1 + Aη2. Continuing this procedure until Wn1 is exhausted we assert that the returned G[n1] = G and Umin[n1] = Umin generate the same module N (n1) and G[n1] is an n1-truncated left Grobner basis of N (n1) and Umin[n1] is a minimal homogeneous generating set of N (n1), as desired. As all elements of U are eventually processed by the second WHILE loop, we conclude that the finally obtained G and Umin have the properties: G generates the module N , G is an n0-truncated left Grobner basis of N , and Umin is a minimal homogeneous generating set of N . (cid:3) Remark. If we are only interested in getting a minimal homogeneous generating set for the submodule N , then Algorithm-MINHGS can indeed be speed up. More precisely, with dgr(ξ1) ≤ dgr(ξ2) ≤ · · · ≤ dgr(ξm) = n0, it follows from the proof above that if we stop executing the algorithm after Sn0 and Wn0 are exhausted, then the resulted Umin[n0] is already the desired minimal homogeneous generating set for N , while G[n0] is an n0-truncated left Grobner basis of N . 4.3.9. Corollary Let U = {ξ1, . . . , ξm} ⊂ L be a finite set of nonzero homogeneous elements of L with dgr(ξ1) = dgr(ξ2) = · · · = dgr(ξm) = n0 (i) If U satisfies LM(ξi) 6= LM(ξj) for all i 6= j, then U is a minimal i=1 Aξi of homogeneous generating set of the graded submodule N =Pm Pm L, and meanwhile U is an n0-truncated left Grobner basis for N . (ii) If U is a minimal left Grobner basis of the graded submodule N = i=1 Aξi (i.e., U is a left Grobner basis of N satisfying LM(ξi) 6= LM(ξj) for all i 6= j), then U is a minimal homogeneous generating set of N . Proof By the assumption, it follows from the second WHILE loop of Algorithm-MINHGS that Umin = U . (cid:3) Let N be an arbitrary nonzero graded submodule of the N-graded free A-module L = ⊕s i=1Aei with dgr(ei) = bi, 1 ≤ i ≤ s, and consider the graded quotient module M = L/N (see previous Section 4.2). Our next goal is to compute a minimal homogeneous generating set for M . 84 Minimal Graded Free Resolutions L0. Let N =Pm elements U = {ξ1, . . . , ξm}, where ξℓ = Ps Since A is Noetherian, N is a finitely generated graded submodule of j=1 Aξj be generated by the set of nonzero homogeneous k=1 fkℓek with fkℓ ∈ A, 1 ≤ ℓ ≤ m. Then, every nonzero fkℓ is a homogeneous element of A such that dgr(ξℓ) = dgr(fkℓek) = dgr(fkℓ) + bk, where bk = dgr(ek), 1 ≤ k ≤ s, 1 ≤ ℓ ≤ m. 4.3.10. Lemma With every ξℓ =Ps i=1 fiℓei as fixed above, 1 ≤ ℓ ≤ m, if the i-th coefficient fij of some ξj is a nonzero constant, say fij = 1 without loss of generality, then for each ℓ = 1, . . . , j − 1, j + 1, . . . , m, Putting U ′ = the element ξ′ j−1, ξ′ {ξ′ m}, there is a graded A-module isomorphism ℓ = ξℓ − fiℓξj does not involve ei. j+1, . . . , ξ′ 1, . . . , ξ′ Proof Since fij = 1 by the assumption, we see that every ξ′ fiℓfkj)ek does not involve ei. Let U ′ = {ξ′ M ′ = L′/N ′ ∼= L/N = M , where L′ = ⊕k6=iAek and N ′ =Pξ′ N ′ = Pξ′ m} and ℓ. Then N ′ ⊂ L′ = ⊕k6=iAek. Again since fij = 1, we have dgr(ξj) = dgr(ei) = bi. It follows from the property (P4) formulated in Subsection 2.1 that ℓ =Pk6=i(fkℓ− j+1, . . . , ξ′ ℓ∈U ′ Aξ′ ℓ. ℓ∈U ′ Aξ′ 1, . . . , ξ′ j−1, ξ′ dgr(fiℓfkjek) = dgr(fiℓ) + dgr(fkjek) = dgr(fiℓ) + dgr(ξj) = dgr(fiℓ) + bi = dgr(fiℓei) = dgr(ξℓ) = dgr(fkℓek). Noticing that dgr(fiℓξj) = dgr(fiℓ) + dgr(ξj), this shows that in the rep- resentation of ξ′ ℓ every nonzero term (fkℓ − fiℓfkj)ek is a homogeneous element of degree dgr(ξℓ) = dgr(fiℓξj), thereby M ′ = L′/N ′ is a graded A-module. Note that N = N ′ + Aξj and that ξj = ei +Pk6=i fkjek. Without making confusion, if we use the same notation ek to denote the coset represented by ek in M ′ and M respectively, it is now clear that the desired graded A-module isomorphism M ′ ϕ −→M is naturally defined by ϕ(ek) = ek, k = 1, . . . , i − 1, i + 1, . . . , s. (cid:3) Let M = L/N be as fixed above with N generated by the set of nonzero homogeneous elements U = {ξ1, . . . , ξm}. Then since A is N- graded with A0 = K, it is well known that the homogeneous generating Minimal Graded Free Resolutions 85 set E = {e1, . . . , es} of M is a minimal homogeneous generating set if and k=1 fkℓek implies dgr(fkℓ) > 0 whenever fkℓ 6= 0, 1 ≤ ℓ ≤ m. only if ξℓ =Ps 4.3.11. Proposition (Compare with ([KR2], Proposition 4.7.24)) With notation as fixed above, the algorithm presented below returns a subset {ei1, . . . , eis′ } ⊂ {e1, . . . , es} and a subset V = {v1, . . . , vt} ⊂ N ∩ L′ such that M ∼= L′/N ′ as graded A-modules, where L′ = ⊕s′ q=1Aeiq with k=1 Avk, and such that {ei1, . . . , eis′ } is a minimal s′ ≤ s and N ′ = Pt homogeneous generating set of M . Algorithm-MINHGSQ INPUT: E = {e1, . . . , es}; U = {ξ1, ..., ξm} k=1 fkℓek with homogeneous fkℓ ∈ A, 1 ≤ ℓ ≤ m OUTPUT: E′ = {ei1 , . . . , eis′ }; V = {v1, . . . , vt} ⊂ N ∩ L′, such that q=1 hqjeiq ∈ L′ = ⊕s′ whenever hqj 6= 0, 1 ≤ j ≤ t q=1Aeiq with hqj 6∈ K ∗ where ξℓ =Ps vj =Ps′ INITIALIZATION: t := m; V := U ; s′ := s; E′ := E BEGIN WHILE there is a vj =Ps′ fkj 6∈ K ∗ for k < i and fij ∈ K ∗ DO k=1 fkjek ∈ V satisfying for T = {1, . . . , j − 1, j + 1, . . . , t} compute ℓ = vℓ − 1 v′ fij fiℓvj, ℓ ∈ T, r = #{ℓ ℓ ∈ T, v′ ℓ = 0} t := t − r − 1 V := {vℓ = v′ ℓ ℓ ∈ T, v′ ℓ 6= 0} = {v1, . . . , vt} (after reordered) s′ := s′ − 1 E′ := E′ − {ei} = {e1, . . . , es′} (after reordered) END END Proof It is clear that the algorithm is finite. The correctness of the algorithm follows immediately from Lemma 4.3.10 and the remark we made before the proposition. 86 Minimal Graded Free Resolutions 4.4. Computation of Minimal Finite Graded Free Resolutions Let A = K[a1, . . . , an] = ⊕p∈NAp be an N-graded solvable polynomial algebra with respect to a positive-degree function d( ), and (B, ≺) a fixed admissible system of A. Let M = ⊕q∈NMq, M ′ = ⊕q∈NM ′ q be ϕ −→ M ′ an A-module homomorphism. N-graded left A-modules, and M If ϕ(Mq) ⊆ M ′ q for all q ∈ N, then ϕ is called a graded homomorphism. In the literature, such graded homomorphisms are also referred to as graded homomorphisms of degree-0 (cf. [NVO]). By the definition it is clear that the identity map of N-graded A-modules is graded homomorphism, and compositions of graded homomorphisms are graded homomorphisms. Thus, all N-graded left A-modules form a subcategory of the category of left A-modules, in which morphisms are the graded homomorphisms as ϕ −→ M ′ is a graded homomorphism, defined above. Furthermore, if M then one checks that the kernel Kerϕ of ϕ is a graded submodule of M , and the image Imϕ of ϕ is a graded submodule of M ′ (See previous Sec- ψ −→ M ′ of tion 4.2). Consequently, the exactness of a sequence N graded homomorphisms in the category of N-graded A-modules is defined as the same as for a sequence of usual A-module homomorphisms, i.e., the sequence satisfies Imϕ = Kerψ. Long exact sequence in the category of N-graded A-modules may be defined in an obvious way. −→ M ϕ Since A is Noetherian and A0 = K, it is theoretically well known that up to a graded isomorphism of chain complexes in the category of graded A-modules, every finitely generated graded A-module M has a unique minimal graded free resolution (cf. [Eis], Chapter 19; [Kr1], Chapter 3; [Li3]). Based on previously obtained results, in this section we establish the algorithmic procedures for constructing minimal finite graded free resolutions over A. All notions, notations and conventions used in previous sections are maintained. Given a finitely generated N-graded A-module M = Ps i=1 Avi with the set of nonzero homogeneous generators {v1, . . . , vs} such that for 1 ≤ i ≤ s, consider the N-graded free A-module dgr(vi) = bi L0 = ⊕s i=1Aei with dgr(ei) = bi, 1 ≤ i ≤ s, as constructed in Section ϕ0−→ M → 0 defined by 4.2. Then, under the N-graded epimorphism L0 ϕ0(ei) = vi, 1 ≤ i ≤ s, there is a graded isomorphism M ∼= L0/N , where N = Kerϕ0. Thus, we may identify M with the graded quotient module Minimal Graded Free Resolutions 87 L0/N and write M = L0/N . Recall from the literature that a minimal graded free resolution of M is an exact sequence by free A-modules and A-module homomorphisms L• · · · ϕi+1−→ Li ϕi−→ · · · ϕ2−→ L1 ϕ1−→ L0 ϕ0−→ M −→ 0 in which each Li is an N-graded free A-module with a finite homogeneous }, and each ϕi is a graded homomorphism, such A-basis Ei = {ei1, . . . , eisi that (1) ϕ0(E0) is a minimal homogeneous generating set of M , Kerϕ0 = N , and (2) for i ≥ 1, ϕi(Ei) is a minimal homogeneous generating set of Kerϕi−1. Pm 4.4.1. Theorem With notation as fixed above, suppose that N = i=1 Aξi with the set of nonzero homogeneous generators U = {ξ1, . . . , ξm}. Then the graded A-module M = L0/N has a minimal graded free reso- lution of length d ≤ n: L• 0 −→ Ld ϕq−→ · · · ϕ2−→ L1 ϕ1−→ L0 ϕ0−→ M −→ 0 which can be constructed by implementing the following procedures: Procedure 1. Run Algorithm-MINHGSQ of Proposition 4.3.11 input data E = {e1, . . . , es} and U = {ξ1, . . . , ξm} with the initial to compute a subset E′ = {ei1 , . . . , eis′ } ⊂ {e1, . . . , es} and a subset 0/N ′ as graded A-modules, V = {v1, . . . , vt} ⊂ N ∩ L′ where L′ k=1 Avk, and such that {ei1, . . . , eis′ } is a minimal homogeneous generating set of M . q=1Aeiq with s′ ≤ s and N ′ = Pt 0 such that M ∼= L′ For convenience, after accomplishing Procedure 1 we may assume that E = E′, U = V and N = N ′. Accordingly we have the short exact sequence 0 = ⊕s′ 0 −→ N −→ L0 ϕ0−→ M −→ 0 such that ϕ0(E) = {e1, . . . , es} is a minimal homogeneous generating set of M . Procedure 2. Choose a left monomial ordering ≺e on the K-basis B(e) of L0 and run Algorithm-MINHGS of Theorem 4.3.8 with the initial input data U = {ξ1, . . . , ξm} to compute a minimal homogeneous generating set Umin = {ξj1, . . . , ξjs1 } and a left Grobner basis G for N ; at the same time, by keeping track of the reductions during executing the 88 Minimal Graded Free Resolutions first WHILE loop and the second WHILE loop respectively, return the matrices Sr×t and Vt×m required by (Theorem 3.1.2, Chapter 3). Procedure 3. By using the division by the left Grobner basis G obtained in Procedure 2, compute the matrix Um×t required by (The- orem 3.1.2, Chapter 3). Use the matrices Sr×t, Vt×m obtained in Pro- cedure 2, the matrix Um×t and (Theorem 3.1.2, Chapter 3) to compute a homogeneous generating set of N1 = Syz(Umin) in the N-graded free A-module L1 = ⊕s1 i=1Aεi, where the gradation of L1 is defined by setting dgr(εk) = dgr(ξjk ), 1 ≤ k ≤ s1. Procedure 4. Construct the exact sequence 0 −→ N1 −→ L1 ϕ1−→ L0 ϕ0−→ M −→ 0 where ϕ1(εk) = ξjk, 1 ≤ k ≤ s1. If N1 6= 0, then repeat Procedure 2 -- Procedure 4 for N1 and so on. Noticing that A is N-graded with the degree-0 homogeneous part A0 = K, A is Noetherian, and that every finitely generated A-module M has finite projective dimension p.dimAM ≤ n by (Theorem 3.3.1, Chapter 3), thereby A is an N-graded local ring of finite global homological dimension. It follows from the literature ([Eis], Chapter 19; [Kr1], Chapter3; [Li3]) that p.dimAM = the length of a minimal graded free resolution of M. Hence, the desired minimal finite graded free resolution L• for M is then obtained after finite times of processing the above procedures. 5. Minimal Finite Filtered Free Resolutions Recall that the N-filtered solvable polynomial algebras with N-filtration determined by the natural length of elements from the PBW K-basis (especially the quadric solvable polynomial algebras are such N-filtered algebras) were studied in ([LW], [Li1]). In this chapter, after specifying the N-filtered structure determined by a positive-degree function d( ) on a solvable polynomial algebra A = K[a1, . . . , an], we specify the correspond- ing N-filtered structure for free A-modules. Then, we introduce minimal filtered free resolutions for finitely generated modules over such N-filtered algebra A by introducing minimal F-bases and minimal standard bases for A-modules and their submodules with respect to good filtration; we show that any two minimal F-bases, respectively any two minimal stan- dard bases, have the same number of elements and the same number of elements of the same filtered-degree, minimal filtered free resolutions are unique up to strict filtered isomorphism of chain complexes in the cate- gory of filtered A-modules, and that minimal filtered free resolutions can be algorithmically computed in case A has a graded monomial ordering ≺gr. Since the standard bases we are going to introduce in terms of good fil- tration are generalization of classical Macaulay bases (see a remark given in Subsection 3.3), while a classical Macaulay basis V is characterized in terms of both the leading homogeneous elements (degree forms) of V and 89 90 Minimal Filtered Free Resolutions the homogenized elements of V (cf. [KR2], P.38, P.55), accordingly, on the basis of Chapter 4, our main idea in reaching the goal of this chapter is to use the filtered-graded transfer strategy (as proposed in [Li1]) by employing both the associated graded algebra (module) and the Rees al- gebra (module) of an N-filtered solvable polynomial algebra (of a filtered module). All notions, notations and conventions introduced in previous chapters are maintained. 5.1. N-Filtered Solvable Polynomial Algebras Comparing with the general theory of Z-filtered rings [LVO], in this sec- tion we formulate the N-filtered structure of solvable polynomial algebras determined by means of positive-degree functions. Let A = K[a1, . . . an] be a solvable polynomial algebra with admissible system (B, ≺), where B = {aα = aα1 n α = (α1, . . . , αn) ∈ Nn} is the PBW K-basis of A and ≺ is a monomial ordering on B, and let d( ) be a positive-degree function on A such that d(ai) = mi > 0, 1 ≤ i ≤ n (see Section 1.1 of Chapter 1). Put 1 · · · aαn FpA = K-span{aα ∈ B d(aα) ≤ p}, p ∈ N, then it is clear that FpA ⊆ Fp+1A for all p ∈ N, A = ∪p∈NFpA, and 1 ∈ F0A = K. 5.1.1. Definition With notation as above, if FpAFqA ⊆ Fp+qA holds for all p, q ∈ N, then we call A an N-filtered solvable polynomial algebra with respect to the positive-degree function d( ), and accordingly we call F A = {FpA}p∈N the N-filtration of A determined by d( ). Note that the N-filtration F A constructed above is clearly separated in the sense that if f is a nonzero element of L, then either f ∈ F0A = K or f ∈ FpA − Fp−1A for some p > 0. Thus, we may define the filtered-degree (abbreviated to fil-degree) of a nonzero f ∈ A, denoted dfil(f ), as follows dfil(f ) =(cid:26) 0, p, if f ∈ F0A = K, if p ∈ FpA − Fp−1A for some p > 0. Minimal Filtered Free Resolutions 91 Bearing in mind the definition of dfil(f ), the following featured prop- erty of F A will very much help us to deal with the associated graded structures of A and filtered A-modules. 5.1.2. Lemma If f = Pi λiaα(i) ∈ A with λi ∈ K ∗ and aα(i) ∈ B, then dfil(f ) = p if and only if d(aα(i′)) = p for some i′ if and only if d(f ) = p = dfil(f ), where d( ) is the given positive-degree function on A. Proof Exercise. (cid:3) Given a solvable polynomial algebra A = K[a1, . . . , an] and a positive- degree function d( ) on A, it follows from Definition 1.1.3 (Section 1, Chapter 1), Definition 5.1.1 and Lemma 5.1.2 above that the next propo- sition is clear now. 5.1.3. Proposition A is an N-filtered solvable polynomial algebra with respect to d( ) if, for 1 ≤ i < j ≤ n, all the relations ajai = λjiaiaj + fji with fji =P µkaα(k) presented in (S2′) (Section 4.1, Chapter 4), satisfy d(aα(k)) ≤ d(aiaj) whenever µk 6= 0. (cid:3) With the proposition presented above, the following examples may be better understood. Example (1) If A = K[a1, . . . , an] is an N-graded solvable polynomial algebra with respect to a positive-degree function d( ), i.e., A = ⊕p∈NAp with the degree-p homogeneous part Ap = K-span{aα ∈ B d(aα) = p} (see Section 4.1 of Chapter 4), then, with respect to the same positive- degree function d( ) on A, A is turned into an N-filtered solvable polynomial algebra with the N-filtration F A = {FpA}p∈N where each FpA = ⊕q≤pAq. Example (2) Let A = K[a1, . . . an] be a solvable polynomial algebra with the admissible system (B, ≺gr), where ≺gr is a graded monomial ordering on B with respect to a given positive-degree function d( ) on A (see the definition of ≺gr given in Section 1.1 of Chapter 1). Then by referring to (Definition 1.1.3 of Chapter 1) and the above proposition, one easily sees that A is an N-filtered solvable polynomial algebra with respect to the same d( ). In the case where ≺gr respects d(ai) = 1 for 1 ≤ i ≤ n, 92 Minimal Filtered Free Resolutions (Definition 1.1.3 of Chapter 1) entails that the generators of A satisfy ajai = λjiaiaj +P λji t at + µji, where 1 ≤ i < j ≤ n, λji ∈ K ∗, λji kℓakaℓ +P λji kℓ, λji t , µji ∈ K. In [Li1] such N-filtered solvable polynomial algebras are referred to as quadric solvable polynomial algebras which include numerous significant algebras such as Weyl algebras and enveloping algebras of finite dimen- sional Lie algebras. One is referred to [Li1] for some detailed study on quadric solvable polynomial algebras. The next example provides N-filtered solvable polynomial algebras in which some generators may have degree ≥ 2. Example (3) Consider the solvable polynomial algebra K-algebra A = K[a1, a2, a3] constructed in (Example (1) of Section 1.4, Chapter 1). Then, one checks that A is turned into an N-filtered solvable polynomial algebra with respect to the lexicographic ordering a3 ≺lex a2 ≺lex a1 and the degree function d( ) such that d(a1) = 2, d(a2) = 1, d(a3) = 4. More- over, one may also check that with respect to the same degree function d( ), the graded lexicographic ordering a3 ≺grlex a2 ≺grlex a1 is another choice to make A into an N-filtered solvable polynomial algebra. Let A be an N-filtered solvable polynomial algebra with respect to a given positive-degree function d( ), and let F A = {FpA}p∈N be the N- filtration of A determined by d( ). Then A has the associated N-graded K-algebra G(A) = ⊕p∈NG(A)p with G(A)0 = F0A = K and G(A)p = FpA/Fp−1A for p ≥ 1, where for f = f + Fp−1A ∈ G(A)p, g = g + Fq−1A, the multiplication is given by f g = f g + Fp+q−1A ∈ G(A)p+q. Another N-graded K-algebra determined by F A is the Rees algebra eA of A, which is defined as eA = ⊕p∈NeAp with eAp = FpA, where the multiplication of eA is induced by FpAFqA ⊆ Fp+qA, p, q ∈ N. For convenience, we fix the following notations once for all: • If h ∈ G(A)p and h 6= 0, then we write dgr(h) for the gr-degree of h as a homogeneous element of G(A), i.e., dgr(h) = p. • If H ∈ eAp and H 6= 0, then we write dgr(H) for the gr-degree of H as a homogeneous element of eA, i.e., dgr(H) = p. Concerning the N-graded structure of G(A), if f ∈ A with dfil(f ) = p, then by Lemma 5.1.2, the coset f + Fp−1A represented by f in G(A)p is a Furthermore, if we write Z for the homogeneous element of degree 1 a nonzero f ∈ FpA, we denote the corresponding homogeneous element eA of A, we note that a nonzero f ∈ FqA represents a homogeneous element of degree q in eAq on one hand, and on the other hand it represents a homogeneous element of degree q1 in eAq1, where q1 = dfil(f ) ≤ q. So, for of degree p in eAp by hp(f ), while we use ef to denote the homogeneous element represented by f in eAp1 with p1 = dfil(f ) ≤ p. Thus, dgr(ef ) = dfil(f ), and we see that hp(f ) = ef if and only if dfil(f ) = p. in eA1 represented by the multiplicative identity element 1, then Z is a central regular element of eA, i.e., Z is not a divisor of zero and is contained in the center of eA. Bringing this homogeneous element Z into play, the homogeneous elements of eA are featured as follows: • If f ∈ A with dfil(f ) = p1 then for all p ≥ p1, hp(f ) = Z p−p1ef . In other words, if H ∈ eAp is a nonzero homogeneous element of degree p, then there is some f ∈ FpA such that H = Z p−dfil(f )ef = ef + (Z p−dfil(f ) − 1)ef . G(A) ∼= eA/hZi as N-graded K-algebras and A ∼= eA/h1− Zi as K-algebras It follows that by sending H to f +Fp−1A and sending H to f respectively, Minimal Filtered Free Resolutions 93 nonzero homogeneous element of degree p. If we denote this homogeneous element by σ(f ) (in the literature it is referred to as the principal symbol of f ), then dfil(f ) = p = dgr(σ(f )). However, considering the Rees algebra (cf. [LVO]). Since a solvable polynomial algebra A is necessarily a domain (Propo- sition 1.1.4, Chapter 1), we summarize two useful properties concerning the multiplication of G(A) and eA respectively into the following lemma. 5.1.4. Lemma With notation as before, let f, g be nonzero elements of A with dfil(f ) = p1, dfil(g) = p2. Then (i) σ(f )σ(g) = σ(f g); (ii) efeg =ff g. If p1 + p2 ≤ p, then hp(f g) = Z p−p1−p2efeg. Proof Exercise. (cid:3) The results given in the next theorem, which are analogues of those concerning quadric solvable polynomial algebras in ([LW], Section 3; [Li1], CH.IV), may be derived in a similar way as in loc. cit. (With the prepa- ration made above, one is also invited to give the detailed proof as an 94 exercise). Minimal Filtered Free Resolutions 5.1.5. Theorem Let A = K[a1, . . . , an] be a solvable polynomial algebra with the admissible system (B, ≺gr), where ≺gr is a graded monomial ordering on B with respect to a given positive-degree function d( ) on A, thereby A is an N-filtered solvable polynomial algebra with respect to the same d( ) by the foregoing Example (2), and let F A = {FpA}p∈N be the corresponding N-filtration of A. Considering the associated graded algebra G(A) as well as the Rees algebra eA of A, the following statements hold. (i) G(A) = K[σ(a1), . . . , σ(an)], G(A) has the PBW K-basis σ(B) = {σ(a)α = σ(a1)α1 · · · σ(an)αn α = (α1, . . . , αn) ∈ Nn}, and, by referring to (Definition 1.1.3, Chapter 1), for σ(a)α, σ(a)β ∈ σ(B) such that aαaβ = λα,βaα+β + fα,β, where λα,β ∈ K ∗, if fα,β = 0 then σ(a)ασ(a)β = λα,βσ(a)α+β, where σ(a)α+β = σ(a1)α1+β1 · · · σ(an)αn+βn; and in the case where fα,β =Pj µα,β σ(a)ασ(a)β = λα,βσ(a)α+β + Xd(aα(k))=d(aα+β ) j aα(j) 6= 0 with µα,β j ∈ K, µα,β j σ(a)α(k). Moreover, the ordering ≺G(A) defined on σ(B) subject to the rule: σ(a)α ≺G(A) σ(a)β ⇐⇒ aα ≺gr aβ, aα, aβ ∈ B, is a graded monomial ordering with respect to the positive-degree function d( ) on G(A) such that d(σ(ai)) = d(ai) for 1 ≤ i ≤ n, that turns G(A) into an N-graded solvable polynomial algebra. (ii) eA = K[ea1, . . . ,ean, Z] where Z is the central regular element of degree 1 in eA1 represented by 1, eA has the PBW K-basis and, by referring to (Definition 1.1.3, Chapter 1), for eaαZ s, ·eaβZ t ∈ eB such that aαaβ = λα,βaα+β + fα,β, where λα,β ∈ K ∗, if fα,β = 0 then n Z m α = (α1, . . . , αn) ∈ Nn, m ∈ N}, eB = {eaαZ m =eaα1 1 · · ·eaαn eaαZ s ·eaβZ t = λα,βeaα+βZ s+t, whereeaα+β =eaα1+β1 1 ; n · · ·eaαn+βn Minimal Filtered Free Resolutions 95 and in the case where fα,β =Pj µα,β j aα(j) 6= 0 with µα,β j ∈ K, where q = d(aα+β) + s + t, mj = d(aα(j)). eaαZ s ·eaβZ t = λα,βeaα+βZ s+t +Pj µα,β j eaα(j)Z q−mj , defined on eB subject to the rule: Moreover, the ordering ≺ eA aα, aβ ∈ B, eaαZ s ≺ eAeaβZ t ⇐⇒ aα ≺gr aβ, or aα = aβ and s < t, is a monomial ordering on eB (which is not necessarily a graded monomial ordering), that turns eA into an N-graded solvable polynomial algebra with respect to the positive-degree function d( ) on eA such that d(Z) = 1 and d(eai) = d(ai) for 1 ≤ i ≤ n. (cid:3) The corollary presented below will be very often used in discussing left Grobner bases and standard bases for submodules of filtered free A- modules and their associated graded free G(A)-modules as well as the 5.1.6. Corollary With the assumption and notations as in Theorem graded free eA-modules (Section 5.3, Section 5.4). 3.1.4, if f = λaα +Pj µjaα(j) with d(f ) = p and LM(f ) = aα, then p = dfil(f ) = dgr(σ(f )) = dgr(ef ), and LM(σ(f )) = σ(a)α = σ(LM(f )); σ(f ) = λσ(a)α +Pd(aα(jk ))=p µjk σ(a)α(jk); ef = λeaα +Pj µjeaα(j)Z p−d(aα(j)); LM(ef ) =eaα = ^ LM(f ), where LM(f ), LM(σ(f )) and LM(ef ) are taken with respect to ≺gr, ≺G(A) and ≺ eA Proof By referring to Lemma 5.1.2 and Lemma 5.1.4, this is a straight- forward exercise. (cid:3) respectively. 5.2. N-Filtered Free Modules Let A = K[a1, . . . , an] be an N-filtered solvable polynomial algebra with the filtration F A = {FpA}p∈N determined by a positive-degree function 96 Minimal Filtered Free Resolutions d( ) on A, and let (B, ≺) be a fixed admissible system of A. Consider a free A-module L = ⊕s i=1Aei with the A-basis {e1, . . . , es}. Then L has the K-basis B(e) = {aαei aα ∈ B, 1 ≤ i ≤ s}. If {b1, . . . , bs} is an arbitrarily fixed subset of N, then, with F L = {FqL}q∈N defined by putting FqL = {0} if q < min{b1, . . . , bs}; otherwise FqL = or alternatively, for q ≥ min{b1, . . . , bs}, sXi=1  Xpi+bi≤q FpiA ei, FqL = K-span{aαei ∈ B(e) d(aα) + bi ≤ q}, L forms an N-filtered free A-module with respect to the N-filtered struc- ture of A, that is, every FqL is a K-subspace of L, FqL ⊆ Fq+1L for all q ∈ N, L = ∪q∈NFqL, FpAFqL ⊆ Fp+qL for all p, q ∈ N, and for each i = 1, . . . , s, ei ∈ F0L if bi = 0; otherwise ei ∈ FbiL − Fbi−1L. Convention. Let A be an N-filtered solvable polynomial algebra with respect to a positive-degree function d( ). Unless otherwise stated, from now on in the subsequent sections if we say that L = ⊕s i=1Aei is a filtered free A-module with the filtration F L = {FqL}q∈N, then F L is always meant the type as constructed above. Let L = ⊕s i=1Aei be a filtered free A-module with the filtration F L = {FqL}q∈N, which is constructed with respect to a given subset {b1, . . . , bs} ⊂ N. Then F L is separated in the sense that if ξ is a nonzero element of L, then either ξ ∈ F0L or ξ ∈ FqL − Fq−1L for some q > 0. Thus, to make the discussion on F L consistent with that on F A in Section 5.1, we define the filtered-degree (abbreviated to fil-degree) of a nonzero ξ ∈ L, denoted dfil(ξ), as follows dfil(ξ) =(cid:26) 0, q, if ξ ∈ F0L, if ξ ∈ FqL − Fq−1L for some q > 0. For instance, we have dfil(ei) = bi, 1 ≤ i ≤ s. Comparing with Lemma 5.1.2 we first note the following Minimal Filtered Free Resolutions 97 5.2.1. Lemma Let ξ ∈ L − {0}. Then dfil(ξ) = q if and only if ξ = Pi,j λijaα(ij )ej, where λij ∈ K ∗ and aα(ij ) ∈ B with α(ij ) = (αij1, . . . , αijn ) ∈ Nn, in which some monomial aα(ij )ej satisfy d(aα(ij )) + bj = q. Proof Exercise. (cid:3) Let L = ⊕s i=1Aei be a filtered free A-module with the filtration F L = {FqL}q∈N such that dfil(ei) = bi, 1 ≤ i ≤ s. Considering the the associated N-graded algebra G(A) of A, the filtered free A module L has the associated N-graded G(A)-module G(L) = ⊕q∈NG(L)q with G(L)q = FqL/Fq−1L, where for f = f + Fp−1A ∈ G(A)p, ξ = ξ + Fq−1L ∈ G(L)q, the module action is given by f · ξ = f ξ + Fp+q−1L ∈ G(L)p+q. As with homogeneous elements in G(A), if h ∈ G(L)q and h 6= 0, then we write dgr(h) for the degree of h as a homogeneous element of G(L), i.e., dgr(h) = q. If ξ ∈ L with dfil(ξ) = q, then the coset ξ + Fq−1L represented by ξ in G(L)q is a nonzero homogeneous element of degree q, and if we denote this homogeneous element by σ(ξ) (in the literature it is referred to as the principal symbol of ξ) then dgr(σ(ξ)) = q = dfil(ξ). and H 6= 0, then we write dgr(H) for the degree of H as a homogeneous q1 = dfil(ξ) ≤ q. So, for a nonzero ξ ∈ FqL we denote the corresponding Furthermore, considering the Rees algebra eA of A, the filtration F L = {FqL}q∈N of L also defines the Rees module eL of L, which is the N-graded eA-moduleeL = ⊕q∈NeLq, whereeLq = FqL and the module action is induced by FpAFqL ⊆ Fp+qL. As with homogeneous elements in eA, if H ∈ eLq element of eL, i.e., dgr(H) = q. Note that any nonzero ξ ∈ FqL represents a homogeneous element of degree q in eLq on one hand, and on the other hand it represents a homogeneous element of degree q1 in eLq1, where homogeneous element of degree q ineLq by hq(ξ), while we useeξ to denote the homogeneous element represented by ξ in eLq1 with q1 = dfil(ξ) ≤ q. Thus, dgr(eξ) = dfil(ξ), and we see that hq(ξ) =eξ if and only if dfil(ξ) = q. eA1 represented by the multiplicative identity element 1, then, similar to isomorphism L ∼= eL/(1 − Z)eL and graded G(A)-module isomorphism G(L) ∼=eL/ZeL. about dfil(f ξ), σ(f ξ), hℓ(f ξ) andff ξ, respectively. We also note that if Z denotes the homogeneous element of degree 1 in the discussion given in the last section, one checks that there are A-module For f ∈ A, ξ ∈ L, the next lemma records several convenient facts 98 Minimal Filtered Free Resolutions 5.2.2. Lemma With notation as above, the following statements hold. (i) dfil(f ξ) = d(f ) + dfil(ξ) holds for all nonzero f ∈ A and nonzero ξ ∈ L. (ii) σ(f )σ(ξ) = σ(f ξ) holds for all nonzero f ∈ A and nonzero ξ ∈ L. (iii) If ξ ∈ L with dfil(ξ) = q ≤ ℓ, then hℓ(ξ) = Z ℓ−qeξ. Furthermore, let f ∈ A with dfil(f ) = p, ξ ∈ L with dfil(ξ) = q. Then efeξ =ff ξ; if p + q ≤ ℓ, then hℓ(f ξ) = Z ℓ−p−qefeξ. Proof Since A is a solvable polynomial algebra, G(A) and eA are N-graded solvable polynomial algebras by Theorem 5.1.5, thereby they are neces- sarily domains (Proposition 1.1.4(ii), Chapter 1). Noticing the definition of dfil(f ) and dfil(ξ), by Lemma 5.2.1, the verification of (i) -- (iii) are then straightforward. (cid:3) 5.2.3. Proposition With notation as fixed before, let L = ⊕s i=1Aei be a filtered free A-module with the filtration F L = {FqL}q∈N such that dfil(ei) = bi, 1 ≤ i ≤ s. The following two statements hold. (i) G(L) is an N-graded free G(A)-module with the homogeneous G(A)- basis {σ(e1), . . . , σ(es)}, that is, G(L) = ⊕s i=1G(A)σ(ei) = ⊕q∈NG(L)q with G(L)q = Xpi+bi=q G(A)piσ(ei) q ∈ N. Moreover, σ(B(e)) = {σ(aαei) = σ(a)ασ(ei) aαei ∈ B(e)} forms a K- basis for G(L). q ∈ N. i=1eAeei = ⊕q∈NeLq with eLq = Xpi+bi=q eApieei, (ii) eL is an N-graded free eA-module with the homogeneous eA-basis {ee1, . . . ,ees}, that is, eL = ⊕s Moreover, gB(e) = {eaαZ meei eaαZ m ∈ eB, 1 ≤ i ≤ s} forms a K-basis for eL, where eB is the PBW K-basis of eA determined in Theorem 5.1.5(ii). i=1(cid:16)Ppi+bi≤q FpiA(cid:17) ei, then dfil(ξ) ≤ q. By Lemma 5.2.2, Ps if ξ = Ps Proof Since dfil(ei) = bi, 1 ≤ i ≤ s, i=1 fiei ∈ FqL = σ(ξ) =Pd(fi)+bi=q σ(fi)σ(ei) ∈Ps i=1 Z q−d(fi)−biefieei ∈Ps hq(ξ) =Ps i=1 eAq−bieei. i=1 G(A)q−bi σ(ei) This shows that {σ(e1), . . . , σ(es)} and {ee1, . . . ,ees} generate the G(A)- module G(L) and the eA-module eL, respectively. Next, since each σ(ei) is Minimal Filtered Free Resolutions 99 a homogeneous element of degree bi, if a degree-q homogeneous element i=1 σ(fi)σ(ei) = 0, where fi ∈ A, dfil(fi) + bi = q, 1 ≤ i ≤ s, then i=1 fiei ∈ Fq−1L and hence each fi ∈ Fq−1−biA by Lemma 5.2.1, a contradiction. It follows that {σ(e1), . . . , σ(es)} is linearly independent Ps Ps over G(A). Concerning the linear independence of {ee1, . . . ,ees} over eA, since each eei is a homogeneous element of degree bi, if a degree-q homo- geneous element Ps i=1 hpi(fi)eei = 0, where fi ∈ FpiA and pi + bi = q, 1 ≤ i ≤ s, thenPs 5.2.1, ξ =Pi,j λijaα(ij )ej with λij ∈ K ∗ and d(aα(ij )) + bj = ℓij ≤ q. It i=1 fiei = 0 in FqL and consequently all fi = 0, thereby hpi(fi) = 0 as desired. Finally, if ξ ∈ FqL with dfil(ξ) = q, then by Lemma follows from Lemma 5.2.2 that σ(ξ) =Pℓik=q λikσ(a)α(ik )σ(ek), eξ =Pi,j λijZ q−ℓijeaα(ij )eej. Therefore, a further application of Lemma 5.2.1 and Lemma 5.2.2 shows that σ(B(e)) and gB(e) are K-bases for G(L) and eL respectively. 5.3. Filtered-Graded Transfer of Grobner Bases for Modules Throughout this section, we let A = K[a1, . . . , an] be a solvable poly- nomial algebra with the admissible system (B, ≺gr), where B = {aα = aα1 n α = (α1, . . . , αn) ∈ Nn} is the PBW K-basis of A and ≺gr is 1 · · · aαn a graded monomial ordering with respect to some given positive-degree function d( ) on A (see Section 1). Thereby A is turned into an N- filtered solvable polynomial algebra with the filtration F A = {FpA}p∈N constructed with respect to the same d( ) (see Example (2) of Section 5.1). In order to compute minimal standard bases by employing both inhomo- geneous and homogenous left Grobner bases in the subsequent Section 5.5, our aim of the current section is to establish the relations between left Grobner bases in a filtered free (left) A-module L and homogeneous left Grobner bases in G(L) as well as homogeneous left Grobner bases in eL, which are just module theory analogues of the results on filtered- graded transfer of Grobner bases for left ideals given in ([LW], [Li1]). All notions, notations and conventions introduced in previous sections are maintained. Let L = ⊕s i=1Aei be a filtered free A-module with the filtration F L = 100 Minimal Filtered Free Resolutions {FqL}q∈N such that dfil(ei) = bi, 1 ≤ i ≤ s. Bearing in mind Lemma 5.2.1, we say that a left monomial ordering on B(e) is a graded left monomial ordering, denoted by ≺e-gr, if for aαei, aβej ∈ B(e), aαei ≺e-gr aβej implies dfil(aαei) = d(aα) + bi ≤ d(aβ) + bj = dfil(aβej). For instance, with respect to the given graded monomial ordering ≺gr on B and the N-filtration F A of A, if {f1, . . . , fs} ⊂ A is a finite subset such that d(fi) = bi = dfil(ei), 1 ≤ i ≤ s, then it is straightforward to check that the Schreyer ordering (see Section 2.1 of Chapter 2) induced by {f1, . . . , fs} subject to the rule: for aαei, aβej ∈ B(e), aαei ≺s-gr aβej ⇐⇒ LM(aαfi) ≺gr LM(aβfj), or LM(aαfi) = LM(aβfj) and i < j. is a graded left monomial ordering on B(e). More generally, let {ξ1, . . . , ξm} ⊂ L be a finite subset, where dfil(ξi) = qi, 1 ≤ i ≤ m, and let L1 = ⊕m i=1Aεi be the filtered free A-module with the filtration F L1 = {FqL1}q∈N such that dfil(εi) = qi, 1 ≤ i ≤ m. Then, given any graded left monomial ordering ≺e-gr on B(e), the Schreyer ordering ≺s-gr defined on the K-basis B(ε) = {aαεi aα ∈ B, 1 ≤ i ≤ m} of L1 subject to the rule: for aαεi, aβεj ∈ B(ε), aαεi ≺s-gr aβεj ⇐⇒ LM(aαξi) ≺e-gr LM(aβξj), or LM(aαξi) = LM(aβξj) and i < j, is a graded left monomial ordering on B(ε). Comparing with Lemma 5.1.2 and Lemma 5.2.1, the lemma given below reveals the intrinsic property of a graded left monomial ordering employed by a filtered free A-module. 5.3.1. Lemma Let L = ⊕s i=1Aei be a filtered free A-module with the filtration F L = {FqL}q∈N such that dfil(ei) = bi, 1 ≤ i ≤ s, and let ≺e-gr be a graded left monomial ordering on B(e). Then ≺e-gr is compatible with the filtration F L of L in the sense that ξ ∈ FqL−Fq−1L, i.e. dfil(ξ) = q, if and only if LM(ξ) = aαei with dfil(aαei) = d(aα) + bi = q. Proof Let ξ = Pi,j λijaα(ij )ej ∈ FqL − Fq−1L. Then by Lemma 5.2.1, there is some aα(iℓ)eℓ such that d(aα(iℓ))+bℓ = q. If LM(ξ) = aα(it)et with Minimal Filtered Free Resolutions 101 respect to ≺e-gr, then aα(ik)ek ≺e-gr aα(it)et for all aα(ik)ek with k 6= t. If ℓ = t, then d(aα(it)) + bt = q; otherwise, since ≺e-gr is a graded left monomial ordering, we have d(aα(ik )) + bk ≤ d(aα(it)) + bt, in particular, q = d(aα(iℓ))+bℓ ≤ d(aα(it))+bt ≤ q. Hence dfil(aα(it)et) = d(aα(it))+bt = q, as desired. Conversely, for ξ = Pi,j λijaα(ij )ej ∈ L, if, with respect to ≺e-gr, LM(ξ) = aα(it)et with dfil(aα(it)et) = d(aα(it))+bt = q, then aα(ik)ek ≺e-gr aα(it)et for all k 6= t. Since ≺e-gr is a graded left monomial ordering, we have d(aα(ik )) + bk ≤ d(aα(it)) + bt = q. It follows from Lemma 5.2.1 that dfil(ξ) = q, i.e., ξ ∈ FqL − Fq−1L. (cid:3) Let L = ⊕s i=1Aei be a filtered free A-module with the filtration F L = {FqL}q∈N such that dfil(ei) = bi, 1 ≤ i ≤ s. Then, by Proposition 5.2.3 we know that the associated graded G(A)-module G(L) of L is an N- graded free module, i.e., G(L) = ⊕s i=1G(A)σ(ei) with the homogeneous G(A)-basis {σ(e1), . . . , σ(es)}, and that G(L) has the K-basis σ(B(e)) = {σ(aαei) = σ(a)ασ(ei) aαei ∈ B(e)}. Furthermore, let ≺e-gr be a graded left monomial ordering on B(e) as defined in the beginning of this section. Then we may define an ordering ≺σ(e)-gr on σ(B(e)) subject to the rule: σ(a)ασ(ei) ≺σ(e)-gr σ(a)βσ(ej ) ⇐⇒ aαei ≺e-gr aβej, aαei, aβej ∈ B(e). 5.3.2. Lemma With the ordering ≺σ(e)-gr defined above, the following statements hold. (i) ≺σ(e)-gr is a graded left monomial ordering on σ(B(e)). (ii) (Compare with Corollary 5.1.6.) LM(σ(ξ)) = σ(LM(ξ)) holds for all nonzero ξ ∈ L, , where the monomial orderings used for taking LM(σ(ξ)) and LM(ξ) are ≺σ(e)-gr and ≺e-gr respectively. Proof (i) Noticing that the given monomial ordering ≺gr on A is a graded monomial ordering with respect to a positive-degree function d( ) on A, it follows from Theorem 5.1.5(i) that G(A) is turned into an N-graded solvable polynomial algebra by using the graded monomial ordering ≺G(A) defined on σ(B) subject to the rule: σ(a)α ≺G(A) σ(a)β ⇐⇒ aα ≺gr aβ, where the positive-degree function on G(A) is given by d(σ(ai)) = d(ai), 1 ≤ i ≤ n. Moreover, since σ(ei) is a homogeneous element of degree bi in G(L), 1 ≤ i ≤ s, by Lemma 5.2.2, it is then straightforward to verify that ≺σ(e)-gr is a graded left monomial ordering on σ(B(e)). 102 Minimal Filtered Free Resolutions (ii) Let ξ = Pi,j λijaα(ij )ej, where λij ∈ K ∗ and aα(ij ) ∈ B with α(ij) = (αij1 , . . . , αijn ) ∈ Nn. i.e., ξ ∈ FqL − Fq−1L, then by Lemma 5.3.1, LM(ξ) = aα(it)et for some t such that dfil(aα(it)et) = d(aα(it)) + bt = q. Since ≺e-gr is a left graded monomial ordering on B(e), by Lemma 5.2.2 we have σ(ξ) = λitσ(a)α(it)σ(et) + If dfil(ξ) = q, Pd(aα(ik ))+bk=q λikσ(a)α(ik )σ(ek). It follows from the definition of ≺σ(e)-gr that LM(σ(ξ)) = σ(a)α(it)σ(et) = σ(LM(ξ)), as desired. (cid:3) 5.3.3. Theorem Let N be a submodule of the filtered free A-module L = ⊕s i=1Aei, where L is equipped with the filtration F L = {FqL}q∈N such that dfil(ei) = bi, 1 ≤ i ≤ s, and let ≺e-gr be a graded left monomial ordering on B(e). For a subset G = {g1, . . . , gm} of N , the following two statements are equivalent. (i) G is a left Grobner basis of N with respect to ≺e-gr. (ii) Putting σ(G) = {σ(g1), . . . , σ(gm)} and considering the filtration F N = {FqN = FqL ∩ N }q∈N of N induced by F L, σ(G) is a left Grobner basis for the associated graded G(A)-module G(N ) of N with respect to the graded left monomial ordering ≺σ(e)-gr defined above. Proof (i) ⇒ (ii) Note that any nonzero homogeneous element of G(N ) is of the form σ(ξ) with ξ ∈ N . If G is a left Grobner basis of N , then there exists some gi ∈ G such that LM(gi)LM(ξ), i.e., there is a monomial aα ∈ B such that LM(ξ) = LM(aαLM(gi)). Since the given left monomial ordering ≺e-gr on B(e) is a graded left monomial ordering, it follows from Lemma 5.2.2 and Lemma 5.3.2 that LM(σ(ξ))) = σ(LM(ξ)) = σ(LM(aαLM(gi))) = LM(σ(aαLM(gi))) = LM(σ(a)ασ(LM(gi))) = LM(σ(a)αLM(σ(gi))). This shows that LM(σ(gi))LM(σ(ξ)), thereby σ(G) is a left Grobner basis for G(N ). (ii) ⇒ (i) Suppose that σ(G) is a left Grobner basis of G(N ) with respect to ≺σ(e)-gr. If ξ ∈ N and ξ 6= 0, then σ(ξ) 6= 0, and there exists a σ(gi) ∈ σ(G) such that LM(σ(gi))LM(σ(ξ)), i.e., there is a monomial σ(a)α ∈ σ(B) such that LM(σ(ξ)) = LM(σ(a)αLM(σ(gi))). Again as ≺e-gr is a left graded monomial ordering on B(e), by Lemma 5.2.2 and Minimal Filtered Free Resolutions 103 Lemma 5.3.2 we have σ(LM(ξ)) = LM(σ(ξ)) = LM(σ(a)αLM(σ(gi))) = LM(σ(a)ασ(LM(gi))) = LM(σ(aαLM(gi))) = σ(LM(aαLM(gi))). This shows that dfil(LM(ξ)) = dfil(LM(aαLM(gi))). Since both LM(ξ) and LM(aαLM(gi)) are monomials in B(e), it follows from the con- struction of F L and Lemma 5.3.1 that LM(ξ) = LM(aαLM(gi)), i.e., LM(gi)LM(ξ). This shows that G is a left Grobner basis for N . (cid:3) Similarly, in light of Proposition 5.2.3 we may define an ordering ≺ee on the K-basis gB(e) = {Z meaαeei Z meaα ∈ eB, 1 ≤ i ≤ s} of the N-graded free i=1eAeei subject to the rule: for Z seaαeei, Z teaβeej ∈ gB(e), eA-module eL = ⊕s Z seaαeei ≺ee Z teaβeej ⇐⇒ aαei ≺e-gr aβej, or aαei = aβej and s < t, where ≺e-gr is a given graded left monomial ordering on B(e). 5.3.4. Lemma With the ordering ≺ee defined above, the following state- ments hold. LM(ξ) holds for all left monomial ordering). (i) ≺ee is a left monomial ordering on gB(e) (but not necessarily a graded (ii) (Compare with Corollary 5.1.6.) LM(eξ) = ^ nonzero ξ ∈ L, , where the monomial orderings used for taking LM(eξ) and LM(ξ) are ≺ee and ≺e-gr respectively. Proof (i) Noticing that the given monomial ordering ≺gr for A is a graded monomial ordering with respect to a positive-degree function d( ) on A, it polynomial algebra by using the monomial ordering ≺ eA subject to the rule: follows from Theorem 5.1.5(ii) that eA is turned into an N-graded solvable defined on eB eaαZ s ≺ eAeaβZ t ⇐⇒ aα ≺gr aβ, or aα = aβ and s < t, where the positive-degree function on eA is given by d(eai) = d(ai) for 1 ≤ i ≤ n, and d(Z) = 1. Moreover, since eei is a homogeneous element of degree bi in eA, 1 ≤ i ≤ s, by Lemma 5.2.2, it is then straightforward to verify that ≺ee is a left monomial ordering on ]B(e). aα, aβ ∈ B, 104 Minimal Filtered Free Resolutions α(ij ) = (αij1 , . . . , αijn) ∈ Nn. If dfil(ξ) = q, i.e., ξ ∈ FqL − Fq−1L, then by Lemma 5.3.1, LM(ξ) = aα(it)et for some t such that dfil(aα(it)et) = d(aα(it)) + bt = q. Since ≺e-gr is a left graded monomial ordering on B(e), (ii) Let ξ = Pi,j λijaα(ij )ej, where λij ∈ K ∗ and aα(ij ) ∈ B with by Lemma 5.2.2 we have eξ = λiteaα(it)eet +Pj6=t λijZ q−ℓijeaα(ij )eej, where that LM(eξ) =eaα(it)eet = ^ ℓij = dfil(aα(ij )ej) = d(aα(ij )) + dj. It follows from the definition of ≺ee (cid:3) 5.3.5. Theorem Let N be a submodule of the filtered free A-module L = ⊕s i=1Aei, where L is equipped with the filtration F L = {FqL}q∈N such that dfil(ei) = bi, 1 ≤ i ≤ s, and let ≺e-gr be a graded left monomial ordering on B(e). For a subset G = {g1, . . . , gm} of N , the following two statements are equivalent. (i) G is a left Grobner basis of N with respect to ≺e-gr. LM(ξ), as desired. ≺ee defined before Lemma 5.3.4. {FqN = FqL ∩ N }q∈N of N induced by F L, τ (G) is a left Grobner basis (ii) Putting τ (G) = {eg1, . . . ,egm} and considering the filtration F N = for the Rees module eN of N with respect to the left monomial ordering Proof (i) ⇒ (ii) Note that any nonzero homogeneous element of eN is of hq(ξ) = Z q−q1eξ. If G is a left Grobner basis of N , then there exists some gi ∈ G such that LM(gi)LM(ξ), i.e., there is a monomial aα ∈ B such that LM(ξ) = LM(aαLM(gi)). It follows from Lemma 5.2.2 and Lemma 5.3.4 that the form hq(ξ) for some ξ ∈ FqN with dfil(ξ) = q1 ≤ q. By Lemma 5.2.2, LM(ξ) LM(gi)) LM(eξ) = ^ = (LM(aαLM(gi)))e = LM((aαLM(gi))e) = LM(eaα ^ = LM(eaαLM(egi)). LM(hq(ξ)) = LM(Z q−q1eξ) = Z q−q1LM(eξ) = Z q−q1LM(eaαLM(egi)) = LM(zq−q1eaαLM(egi)). Hence, noticing the definition of ≺ee we have This shows that LM(egi)LM(hq(ξ)), thereby τ (G) is a left Grobner basis of eN . Minimal Filtered Free Resolutions 105 (ii) ⇒ (i) If ξ ∈ N and ξ 6= 0, then eξ 6= 0 and LM(eξ) = ^ Lemma 5.3.4. Suppose that τ (G) is a left Grobner basis of eN with respect to ≺ee. Then there exists someegi ∈ τ (G) such that LM(egi)LM(eξ), i.e., there is a monomial Z meaγ ∈ eB such that LM(eξ) = LM(Z meaγ LM(egi)). Since the given left monomial ordering ≺e-gr on B(e) is a graded left monomial ordering, it follows from Lemma 5.2.2, the definition of ≺ee and Lemma 5.3.2 that LM(ξ) by gaαej = ^ LM(ξ) = LM(eξ) = LM(Z meaγLM(egi)) = Z m(LM((aγ LM(gi))e)) = Z m(LM(aγLM(gi)))e. Noticing the discussion oneL and the role played by Z given before Lemma 5.2.2, we must have m = 0, thereby LM(ξ) = LM(aγ LM(gi)). This shows that G is a left Grobner basis for N . Remark. It is known that Grobner bases for ungraded ideals in both a commutative polynomial algebra and a noncommutative free algebra can be obtained via computing homogeneous Grobner bases for graded ideals in the corresponding homogenized (graded) algebras (cf. [Frob], [Li2]). In our case here for an N-filtered solvable polynomial algebra A with respect to a positive-degree function d( ), by using a (de)homogenization-like trick theory and strategy similar to that given in (Section 3.6 and Section 3.7 of [Li2]), so that left Grobner bases of submodules (left ideals) in L (in A) can be obtained via computing homogeneous left Grobner bases of with respect to the central regular element Z in eA, the discussion on eA andeL presented in previous Section 5.1 indeed enables us to have a whole graded submodules (graded left ideals) in eL (in eA). Since such a topic is beyond the scope of this text, we omit the detailed discussion here. 5.4. F-Bases and Standard Bases with Respect to Good Filtration Let A = K[a1, . . . , an] be an N-filtered solvable polynomial algebra with admissible system (B, ≺) and the N-filtration F A = {FpA}p∈N con- structed with respect to a given positive-degree function d( ) on A (see Section 5.1). In this section, we introduce F-bases and standard bases re- spectively for N-filtered left A-modules and their submodules with respect 106 Minimal Filtered Free Resolutions to good filtration, and we show that any two minimal F-bases, respec- tively any two minimal standard bases have the same number of elements and the same number of elements of the same fil-degree. Moreover, we show that a standard basis for a submodule N of a filtered free A-module L can be obtained via computing a left Grobner basis of N with respect to a graded left monomial ordering. All notions, notations and conventions used before are maintained. Let M be a left A-module. We say that M is an N-filtered A-module if M has a filtration F M = {FqM }q∈N, where each FqM is a K-subspace of M , such that M = ∪q∈NFqM , FqM ⊆ Fq+1M for all q ∈ N, and FpAFqM ⊆ Fp+qM for all p, q ∈ N. Convention. Unless otherwise stated, from now on in the subsequent sections a filtered A-module M is always meant an N-filtered module with a filtration of the type F M = {FqM }q∈N as described above. Let G(A) be the associated graded algebra of A, eA the Rees algebra of A, and Z the homogeneous element of degree 1 in eA1 represented by the multiplicative identity 1 of A (see Section 5.1). If M is a filtered A-module with the filtration F M = {FqM }q∈N, then, actually as with a filtered free A-module in Section 5.2, M has the associated graded G(A)-module G(M ) = ⊕q∈NG(M )q with G(M )0 = F0M and G(M )q = FqM/Fq−1M for q ≥ 1, and the Rees module of M is defined as the define the fil-degree of a nonzero ξ ∈ M , that is, dfil(ξ) = 0 if ξ ∈ F0M , and if ξ ∈ FqM − Fq−1M for some q > 0, then dfil(ξ) = q. For a nonzero ξ ∈ M with dfil(ξ) = q ≥ 0, if we write σ(ξ) for the nonzero graded eA-module fM = ⊕q∈NfMq with each fMq = FqM . Also, we may homogeneous element of degree q represented by ξ in G(M )q, eξ for the degree-q homogeneous element represented by ξ in fMq, and hq′(ξ) for the degree-q′ homogeneous element represented by ξ in fMq′ with q < q′, then dfil(ξ) = q = dgr(σ(ξ)) = dgr(eξ), and dgr(hq′(ξ)) = q′. Finally, as for a filtered free A-module in Section 5.2, we have fM /ZfM ∼= G(M ) as graded G(A)-modules, and fM /(1 − Z)fM ∼= M as A-modules. With notation as fixed above, the lemma presented below is a version of ([LVO], Ch.I, Lemma 5.4, Theorem 5.7) for N-filtered modules. 5.4.1. Lemma Let M be a filtered A-module with the filtration F M = Minimal Filtered Free Resolutions 107 {FqM }q∈N, and V = {v1, . . . , vm} a finite subset of nonzero elements in M . The following statements are equivalent: (i) There is a subset S = {n1, . . . , nm} ⊂ N such that FqM = mXi=1  Xpi+ni≤q FpiA vi, q ∈ N; i=1 G(A)σ(vi); (ii) G(M ) =Pm (iii) fM =Pm i=1 eAevi. (cid:3) 5.4.2. Definition Let M be a filtered A-module with the filtration F M = {FqM }q∈N, and let V = {v1, . . . , vm} ⊂ M be a finite subset of nonzero elements. If V satisfies one of the equivalent conditions of Lemma 5.4.1, then we call V an F-basis of M with respect to F M . Let M be a filtered A-module with the filtration F M = {FqM }q∈N. If V is an F-basis of M with respect to F M , then it is necessary to note that (1) since M = ∪q∈NFqM , it is clear that V is certainly a generating set of the A-module M , i.e., M =Pm (2) due to Lemma 5.4.1(i), the filtration F M is usually referred to as a good filtration of M in the literature concerning filtered module theory (cf. [LVO]). i=1 Avi; Indeed, if M is a finitely generated A-module, then any finite generating set U = {u1, . . . , ut} of M can be turned into an F-basis with respect to some good filtration F M . More precisely, let S = {n1, . . . , nt} be an arbitrarily chosen subset of N, then the required good filtration F M = {FqM }q∈N can be defined by setting FqM = {0} if q < min{n1, . . . , nm}; FqM =Pt i=1(cid:16)Ppi+ni≤q FpiA(cid:17) ui otherwise. q ∈ N. In particular, if L = ⊕s i=1Aei is a filtered free A-module with the filtration F L = {FqL}q∈N as constructed in Subsection 5.2 such that dfil(ei) = bi, 1 ≤ i ≤ s, then {e1, . . . , es} is an F-basis of L with respect to the good filtration F L. 108 Minimal Filtered Free Resolutions 5.4.3. Definition Let M be a filtered A-module with the filtration F M = {FqM }q∈N, and suppose that M has an F-basis V = {v1, . . . , vm} with respect to F M . If any proper subset of V cannot be an F-basis of M with respect to F M , then we say that V is a minimal F-basis of M with respect to F M . Note that A is an N-filtered K-algebra such that G(A) = ⊕p∈NG(A)p with G(A)0 = K, eA = ⊕p∈NeAp with eA0 = K, while K is a field. By Lemma 5.4.1 and the well-known result on graded modules over an N- graded algebra with the degree-0 homogeneous part a field (cf. [Eis], Chapter 19; [Kr1], Chapter 3; [Li3]), we have immediately the following 5.4.4. Proposition Let M be a filtered A-module with the filtration F M = {FqM }q∈N, and V = {v1, . . . , vm} ⊂ M a subset of nonzero ele- ments. Then V is a minimal F-basis of M with respect to F M if and only if σ(V ) = {σ(v1), . . . , σ(vm)} is a minimal homogeneous generating set of G(M ) if and only if τ (V ) = {ev1, . . . ,evm} is a minimal homogeneous gen- erating set of fM . Hence, any two minimal F-bases of M with respect to F M have the same number of elements and the same number of elements of the same fil-degree. (cid:3) Let M be an N-filtered A-module with the filtration F M = {FqM }q∈N, and let N be a submodule of M with the filtration F N = {FqN = N ∩ FqM }q∈N induced by F M . Then, as with a filtered free A-module in Section 5.2, the associated graded G(A)-module G(N ) = ⊕q∈NG(N )q of N with G(N )q = FqN/Fq−1N is a graded submodule of G(M ), and the Rees module eN = ⊕q∈NeNq of N with eNq = FqN is a graded submodule of fM . 5.4.5. Definition Let M be a filtered A-module with the filtration F M = {FqM }q∈N, and let N be a submodule of M . Consider the filtration F N = {FqN = N ∩ FqM }q∈N of N induced by F M . If W = {ξ1, . . . , ξs} ⊂ N is an F-basis with respect to F N in the sense of Definition 5.4.2, then we call W a standard basis of N . Remark. By referring to Lemma 5.4.1, one may check that our def- inition 5.4.5 of a standard basis coincides with the classical Macaulay Minimal Filtered Free Resolutions 109 basis provided A = K[x1, . . . , xn] is the commutative polynomial K- algebra (cf. [KR2], Definition 4.2.13, Theorem 4.3.19), for, taking the N-filtration F A with respect to an arbitrarily chosen positive-degree func- tion d( ) on A, there are graded algebra isomorphisms G(A) ∼= A and eA ∼= K[x0, x1, . . . , xn], where d(x0) = 1 and x0 plays the role that the central regular element Z of degree 1 does in eA. Moreover, if two-sided ideals of an N-filtered solvable polynomial algebra A are considered, then one may see that our definition 5.4.5 of a standard basis coincides with the standard basis defined in [Gol]. 5.4.6. Definition Let M be a filtered A-module with the filtration F M = {FqM }q∈N, and N a submodule of M with the filtration F N = {FqN = N ∩ FqM }q∈N induced by F M . Suppose that N has a standard basis W = {ξ1, . . . , ξm} with respect to F N . If any proper subset of W cannot be a standard basis for N with respect to F N , then we call W a minimal standard basis of N with respect to F N . If N is a submodule of some filtered A-module M with filtration F M , then since a standard basis of N is defined as an F-basis of N with respect to the filtration F N induced by F M , the next proposition follows from Proposition 5.4.4. 5.4.7. Proposition Let M be a filtered A-module with the filtration F M = {FqM }q∈N, and N a submodule of M with the induced filtra- tion F N = {FqN = N ∩ FqM }q∈N. A finite subset of nonzero elements W = {ξ1, . . . , ξs} ⊂ N is a minimal standard basis of N with respect to F N if and only if σ(W ) = {σ(ξ1), . . . , σ(ξm)} is a minimal homogeneous generating set of G(N ) if and only if τ (W ) = {eξ1, . . . ,eξm} is a minimal homogeneous generating set of eN . Hence, any two minimal standard bases of N have the same number of elements and the same number of elements of the same fil-degree. (cid:3) Since A, G(A) and eA are all Noetherian domains (Proposition 1.1.4(ii) of Chapter 1, Theorem 5.1.5 of the current chapter), if a filtered A-module M has an F-basis V with respect to a given filtration F M , then the existence of a standard basis for a submodule N of M follows immediately from Lemma 5.4.1. Our next theorem shows that a standard basis for a 110 Minimal Filtered Free Resolutions submodule N of a filtered free A-module L can be obtained via computing a left Grobner basis of N with respect to a graded left monomial ordering. the sense of Definition 5.4.5. Proof If ξ ∈ FqN = FqL ∩ N and ξ 6= 0, then dfil(ξ) ≤ q and ξ has a 5.4.8. Theorem Let L = ⊕s i=1Aei be a filtered free A-module with the filtration F L = {FqL}q∈N such that dfil(ei) = bi, 1 ≤ i ≤ s, and let ≺e-gr be a graded left monomial ordering on B(e) (see Section 5.3). If G = {g1, . . . , gm} ⊂ L is a left Grobner basis for the submodule N = i=1 Agi of L with respect to ≺e-gr, then G is a standard basis for N in Pm left Grobner representation by G, that is, ξ = Pi,j λijaα(ij )gj, where λij ∈ K ∗, aα(ij ) ∈ B with α(ij ) = (αij1, . . . , αijn ) ∈ Nn, satisfying LM(aα(ij )gj) (cid:22)e-gr LM(ξ). Suppose dfil(gj) = nj, 1 ≤ j ≤ m. Since ≺e-gr is a graded left monomial ordering on B(e), by Lemma 5.3.1 we may assume that LM(gj) = aβ(j)etj with β(j) = (βj1, . . . , βjn) ∈ Nn and 1 ≤ tj ≤ s, such that d(aβ(j)) + btj = nj, where d( ) is the given positive- degree function on A. Furthermore, by (Lemma 2.1.2(ii) of Chapter 2), we have LM(aα(ij )gj) = LM(aα(ij )aβ(j)etj ) = aα(ij )+β(j)etj , and it follows from Lemma 5.1.2, Lemma 5.2.2 and Lemma 5.3.1 that d(aα(ij )) + nj = d(aα(ij )) + d(aβ(j)) + btj = d(aα(ij )+β(j)) + btj ≤ q. Hence ξ ∈ Pm j=1(cid:16)Ppj+nj≤q Fpj A(cid:17) gj, i.e., G is a standard basis for N . Pm j=1(cid:16)Ppj+nj≤q Fpj A(cid:17) gj. This shows that FqN = 5.5. Computation of Minimal F-Bases and Mini- mal Standard Bases Let A = K[a1, . . . , an] be an N-filtered solvable polynomial algebra with admissible system (B, ≺) and the N-filtration F A = {FpA}p∈N con- structed with respect to a positive-degree function d( ) on A (see Section 5.1). In this section we show how to algorithmically compute minimal F-bases for quotient modules of a filtered free left A-module L, and how to algorithmically compute minimal standard bases for submodules of L in the case where a graded left monomial ordering ≺e-gr on L is employed. All notions, notations and conventions used before are maintained. Minimal Filtered Free Resolutions 111 We start by a little more preparation. Let M and M ′ be N-filtered left A-modules with the filtration F M = {FqM }q∈N and F M ′ = {FqM ′}q∈N ϕ −→ M ′ an A-module homomorphism. If ϕ(FqM ) ⊆ respectively, and M FqM ′ for all q ∈ N, then we call ϕ a filtered homomorphism. In the literature, such filtered homomorphisms are also referred to as filtered homomorphism of degree-0 (cf. [NVO], [LVO]). By the definition it is clear that the identity map of N-filtered A-modules is filtered homomorphism, and compositions of filtered homomorphisms are filtered homomorphisms. Thus, all N-filtered left A-modules form a subcategory of the category of left A-modules, in which morphisms are the filtered homomorphisms as ϕ −→ M ′ is a filtered homomorphism defined above. Furthermore, if M with kernel Kerϕ = N , then N is an N-filtered submodule of M with the induced filtration F N = {FqN = N ∩ FqM }q∈N, and the image ϕ(M ) of ϕ is an N-filtered submodule of M ′ with the filtration F ϕ(M ) = {Fqϕ(M ) = ϕ(FqM )}q∈N. Consequently, the exactness of a sequence ψ −→ M ′ of filtered homomorphisms in the category of N-filtered N A-modules is defined as the same as for a sequence of usual A-module homomorphisms, i.e., the sequence satisfies Imϕ = Kerψ. Long exact sequence in the category of N-filtered A-modules may be defined in an obvious way. −→ M ϕ Let G(A) be the associated N-graded algebra of A and eA the Rees ϕ −→ M ′ in- algebra of A. Then naturally, a filtered homomorphism M G(ϕ) −→ G(M ′), where duces a graded G(A)-module homomorphism G(M ) if ξ ∈ FqM and ξ = ξ + Fq−1M is the coset represented by ξ in G(M )q = FqM/Fq−1M , then G(ϕ)(ξ) = ϕ(ξ) + Fq−1M ′ ∈ G(M ′)q = eϕ −→fM ′, if M ϕ −→ M ′ Furthermore, recall that a filtered homomorphism M where if ξ ∈ FqM and hq(ξ) is the homogeneous element of degree q q = FqM ′. Moreover, ψ −→ M ′′ is a sequence of filtered homomorphisms, then FqM ′/Fq−1M ′, and ϕ induces a graded eA-module homomorphismfM in fMq = FqM , then eϕ(hq(ξ)) = hq(ϕ(ξ)) ∈ fM ′ G(ψ) ◦ G(ϕ) = G(ψ ◦ ϕ) and eψ ◦eϕ = ]ψ ◦ ϕ. ϕ −→ M ′ is called a strict filtered homomorphism if ϕ(FqM ) = ϕ(M ) ∩ FqM ′ for all q ∈ N. Note that if N is a submodule of M and M = M/N , then, consid- ering the induced filtration F N = {FqN = N ∩ FqM }q∈N of N and the induced filtration F (M ) = {FqM = (FqM + N )/N }q∈N of M , the inclusion map N ֒→ M and the canonical map M → M are strict fil- tered homomorphisms. Concerning strict filtered homomorphisms and 112 Minimal Filtered Free Resolutions the induced graded homomorphisms, the next proposition is quoted from ([LVO], CH.I, Section 4). 5.5.1. Proposition Given a sequence of filtered homomorphisms (∗) N ϕ −→ M ψ −→ M ′, such that ψ ◦ ϕ = 0, the following statements are equivalent. (i) The sequence (∗) is exact and ϕ, ψ are strict filtered homomorphisms. (ii) The induced sequence G(N ) eϕ G(ψ)) −→ G(M ′) is exact. G(ϕ) −→ G(M ) eψ (cid:3) (iii) The induced sequence eN −→ fM −→ fM ′ is exact. Let L = ⊕m i=1Aei be a filtered free A-module with the filtration F L = {FqL}q∈N such that dfil(ei) = bi, 1 ≤ i ≤ m. Then as we have noted in Section 5.4, {e1, . . . , em} is an F-basis of L with re- spect to the good filtration F L. Let N be a submodule of L, and let the quotient module M = L/N be equipped with the filtration F M = {FqM = (FqL + N )/N }q∈N induced by F L. Without loss of generality, we assume that ei 6= 0 for 1 ≤ i ≤ m, where each ei is the coset represented by ei in M . Then we see that {e1, . . . , em} is an F-basis of M with respect to F M . j=1 Aξj be generated by the set of nonzero elements U = {ξ1, . . . , ξs}, where k=1 fkℓek with fkℓ ∈ A and dfil(ξℓ) = qℓ, 1 ≤ ℓ ≤ s. The following 5.5.2. Lemma Let M = L/N be as fixed above, and let N =Ps ξℓ =Ps ℓ = ξℓ − fiℓξj does not involve ei. j+1, . . . , ξ′ statements hold. (i) If for some j, ξj has a nonzero term fijei such that dfil(fijei) = dfil(ξj) = qj and the coefficient fij is a nonzero constant, say fij = 1 without loss of generality, then for each ℓ = 1, . . . , j − 1, j + 1, . . . , s, the element ξ′ Putting U ′ = {ξ′ j−1, ξ′ ℓ, and considering the filtered free A-module L′ = ⊕k6=iAek with the filtration F L′ = {FqL′}q∈N in which each ek has the same filtered degree as it is in L, i.e., dfil(ek) = bk, if the quotient module M ′ = L′/N ′ is equipped with the filtration F M ′ = {FqM ′ = (FqL′ + N ′)/N ′}q∈N induced by F L′, then there is a strict filtered isomorphism ϕ: M ′ ∼= M , i.e., ϕ is an A-module isomor- phism such that ϕ(FqM ′) = FqM for all q ∈ N. s}, N ′ =Pξ′ ℓ∈U ′ Aξ′ 1, . . . , ξ′ Minimal Filtered Free Resolutions 113 (ii) With the assumptions and notations as in (i), if U = {ξ1, . . . , ξs} is a standard basis of N with respect to the filtration F N induced by s} is a standard basis of N ′ with F L, then U ′ = {ξ′ respect to the filtration F N ′ induced by F L′. j+1, . . . , ξ′ 1, . . . , ξ′ j−1, ξ′ j−1, ξ′ 1, . . . , ξ′ ℓ∈U ′ Aξ′ j+1, . . . , ξ′ Proof (i) Since fij = 1 by the assumption, we see that every ξ′ fiℓξj does not involve ei. Let U ′ = {ξ′ ℓ = ξℓ − s} and N ′ = ℓ. Then N ′ ⊂ L′ = ⊕k6=iAek ⊂ L and N = N ′+Aξj. Since ξj = Pξ′ ei +Pk6=i fkjek, the naturally defined A-module homomorphism M ′ = L′/N ′ ϕ −→L/N = M with ϕ(ek) = ek, k = 1, . . . , i − 1, i + 1, . . . , m, is an isomorphism, where, without confusion, ek denotes the coset represented by ek in M ′ and M respectively. It remains to see that ϕ is a strict filtered isomorphism. Note that {e1, . . . , em} is an F-basis of L with respect to F L such that dfil(ei) = bi, 1 ≤ i ≤ m, i.e., FqL = mXi=1 FqL′ =Xk6=i  Xpi+bi≤q  Xpi+bk≤q FpiA ei, FpiA ek, q ∈ N, q ∈ N, that {e1, . . . , ei−1, ei+1, . . . , em} is an F-basis of L′ with respect to F L′ such that dfil(ek) = bk, where k 6= i, i.e., desired. (ii) Note that ξ′ and that ξj = ei +Pk6=i fkjek with qj = dfil(ξj) = dfil(ei) = bi such that dfil(fkj) + bk ≤ qj for all fkj 6= 0. It follows thatPk6=i fkjek ∈ Fqj M ′ and ϕ(Pk6=i fkjek) = ei ∈ Fqj M , thereby ϕ(FqM ′) = FqM for all q ∈ N, as ℓ = ξℓ − fiℓξj. By the assumption on ξj, if fiℓ 6= 0 and dfil(fiℓei) = dfil(ξℓ) = qℓ, then since dfil(ξj) = dfil(ei) we have dfil(fiℓξj) = dfil(ξℓ) = qℓ. It follows that if we equip N with the filtration F N = {FqN = N ∩ FqL}q∈N induced by F L and consider the associated graded module G(N ) of N , then dgr(σ(ξℓ)) = dgr(σ(fiℓξj)) = dgr(σ(fiℓ)σ(ξj)) in G(N ), i.e., σ(ξℓ) − σ(fiℓ)σ(ξj ) ∈ G(N )qℓ . So, if σ(ξℓ) − σ(fiℓ)σ(ξj ) 6= 0 then dfil(ξ′ ℓ) = qℓ and thus σ(ξ′ ℓ) = σ(ξℓ − fiℓξj) = σ(ξℓ) − σ(fiℓ)σ(ξj). (1) If fiℓ 6= 0 and dfil(fiℓei) < dfil(ξℓ) = qℓ, then σ(ξℓ) =Pd(fkℓ)+bk=qℓ does not involve σ(ei). Also since dfil(ξj) = dfil(ei), we have dfil(fiℓξj) < σ(fkℓ)σ(ek) 114 Minimal Filtered Free Resolutions dfil(ξℓ) = qℓ. Hence dfil(ξ′ ℓ) = dfil(ξℓ) = qℓ and thus σ(ξ′ ℓ) = σ(ξℓ − fiℓξj) = σ(ξℓ). (2) ℓ∈U ′ G(A)σ(ξ′ ℓ) + G(A)σ(ξj ) where the σ(ξ′ If fiℓ = 0, then the equality (1) is the same as equality (2). Now, if U = {ξ1, . . . , ξs} is a standard basis of N with respect to the induced ℓ=1 G(A)σ(ξℓ) by Lemma 5.4.1. But since ℓ) are those nonzero homogeneous elements obtained according to the above equalities (1) and (2), it follows from Lemma 5.4.1 that U ′ ∪ {ξj} is a standard basis of N with respect to the induced filtration F N . filtration F N , then G(N ) =Ps we have also G(N ) =Pξ′ We next prove that U ′ is a standard basis of N ′ =Pξ′ ℓ with respect to the filtration F N ′ = {FqN ′ = N ′ ∩ FqL′}q∈N induced by F L′. Since {e1, . . . , ei−1, ei+1, . . . , em} is an F-basis of L′ with respect to the filtration F L′ such that each ek has the same filtered degree as it is in L, i.e., dfil(ek) = bk, it is clear that FqL′ = L′ ∩ FqL, q ∈ N, i.e., the filtration F L′ is the one induced by F L. Considering the filtration F N ′ of N ′ induced by F L′, it turns out that ℓ∈U ′ Aξ′ FqN ′ = N ′ ∩ FqL′ = N ′ ∩ FqL ⊆ N ∩ FqL = FqN, q ∈ N. (3) If ξ ∈ FqN ′, then since U ′ ∪ {ξj} is a standard basis of N with respect to the induced filtration F N , the formula (3) entails that fℓξ′ ℓ+fjξj with fℓ, fj ∈ A, d(fℓ)+dfil(ξ′ ℓ) ≤ q, d(fj)+dfil(ξj) ≤ q. ξ = Xξ′ ℓ∈U ′ (4) Note that every ξ′ ℓ does not involve ei and consequently ξ does not involve ei. Hence fj = 0 in (4) by the assumption on ξj, and thus ℓ. Therefor we conclude that U ′ is a stan- dard basis for N ′ with respect to the induced filtration F N ′, as desired. (cid:3) ℓ∈U ′(cid:16)Ppi+qℓ≤q FpiA(cid:17) ξ′ ξ ∈Pξ′ Combining Proposition 5.4.4, we now show that for quotient modules of filtered free A-modules, a result similar to Proposition 4.3.11 holds true. 5.5.3. Proposition Let L = ⊕m i=1Aei, M = L/N be as in Lemma 5.5.2, and suppose that U = {ξ1, ..., ξs} is now a standard basis of N with Minimal Filtered Free Resolutions 115 q=1Aeiq such that respect to the filtration F N induced by F L. The algorithm presented below computes a subset {ei1 , . . . , eim′ } ⊂ {e1, . . . , em} and a subset V = {v1, . . . , vt} ⊂ N ∩ L′, where m′ ≤ m and L′ = ⊕m′ (i) there is a strict filtered isomorphism L′/N ′ = M ′ ∼= M , where ℓ=1 Avℓ, and M ′ has the filtration F M ′ = {FqM ′ = (FqL′ + N ′)/N ′}q∈N induced by the good filtration F L′ determined by the F-basis {ei1, . . . , eim′ } of L′; N ′ = Pt (ii) V = {v1, . . . , vt} is a standard basis of N ′ =Pt to the filtration F N ′ induced by F L′, such that each vℓ =Pm′ has the property that hkℓ ∈ K ∗ implies dfil(eik ) = bik < dfil(vℓ); (iii) {ei1, . . . , eim′ } is a minimal F-basis of M with respect to the filtration F M . Algorithm-MINFB ℓ=1 Avℓ with respect k=1 hkℓeik INPUT: E = {e1, . . . , em}; U = {ξ1, ..., ξs}, where ξℓ =Pm k=1 fkℓek with fkℓ ∈ A, and d(fkℓ) + bk ≤ qℓ = dfil(ξℓ), 1 ≤ ℓ ≤ s OUTPUT: E′ = {ei1, . . . , eim′ }; V = {v1, . . . , vt} ⊂ N ∩ L′, k=1Aeik where L′ = ⊕m′ INITIALIZATION: t := s; V := U ; m′ := m; E′ := E BEGIN WHILE there is a vj =Pm′ k=1 fkjek ∈ V and i is the least index such that fij ∈ K ∗ with d(fij) + bi = dfil(vj) DO set T = {1, . . . , j − 1, j + 1, . . . , t} and compute ℓ = vℓ − 1 v′ fiℓvj, ℓ ∈ T, fij r = #{ℓ ℓ ∈ T, vℓ = 0} t := t − r − 1 V := {vℓ = v′ ℓ ℓ ∈ T, v′ ℓ 6= 0} = {v1, . . . , vt} (after reordered) m′ := m′ − 1 E′ := E′ − {ei} = {e1, . . . , em′ } (after reordered) END END Proof First note that for each ξℓ ∈ U , dfil(ξℓ) is determined by Lemma 5.2.1. Since the algorithm is clearly finite, the conclusions (i) and (ii) follow from Lemma 5.5.2. 116 Minimal Filtered Free Resolutions the filtration F N ′ has the property that hkℓ ∈ K ∗ implies dfil(eik ) = bik < dfil(vℓ). induced by F L′ such that each vℓ = Pm′ To prove the conclusion (iii), by the strict filtered isomorphism M ′ = L′/N ′ ∼= M (or the proof of Lemma 5.5.2(i)) it is sufficient to show that {ei1, . . . , eim′ } is a minimal F-basis of M ′ with respect to the filtration F M ′. By the conclusion (ii), V = {v1, . . . , vt} is a ℓ=1 Avℓ of L′ with respect to k=1 hkℓeik It k=1 G(A)σ(vk) in which standard basis of the submodule N ′ = Pt follows from Lemma 5.4.1 that G(N ′) = Pt each σ(vk) = Pd(hkℓ)+bik =dfil(vk) σ(hkℓ)σ(eik ) and all the coefficients σ(hkℓ) satisfy dgr(σ(hkℓ)) > 0 (please note the discussion in Section 5.2). Since G(A)0 = K, G(L′) = ⊕m′ k=1G(A)σ(eik ) (Proposition 5.2.3) and G(N ′) is the graded syzygy module of the graded quotient mod- ule G(L′)/G(N ′), by the well-known result on finitely generated graded modules over N-graded algebras with the degree-0 homogeneous part a field (cf. [Li3]), we conclude that {σ(ei1 ), . . . , σ(eim′ )} is a minimal homogeneous generating set of G(L′)/G(N ′). On the other hand, considering the naturally formed ex- act sequence of strict filtered homomorphisms [Eis], Chapter 19; [Kr1], Chapter 3; 0 −→ N ′ −→ L′ −→ M ′ = L′/N ′ −→ 0, by Proposition 5.5.1 we have the N-graded G(A)-module isomorphism G(L′)/G(N ′) ∼= G(L′/N ′) = G(M ′) with σ(eik ) 7→ σ(eik ), 1 ≤ k ≤ m′. Now, by means of Proposition 5.4.4, we conclude that {ei1, . . . , eim′ } is a minimal F-basis of M ′ with respect to the filtration F M ′, as desired. (cid:3) Finally, let L = ⊕s i=1Aei be a filtered free A-module with the filtra- tion F L = {FqL}q∈N such that dfil(ei) = bi, 1 ≤ i ≤ s, and let ≺e-gr be a graded left monomial ordering on the K-basis B(e) = {aαei aα ∈ B, 1 ≤ i ≤ s} of L (see Section 5.3). Combining Theorem 5.3.3 and (The- orem 4.3.8 of Chapter 4), the next theorem shows how to algorithmically compute a minimal standard basis. 5.5.4. Theorem Let N =Pc i=1 Aθi be a submodule of L generated by the subset of nonzero elements Θ = {θ1, . . . , θc}, and let F N = {FqN = FqL ∩ N }q∈N be the filtration of N induced by F L. Then a minimal standard basis of N with respect to F N can be obtained by implementing the following procedures: Procudure 1. With the initial input data Θ = {θ1, . . . , θc}, run Minimal Filtered Free Resolutions 117 Algorithm-LGB presented in (Section 2.3 of Chapter 2) to compute a left Grobner basis U = {ξ1, . . . , ξm} for N with respect to ≺e-gr on B(e). Procudure 2. Let G(N ) be the associated graded G(A)-module of N determined by the induced filtration F N . Then G(N ) is a graded submodule of the associated grade free G(A)-module G(L), and it follows from Theorem 5.3.3 that σ(U ) = {σ(ξ1), . . . , σ(ξm)} is a homogeneous left Grobner basis of G(N ) with respect to ≺σ(e)-gr on σ(B(e)). With the initial input data σ(U ), run Algorithm-MINHGS presented in (Theo- rem 4.3.8 of Chapter 4) to compute a minimal homogeneous generating set {σ(ξj1), . . . , σ(ξjt)} ⊆ σ(U ) for G(N ). Procudure 3. Write down W = {ξj1, . . . , ξjt}. Then W is a Minimal standard basis for N by Proposition 5.4.7. (cid:3) Remark. (i) Note that the initial input data σ(U ) in Procedure 2 above is already a left Grobner basis for G(N ). By (Theorem 4.3.8 of Chapter 4), the finally returned left Grobner basis by Algorithm-MINHGS is indeed a minimal left Grobner basis for G(N ). (ii) By Theorem 5.3.5 and Proposition 5.4.7 it is clear that we can also obtain a minimal standard basis of the submodule N via computing a minimal homogeneous generating set for the Rees module eN of N , which is a graded submodule of the Rees module eL of L. However, noticing the structure of eA and eL (see Theorem 5.1.5, Proposition 5.2.3) it is equally clear that the cost of working on eA will be much higher than working on G(N ). 5.6. Minimal Filtered Free Resolutions and Their Uniqueness Let A = K[a1, . . . , an] be an N-filtered solvable polynomial algebra with admissible system (B, ≺) and the N-filtration F A = {FpA}p∈N con- structed with respect to a positive-degree function d( ) on A (see Section 5.1). In this section, by using minimal F-bases and minimal standard bases in the sense of Definition 5.4.3 and Definition 5.4.6, we define mini- mal filtered free resolutions for finitely generated left A-modules, and we show that such minimal resolutions are unique up to strict filtered iso- morphism of chain complexes in the category of filtered A-modules. All 118 Minimal Filtered Free Resolutions notions, notations and conventions used before are maintained. Let M = Pm i=1 Aξi be an arbitrary finitely generated A-module. Then, as we have noted in section 5.4, M may be endowed with a good filtration F M = {FqM }q∈N with respect to an arbitrarily chosen subset {n1, . . . , nm} ⊂ N, where FqM = {0} if q < min{n1, . . . , nm}; FqM =Pt i=1(cid:16)Ppi+ni≤q FpiA(cid:17) ξi otherwise, q ∈ N. 5.6.1. Proposition Let M and F M be as above. Consider the filtered free A-module L = ⊕m i=1Aei with the good filtration F L = {FqL}q∈N such that dfil(ei) = ni, 1 ≤ i ≤ m, and the exact sequence of A-module homomorphisms (∗) 0 −→ N ι−→ L ϕ −→ M −→ 0, in which ϕ(ei) = ξi, 1 ≤ i ≤ m, N = Kerϕ with the induced filtration F N = {FqN = N ∩ FqL}q∈N, and ι is the inclusion map. The following statements hold. (i) The A-module homomorphisms ι and ϕ are strict filtered homomor- phisms. Equipping L = L/N with the induced filtration F L = {FqL = (FqL+N )/N }q∈N, the induced A-module isomorphism L −→ M is a strict filtered isomorphism, that is, L ∼= M and ϕ satisfies ϕ(FqL) = FqM for all q ∈ N. (ii) The induced sequence ϕ G(∗) 0 −→ G(N ) G(ι) −→ G(L) G(ϕ) −→ G(M ) −→ 0 is an exact sequence of graded G(A)-module homomorphisms, thereby G(L)/G(N ) ∼= G(M ) ∼= G(L) = G(L/N ) as graded G(A)-modules. (iii) The induced sequence f(∗) 0 −→ eN eι−→ eL is an exact sequence of graded eA-module homomorphisms, thereby eL/eN ∼= fM ∼=eL = ]L/N as graded eA-modules. Proof By the construction of F L (see section 5.2) and Proposition 5.5.1, the proof of all assertions is a straightforward exercise. (cid:3) eϕ −→ fM −→ 0 Minimal Filtered Free Resolutions 119 Proposition 5.6.1(i) enables us to make the following In what follows we shall always assume that a finitely Convention. generated A-module M is of the form as presented in Proposition 5.6.1(i), i.e., M = L/N , and M has the good filtration F M = {FqM = (FqL + N )/N }q∈N. Comparing with the classical minimal graded free resolutions defined for finitely generated graded modules over finitely generated N-graded algebras with the degree-0 homogeneous part a field (cf. [Eis], Chapter 19; [Kr1], Chapter 3; [Li3]), the results obtained in previous sections and the preliminary we made above naturally motivate the following 5.6.2. Definition Let L0 = ⊕m i=1Aei be a filtered free A-module with the filtration F L0 = {FqL0}q∈N such that dfil(ei) = bi, 1 ≤ i ≤ m, let N be a submodule of L0, and let the A-module M = L0/N be equipped with the filtration F M = {FqM = (FqL0 + N )/N }q∈N. A minimal filtered free resolution of M is an exact sequence of filtered A-modules and filtered homomorphisms L• · · · ϕi+1−→ Li ϕi−→ · · · ϕ2−→ L1 ϕ1−→ L0 ϕ0−→ M → 0 satisfying (1) ϕ0 is the canonical epimorphism, i.e., ϕ0(ei) = ei for ei ∈ E0 = {e1, . . . , em} (where each ei denotes the coset represented by ei in M ), such that ϕ0(E0) is a minimal F-basis of M with respect to F M (in the sense of Definition 5.4.3); (2) for i ≥ 1, each Li is a filtered free A-module with finite A-basis Ei, and each ϕi is a strict filtered homomorphism, such that ϕi(Ei) is a minimal standard basis of Kerϕi−1 with respect to the filtration induced by F Li−1 (in the sense of Definition 5.4.6). To see that the minimal filtered free resolution introduced above is an appropriate definition of "minimal free resolution" for finitely generated modules over the N-filtered solvable polynomial algebras with filtration determined by positive-degree functions, we now show that such a reso- lution is unique up to a strict filtered isomorphism of chain complexes in the category of N-filtered A-modules. 120 Minimal Filtered Free Resolutions 5.6.3. Theorem Let L• be a minimal filtered free resolution of M as presented in Definition 5.6.2. The following statements hold. (i) The induced sequence of graded G(A)-modules and graded G(A)- module homomorphisms G(L•) · · · G(ϕi+1) −→ G(Li) G(ϕi) −→ · · · G(ϕ2) −→ G(L1) G(ϕ1) −→ G(L0) G(ϕ0) −→ G(M ) → 0 is a minimal graded free resolution of the finitely generated graded G(A)- module G(M ). · · · (ii) The induced sequence of graded eA-modules and graded eA-module homomorphisms eϕi+1−→ eLi eL• is a minimal graded free resolution of the finitely generated graded eA- module fM . (iii) L• is uniquely determined by M in the sense that if M has another minimal filtered free resolution eϕ0−→ fM → 0 eϕ1−→ eL0 eϕ2−→ eL1 eϕi−→ · · · L′ • · · · ϕ′ i+1−→ L′ i ϕ′ i−→ · · · ϕ′ 2−→ L′ 1 ϕ′ 1−→ L0 ϕ0−→ M → 0 then for each i ≥ 1, there is a strict filtered A-module isomorphism χi: Li → L′ i such that the diagram ∼= Li χiy L′ i ϕi−→ Li−1 ∼= χi−1y ϕ′ i−→ L′ i−1 is commutative, thereby {χi i ≥ 1} gives rise to a strict filtered iso- morphism of chain complexes L• ∼= L′ • in the category of N-filtered A- modules. Proof (i) and (ii) follow from Proposition 5.4.4, Proposition 5.4.7, and Proposition 5.6.1. To prove (iii), let the sequence L′ • be as presented above. By the assertion (i), G(L′ •) is another minimal graded free resolution of G(M ). It follows from the well-known result on minimal graded free resolutions ([Eis], Chapter 19; [Kr1], Chapter 3; [Li3]) that there is a graded iso- morphism of chain complexes G(L•) ∼= G(L′ •) in the category of graded Minimal Filtered Free Resolutions 121 G(A)-modules, i.e., for each i ≥ 1, there is a graded G(A)-modules iso- morphism ψi: G(Li) → G(L′ i) such that the diagram G(Li) G(ϕi) −→ G(Li−1) G(L′ i) G(ϕ′ i) −→ G(L′ i−1) ∼= ψiy ∼= ψi−1y χj−→ L′ j, such that G(χj) = ψj and χj−1ϕj = ϕ′ is commutative. Our aim below is to construct the desired strict fil- tered isomorphisms χi by using the graded isomorphisms ψi carefully. So, starting with L0, we assume that we have constructed the strict filtered isomorphisms Lj jχj, 1 ≤ j ≤ i − 1. Let Li = ⊕si is a graded j=1Aeij . isomorphism, we have ψi(σ(eij )) = σ(ξ′ i satisfying dfil(ξ′ It follows that if we construct the filtered A-module homomorphism Li i by setting τi(eij ) = ξ′ j, 1 ≤ j ≤ si, then G(τi) = ψi. Hence, τi is a strict filtered isomorphism by Proposition 5.5.1. Since ψi−1 = G(χi−1), ψi = G(τi), and thus j)) = dgr(σ(eij )) = dfil(eij ). j) for some ξ′ j) = dgr(σ(ξ′ Since each ψi τi−→ L′ j ∈ L′ ψi−1G(ϕi) = G(χi−1)G(ϕi) = G(χi−1ϕi) G(ϕ′ iτi), i)G(τi) = G(ϕ′ for each q ∈ N, by the strictness of the ϕj, ϕ′ i)ψi = G(ϕ′ j, χj and τi, we have G(χi−1ϕi)(G(Li)q) = ((χi−1ϕi)(FqLi) + Fq−1L′ i−1)/Fq−1L′ i−1 = (χi−1(ϕi(Li) ∩ FqLi−1) + Fq−1L′ ⊆ ((χi−1ϕi)(L − i) ∩ χi−1(FqLi−1) + Fq−1L′ = ((χi−1ϕi)(Li) ∩ FqL′ i−1 + Fq−1L′ i−1)/Fq−1L′ i−1)/Fq−1L′ i−1 i−1, i−1)/Fq−1L′ i−1 G(ϕ′ iτi)(G(Li)q) = ((ϕ′ = (ϕ′ = (ϕ′ = ((ϕ′ i) + Fq−1L′ iτi)(FqLi) + Fq−1L′ i(FqL′ i(L′ iτi)(Li) ∩ FqL′ i) ∩ FqL′ i−1)/Fq−1L′ i−1 i−1)/Fq−1L′ i−1 i−1 + Fq−1L′ i−1)/Fq−1L′ i−1 + Fq−1L′ i−1 i−1)/Fq−1L′ i−1. Note that by the exactness of L• and L′ tativity χi−2ϕi−1 = ϕ′ (ϕ′ and (ϕ′ tativity ψi−1G(ϕi) = G(ϕ′ •, as well as the commu- i) = iτi)(Li). Considering the filtration of the submodules (χi−1ϕi)(Li) i−1, the commu- i)ψi and the formulas derived above show i−1χi−1, we have (χi−1ϕi)(Li) ⊆ ϕ′ iτi)(Li) induced by the filtration F L′ i−1 of L′ i(L′ 122 Minimal Filtered Free Resolutions that both submodules have the same associated graded module, G((χi−1ϕi)(Li)) = G((ϕ′ iτi)(Li)). ([LVO], Theorem 5.7 on P.49) that i.e., It follows from a similar proof of (χi−1ϕi)(Li) = (ϕ′ iτi)(Li). (1) Clearly, the equality (1) does not necessarily mean the commutativity of the diagram ∼= Li τiy L′ i ϕi−→ Li−1 ∼= χi−1y ϕ′ i−→ L′ i−1 To remedy this problem, we need to further modify the filtered isomor- phism τi. Since G(χi−1ϕi)(σ(eij )) = G(ϕ′ iτi)(σ(eij )), 1 ≤ j ≤ si, if dfil(eij ) = bj, then by the equality (1) and the strictness of τi and ϕi we have (χi−1ϕi)(eij ) − (ϕ′ iτi)(eij ) ∈ (ϕ′ = ϕ′ = ϕ′ = (ϕ′ i) ∩ Fbj −1L′ iτi)(Li) ∩ Fbj −1L′ i(L′ i−1 i(Fbj −1L′ i) iτi)(Fbj −1Li), i−1 (2) and furthermore from (2) we have a ξj ∈ Fbj −1Li such that dfil(eij − ξj) = bj and (χi−1ϕi)(eij ) = (ϕ′ iτi)(eij − ξj), 1 ≤ j ≤ si. (3) Now, if we construct the filtered homomorphism Li χi(eij ) = τi(eij − ξj), 1 ≤ j ≤ si, then since τi(ξj) ∈ Fbj −1L′ that χi−→ L′ i by setting i, it turns out G(χi)(σ(eij )) = G(τi)(σ(eij −ξj)) = G(τi)(σ(eij )) = ψi(σ(eij )), 1 ≤ j ≤ si, thereby G(χi) = ψi. Hence, χi is a strict filtered isomorphism by Propo- sition 5.5.1, and moreover, it follows from (3) that we have reached the following diagram · · · · · · ϕi−1−→ · · · ϕ′ i−1−→ · · · ϕi+1−→ Li ϕi−→ Li−1 ∼= χiy ∼= χi−1y ϕ′ i+1−→ L′ i ϕ′ i−→ L′ i−1 in which χi−1ϕi = ϕ′ desired χi+1 and so on, the proof is thus finished. iχi. Repeating the same process to getting the Minimal Filtered Free Resolutions 123 5.7. Computation of Minimal Finite Filtered Free Resolutions Let A = K[a1, . . . , an] be a solvable polynomial algebra with the admis- sible system (B, ≺gr) in which ≺gr is a graded monomial ordering with respect to some given positive-degree function d( ) on A (see section 1.1 of Chapter 1). Thereby A is turned into an N-filtered solvable polynomial algebra with the filtration F A = {FpA}p∈N constructed with respect to the same d( ) (see Example (2) given in Section 5.1). Note that Theorem 3.1.1, Theorem 3.1.2, and Theorem 3.2.2 given in Chapter 3 hold true for any solvable polynomial algebra. Combining the results of Chapter 4 and previous sections of the current chapter, we are now able to work out the algorithmic procedures for computing minimal finite filtered free resolu- tions over A (in the sense of Definition 5.6.2) with respect to any graded left monomial ordering on free left modules. All notions, notations and conventions used before are maintained. 5.7.1. Theorem Let L0 = ⊕m i=1Aei be a filtered free A-module with the filtration F L0 = {FqL0}q∈N such that dfil(ei) = bi, 1 ≤ i ≤ m. If N = i=1 Aξi is a finitely generated submodule of L0 and the quotient module M = L0/N is equipped with the filtration F M = {FqM = (FqL0 + N )/N }q∈N, then M has a minimal filtered free resolution of length d ≤ n: Ps L• 0 −→ Ld ϕq−→ · · · ϕ2−→ L1 ϕ1−→ L0 ϕ0−→ M −→ 0 which can be constructed by implementing the following procedures: Procedure 1. Fix a graded left monomial ordering ≺e-gr on the K- basis B(e) of L0 (see Section 5.3), and run Algorithm-LGB (presented in Section 2.3 of Chapter 2) with the initial input data U = {ξ1, . . . , ξs} to compute a left Grobner basis G = {g1, . . . , gz} for N , so that N has the standard basis G with respect to the induced filtration F N = {FqN = N ∩ FqL0}q∈N (Theorem 5.4.8). Procedure 2. Run Algorithm-MINFB (presented in Proposition 5.5.3) with the initial input data E = {e1, . . . , em} and G = {g1, . . . , gz} to compute a subset E ′ 0 = {ei1, . . . , eim′ } ⊂ E0 = {e1, . . . , em} and a subset V = {v1, . . . , vt} ⊂ N ∩ L′ 0 such that there is a strict filtered q=1Aeiq with m′ ≤ m and isomorphism L′ k=1 Avk, and such that {ei1, . . . , eim′ } is a minimal F-basis of M 0/N ′ = M ′ ∼= M , where L′ 0 = ⊕m′ N ′ =Pt with respect to the filtration F M . 124 Minimal Filtered Free Resolutions For convenience, after accomplishing Procedure 2 we may assume that 0, U = V and N = N ′. Accordingly we have the short exact E0 = E ′ sequence 0 −→ N ι−→ L0 ϕ0−→ M −→ 0 such that ϕ0(E0) = {e1, . . . , em} is a minimal F-basis of M with respect to the filtration F M , where ι is the inclusion map. Procedure 3. With the initial input data U = V , implements the procedures presented in Theorem 5.5.4 to compute a minimal standard basis W = {ξj1, . . . , ξjm1 } for N with respect to the induced filtration F N . Procedure 4. Computes a generating set U1 = {η1, . . . , ηs1} of N1 = i=1Aεi by running Algorithm- Syz(W ) in the free A-module L1 = ⊕m1 LGB with the initial input data W and using Theorem 3.1.2. Procedure 5. Construct the strict filtered exact sequence 0 −→ N1 −→ L1 ϕ1−→ L0 ϕ0−→ M −→ 0 where the filtration F L1 of L1 is constructed by setting dfil(εk) = dfil(ξjk ), 1 ≤ k ≤ m1, and ϕ1 is defined by setting ϕ1(εk) = ξjk , 1 ≤ k ≤ m1. If N1 6= 0, then, with the initial input data U = U1, repeat Procedure 3 -- Procedure 5 for N1 and so on. By Theorem 5.6.3, a minimal filtered free resolution L• of M gives rise to a minimal graded free resolution G(L•) of G(M ). Since G(A) = K[σ(a1), . . . , σ(an)] is a solvable polynomial algebra by Theorem 5.1.5, it follows from Theorem 4.4.1 that G(L•) terminates at a certain step, i.e., KerG(ϕd) = 0 for some d ≤ n. But KerG(ϕd) = G(Kerϕd) by Proposition 5.5.1, where Kerϕd has the filtration induced by F Ld, thereby G(Kerϕd) = 0. Consequently Kerϕd = 0 since all filtration we are dealing with are separated, thereby a minimal finite filtered free resolution of length d ≤ n is achieved for M . References [AF] F.W. Anderson and K.R. Fuller, Rings and categories of Mod- ules. Springer-Verlag, 1974. [AL1] J. Apel and W. Lassner, An extension of Buchberger's algorithm and calculations in enveloping fields of Lie algebras. J. Symbolic Comput., 6(1988), 361 -- 370. Minimal Filtered Free Resolutions 125 [AL2] W. W. Adams and P. Loustaunau, An Introduction to Grobner Bases. Graduate Studies in Mathematics, Vol. 3. American Mathematical Society, 1994. [AVV] M. J. Asensio, et al, A new algebraic approach to microlocaliza- tion of filtered rings. Trans. Amer. math. Soc, 2(316)(1989), 537-555. [Ber1] R. Berger, The quantum Poincar´e-Birkhoff-Witt theorem. Comm. Math. Physics, 143(1992), 215 -- 234. [Ber2] G. Bergman, The diamond lemma for ring theory. Adv. Math., 29(1978), 178 -- 218. [Bu1] B. Buchberger, Ein Algorithmus zum Auffinden der Basise- lemente des Restklassenringes nach einem nulldimensionalen polynomideal. PhD thesis, University of Innsbruck, 1965. [Bu2] B. Buchberger, Grobner bases: An algorithmic method in polynomial ideal theory. In: Multidimensional Systems Theory (Bose, N.K., ed.), Reidel Dordrecht, 1985, 184 -- 232. [BW] T. Becker and V. Weispfenning, Grobner Bases. Springer- Verlag, 1993. [CDNR] A. Capani, et al, Computing minimal finite free resolutions. Journal of Pure and Applied Algebra, (117& 118)(1997), 105 -- 117. [DGPS] W. Decker, et al, Singular 3-1-3 -- A computer algebra system for polynomial computations. Available at http://www.singular.uni-kl.de(2011). [Dir] P. A. M. Dirac, On quantum algebra. Proc. Camb. Phil. Soc., 23(1926), 412 -- 418. [Eis] D. Eisenbud, Commutative Algebra with a View Toward to Al- gebraic Geometry, GTM 150. Springer, New York, 1995. [EPS] D. Eisenbud, et al, Non-commutative Grobner bases for com- mutative algebras. Proc. Amer. Math. Soc., 126, 1998, pp. 687-691. [Fau] J.-C. Faug´ere. A new efficient algorithm for computing Groobner bases without reduction to zero (F5). In: proc. ISSAC'02, ACM Press, New York, USA, 75-82, 2002. [Frob] R. Froberg, An Introduction to Grobner Bases. Wiley, 1997. [Gal] A. Galligo, Some algorithmic questions on ideals of differential operators. Proc. EUROCAL'85, LNCS 204, 1985, 413 -- 421. [Gol] E. S. Golod, Standard bases and homology, in: Some Current 126 Minimal Filtered Free Resolutions Trends in Algebra. (Varna, 1986), Lecture Notes in Mathemat- ics, 1352, Springer-Verlag, 1988, 88-95. [Gr] E. Green, Noncommutative Grobner Bases, A Computational and Theoretical Tool. Lecture notes, Dec. 15, 1996. Available at www.math.unl.edu/ shermiller2/hs/green2.ps [GV] J. Gago-Vargas, Bases for projective modules in An(k). Journal of Symbolic Computation, 36(2003), 845C853. [Hay] T. Hayashi, Q-analogues of Clifford and Weyl algebras-Spinor and oscillator representations of quantum enveloping algebras. Comm. Math. Phys., 127(1990), 129 -- 144. [Hu] J. E. Humphreys, Introduction to Lie Algebras and Representa- tion Theory. Springer, 1972. [Jat] V.A. Jategaonkar, A multiplicative analogue of the Weyl alge- bra. Comm. Alg., 12(1984), 1669 -- 1688. [JBS] A. Jannussis, et al, Remarks on the q-quantization. Lett. Nuovo Cimento, 30(1981), 123 -- 127. [Kr1] U. Krahmer, Notes on Koszul algebras. 2010. Available at http://www.maths.gla.ac.uk/ ukraehmer/connected.pdf [Kr2] H. Kredel, Solvable Polynomial Rings. Shaker-Verlag, 1993. [KP] H. Kredel and M. Pesch, MAS, modula-2 algebra system. 1998. Available at http://krum.rz.uni-mannheim.de/mas/ [KR1] M. Kreuzer, L. Robbiano, Computational Commutative Algebra 1. Springer, 2000. [KR2] M. Kreuzer, L. Robbiano, Computational Commutative Algebra 2. Springer, 2005. [K-RW] A. Kandri-Rody and V. Weispfenning, Non-commutative Grobner bases in algebras of solvable type. 9(1990), 1 -- 26. J. Symbolic Comput., [Kur] M.V. Kuryshkin, Op´erateurs quantiques g´en´eralis´es de cr´eation et d'annihilation. Ann. Fond. L. de Broglie, 5(1980), 111 -- 125. [Lev] V. Levandovskyy, Non-commutative Computer Algebra for Poly- nomial Algebra: Grobner Bases, Applications and Implementa- tion. Ph.D. Thesis, TU Kaiserslautern, 2005. [Li1] H. Li, Noncommutative Grobner Bases and Filtered-Graded Transfer. Lecture Note in Mathematics, Vol. 1795, Springer- Verlag, 2002. [Li2] H. Li, Grobner Bases in Ring Theory. World Scientific Publish- ing Co., 2011. Minimal Filtered Free Resolutions 127 [Li3] H. Li, On monoid graded local rings. Journal of Pure and Ap- plied Algebra, 216(2012), 2697 -- 2708. [Li4] H. Li, A Note on Solvable Polynomial Algebras. Computer Sci- ence Journal of Moldova, 1(64)(2014), 99 -- 109. Available at arXiv:1212.5988 [math.RA]. [Li5] H. Li, Computation of minimal graded free resolutions over N- graded solvable polynomial algebras. Available at arXiv:1401.5206 [math.RA] [Li6] H. Li, Computation of minimal filtered free resolutions over N- filtered solvable polynomial algebras. Available at arXiv:1401.5464 [math.RA] [Li7] H. Li, On Computation of Minimal Free Resolutions over Solv- able Polynomial Algebras. (pp66) accepted by Commentationes Mathematicae Universitatis Carolinae, to appear. [LVO] H. Li and F. Van Oystaeyen, Zariskian Filtrations. K- Monograph in Mathematics, Vol.2. Kluwer Academic Publish- ers, 1996. [LW] H. Li and Y. Wu, Filtered-graded transfer of Grobner basis computation in solvable polynomial algebras. Communications in Algebra, 1(28)(2000), 15 -- 32. [Man] Yu.I. Manin, Quantum Groups and Noncommutative Geome- try. Les Publ. du Centre de R´echerches Math., Universite de Montreal, 1988. [Mor] T. Mora, An introduction to commutative and noncommutative Grobner Bases, Theoretic Computer Science, 134(1994), 131 -- 173. [MP] J.C. McConnell and J.J. Pettit, Crossed products and multi- plicative analogues of Weyl algebras. J. London Math. Soc., (2)38(1988), 47 -- 55. [MR] J.C. McConnell and J.C. Robson, Noncommutative Noetherian Rings. Wiley-Interscience Publication, 1987. [NVO] C. Nastasescu and F. Van Oystaeyen, Graded ring theoey, Math. Library 28, North Holland, Amsterdam, 1982. [Ros] A.L. Rosenberg, Noncommutative Algebraic Geometry and Rep- resentations of Quantized Algebras. Kluwer Academic Publish- ers, 1995. [Rot] J.J. Rotman, An introduction to homological algebra. Academic Press, 1979. 128 Minimal Filtered Free Resolutions [Sch] F.O. Schreyer, Die Berechnung von Syzygien mit dem verallge- meinerten Weierstrasschen Divisionsatz. Diplomarbeit, Ham- burg, 1980. [Sm] S.P. Smith, Quantum groups: an introduction and survey for in: Noncommutative rings, S. Montgomery and ring theorists. L. Small eds., MSRI Publ. 24(1992), Springer-Verlag, New York, 131 -- 178. [SWMZ] Y. Sun, et al, A signature-based algorithm for computing Grobner bases in solvable polynomial algebras. In: Proc. IS- SAC'12, ACM Press, 351-358, 2012. [Uf] V. Ufnarovski, Introduction to noncommutative Grobner ba- in: Grobner Bases and Applications (Linz, 1998), sis theory. London Math. Soc. Lecture Notes Ser., 251, Cambridge Univ. Press, Cambridge, 1998, 259 -- 280. [Wal] R. Wallisser, Rationale approximation der q-analogues der expo- nentialfunktion und Irrationalitatsaussagen fur diese Funktion. Arch. Math., 44(1985), 59 -- 64. [Wik1] Decision problem (Solvable problem). Available at https://en.wikipedia.org/wiki/Solvable−problem [Wik2] Confluence (abstract rewriting). Available at https://en.wikipedia.org/wiki/Confluence−(term−rewriting)
1104.2460	1	1104	2011-04-13T12:07:24	Morita equivalence of inverse semigroups	[ "math.RA" ]	We describe how to construct all inverse semigroups Morita equivalent to a given inverse semigroup. This is done by taking the maximum inverse images of the regular Rees matrix semigroups over the inverse semigroup where the sandwich matrix satisfies what we call the McAlister conditions.	math.RA	math	MORITA EQUIVALENCE OF INVERSE SEMIGROUPS B. AFARA AND M. V. LAWSON Abstract. We describe how to construct all inverse semigroups Morita equivalent to a given inverse semigroup S. This is done by taking the maximum inverse images of the regular Rees matrix semigroups over S where the sandwich matrix satisfies what we call the McAlister conditions. 1 1 0 2 r p A 3 1 ] . A R h t a m [ 1 v 0 6 4 2 . 4 0 1 1 : v i X r a 1. Introduction The Morita theory of monoids was introduced independently by Ba- naschewski [1] and Knauer [6] as the analogue of the classical Morita theory of rings [8]. This theory was extended to semigroups with local units by Talwar [19, 20, 21]; a semigroup S is said to have local units if for each s ∈ S there exist idempotents e and f such that s = esf . Inverse semigroups have local units and the definition of Morita equiv- alence in their case assumes the following form. Let S be an inverse semigroup. If S acts on a set X in such a way that SX = X we say that the action is unitary. We denote by S-mod the category of uni- tary left S-sets and their left S-homomorphisms. Inverse semigroups S and T are said to be Morita equivalent if the categories S-mod and T -mod are equivalent. There have been a number of recent papers on this topic [3, 12, 13, 18] and ours takes the development of this theory a stage further. Rather than taking the definition of Morita equivalence as our start- ing point, we shall use instead two characterizations that are much easier to work with. We denote by C(S) the Cauchy completion of the semigroup S. This is the category with elements triples of the form (e, s, f ), where s = esf and e and f are idempotents, and multiplica- tion given by (e, s, f )(f, t, g) = (e, st, g). The first characterization is the following [3]. The first author would like to thank Aleppo University, Syria and the British Council for their support. The second author would like to thank Prof Gracinda Gomes and the CAUL project ISFL-1-143 supported by FCT . 1 2 B. AFARA AND M. V. LAWSON Theorem 1.1. Let S and T be semigroups with local units. Then S and T are Morita equivalent if and only if their Cauchy completions are equivalent. To describe the second characterization we shall need the following definition from [18]. Let S and T be inverse semigroups. An equivalence biset from S to T consists of an (S, T )-biset X equipped with surjective functions h−, −i : X × X → S , and [−, −] : X × X → T such that the following axioms hold, where x, y, z ∈ X, s ∈ S, and t ∈ T : (M1): hsx, yi = shx, yi (M2): hy, xi = hx, yi−1 (M3): hx, xix = x (M4): [x, yt] = [x, y]t (M5): [x, y] = [y, x]−1 (M6): x[x, x] = x (M7): hx, yiz = x[y, z]. Observe that by (M6) and (M7), we have that hx, xix = x[x, x] = x. Recall that a weak equivalence from one category to another is a functor that is full, faithful and essentially surjective. By the Ax- iom of Choice, categories are equivalent if and only if there is a weak equivalence between them. It is not hard to see, Theorem 5.1 of [18], that if there is an equivalence biset from S to T then there is a weak equivalence from C(S) to C(T ) and so by Theorem 1.1, the inverse semigroups S and T are Morita equivalent. In fact, the converse is true by Theorem 2.14 of [3]. Theorem 1.2. Let S and T be inverse semigroups. Then S and T are Morita equivalent if and only if there is an equivalence biset from S to T . The goal of this paper can now be stated: given an inverse semi- group S how do we construct all inverse semigroups T that are Morita equivalent to S? We shall show how to do this. This paper can be seen as a generalization and completion of some of the results to be found in [10]. Our main reference for general semigroup theory is Howie [4] and for inverse semigroups Lawson [11]. Since categories play a role, it is worth stressing, to avoid confusion, that a semigroup S is (von Neumann) regular if each element s ∈ S has an inverse t such that s = sts and t = tst. The set of inverses of s is denoted by V (s). Inverse semigroups are the regular semigroups in which each element has a unique inverse. MORITA EQUIVALENCE OF INVERSE SEMIGROUPS 3 2. The main construction Our main tool will be Rees matrix semigroups. These can be viewed as the semigroup analogues of matrix rings and, the reader will recall, matrix rings play an important role in the Morita theory of unital rings [8]. If S is a regular semigroup then a Rees matrix semigroup M(S; I, Λ; P ) over S need not be regular. However, we do have the following. Lemma 2.1 (Lemma 2.1 of [16]). Let S be a regular semigroup. Let RM(S; I, Λ; P ) be the set of regular elements of M(S; I, Λ; P ). Then RM(S; I, Λ; P ) is a regular semigroup. The semigroup RM(S; I, Λ; P ) is called a regular Rees matrix semi- group over S. Recall that a local submonoid of a semigroup S is a subsemigroup of the form eSe where e is an idempotent. A regular semigroup S is said to be locally inverse if each local submonoid is in- verse. Regular Rees matrix semigroups over inverse semigroups need not be inverse, but we do have the following. The proof follows by showing that each local submonoid of RM(S; I, Λ; P ) is isomorphic to a local submonoid of S. Lemma 2.2 (Lemma 1.1 of [17]). Let S be an inverse semigroup. Then a regular Rees matrix semigroup over S is locally inverse. Regular Rees matrix semigroups over inverse semigroups are locally inverse but not inverse. To get closer to being an inverse semigroup we need to impose more conditions on the Rees matrix semigroup. First, we shall restrict our attention to square Rees matrix semigroups: those semigroups where I = Λ. In this case, we shall denote our Rees matrix semigroup by M(S, I, p) where p : I × I → S is the function giving the entries of the sandwich matrix P . Next, we shall place some conditions on the sandwich matrix P : (MF1): pi,i is an idempotent for all i ∈ I. (MF2): pi,ipi,jpj,j = pi,j. (MF3): pi,j = p−1 j,i . (MF4): pi,jpj,k ≤ pi,k. (MF5): For each e ∈ E(S) there exists i ∈ I such that e ≤ pi,i. We shall call functions satisfying all these conditions McAlister func- tions. Our choice of name reflects the fact that McAlister was the first to study functions of this kind in [17]. The following is essentially Theorem 6.7 of [10] but we include a full proof for the sake of completeness. Lemma 2.3. Let M = M(S, I, p) where p satisfies (M1) -- (M4). 4 B. AFARA AND M. V. LAWSON (1) (i, s, j) is regular if and only if s−1s ≤ pj,j and ss−1 ≤ pi,i. (2) If (i, s, j) is regular then one of its inverses is (j, s−1, i). (3) (i, s, j) is an idempotent if and only if s ≤ pi,j. (4) The idempotents form a subsemigroup. Proof. (1). Suppose that (i, s, j) is a regular element. Then there is an element (k, t, l) such that (i, s, j) = (i, s, j)(k, t, l)(i, s, j) and (k, t, l) = (k, t, l)(i, s, j)(k, t, l). Thus, in particular, s = spj,ktpl,is. Now pj,js−1s = pj,js−1spj,ktpl,is = s−1spj,jpj,ktpl,is using the fact that pj,j is an idempotent. But pj,jpj,k = pj,k and so pj,js−1s = s−1spj,ktpl,is = s−1s. Thus s−1s ≤ pj,j. By symmetry, ss−1 ≤ pi,i. (2) This is a straightforward verification. (3). Suppose that (i, s, j) is an idempotent. Then s = spj,is. It follows that s−1 = s−1spj,iss−1 ≤ pj,i and so s ≤ pi,j. Conversely, suppose that s ≤ pi,j. Then s−1 ≤ pj,i and so s−1 = s−1spj,iss−1 which gives s = spj,is. This implies that (i, s, j) is an idempotent. (4). Let (i, s, j) and (k, t, l) be idempotents. Then by (2) above we have that s ≤ pi,j and t ≤ pk,l. Now (i, s, j)(k, t, l) = (i, spj,kt, l). But spj,kt ≤ pi,jpj,kpk,l ≤ pi,l. It follows that (i, s, j)(k, t, l) is an idempo- tent. (cid:3) A regular semigroup is said to be orthodox if its idempotents form a subsemigroup. Inverse semigroups are orthodox. An orthodox locally inverse semigroup is called a generalized inverse semigroup. They are the orthodox semigroups whose idempotents form a normal band. Corollary 2.4. Let S be an inverse semigroup. If M = M(S, I, p) where p satisfies (M1) -- (M4) then RM(S, I, p) is a generalized inverse semigroup. Let S be a regular semigroup. Then the intersection of all congru- ences ρ on S such S/ρ is inverse is a congruence denoted by γ; it is called the minimum inverse congruence. Lemma 2.5 (Theorems 6.2.4 and 6.2.5 of [4]). Let S be an orthodox semigroup. Then the following are equivalent: (1) s γ t. (2) V (s) ∩ V (t) 6= ∅. (3) V (s) = V (t). Lemma 2.6. Let RM = RM(S, I, p) where p satisfies (M1) -- (M4). Then (i, s, j)γ(k, t, l) if and only if s = pi,ktpl,j and t = pk,lspj,l. MORITA EQUIVALENCE OF INVERSE SEMIGROUPS 5 Proof. Lemma 2.5 forms the backdrop to this proof. Suppose that (i, s, j)γ(k, t, l). Then the two elements have the same sets of inverses. Now (j, s−1, i) is an inverse of (i, s, j) and so by assumption it is an inverse of (k, t, l). Thus t = tpl,js−1pi,kt and s−1 = s−1pi,ktpl,js−1. It follows that so that Now s ≤ pi,ktpl,j and t−1 ≤ pl,js−1pi,k t ≤ pk,ispj,l. s ≤ pi,ktpl,j ≤ pi,kpk,ispj,lpl,j ≤ pi,ispj,j = s. Thus s = pi,ktpl,j. Similarly, t = pk,ispj,l. Conversely, suppose that s = pi,ktpl,j and t = pk,lspj,l. We shall prove that V (i, s, j) ∩ V (k, t, l) 6= ∅. To do this, we shall prove that (j, s−1, i) is an inverse of (k, t, l). We calculate tpl,js−1pi,kt = t(pi,js−1pi,k)t = t(pk,ispj,l)−1t = tt−1t = t. Similarly, s−1 = s−1pi,ktpl,js−1. The result now follows. (cid:3) With the assumptions of the above lemma, put IM(S, I, p) = RM(S, I, p)/γ. We call IM(S, I, p) the inverse Rees matrix semigroup over S. A homomorphism θ : S → T between semigroups with local units is said to be a local isomorphism if the following two conditions are satisfied: (LI1): θ eSf : eSf → θ(e)T θ(f ) is an isomorphism for all idem- potents e, f ∈ S. (LI2): For each idempotent i ∈ T there exists an idempotent e ∈ S such that iDθ(e). This definition is a slight refinement of the one given in [12]. Lemma 2.7. Let θ : S → T be a surjective homomorphism between regular semigroups. Then θ is a local isomorphism if and only if θ eSe : eSe → θ(e)T θ(e) is an isomorphism for all idempotents e ∈ S. Proof. The homomorphism is surjective and so (LI2) is automatic. We need only prove that (LI2) follows from the assumption that θ eSe : eSe → θ(e)T θ(e) is an isomorphism for all idempotents e ∈ S. This follows from Lemma 1.3 of [17]. (cid:3) 6 B. AFARA AND M. V. LAWSON Lemma 2.8 (Proposition 1.4 of [17]). Let S be a regular semigroup. Then the natural homomorphism from S to S/γ is a local isomorphism if and only if S is a generalized inverse semigroup. Our next two results bring Morita equivalence into the picture via Theorem 1.1. Lemma 2.9. Let S and T be inverse semigroups. If θ : S → T is a surjective local isomorphism then S and T are Morita equivalent. Proof. Define Θ : C(S) → C(T ) by Θ(e, s, f ) = (θ(e), θ(s), θ(f )). Then Θ is a functor, and it is full and faithful because θ is a local isomor- phism. Identities in C(T ) have the form (i, i, i) where i is an idempotent in T . Because θ is surjective and S is inverse there is an idempotent e ∈ S such that θ(e) = i. Thus every identity in C(T ) is the image of an identity in C(S). It follows that Θ is a weak equivalence. Thus the categories C(S) and C(T ) are equivalent and so, by Theorem 1.1, the semigroups S and T are Morita equivalent. (cid:3) Lemma 2.10. Let M = M(S, I, p) where p satisfies (MF1) -- (MF5). Then S is Morita equivalent to RM(S, I, p). Proof. We shall construct a weak equivalence from C(RM(S, I, p)) to C(S). By Theorem 1.1 this implies that S is Morita equivalent to RM(S, I, p). A typical element of C(RM(S, I, p)) has the form s = [(i, a, j), (i, s, k), (l, b, k)] where (i, s, j) is regular and (i, a, j) and (l, b, k) are idempotents and (i, a, j)(i, s, k)(l, b, k) = (i, s, k). Observe that both apj,i and bpk,l are idempotents and that (apj,i)spk,l(bpk,l) = spk,l. It follows that (apj,i, spk,l, bpk,l) is a well-defined element of C(S). We may therefore define Ψ : C(RM(S, I, p)) → C(S) by Ψ[(i, a, j), (i, s, k), (l, b, k)] = (apj,i, spk,l, bpk,l). It is now easy to check that Ψ is full and faithful. Let (e, e, e) be an arbitrary identity of C(S). Then e is an idempotent in S. By (MF5), there exists i ∈ I such that e ≤ pi,i. It follows that (i, e, i) is an idempotent in RM(S, I, p). Thus is an identity in C(RM(S, I, p)). But [(i, e, i), (i, e, i), (i, e, i)] Ψ[(i, e, i), (i, e, i), (i, e, i)] = (epi,i, epi,i, epi,i) = (e, e, e). MORITA EQUIVALENCE OF INVERSE SEMIGROUPS 7 Thus every identity in C(S) is the image under Ψ of an identity in C(RM(S, I, p)). In particular, Ψ is essentially surjective. (cid:3) We may summarize what we have found so far in the following result. Proposition 2.11. Let S be an inverse semigroup and let p : I ×I → S be a McAlister function. Then S is Morita equivalent to the inverse Rees matrix semigroup IM(S, I, p). 3. The main theorem Our goal now is to prove that all inverse semigroups Morita equiva- lent to S are isomorphic to inverse Rees matrix semigroups IM(S, I, p). We shall use Theorem 1.2. We begin with some results about equiva- lence bisets all of which are taken from [18]. The following is part of Proposition 2.3 [18]. Lemma 3.1. Let (S, T, X, h−, −i, [−, −]) be an equivalence biset. (1) For each x ∈ X both hx, xi and [x, x] are idempotents. (2) hx, yihz, wi = hx[y, z], wi. (3) [x, y][z, w] = [x, hy, ziw]. (4) hxt, yi = hx, yt−1i. (5) [sx, y] = [x, s−1y]. Lemma 3.2. Let (S, T, X, h, i, [, ]) be an equivalence biset from S to T . (1) For each x ∈ X there exists a homomorphism ǫx : E(S) → E(T ) such that ex = xǫx(e) for all e ∈ E(S). (2) For each x ∈ X there exists a homomorphism ηx : E(S) → E(T ) such that xf = ηx(f )x for all e ∈ E(S). Proof. We prove (1); the proof of (2) follows by symmetry. Define ǫx by ǫx(e) = [ex, ex]. By Proposition 2.4 of [18], this is a semigroup homomorphism. Next we use the argument from Proposition 3.6 of [18]. We calculate x[ex, ex] as follows x[ex, ex] = hx, exiex = hx, xiex = ehx, xix = ex, as required. (cid:3) Lemma 3.3. Let (S, T, X, h, i, [, ]) be an equivalence biset from S to T . Define p : X × X → S by px,y = hx, yi. Then p is a McAlister function. Proof. (MF1) holds. By Lemma 3.1(1), hx, xi is an idempotent. (MF2) holds. By Lemma 3.1(2), hx, xihx, yi = hx[x, x], yi. But x[x, x] = x by (M6), and so hx, xihx, yi = hx, yi. The other result holds dually. (MF3) holds. This follows from (M2). 8 B. AFARA AND M. V. LAWSON (MF4) holds. By Lemma 3.1(2), we have that hx, yihy, zi = hx[y, y], zi. By Lemma 3.2, we have that x[y, y] = ηx([y, y])x = f x. Thus hx[y, y], zi = hf x, xi = f hx, zi ≤ hx, zi. (MF5) holds. Let e ∈ E(S). Then since h−, −i is surjective, there exists x, y ∈ X such that e = hx, yi. But then e = hx, yihy, xi ≤ hx, xi = px,x. (cid:3) Lemma 3.4. Let (S, T, X, h, i, [, ]) be an equivalence biset from S to T . Define p : X × X → S by px,y = hx, yi. Form the regular Rees matrix semigroup R = RM(S, X, p). Define θ : RM(S, X, p) → T by θ(x, s, y) = [x, sy]. Then θ is a surjective homomorphism with kernel γ. Proof. We show first that θ is a homomorphism. By definition Thus whereas (x, s, y)(u, t, v) = (x, shy, uit, v). θ((x, s, y)(u, t, v)) = [x, shy, uitv], θ(x, s, y)θ(u, t, v) = [x, sy][u, tv]. By Lemma 3.1(3), we have that [x, sy][u, tv] = [x, hsy, uitv] but by (M1), hsy, ui = shy, ui. It follows that θ is a homomorphism. Next we show that θ is surjective. Let t ∈ T . Then there exists (x, y) ∈ X×X such that [x, y] = t. Consider the element (x, hx, xihy, yi, y) of M(S, I, p). This is in fact an element of RM(S, X, p). The image of this element under θ is [x, hx, xihy, yiy] = [x, hx, xiy] since hy, yiy = y. But by Lemma 3.1(5), we have that [x, hx, xiy] = [hx, xix, y] = [x, y] = t, as required. It remains to show that the kernel of θ is γ. Let (x, s, y), (u, t, v) ∈ RM(S, X, p). Suppose first that θ(x, s, y) = θ(u, t, v). By definition, [x, sy] = [u, tv]. Then s = hx, xishy, yi = hx, xihsy, yi = hx[x, sy], yi by Lemma 3.1(2). But [x, sy] = [u, tv]. Thus s = hx[u, tv], yi = hx, uihtv, yi = hx, uithv, yi. By symmetry and Lemma 2.6, we deduce that (x, s, y)γ(u, t, v). MORITA EQUIVALENCE OF INVERSE SEMIGROUPS 9 Suppose now that (x, s, y)γ(u, t, v). Then by Lemma 2.6 s = hx, uithv, yi and t = hu, xishy, vi. Now [x, sy] = [x, hx, uithv, yiy] = [x, hx, uitv[y, y]] = [u[x, x], tv[y, y]] = [x, x][u, tv][y, y] using Lemma 3.1. This gives [x, sy] ≤ [u, tv]. A symmetric argument shows that [u, tv] ≤ [x, sy]. Hence [x, sy] = [u, tv], as required. (cid:3) We may now state our main theorem. Theorem 3.5. Let S be an inverse semigroup. For each McAlister function p : I ×I → S the inverse Rees matrix semigroup IM(S, I, p) is Morita equivalent to S, and every inverse semigroup Morita equivalent to S is isomorphic to one of this form. Remarks 3.6. (1) Let S be an inverse monoid and suppose that p : I × I → S is a function satisfying (MF1) -- (MF5). Condition (MF5) says that For each e ∈ E(S) there exists i ∈ I such that e ≤ pi,i. Thus,in particular, there exists i0 ∈ I such that 1 ≤ pi0,i0. But pi0,i0 is an idempotent and so 1 = pi0,i0. Suppose now that p : I ×I → S is a function satisfying (MF1) -- (MF4) and there exists i0 ∈ I such that 1 = pi0,i0. Every idempotent e ∈ S satisfies e ≤ 1. It follows that (MF5) holds. Thus in the monoid case, the functions p : I × I → S satisfying (MF1) -- (MF5) are precisely what we called normalized, pointed sandwich functions in [10]. Furthermore, the inverse semigroups Morita equivalent to an inverse monoid are precisely the enlargements of that monoid [3, 12]. Thus the theory developed in pages 446 -- 450 of [10] is the monoid case of the theory we have just developed. (2) McAlister functions are clearly examples of the manifolds de- fined by Grandis [7] and so are related to the approach to sheaves based on Lawvere's paper [15] and developed by Walters [22]. See Section 2.8 of [2]. (3) The Morita theory of inverse semigroups is initmately connected to the theory of E-unitary covers and almost factorizability [9]. It has also arisen in the solution of concrete problems [5]. (4) In the light of (2) and (3) above, an interesting special case to consider would be where the inverse semigroup is complete and infinitely distributive. 10 B. AFARA AND M. V. LAWSON References [1] B. Banaschewski, Functors into categories of M -sets, Abh. Math. Sem. Univ. Hamburg 38 (1972), 49 -- 64. [2] F. Borceux, Handbook of categorical algebra 3, CUP, 1994. [3] J. Funk, M. V. Lawson, B. Steinberg, Characterizations of Morita equivalence of inverse semigroups, Preprint, 2010. [4] J. M. Howie, Fundamentals of semigroup theory, Clarendon Press, Oxford, 1995. [5] K. Kaarli, L. M´arki, A characterization of the inverse monoid of bi-congruences of certain algebras, Internat. J. Algebra Comput. 19 (2009), 791 -- 808. Periodica Math. Hungar. 40 (2000), 85 -- 107. Proc. Edinb. Math. Soc. 44 (2001), 173 -- 186. [6] U. Knauer, Projectivity of acts and Morita equivalence of monoids, Semigroup Forum 3 (1972), 359 -- 370. [7] M. Grandis, Cohesive categories and manifolds, Annali di Matematica pura ed applicata CLVII (1990), 199 -- 244. Errata corrige, ibid 179 (2001), 471 -- 472. [8] T. Y. Lam, Lectures on rings and modules, Springer-Verlag, New York, 1999. [9] M. V. Lawson, Almost factorisable inverse semigroups, Glasgow Math. J. 36 (1994), 97 -- 111. [10] M. V. Lawson, Enlargements of regular semigroups, Proc. Edinb. Math. Soc. 39 (1996), 425 -- 460. [11] M. V. Lawson, Inverse semigroups, World Scientific, 1998. [12] M. V. Lawson, Morita equivalence of semigroups with local units, to appear in J. Pure and Applied Algebra. [13] M. V. Lawson, Generalized heaps, inverse semigroups and Morita equivalence, accepted by Algebra Universalis. [14] M. V. Lawson, L. M´arki, Enlargements and coverings by Rees matrix semi- groups, Monatsh. Math. 129 (2000), 191 -- 195. [15] F. W. Lawvere, Metric spaces, generalized logic and closed categories, Rend. Sem. Mat. e Fis. di Milano 43 (1973), 135 -- 166. [16] D. B. McAlister, Regular Rees matrix semigroups and regular Dubreil-Jacotin semigroups, J. Aust. Math. Soc. (Series A) 31 (1981), 325 -- 336. [17] D. B. McAlister, Rees matrix covers for locally inverse semigroups, Trans Amer. Math. Soc. 277 (1983), 727 -- 738. [18] B. Steinberg, Strong Morita equivalence of inverse semigroups, To appear in Houston J. Math. [19] S. Talwar, Morita equivalence for semigroups, J. Aust. Math. Soc. (Series A) 59 (1995), 81 -- 111. [20] S. Talwar, Strong morita equivalence and a generalisation of the Rees theorem, J. Algebra 181 (1996), 371 -- 394. [21] S. Talwar, Strong morita equivalence and the synthesis theorem, Internat. J. Algebra Comput. 6 (1996), 123 -- 141. [22] R. F. C. Walters, Sheaves and Cauchy-complete categories, Cah. Topol. G´eom. Diff´er. Cat´eg. 22 (1981), 283 -- 286. MORITA EQUIVALENCE OF INVERSE SEMIGROUPS 11 Department of Mathematics, Heriot-Watt University, Riccarton, Edinburgh EH14 4AS, Scotland E-mail address: [email protected] Department of Mathematics and the Maxwell Institute for Mathe- matical Sciences, Heriot-Watt University, Riccarton, Edinburgh EH14 4AS, Scotland E-mail address: [email protected]
1804.05206	1	1804	2018-04-14T11:29:57	Three limit representations of the core-EP inverse	[ "math.RA" ]	In this paper, we present three limit representations of the core-EP inverse. The first approach is based on the full-rank decomposition of a given matrix. The second and third approaches, which depend on the explicit expression of the core-EP inverse, are established. The corresponding limit representations of the dual core-EP inverse are also given. In particular, limit representations of the core and dual core inverse are derived	math.RA	math	Three limit representations of the core-EP inverse Mengmeng Zhoua, Jianlong Chenb,∗, Tingting Lic, Dingguo Wangd aSchool of Mathematics, Southeast University, Nanjing, Jiangsu 210096, China bSchool of Mathematics, Southeast University, Nanjing, Jiangsu, 210096, China cSchool of Mathematics, Southeast University, Nanjing, Jiangsu 210096, China dSchool of Mathematical Sciences, Qufu Normal University, Qufu, Shandong, 273165, China Abstract In this paper, we present three limit representations of the core-EP inverse. The first approach is based on the full-rank decomposition of a given matrix. The second and third approaches, which depend on the explicit expression of the core-EP inverse, are established. The corresponding limit representations of the dual core-EP inverse are also given. In particular, limit representations of the core and dual core inverse are derived. Keywords: Core-EP inverse; Core inverse; Limit representation; 2017 MSC: 15A09 1. Introduction In 1974, limit representation of the Drazin inverse was established by Meyer [12]. In 1986, Kalaba and Rasakhoo [8] introduced limit representation of the Moore-Penrose inverse. In 1994, An alternative limit representation of the 5 Drazin inverse was given by Ji [7]. It is well known that the six kinds of classical generalized inverses: the Moore-Penrose inverse, the weighted Moore-Penrose inverse, the group inverse, the Drazin inverse, the Bott-Duffin inverse and the generalized Bott-Duffin inverse can be presented as particular generalized in- verse A(2) T,S with prescribed range and null space (see, for example, [2, 3, 5, 14]). ∗Corresponding author Email addresses: [email protected]) (Mengmeng Zhou), [email protected]) (Jianlong Chen), [email protected]) (Tingting Li), [email protected] (Dingguo Wang) Preprint submitted to Journal of Filomat April 17, 2018 Let T be a subspace of Cn and let S be a subspace of Cm. The generalized inverse A(2) T,S [2] of matrix A ∈ Cm×n is the matrix G ∈ Cn×m satisfying GAG = G, R(G) = T, N (G) = S. In 1998, Wei [18] established a unified limit representation of the generalized inverse A(2) T,S . Let A ∈ Cm×n be matrix of rank r, let T be a subspace of Cn of dimension s ≤ r, and let S be a subspace of Cm of dimension m − s. Suppose G ∈ Cn×m such that R(G) = T and N (G) = S. If A(2) T,S exists, then A(2) T,S = lim λ→0 (GA − λI)−1G = lim λ→0 G(AG − λI)−1, 10 where R(G) and N (G) denote the range space and null space of G, respectively. In 1999, Stanimirovi´c [15] introduced a more general limit formula. Let M and N are two arbitrary p × q matrices, then lim λ→0 (M ∗N + λI)−1M ∗ = lim λ→0 M ∗(N M ∗ + λI)−1. (1.1) The limit representation of generalized inverse A(2) T,S is a special case of the above general formula. In 2012, Liu et al. [10] introduced limit representation of the generalized inverse A(2) R(B),N (C) in Banach space. Let A ∈ Cm×n be matrix of rank r. Let T be a subspace of Cn of dimension s ≤ r, and let S be a subspace of Cm of dimension m − s. Suppose B ∈ Cn×s and C ∈ Cs×m such that R(B) = T and N (C) = S. If A(2) T,S exists, then A(2) T,S = lim λ→0 B(CAB + λI)−1C. 15 In 2010, the core and dual core inverse were introduced by Baksalary and Trenkler for square matrices of index at most 1 in [1]. In 2014, the core inverse was extended to the core-EP inverse defined by Manjunatha Prasad and Mohana [11]. The core-EP inverse coincides with the core inverse if the index of a given matrix is 1. In this paper, the dual conception of the core-EP inverse 20 was called the dual core-EP inverse. The characterizations of the core-EP and 2 core inverse were investigated in complex matrices and rings (see, for example, [4, 6, 9, 13, 16, 17, 19]). From the above mentioned limit representations of the generalized inverse, we know that limit representation of the core-EP inverse similar to the form of 25 (1.1) has not been investigated in the literature. The purpose of this paper is to establish three limit representations of the core-EP inverse. The first approach is based on the full-rank decomposition of a given matrix. The second and third approaches, which depend on the explicit expression of the core-EP inverse, are represented. The corresponding limit 30 representations of the dual core-EP inverse are also given. In particular, limit representations of the core and dual core inverse are derived. 2. Preliminaries In this section, we give some auxiliary definitions and lemmas. For arbitrary matrix A ∈ Cm×n, the symbol Cm×n denotes the set of all 35 complex m × n matrices. A∗ and rk(A) denote the conjugate transpose and rank of A, respectively. I is the identity matrix of an appropriate order. If k is the smallest nonnegative integer such that rk(Ak) = rk(Ak+1), then k is called the index of A and denoted by ind(A). Definition 2.1. [2] Let A ∈ Cm×n. The unique matrix A† ∈ Cn×m is called the Moore-Penrose inverse of A if it satisfies AA†A = A, A†AA† = A†, (AA†)∗ = AA†, (A†A)∗ = A†A. Definition 2.2. [2] Let A ∈ Cn×n. The unique matrix AD ∈ Cn×n is called the Drazin inverse of A if it satisfies Ak+1AD = Ak, ADAAD = AD, AAD = ADA, where k = ind(A). When k = 1, the Drazin inverse reduced to the group inverse 40 and it is denoted by A#. 3 Definition 2.3. [1] A matrix A #(cid:13) ∈ Cn×n is called the core inverse of A ∈ Cn×n if it satisfies AA #(cid:13) = PA and R(A #(cid:13)) ⊆ R(A). Dually, A matrix A #(cid:13) ∈ Cn×n is called the dual core inverse of A ∈ Cn×n if it satisfies A #(cid:13)A = PA∗ and R(A #(cid:13)) ⊆ R(A∗). Definition 2.4. [11] A matrix X ∈ Cn×n, denoted by A †(cid:13), is called the core-EP inverse of A ∈ Cn×n if it satisfies XAX = X and R(X) = R(X ∗) = R(AD). Dually, A matrix X ∈ Cn×n, denoted by A †(cid:13), is called the dual core-EP inverse of A ∈ Cn×n if it satisfies XAX = X and R(X) = R(X ∗) = R((A∗)D). The core-EP inverse was extended from complex matrices to rings by Gao and Chen in [6]. Lemma 2.5. [6] Let A ∈ Cn×n with ind(A) = k and let m be a positive integer with m ≥ k. Then A †(cid:13) = ADAm(Am)†. 45 Clearly, If A ∈ Cn×n with ind(A) = k, then it has a unique core-EP inverse. So according to Lemma 2.5, we have A †(cid:13) = ADAk(Ak)† = ADAk+1(Ak+1)† = Ak(Ak+1)†. (2.1) As for an arbitrary matrix A ∈ Cn×n. If A is nilpotent, then AD = 0. In this case, A †(cid:13) = A †(cid:13) = 0. This case is considered to be trivial. So we restrict the matrix A to be non-nilpotent in this paper. 50 Lemma 2.6. [2] Let A ∈ Cn×n with ind(A) = k. If A = B1G1 is a full- rank decomposition and GiBi = Bi+1Gi+1 are also full-rank decompositions, i = 1, 2, ..., k − 1. Then the following statements hold: 4 (1) GkBk is invertible. (2) Ak = B1B2 · · · BkGk · · · G2G1. 55 (3) AD = B1B2 · · · Bk(GkBk)−k−1Gk · · · G1. In particular, for k = 1, then G1B1 is invertible and A# = B1(G1B1)−2G1. According to [2], it is also known that A† = G∗ 1(G1G∗ 1)−1(B∗ 1 B1)−1B∗ 1 . Lemma 2.7. [17] Let A ∈ Cn×n with ind(A) = 1. If A = M N is a full-rank decomposition, then A #(cid:13) = M (N M )−1(M ∗M )−1M ∗. Lemma 2.8. [16] Let A ∈ Cn×n with ind(A) = k. Then A can be written as the sum of matrices A1 and A2, i.e., A = A1 + A2, where (1) rk(A2 1) = rk(A1). 60 (2) Ak 2 = 0. (3) A∗ 1A2 = A2A1 = 0. 3. The first approach In this section, we present limit representations of the core-EP and dual core- EP inverse, which depend on the full-rank decomposition of a given matrix. In 65 particular, limit representations of the core and dual core inverse are also given. Theorem 3.1. Let A ∈ Cn×n with ind(A) = k, and the full-rank decomposition of A be as in Lemma 2.6. Then A †(cid:13) = lim λ→0 = lim λ→0 B(BGkBk)∗(BGkBk(BGkBk)∗ + λI)−1 B((BGkBk)∗BGkBk + λI)−1(BGkBk)∗ 5 and A †(cid:13) = lim λ→0 = lim λ→0 ((GkBkG)∗GkBkG + λI)−1(GkBkG)∗G (GkBkG)∗(GkBkG(GkBkG)∗ + λI)−1G, where B = B1B2 · · · Bk and G = Gk · · · G2G1. 70 Proof. Let B = B1B2 · · · Bk, G = Gk · · · G2G1 and X = B(BGkBk)†. Suppose that A ∈ Cn×n with ind(A) = k, and the full-rank decomposition of A be as in Lemma 2.6. From [15], we know that A† = lim λ→0 A∗(AA∗ + λI)−1 = lim λ→0 (A∗A + λI)−1A∗. (3.1) So it is sufficient to verify X = A †(cid:13). Since AB = B1G1B1 · · · Bk = B1B2G2B2 · · · Bk = · · · = BGkBk, (3.2) XAX = B(BGkBk)†AB(BGkBk)† (3.2) = B(BGkBk)† = X. From Lemma 2.6,we obtain B(BGkBk)† = B(GkBk)−1(B∗B)−1B∗. Therefore, X ∗ = B(B∗B)−1((GkBk)−1)∗B∗. It is easy verify that B∗B, GG∗ and GkBk are invertible. Hence rk(B) = rk(B(GkBk)∗B∗B) ≤ rk(B(BGkBk)∗) ≤ rk(B), rk(B) = rk(B(B∗B)−1((GkBk)−1)∗B∗B) ≤ rk(X ∗) ≤ rk(B), rk(B) = rk(B(GkBk)−k−1GG∗) ≤ rk(B(GkBk)−k−1G) ≤ rk(B). Thus, we have the following equalities: R(X) = R(B(BGkBk)†) = R(B(BGkBk)∗) = R(B), R(X ∗) = R(B(B∗B)−1((GkBk)−1)∗B∗) = R(B), 6 R(AD) = R(B(GkBk)−k−1G) = R(B). Namely, R(X) = R(X ∗) = R(AD). Similarly, we can verify A †(cid:13) = (GkBkG)†G. This completes the proof. 75 Let k = 1 in Theorem 3.1. Then we obtain the following corollary. Corollary 3.2. Let A ∈ Cn×n with ind(A) = 1. If A = BG is a full-rank decomposition, then and A #(cid:13) = lim λ→0 = lim λ→0 B(AB)∗(AB(AB)∗ + λI)−1 B((AB)∗AB + λI)−1(AB)∗ A #(cid:13) = lim λ→0 ((GA)∗GA + λI)−1(GA)∗G = lim λ→0 (GA)∗(GA(GA)∗ + λI)−1G. 4. The second and third approaches 80 In this section, we present two types of limit representations of the core-EP and dual core-EP inverse, which depend on their own explicit representation. In particular, limit representations of the core and dual core inverse are also given. Based on Definition 2.4 and Lemma 2.5, we present the second approach by using the following equation firstly: A †(cid:13)Ak+1 = ADAk(Ak)†Ak+1 = Ak, where k = ind(A). Theorem 4.1. Let A ∈ Cn×n with ind(A) = k. Then 85 (1) A †(cid:13) = lim λ→0 Ak(Ak)∗(Ak+1(Ak)∗ + λI)−1, (2) A †(cid:13) = lim λ→0 ((Ak)∗Ak+1 + λI)−1(Ak)∗Ak. 7 Proof. Suppose that A ∈ Cn×n with ind(A) = k. Let the core-EP decomposition be as in Lemma 2.8. From [16], we know that there exists a unitary matrix U such that A = U   T S 0 N   U ∗, where T is a nonsingular matrix, rk(T ) = rk(Ak) and N is a nilpotent matrix of index k. The core-EP inverse of A is A †(cid:13) = U   T −1 0  0  0 U ∗. By a direct computation, we obtain Ak = U   T k 0 T 0   U ∗, Ak+1 = U  T k+1 T T  0 0   U ∗, (Ak)∗ = U   0 0 T ∗ (T k)∗ U ∗, where T =   Ak(Ak)∗ = U  T k(T k)∗ + T ( T )∗  0 k−1 X i=0 T iSN k−1−i. (4.1)   0 0 U ∗, (4.2) Ak+1(Ak)∗ + λIn = U  T k+1(T k)∗ + T T ( T )∗ + λIrk(T )  0 0 λIn−rk(T )   U ∗. (4.3) Since T is nonsingular, T k(T k)∗ + T ( T )∗ is positive definite matrix. Combining 90 with (4.2)and (4.3), we have × 0 0    0  0 U ∗. lim λ→0 Ak(Ak)∗(Ak+1(Ak)∗ + λI)−1 = lim λ→0 U  T k(T k)∗ + T ( T )∗  0  T k+1(T k)∗ + T T ( T )∗ + λIrk(T )  0 0 λIn−rk(T ) −1   U ∗ = U  T −1  0 (2) It is analogous. Let k = 1, we have the following corollary. Corollary 4.2. Let A ∈ Cn×n with ind(A) = 1. Then (1) A #(cid:13) = lim λ→0 AA∗(A2A∗ + λI)−1, 8 95 (2) A #(cid:13) = lim λ→0 (A∗A2 + λI)−1A∗A. Next, we present the third approach of limit representations of the core-EP inverse. From Definition 2.4 and equation (2.1), it is easy to know that A †(cid:13) = Ak(Ak+1)† and A †(cid:13) = (Ak+1)†Ak, where k = ind(A). Combining with limit representation of the Moore-Penrose inverse of A, we have the following theorem. Theorem 4.3. Let A ∈ Cn×n with ind(A) = k. Then and A †(cid:13) = lim λ→0 = lim λ→0 Ak(Ak+1)∗(Ak+1(Ak+1)∗ + λI)−1 Ak((Ak+1)∗Ak+1 + λI)−1(Ak+1)∗ A †(cid:13) = lim λ→0 = lim λ→0 ((Ak+1)∗Ak+1 + λI)−1(Ak+1)∗Ak (Ak+1)∗(Ak+1(Ak+1)∗ + λI)−1Ak 100 Let k = 1, we have the following corollary. Corollary 4.4. Let A ∈ Cn×n with ind(A) = 1. Then and A #(cid:13) = lim λ→0 = lim λ→0 A(A2)∗(A2(A2)∗ + λI)−1 A((A2)∗A2 + λI)−1(A2)∗ A #(cid:13) = lim λ→0 = lim λ→0 ((A2)∗A2 + λI)−1(A2)∗A (A2)∗(A2(A2)∗ + λI)−1A. Let k = 1. From [1], we know that A #(cid:13) = (A2A†)†. According to the above corollary, it is known that A #(cid:13) = A(A2)†. Since the core inverse is unique, we 105 obtain the following corollary. Corollary 4.5. Let A ∈ Cn×n with ind(A) = 1. Then (A2A†)† = A(A2)†. 9 5. Examples In this section, we present two examples to illustrate the efficacy of the established limit representations in this paper. Example 5.1. Let A =   1 1 2 5 0 1 1 1 0 3 3 1 1 0 1 4   , where ind(A) = 2 and rk(A) = 3. Let B1 =  1 0  0 1 1 5 1 1 3 1 0 4   , G1 =  1 0  0 0 1 1 1 0 0  0 0  1 , B2 =  1 1  0  1 0  1 , G2 =  0  1 4 2 0 4   . The exact core-EP inverse of A is equal to . Set B = B1B2 and E = G2B2. 20 189 2 189 − 2 189   2 21 20 189 2 189 − 2 189 2 21 20 189 2 189 − 2 189 2 21     1 189 13 756 25 378 − 1 84 − 2 63 4 63 17 63 − 2 21 19 189 − 5 756 − 29 378 3 28   , 1 189 13 756 25 378 − 1 84 1 189 13 756 25 378 − 1 84 − 2 63 4 63 17 63 − 2 21 − 2 63 4 63 17 63 − 2 21 19 189 − 5 756 − 29 378 3 28 19 189 − 5 756 − 29 378 3 28     , . A †(cid:13) = 1 756   80 8 −8 4 −24 48 13 50 204 −58 76 −5   72 −9 −72 81 Using matlab, we have B(BE)∗(BE(BE)∗ + λI)−1 = lim λ→0 A2(A2)∗(A3(A2)∗ + λI)−1 = lim λ→0 A2(A3)∗(A3(A3)∗ + λI)−1 = lim λ→0 10 Example 5.2. Let A =   1 0 4 0 2 0  3 2  1 , where rk(A) = 2 and ind(A) = 1. Let B =  1 4  2  3 2  1 ,G =  1  0 0 0 0 1   . According to Lemma 2.7, the exact core inverse of A is equal to A #(cid:13) = B(B∗BGB)−1B∗ = Using matlab, we obtain    −0.2000  0.8000 −0.1600 −0.0800 0.1200 0.2400 . 0.4000 −0.0800 −0.0400 B(AB)∗(AB(AB)∗ + λI)−1 = lim λ→0  − 1 5  4 5 2 5 6 25 − 4 25 − 2 25 3 25 − 2 25 − 1 25   , AA∗(A2A∗ + λI)−1 = lim λ→0  − 1 5  4 5 2 5 6 25 − 4 25 − 2 25 3 25 − 2 25 − 1 25   , A(A2)∗(A2(A2)∗ + λI)−1 = lim λ→0  − 1 5  4 5 2 5 6 25 − 4 25 − 2 25 3 25 − 2 25 − 1 25   . 110 Acknowledgements This research is supported by the National Natural Science Foundation of China (No.11771076, No.11471186). References References 115 [1] O.M. Baksalary, G. Trenkler, Core inverse of matrices, Linear Multilinear Algebra 58 (2010) 681-697. 11 [2] A. Ben-Israel, T.N.E. Greville, Generalized Inverses: Theory and Applica- tions, second ed., Springer-Verlag, New York, 2003. [3] S.L. Campbell, C.D. Meyer Jr., Generalized Inverse of Linear Transforma- 120 tion, Pitman London, 1979, Dover, New York, 1991. [4] J.L. Chen, H.H. Zhu, P. Patri´cio, Y.L. Zhang, Characterizations and rep- resentations of core and dual core inverses, Canad. Math. Bull. 60 (2017) 269-282. [5] Y.L. Chen, The generalized Bott-Duffin inverse and its applications, Linear 125 Algebra Appl. 134 (1990) 71-91. [6] Y.F. Gao, J.L. Chen, Pseudo core inverses in rings with involution, Comm. Algebra 46 (2018) 38-50. [7] J. Ji, An alternative limit expression of Drazin inverse and its application, Appl. Math. Comput. 61 (1994) 151-156. 130 [8] R.E. Kalaba, N. Rasakhoo, Algorithm for generalized inverse. J. Optimiza- tion Theory and Appl. 48 (1986) 427-435. [9] T.T. Li, J.L. Chen, Characterizations of core and dual core inverses in rings with involution, Linear Multilinear Algebra Doi:10.108003081087.2017.1320963. 135 [10] X.J. Liu, Y.M. Yu, J. Zhong, Y.M. Wei, Integral and limit representations of the outer inverse in Banach space, Linear Multilinear Algebra 60 (2012) 333-347. [11] K. Manjunatha Prasad, K.S. Mohana, Core-EP inverse, Linear Multilinear Algebra 62 (2014) 792-802. 140 [12] C.D. Meyer, Limits and the index of a square matrix, SIAM J. Appl. Math. 26 (1974) 469-478. 12 [13] D.S. Raki´c, N. C. Dinci´c, D.S. Djordjevi´c, Group, Moore-Penrose, core and dual core inverse in rings with involution, Linear Algebra Appl. 463 (2014) 115-133. 145 [14] C.R. Rao, S.K. Mitra, Generalized Inverse of Matrices and its Applications, Wiley, New York, 1971. [15] P.S. Stanimirovi´c, Limit representations of generalized inverses and related methods, Appl. Math. Comput. 103 (1999) 51-68. [16] H.X. Wang, Core-EP decomposition and its applications, Linear Algebra 150 Appl. 508 (2016) 289-300. [17] H.X. Wang, X.J. Liu, Characterizations of the core inverse and the core partial ordering, Linear Multilinear Algebra 63 (2015) 1829-1836. [18] Y.M. Wei, A characterization and representation of the generalized inverse A(2) T,S and its applications, Linear Algebra Appl. 280 (1998) 87-96. 155 [19] S.Z. Xu, J.L. Chen, X.X. Zhang, New characterizations for core inverses in rings with involution, Front. Math. China 12 (2017) 231-246. 13
1503.03958	1	1503	2015-03-13T05:30:32	On subalgebras of an evolution algebra of a "chicken" population	[ "math.RA" ]	We consider an evolution algebra which corresponds to a bisexual population with a set of females partitioned into finitely many different types and the males having only one type. For such algebras in terms of its structure constants we calculate right and plenary periods of generator elements. Some results on subalgebras of EACP and ideals on low-dimensional EACP are obtained.	math.RA	math	ON SUBALGEBRAS OF AN EVOLUTION ALGEBRA OF A "CHICKEN" POPULATION B.A. OMIROV, U. A. ROZIKOV Abstract. We consider an evolution algebra which corresponds to a bisexual popula- tion with a set of females partitioned into finitely many different types and the males having only one type. For such algebras in terms of its structure constants we calculate right and plenary periods of generator elements. Some results on subalgebras of EACP and ideals on low-dimensional EACP are obtained. Key words. Evolution algebra; bisexual population; associative algebra; subalgebra. Mathematics Subject Classifications (2010). 17D92; 17D99; 60J10. 1. Introduction In recent years the non-commutative and non-associative analogies of the classical con- structions become interesting in the connection with their applications in many branches of mathematics, biology (population, genetics, etc.) and physics (string theory, quantum field theory, etc.). An algebraic approach in Genetics consists of the study of various types of genetic algebras (like algebras of free, "self-reproduction" and bisexual populations, Bernstein algebras) [4], [6]. Mendel exploited symbols that are quite algebraically suggestive to express his genetic laws. The evolution of a population comprises a determined change of state in the next generations as a result of reproduction and selection [6], [7]. The main problem for a given algebra of a sex linked population is to carefully examine how the basic algebraic model must be altered in order to compensate for this lack of symmetry in the genetic inheritance system. In [2] Etherington began the study of this kind of algebras with the simplest possible case. Recently in [4] an evolution algebra B is introduced identifying the coefficients of inheritance of a bisexual population as the structure constants of the algebra. The basic properties of the algebra are studied. Moreover a detailed analysis of a special case of the evolution algebra (of bisexual population in which type "1" of females and males have preference) is given. Since the structural constants of the algebra B are given by two cubic matrices, the study of this algebra is quite difficult. To avoid such difficulties one has to consider an algebra of bisexual population with a simplified form of matrices of structural constants. In [5] a such simplified model of bisexual population is considered and basic properties of corresponding evolution algebra (called evolution algebras of a "chicken" population (CEACP)) are studied. In [8] a notion of chain of 1 2 B.A. OMIROV, U. A. ROZIKOV EACP is introduced and several examples (time homogenous, time non-homogenous, periodic, etc.) of such chains are given. In this paper we calculate right and plenary periods for generator elements of EACP and establish that natural basis of any subalgebra of EACP (which is also a EACP) can be extended to a natural basis of whole algebra. Moreover, we describe one-dimensional subalgebras (in ordinary sense) of EACP. Finally, simplicity of low-dimensional EACP is investigated. 2. Basic definitions Following [5] we consider a set {hi, i = 1, . . . , n} (the set of "hen"s) and r (a "rooster"). Definition 1. [5] Let (E,·) be an algebra over a field K. If it admits a basis {h1, . . . , hn, r}, such that hir = rhi =Pn then this algebra is called an evolution algebra of a "chicken" population (EACP). We call the basis {h1, . . . , hn, r} a natural basis. i, j = 1, . . . , n; rr = 0 j=1 aijhj + bir, hihj = 0, (2.1) Thus an algebra EACP, E, is defined by a rectangular n × (n + 1)-matrix M = a11 a12 a21 a22 ... ... an1 an2 . . . . . . . . . . . . a1n a2n ... ann b1 b2 ... bn   ,   which is called the matrix of structural constants of the algebra E. (b1, . . . , bn). We write the matrix M in the form M = A ⊕ b where A = (aij)i,j=1,...,n and bT = Assume we have two rectangular n × (n + 1)-matrices M = A ⊕ b and H = B ⊕ c. Then we define multiplication of such matrices by (2.2) We note that this multiplication agrees with usual multiplication of (n + 1) × (n + 1)- matrices with zero (n + 1)-th row. M H = AB ⊕ Ac, HM = BA ⊕ Bb. Let E be a commutative algebra, define principal power of a ∈ E as a2 = a · a, a3 = a2 · a, . . . , an = an−1 · a; and plenary powers of a as Define right multiplication operator by a[1] = a · a, a[n] = a[n−1]a[n−1], n ≥ 2. Ra(x) = xa. EVOLUTION ALGEBRA OF A "CHICKEN" POPULATION 3 Let E be an EACP with the generator set {h1, h2, . . . , hn, r}. We say hi (or r) occurs in x ∈ E if the coefficient αi (or a) in x = Pn i=1 αihi + ar is non-zero. Write hi ≺ x (r ≺ x). Definition 2. Let hj be a generator of an EACP, the right period pj of hj is defined by pj = min{m ∈ N : hj ≺ Rm r (hj)}. If pj = 1, we say hj is aperiodic; if the set is empty we define pj = ∞. Definition 3. Let hj be a generator of an EACP, the plenary period qj of hj is defined by If qj = 1, we say hj is aperiodic; if the set is empty we define qj = ∞. qj = min{m ∈ N : hj ≺ (hj r)[m]}. 3. Conditions of periodicity (i) Rm r (hi) = (Amh)i + (Am−1b)ir; Proposition 1. For any m ≥ 1 and for any i = 1, . . . , n the following identities hold (ii) (hir)[m] = γm(cid:2)(Am+1h)i + (Amb)ir(cid:3) , where h = {h1, . . . , hn} and γm satisfies the recurrent equation: γm+1 = 2γ2 m(Amb)i, with γ1 = 2bi. Proof. (i) Compute actions of Rr to the set h: Rr(h) = {Rr(h1), . . . , Rr(hn)} = {h1r, . . . , hnr} = {(M h)1, . . . , (M h)n}, where Also we have (M h)i = n Xj=1 aijhj + bir = (Ah)i + bir, i = 1, . . . , n. n n n R2 r(hi) = Rr((M h)i) = aijajshs + aijbjr = (A2h)i + (Ab)ir. Xs=1 Xj=1 Xj=1 Using induction by m we get Rm r (h) = {(M mh)1, . . . , (M mh)n}, r (hi) = (Amh)i + (Am−1b)ir, (M mh)i = Rm where (ii) Use induction by m ≥ 1. For m = 1 we have i = 1, . . . , n. (hir)[1] =  n Xj=1 2 aijhj + bir  = 2bi(cid:2)(A2h)i + (Ab)ir(cid:3) . Assume now that the formula (ii) is true for m and prove it for m + 1: (hir)[m+1] =(cid:0)γm(cid:2)(Am+1h)i + (Amb)ir(cid:3)(cid:1)2 = 2γ2 m(Amb)i((Am+1h)ir). (3.1) 4 B.A. OMIROV, U. A. ROZIKOV Let Am = (a(m) ij )ij=1,...,n. Then from (3.1) we get n (hir)[m+1] = 2γ2 m(Amb)i( Xj=1 a(m+1) ij hjr) = γm+1(cid:2)(Am+2h)i + (Am+1b)ir(cid:3) . (3.2) (cid:3) As a corollary of this proposition we have Proposition 2. 1) The right period of hi is pi = min{m ∈ N : a(m) ii 6= 0}. 2) If bi = 0 then qi = ∞, otherwise the plenary period of hi is (Ajb)i 6= 0}, qi = min{m ∈ N : a(m+1) m−1 ii Yj=0 where A0 = id. Proof. 1) This simply follows from the part (i) of Proposition 1. 2) Using part (ii) of Proposition 1 we get (hir)[m] = 22m−1 Thus the coefficient of hi is m−1 (Ajb)2m−j−1 i Yj=0 (cid:2)(Am+1h)i + (Amb)ir(cid:3) . 22m −1 m−1 Yj=0 (Ajb)2m−j−1 i a(m+1) ii . This completes the proof. The following proposition reduces an EACP to a simple one. (cid:3) Proposition 3. that C on this basis is represented by the table of multiplication as follows [1] Let C be an EACP, then there exists a basis {h1, h2, . . . , hn, r} such n n h1r = Xj=1 a1jhj + δr, δ ∈ {0, 1}, hir = aijhj, 2 ≤ i ≤ n. Xj=1 Using this proposition by Proposition 2 we get Corollary 1. For EACP mentioned in Proposition 3 the following hold a) If δ = 0 then qi = 1 or ∞, b) If δ = 1 then the plenary period of hi is qi ∈( {1, 2,∞} if if {1,∞}, i = 1 i 6= 1. EVOLUTION ALGEBRA OF A "CHICKEN" POPULATION 5 Proof. a) If hi is present in hir then qi = 1, otherwise since (hir)[m] = 0 for all m ≥ 2 we get qi = ∞. b) Case i = 1. If h1 is present in h1r then q1 = 1, otherwise consider (h1r)[2] if this contains h1 then q1 = 2, if (h1r)[2] does not contain h1 then since (h1r)[m] = 0 for all m ≥ 3 we get qi = ∞. (cid:3) Case i 6= 1 is similar to part a). By definition of an EACP we know that this algebra depends on a natural basis 4. Subalgebras of an EACP {h1, h2, . . . , hn, r}. [5] Definition 4. ideal of a CP. 1) Let C be an EACP and C1 be a subspace of C. If C1 has a natural basis {h′ 2, . . . , h′ with multiplication table like (2.1), then we call C1 an evolution subalgebra of a CP. 2) Let I ⊂ C be an evolution subalgebra of a CP. If CI ⊆ I, we call I an evolution 3) Let C and D be EACPs, we say a linear homomorphism f from C to D is an evolution homomorphism, if f is an algebraic map and for a natural basis {h1, . . . , hn, r} of C, {f (h1), . . . , f (hn), f (r)} spans an evolution subalgebra of a CP in D. Furthermore, if an evolution homomorphism is one to one and onto, it is an evolution isomorphism. 1, h′ m, r′} 4) An EACP C is called simple if it has no proper evolution ideals. 5) C is called irreducible if it has no proper subalgebras. In fact, for linear subspace C1 of an EACP C we can consider three type of subalgebras: (i) C1 is a subalgebra in ordinary sense; (ii) C1 is subalgebra and there exists a natural basis of C1; (iii) C1 is subalgebra and there exist a natural basis of C1 which can be extended to a Note that Definition 4 agrees with the second type of subalgebra. The following proposition gives equivalence of (ii) and (iii). natural basis of C. Proposition 4. Definitions (ii) and (iii) are equivalent. Proof. Part (iii) ⇒ (ii) is straightforward. We shall prove (ii) ⇒ (iii). Let C1 = {f1, f2, . . . , fm, r′} be a subalgebra of C = {h1, . . . , hn, r} in sense (ii). We shall show that the natural basis of C1 can be extended to a natural basis of C. We have j=1 αijhj + γir, j=1 βjhj + γr. i = 1, . . . , m, (4.1) fi =Pn r′ =Pn f ′ i = fi − Case γ 6= 0. Take the following change of the basis 1 ≤ i ≤ m, γi γ r′, r′′ = r′. 6 B.A. OMIROV, U. A. ROZIKOV 1, . . . , f ′ This new basis also is a natural basis, moreover the vectors f ′ i do not contain r in their decompositions. Thus vectors {f ′ m} generate a subspace in the vector space generated by {h1, . . . , hn}. Then using theorem about change of basis (see [10]) we can m}. Moreover r can be replaced by r′. Hence for replace {hi1, . . . , him} by {f ′ γ 6= 0 we can extend the natural basis of C1 to the natural basis of C. Case γ = 0 and γi = 0 for all i. In this case all fi and r′ do not depend on r. So we can again use theorem about change of basis and replace {hi1 , . . . , him , him+1} by {f1, . . . , fm, r′}. Case γ = 0 and γi 6= 0 for some i. By change Xi = r′; Xj = fj, j 6= i; r′′ = fi we reduce this case to the first case. This completes the proof. 1, . . . , f ′ The following is an example of a subalgebra (as in (i)) of C, which is not an evolution subalgebra of a CP (as in (ii)). (cid:3) [5] Let C be an EACP over a field K with basis {h1, h2, h3, r} and multi- Example 1. plication defined by hir = hi + r, i = 1, 2, 3. Take u1 = h1 + r, u2 = h2 + r. Then (au1 +bu2)(cu1 +du2) = acu2 2 = (2ac+ad+bc)u1 +(2bd+ad+bc)u2. Hence, F = Ku1 + Ku2 is a subalgebra of C, but it is not an evolution subalgebra of a CP. Indeed, assume v1, v2 be a basis of F . Then v1 = au1 + bu2 and v2 = cu1 + du2 for some 1 = (2a2 + 2ab)u1 + (2b2 + 2ab)u2 a, b, c, d ∈ K such that D = ad − bc 6= 0. We have v2 and v2 2 = (2c2 + 2cd)u1 + (2d2 + 2cd)u2. We must have v2 1 +(ad+bc)u1u2 +bdu2 1 = v2 2 = 0, i.e. a2 + ab = 0, b2 + ab = 0, c2 + cd = 0, d2 + cd = 0. From this we get a = −b and c = −d. Then D = 0, a contradiction. If a = 0 then b = 0 (resp. c = 0 then d = 0), we reach the same contradiction. Hence v2 1 6= 0 and v2 2 6= 0, and consequently F is not an evolution subalgebra of a CP. In sequel for a subalgebra we mean a subalgebra in the sense (iii). Proposition 5. Let C be an EACP over R with basis {h1, . . . , hn, r} and matrix of structural constants M = A ⊕ b. If rankA = n, then any subalgebra of C has the form {f1, . . . , fm, ar}, where 0 ≤ m ≤ n, a ∈ {0, 1} and n fi = αijhj, αij ∈ R, i = 1, . . . , m. Xj=1 n Proof. Let C = {ϕ1, . . . , ϕm} be a subalgebra of C. Then we have ϕi = Since ϕ2 i = 0, then we have βikhk + βir, i = 1, . . . , m. Xk=1 2βi n Xk=1 βikhkr = 2βi n Xk=1 n Xs=1 βikakshs + βikbkr! = 0. n Xk=1 (4.2) EVOLUTION ALGEBRA OF A "CHICKEN" POPULATION 7 Hence βi = 0 or βikaks = 0 for any s and n Xk=1 n Xk=1 βikbk = 0. (4.3) k=1 βikhk. This completes the proof. Since rankA = n from (4.3) we get βik = 0 for all k. Hence ϕi is equal to βir or to (cid:3) Pn Proposition 6. Let C be an EACP with matrix of structural constants M = A ⊕ b. Then X = hxi, where 0 6= x = y + βr = Pn i=1 αihi + βr generates an one-dimensional subalgebra if one of the following conditions is satisfied a. β = 0 or Ay = 0, by = 0. b. β 6= 0, by = 1 and y is an eigenvector of A with eigenvalue 1/β. Proof. An arbitrary x = Pn i=1 αihi + βr generates a subalgebra iff x2 = cx for some c. Here one can consider only the case c = 0 and c = 1. Thus x generates a subalgebra iff it is an absolute nilpotent or idempotent of C. Now the proof follows from Propositions 3.4 and 3.5 of [5]. (cid:3) Proposition 7. Let C be an EACP as in Proposition 3, δ = 1 and with matrix of structural constants M = A⊕ b. Then X = hxi, where x =Pn i=1 αihi + βr generates an one-dimensional ideal iff one of the following conditions is satisfied i=2 ai1αi = 0 and x (with α1 = 0) is an eigenvector of A1 with a real eigenvalue, where A1 = (aij)i,j=2,...,n is the minor of the matrix A. b. β = 1 and αj = a1j and akj = 0, for all k = 2, . . . , n, j = 1, . . . , n. a. β = α1 =Pn Proof. Take an arbitrary element y = Pn there exists c such that xy = cx. The last equality is equivalent to i=1 γihi + νr ∈ C we should have xy ∈ X , i.e. i=1(ναi + βγi)aij = cαj; j = 1, 2, . . . , n (4.4) ( Pn να1 + βγ1 = cβ. a. For case β = 0 if ν = 0 then in (4.4) one can take c = 0. If ν 6= 0 then α1 = 0 and i=2 αiaij = cαj; j = 2, . . . , n νPn Pn i=2 αiai1 = 0. This completes the proof of a. b. In the case β 6= 0 one can take β = 1. For y = hk, k = 2, . . . , n from (4.4) for some c = ck we get the system akj = ckαj, j = 1, . . . , n and ck = 0. This implies akj = 0 for all k = 2, . . . , n and j = 1, . . . , n. In case y = h1 we get the system a1j = c1αj, j = 1, . . . , n and c1 = 1. Hence a1j = αj. Taking into account the above obtained results, for y = r 8 B.A. OMIROV, U. A. ROZIKOV we get α1a1j = cαj and α1 = c. Thus we proved that if A has the following form A = α1 α2 . . . αn 0 ... 0 0 ... 0 . . . ... . . . 0 ... 0     then there are ck and c such that xy = ckx if y = hk and xy = cx if y = r. Using this result for an arbitrary y =Pn i=1 γihi + νr ∈ C we obtain Xi=1 γixhi + νxr = ( Xi=1 xy = n n γici + c)x = Cx. Thus X = hx =Pn i=1 αihi + ri is an ideal of the algebra C with matrix M = α1 α2 . . . αn 1 0 ... 0 0 ... 0 . . . ... . . . 0 ... 0 0 ... 0     (cid:3) 5. Simple three-dimensional complex EACPs In the following theorem the classification of three dimensional EACP is presented. Theorem 1. 1. [5] Any 2-dimensional, non-trivial EACP C is isomorphic to one rh = hr = h, h2 = r2 = 0, rh = hr = 1 of the following pairwise non isomorphic algebras: C1: C2: [1] An arbitrary three dimensional complex EACP C is isomorphic to one of the following pairwise non-isomorphic algebras If dim C2 = 1 then 2 (h + r), h2 = r2 = 0. 2. C1 : h1r = 1 2 r C2 : h1r = 1 2 h2; 2 h1 + 1 C3 : h1r = 1 2 r. EVOLUTION ALGEBRA OF A "CHICKEN" POPULATION 9 If dim C2 = 2 then h1r = 1 C4 : h1r = 1 C5(β) : C6(α, β) : h1r = 1 h1r = 1 C7(α) : h1r = 1 C8 : h2r = 1 2 (h1 + h2), h2r = β 2 h1, 2 (αh1 + βh2 + r), h2r = 1 h2r = 1 2 (αh1 + r), h2r = 1 2 (h1 + h2 + r), 2 h2; 2 h2, β 6= 0; 2 h1; 2 h2; 2 h2. where one of non-zero parameter α, β in the algebra C6(α, β) can be assumed to be equal to 1. The following theorem describes simple and not simple EACP listed in Theorem 1. Theorem 2. a. The two-dimensional algebra C1 and C2 are not simple. b. The three-dimensional algebra Ci is not simple for i = 1, 2, 3, 4, 5, 7, 8 and i = 6 for β = 0. Moreover, C6(α, β) is simple for β 6= 0. Proof. a. It is easy to see that hhi ⊳ C1 and hh + ri ⊳ C2. b. Consider some possible subalgebras (in sense (iii)) of C = {h1, h2, r}: D1 = {h1}, D2 = {h1, h2}, D3 = {h1, r}, D4 = {h2}, D5 = {h2, r}, D6 = {r}. It is easy to check that if j = 1, 2; if j = 1, 6; is ideal f or C1 if j = 3, 4, 5, 6 and is not ideal is ideal f or C2 if j = 2, 4, 5 and is not ideal is ideal f or C3 if j = 3, 4 and is not ideal is ideal f or C4 if j = 2, 4, 5 and is not ideal is ideal f or C5 if j = 1, 2, 4, and is not ideal is not ideal f or C6 if j = 1, 2, 4, 6; is ideal f or C7 if j = 4 and is not ideal if j = 1, 2, 3, 5, 6; is ideal f or C8 if j = 4 and is not ideal f or j = 1, 2, 5, 6. if j = 1, 2, 5, 6; if j = 1, 6; if j = 3, 5, 6; Dj =   Now consider C6: Case β = 0. In this case D3 will be an ideal, i.e. C6(α, 0) is not simple. Case β 6= 0. This β can be reduced to β = 1. We have rankA = 2. So we can use Proposition 6: consider a general subalgebra C6 = {ah1 + bh2, δr}. For δ = 0 it is easy to see that C6C6 6⊂ C6. If δ = 1 then C6C6 = {(aα + b)h1 + ah2 + ar, αh1 + h2 + r, h1}. 2 · (α ∓√α2 + 4). For Simple calculations show that (aα + b)h1 + ah2 + ar ∈ C6 iff b = − a this value of b one gets αh1 + h2 + r ∈ C6 iff α√α2 + 4 = α2 + 2. But the last equation has not solution. Hence C6(α, β) is simple for any β 6= 0. (cid:3) 10 B.A. OMIROV, U. A. ROZIKOV Acknowledgements The work supported by the Grant No.0251/GF3 of Education and Science Ministry of Republic of Kazakhstan. U.Rozikov thanks Aix-Marseille University Institute for Advanced Study IM´eRA (Marseille, France) for support by a residency scheme. References [1] A. Dzhumadil'daev, B.A. Omirov, U.A. Rozikov, On a class of evolution algebras of "chicken" population. Inter. Jour. Math. 25(8) (2014) 1450073 (19 pages). [2] I.M.H. Etherington, Non-associative algebra and the symbolism of genetics, Proc. Roy. Soc. Edin- burgh, 61 (1941) 24 -- 42. [3] R.N. Ganikhodzhaev, F.M. Mukhamedov, U.A. Rozikov, Quadratic stochastic operators and pro- cesses: results and open problems. Inf. Dim. Anal. Quant. Prob. Rel. Fields., 14(2) (2011), 279 -- 335. [4] M. Ladra, U.A. Rozikov, Evolution algebra of a bisexual population, J. Algebra. 378 (2013) 153 -- 172. [5] A. Labra, M.Ladra, U.A. Rozikov, An evolution algebra in population genetics. Linear Algebra Appl. 457 (2014) 348 -- 362. [6] Y.I. Lyubich, Mathematical structures in population genetics, Springer-Verlag, Berlin, 1992. [7] M.L. Reed, Algebraic structure of genetic inheritance, Bull. Amer. Math. Soc. (N.S.) 34 (2) (1997) 107 -- 130. [8] U.A. Rozikov, Sh.N. Murodov, Chain of evolution algebras of "chicken" population. Linear Algebra Appl. 450 (2014) 186 -- 201. [9] J.P. Tian, Evolution algebras and their applications, Lecture Notes in Mathematics, 1921, Springer- Verlag, Berlin, 2008. [10] B.L.van der Waerden. Algebra. Springer-Verlag, 1971. B. A. Omirov and U. A. Rozikov, Institute of mathematics, 29, Do'rmon Yo'li str., 100125, Tashkent, Uzbekistan. E-mail address: [email protected], [email protected]
1805.04450	1	1805	2018-05-11T15:31:50	Representability of affine algebras over an arbitrary field	[ "math.RA" ]	In a series of papers, we used full quivers as tools in describing PI-varieties of algebras and providing a complete proof of Belov's solution of Specht's problem for affine algebras over an arbitrary Noetherian ring. In this paper, utilizing ideas from that work, we give a full exposition of Belov's theorem that relatively free affine PI-algebras over an arbitrary field are representable. (Kemer proved the theorem over an infinite field.)	math.RA	math	REPRESENTABILITY OF AFFINE ALGEBRAS OVER AN ARBITRARY FIELD ALEXEI BELOV-KANEL, LOUIS ROWEN, AND UZI VISHNE Abstract. In a series of papers culminating in [16], summarized in [17], we used full quivers as tools in describing PI-varieties of algebras and providing a complete proof of Belov's solution of Specht's problem for affine algebras over an arbitrary Noetherian ring. In this paper, utilizing ideas from that work, we give a full exposition of Belov's theorem [6] that relatively free affine PI-algebras over an arbitrary field are representable. 8 1 0 2 y a M 1 1 ] . A R h t a m [ 1 v 0 5 4 4 0 . 5 0 8 1 : v i X r a Contents Introduction 1. 1.1. Representability 2. Preliminaries 2.1. Linearization and quasi-linearization 2.2. Full quivers 3. The Canonization Theorem for Polynomials 4. The proof of the Canonization Theorem for Polynomials 4.1. Unmixed case 4.2. The mixed case: the hiking procedure 5. Details of hiking 5.1. Preliminary hiking 5.2. First stage of hiking 5.3. Second stage of hiking 6. Characteristic coefficient-absorbing polynomials inside T-ideals 7. Resolving ambiguities for nonhomogeneous polynomials 7.1. Removing ambiguity of matrix degree for nonhomogeneous polynomials 7.2. Removing ambiguities for nonhomogeneous polynomials having molecules of the same matrix degree 8. Application of Shirshov's theorem 9. Conclusion of the proof of Theorem 1.1 References 2 2 4 4 5 7 9 9 9 10 11 11 12 14 16 16 17 17 18 18 Date: May 14, 2018. 2010 Mathematics Subject Classification. Primary: 16R10 Secondary: 16R30, 17A01, 17B01, 17C05. Key words and phrases. Kemer's representability theorem, Specht's question, hiking, Shirshov's Theorem, polynomial identities, T-ideal, affine algebra, relatively free, representable. This research was supported by the Israel Science Foundation, (grant No. 1178/06). 1 2 ALEXEI BELOV-KANEL, LOUIS ROWEN, AND UZI VISHNE 1. Introduction In this paper, utilizing ideas from [16], we give a full exposition of Belov's theorem [6] that relatively free affine PI-algebras over an arbitrary field are representable. As in [16], the main tool in utilizing the combinatorics of polynomials is "hiking," which however is more complicated here since it involves non-homogeneous polynomials, and is described below in several stages. We work with algebras over a field F , with special emphasis to the possibility that F is finite. A (noncommutative) polynomial is an element of the free associative algebra F {x} on countably many generators. A polynomial identity (PI) of an algebra A over F is a noncommutative polynomial which vanishes identically for any substitution in A. We use [10] as a general reference for PIs. A T-ideal of F {x} is an ideal I of F {x} closed under all algebra endomorphisms F {x} → F {x}. We write id(A) for the T-ideal of PIs of an algebra A. Conversely, for any T-ideal I of F {x}, each element of I is a PI of the algebra F {x}/I, and F {x}/I is relatively free, in the sense that for any PI-algebra A with id(A) ⊇ I, and any a1, a2, . . . ∈ A, there is a natural homomorphism F {x}/I → A sending xi 7→ ai for i = 1, 2, . . . . 1.1. Representability. An F -algebra A is called representable if it is embeddable as an F -subalgebra of Mn(K) for a suitable field K ⊇ F . Obviously any representable algebra is PI, but an easy counting argument of Lewin [30] leads to the existence of non-representable affine PI-algebras. Nevertheless, the representability question of relatively free affine algebras has con- siderable independent interest, and the purpose of this paper is to give a full, self- contained proof of the following result proved by Kemer over any infinite field, and to elaborate Belov [6] (over a finite field): Theorem 1.1. Every relatively free affine PI-algebra over an arbitrary field is repre- sentable. Kemer obtained Theorem 1.1 over infinite fields by means of the following amazing results: Theorem 1.2 ([10, Theorem 4.66], [27]). (1) Every affine PI-algebra over an infinite field (of arbitrary characteristic) is PI- equivalent to a finite dimensional (f.d.) algebra. (2) Every PI-algebra of characteristic 0 is PI-equivalent to the Grassmann envelope of a finite dimensional (f.d.) algebra. In [12]–[16] we have provided a complete proof for the affine case of Specht's problem in arbitrary positive characteristic. (The non-affine case has counterexamples, cf. [4, 5].) Together with Kemer's solution in characteristic 0, this leads to: Theorem 1.3. Any affine PI-algebra over an arbitrary commutative Noetherian ring satisfies the ACC on T-ideals. Remark 1.4. For graded associative algebras [1] and various nonassociative affine algebras of characteristic 0, the finite basis of T-ideals has been established in the case when the operator algebra is PI (Iltyakov [24, 25] for alternative and Lie algebras, and REPRESENTABILITY OF AFFINE ALGEBRAS OVER AN ARBITRARY FIELD (VER. VI:34.0) 3 Vais and Zelmanov [34] for Jordan algebras) but the representability question remains open for nonassociative algebras, so representability presumably is more difficult. The obstacle is getting started via some analog of Lewin's theorem [30], which is not yet available. Belov [9] proved representability of alternative or Jordan algebras satisfying all identities of some finite dimensional algebra. 1.1.1. Plan of the proof of Theorem 1.1. An immediate consequence of Theorem 1.2(1) is that every relatively free affine PI- algebra over an infinite field is representable, since it can be constructed with generic elements obtained by adjoining commutative indeterminates to the f.d. algebra. Kemer deduced from this the finite basis of T-ideals (the solution of Specht's problem) over an infinite field, in which he applied combinatorial techniques to representable algebras. The approach here for positive characteristic, following [6], is the reverse, where one starts with the solution of Specht's problem and applies Noetherian induction to prove representability of affine relatively free PI-algebras over arbitrary fields (including finite fields). (These methods also work in characteristic 0, but rely on Kemer's solution of Specht's problem in characteristic 0, which in turn relies on his representability theorem in characteristic 0.) Remark 1.5. We fix the following notation: We start with the free affine algebra F {x} = F {x1, . . . , xℓ} in ℓ indeterminates, and a T-ideal I. This gives us the relatively free algebra A = F {x}/I. We say that the T-ideal I is representable if the affine algebra A is representable. The proof of Theorem 1.1 goes along the following version of Noetherian induction: We aim to show that every T-ideal I is representable. In view of Lewin's theorem [30], I contains a representable T-ideal I0, so we assume that A0 := F {x}/I0 is repre- sentable. In view of Theorem 1.3, we have a maximal representable T-ideal I1 ⊇ I0 of A contained in I, which we aim to show is I. Assuming on the contrary that I1 ⊂ I, we replace I0 by I1 and A0 by A0/I1. This reduces us to the case where A0 is rep- resentable but every nonzero T-ideal of A0 contained in I is not representable. Our goal is to arrive at a contradiction by finding a representable T-ideal J ⊆ I which strictly contains I1. We will do this by taking some f ∈ I \ I1, i.e., f /∈ id(A0), and finding J inside the T-ideal generated by f . (This process will terminate because of the solution to Specht's problem given in [16]. For this reason, we do not need to introduce parameters of the induction.) The rest of this paper consists of the proof of Theorem 1.1 by means of Remark 1.5. The proof relies on the ideas of the proof of Kemer's representability theorem given in [10, 11]. Much of this paper is devoted to elaborating the theory of [13] and [14], as described in [16, 17], and there is a considerable overlap with [6]. The proof of Theorem 1.3 in [16] is somewhat different from Kemer's proof. In [13] we considered the full quiver of a representation of an associative algebra over a field, and determined properties of full quivers by means of a close examination of the structure of Zariski closed algebras, studied in [12]. Then we modified f by means of a "hiking procedure" in order to force f to have certain combinatorial properties, and used this to carve out a T-ideal J from inside a given T-ideal; modding out J lowers the quiver in some sense, and then one obtains Specht's conjecture by induction. 4 ALEXEI BELOV-KANEL, LOUIS ROWEN, AND UZI VISHNE Our approach here is similar, but with some variation. Here we need not mod out by J , but need J to be representable. We start the same way, but one of the key steps fails, and we need a way of getting around it. In both instances, our techniques rely on the theory of full quivers of a f.d. algebra over a field, to be recalled, followed by adjunction of characteristic coordinates. We introduce "critical" polynomials (Definition 4.3) which enable us to calculate characteristic coefficients using the combinatoric properties of polynomials, leading to our main tool: Theorem 1.6 (Canonization Theorem for Polynomials). Suppose f (x1, . . . , xt) is a nonidentity of A0 whose nonzero evaluation passes through all the blocks of the quiver, via the dominant branch. Then the T-ideal of f contains a critical nonidentity. The proof of the Canonization Theorem for Polynomials is based on applying hiking to obtain more and more complicated polynomials while preserving the two "Kemer invariants" of the polynomial described in [10, 11], which underly the computational study of T-ideals. The basic operations of hiking, namely multiplying by a Capelli polynomial or replacing a radical element by a commutator element of the same form preserves the hypotheses of [11, Lemma 6.7.3], so we can measure the dimension of the semisimple part and the nilpotence index of the radical in terms of the Kemer invariants. The Canonization Theorem for Polynomials will enable us to replace multiplication by characteristic coefficients in the Shirshov extension, with multiplication by elements of A0. Let us review some of the techniques we need for the proof. The reader can refer to 2. Preliminaries [17] for further details of all of this material. 2.1. Linearization and quasi-linearization. The well-known linearization process of a polynomial can be described in two stages: First, writing a polynomial f (x1, . . . , xn) as f (0, x2, . . . , xn) + (f (x1, . . . , xn) − f (0, x2, . . . , xn)), one sees by iteration that any T-ideal is additively spanned by T-ideals of polynomials for which each indeterminate appearing nontrivially appears in each of its monomials, cf. [31, Exercise 2.3.7]. Then we define the linearization process by introducing a new indeterminate x′ i and passing to f (x1, . . . , xi + x′ i, . . . , xm) − f (x1, . . . , xi, . . . , xm) − f (x1, . . . , x′ i, . . . , xm). This process, applied repeatedly, yields a multilinear polynomial in the same T-ideal. In characteristic 0 the linearization process can be reversed by taking x′ i = xi, implying that every T-ideal is generated by multilinear polynomials. But this fails in positive characteristic, as exemplified by the Boolean identity x2 − x, so we need an alternative. To handle characteristic p > 0, Kemer [28] took a closer look, which we review from [17]. Definition 2.1. A function f is i-quasi-linear on A if f (. . . , ai + a′ i, . . . ) = f (. . . , ai, . . . ) + f (. . . , a′ i, . . . ) REPRESENTABILITY OF AFFINE ALGEBRAS OVER AN ARBITRARY FIELD (VER. VI:34.0) 5 for all ai, a′ understood, we just say quasi-linear. i ∈ A; f is A-quasi-linear if f is i-quasi-linear on A for all i. When A is Suppose f (x1, x2, . . . ) ∈ F {x} has degree di in xi. The i-partial linearization of f is ∆if := f (x1, x2, . . . , xi,1 + · · · + xi,di, . . . ) − di Xj=1 f (x1, x2, . . . , xi,j, . . . ) (1) where the substitutions were made in the i component, and xi,1, . . . , xi,di are new variables. When ∆if (A) = 0, then f is i-quasi-linear on A, so we apply (1) at most degi f times repeatedly, if necessary, to each xi in turn, to obtain a nonzero polynomial that is A-quasi-linear in the T-ideal of f . Proposition 2.2 ([15, Corollary 2.13]). Assume char F = p > 0. For any polynomial f which is not an identity of A0, the T-ideal generated by f contains a quasi-linear non- identity for which the degree in each indeterminate is a p-power. 2.2. Full quivers. In this subsection A0 is a representable affine algebra over a field F , i.e., A0 ⊂ Mn(K) with K finite or algebraically closed, and we fix this particular representation. The closure of A0 in Mn(K) with respect to the Zariski topology [17, § 3.1] is PI-equivalent to A0, so we assume throughout that A0 is Zariski closed. In particular, when F is infinite then we may assume F = K, cf. [17, Remark 3.1]. By Wedderburn's Principal Theorem [32, Theorem 2.5.37], A0 = S ⊕J as vector spaces, where J is the radical of A0 and S ∼= A0/J is a semisimple subalgebra of A0. Thus S is a direct product of matrix algebras R1 × · · · × Rk, which we want to view along the diagonal of Mn(K), although perhaps with identification of coordinates, which are to be described graphically. By the Braun-Kemer-Razmyslov theorem, cf. [20], J is nilpotent, so we take t = tA0 maximal such that J t 6= 0. We need an explicit description, but which may distinguish Morita equivalent al- gebras since matrix algebras of different size are not PI-equivalent. The full quiver of A0 is a directed graph Γ, having neither loops, double edges, nor cycles, with the following information attached to the vertices and edges: The vertices of the full quiver of A0 correspond to the diagonal matrix blocks arising in the semisimple part S, whereas the arrows come from the radical J. Every vertex likewise corresponds to a central idempotent in a corresponding matrix block of Mn(K). • The vertices are ordered, say from 1 to k, and an edge always takes a vertex to a vertex of higher order. There are identifications of vertices, called diagonal gluing, and identification of edges, called off-diagonal gluing. Gluing of vertices in full quivers is identical or Frobenius, as in (cid:26)(cid:18)α 0 0 αq(cid:19) : α ∈ K(cid:27) where F = q and K = F . • Each vertex is labeled with a roman numeral (I, II etc.); glued vertices are labeled with the same roman numeral. A vertex can be either filled or empty. The first vertex listed in a glued matrix block is also given a pair of sub- scripts - the matrix degree ni and the cardinality of the corresponding field extension of F (which, when finite, is denoted as a power qti of q = F ). 6 ALEXEI BELOV-KANEL, LOUIS ROWEN, AND UZI VISHNE • When the base field F is finite, superscripts indicate the Frobenius twist between glued vertices, induced by the Frobenius automorphism a 7→ aq; this could identify aq1 with aq2 for powers q1, q2 of q (or equivalently a with aq2/q1 when q1 < q2); we call this (q1, q2)-Frobenius gluing. • Off-diagonal gluing (i.e., gluing among the edges) includes Frobenius gluing (which only exists in nonzero characteristic) and proportional gluing with an accompanying scaling factor ν. Examples are given in [14]. Now we take some non-identity of A0, say f (x1, . . . , xm) = P gj(x1, . . . , xm) ∈ I for monomials gj. An easy technical condition: Since the full quiver Γ of A0 could be replaced by the subquiver corresponding to the algebra gener- ated by evaluations of all polynomials in I, and then f could be replaced by a sum of polynomials in I, we may assume that Γ passes through all blocks. The numbers dimF A0 and t are crucial to the description of quivers, so we want these numbers to be reflected in the polynomial f . This is achieved by means of Kemer's First Lemma ([11, Proposition 6.5.2]) and Kemer's Second Lemma ([11, Proposition 6.6.31]). On the other hand, we need f to be full on the f.d. algebra A0 in the sense that some nonzero evaluation of f passes through all the blocks of the quiver, via the dominant branch B. This is achieved by means of Lemma [11, Proposition 6.7.3], called the Phoenix property. Applying these results to f after hiking (to be described below), we assume throughout that f is full, and that the conclusion of Kemer's First Lemma and Kemer's Second Lemma hold. In view of Proposition 2.2 we may assume that f is quasi-linear. When specializing xi to A0, we write the substitutions ¯xi as sums of radical and semisimple elements; since f (x1, . . . , xm) is quasi-linear, we reduce the substitutions in S + J to their component parts in S ∪ J; we call these substitutions pure. Thus f has a nonzero specializa- tion where all substitutions ¯xi are pure. We fix this specialization and the notation ¯x1, . . . , ¯xm. Any other specialization is denoted ¯x′ i. Any pure semisimple substitution ¯xi is in S and thus in a block (or in glued blocks) of some degree ni, which we also call the degree of ¯xi. A radical substitution ¯xi is somewhat more subtle. It is viewed as an edge connecting two vertices in blocks, say of degrees ni1 and ni2. If these blocks are not glued, then we call this substitution a bridge of degrees ni1 and ni2. A bridge is proper if ni1 6= ni2. A proper bridge connecting vertices of degree ni 6= nj is an n-bridge if ni or nj is n. But there also is the possibility that a radical substitution connects two glued blocks of the same degree n, in which case we call it n-internal. 2.2.1. Review of the three canonization theorems for quivers. Since arbitrary gluing is difficult to describe, we need some "canonization" theorems to "improve" the gluing. The first theorem shows that we have already specified enough kinds of gluing. Theorem 2.3 (First Canonization Theorem, cf. [13, Theorem 6.12]). The Zariski closure of any representable affine PI-algebra A0 has a representation for whose full quiver all gluing is proportional Frobenius. For the Second Canonization Theorem we grade paths according to the following rule: REPRESENTABILITY OF AFFINE ALGEBRAS OVER AN ARBITRARY FIELD (VER. VI:34.0) 7 Definition 2.4. When F = q < ∞, we write M∞ for the multiplicative monoid {1, q, q2, . . . , ǫ}, where ǫa = ǫ for every a ∈ M∞. (In other words, ǫ is the zero element adjoined to the multiplicative monoid hqi.) Let M be the semigroup M∞/∼ where ∼ is the equivalence relation obtained by matching the degrees of glued variables: When two vertices have a (q1, q2)-Frobenius twist, we identify 1 with qk = q1 in the respective q2 matrix blocks, and use M to grade the paths. Definition 2.5. A full quiver is basic if it has a unique initial vertex r and unique terminal vertex s, and all of its gluing above the diagonal is proportional Frobenius. A basic full quiver Γ is canonical if any two paths from the vertex r to the vertex s have the same grade. (Our notion of basic quiver has nothing to do with the notion of basic algebra in representation theory.) Theorem 2.6 (Second Canonization Theorem, cf. [14, Theorem 3.7]). Any relatively free algebra is a subdirect product of algebras whose full quivers are basic. Any basic full quiver Γ of a representable relatively free algebra can be modified (via a change of base) to a canonical full quiver of an isomorphic algebra (i.e., relatively free algebra of the same variety). In view of this result, we may reduce to the case that the full quiver of our polyno- mial f is basic. The Third Canonization Theorem [14, Theorem 3.12] describes what happens when one mods out a "nice" T-ideal, so is not relevant, since all we need is to find a repre- sentable T-ideal, which we do later by another method. 3. The Canonization Theorem for Polynomials We have two languages: quivers and their representations on one hand, versus the combinatorial language of identities on the other hand. First we consider the geometri- cal aspect. A branch is a path B in the quiver. The length of B is its number of arrows, excluding loops, which equals its number of vertices (say k) minus 1. Thus, a typical branch has vertices of various matrix degree nj, j = 1, 2, . . . , k. We call (n1, . . . , nk) the degree vector [16, Definition 2.32] of the branch B. The descending degree vec- tor is obtained by ordering the entries of the degree vector to put them in descending order lexicographically (according to the largest nj which appears in the distinct glued matrix blocks, excluding repetitions, taking the multiplicity into account in the case of Frobenius gluing). We write the descending degree vector as (π(n)1, . . . , π(n)k). Thus, π(n)1 = max {n1, . . . , nk}. If B appears in a nonzero specialization of a monomial of f , we call B a branch of f . We denote the largest nj appearing in the quiver as n. Definition 3.1. A branch B is dominant if it has the maximal possible number of n-bridges, has maximal possible length k with regard to this property, has the maximal possible number of vertices of n-bridges among these in the lexicographic order, and then we continue down the line to n − 1 etc. The depth of a dominant branch B is the number of times n appears in its degree vector. 8 ALEXEI BELOV-KANEL, LOUIS ROWEN, AND UZI VISHNE Our goal is somehow to force every nonzero evaluation of f into a dominant branch by considering each degree from n down in turn. Throughout, cm here denotes the Capelli polynomial in 2m2 indeterminates (denoted c2m2 in [10]), which is alternating in m indeterminates and an identity of Mm−1(F ) for any field F ; hm,i(y) denotes a multilinear central polynomial hm,i(yi1, . . . , yi′ m) for Mm(F ), in specific indeterminates yi1, . . . , yim which are all distinct. Evaluating hm,i on semisimple matrices of degree < m is 0. We put hm = hm,1hm,2 · · · hm,t+1, the product of t + 1 copies of distinct Capelli polynomials of the same degree 2m2, and call the hm,j the respective components of h. We focus first on semisimple substitutions having matrix degree n, and put h = hn. Lemma 3.2. Any nonzero specialization of h has a component of solely semisimple substitutions (all of the same degree). Proof. Otherwise every component has a radical substitution, so we have a product of t + 1 radical elements, which is 0 by definition of t. (cid:3) Viewing a substitution of xi as corresponding to an edge in the quiver, we have two degrees, one for each vertex. Definition 3.3. An m-right substitution of xi is one having degree including m. An m-wrong substitution xi of xi is one both of whose degrees differ from m, where m appears as a degree of the substitution xi. We write right (resp. wrong) for n-right (resp. n-wrong). In view of Lemma 3.2, a wrong substitution would lead to h having a component of semisimple substitutions in a matrix block of the wrong degree. One delicate point: An internal radical bridge say from one matrix block of degree m to a different matrix block of degree m is technically "right" according to this definition, but must be dealt with. Remark 3.4. Suppose f (x1, . . . , xℓ) is a full nonidentity of A0 whose nonzero eval- uation passes through all the blocks of the quiver, via the dominant branch B say of degrees m1, . . . , mk having some number k of bridges, and k′ internal radical substi- tutions. By Theorem 2.6, any wrong nonzero substitution may be assumed to have k bridges since otherwise we apply induction to the number of semisimple components in the full quiver. On the other hand, we can take the nonzero evaluation with k′ maximal, so then any wrong substitution has at most k′ internal radical substitutions. We work with a dominant branch B of Γ in f . Our objective is to modify f to a nonidentity containing a Capelli component which enables us to use combinatorial methods to calculate characteristic coefficients in a Shirshov extension, with multipli- cation by elements of A0. Here is one of our main results, enabling us to correspond quivers with properties of polynomials, and which leads directly to the representability theorem. Theorem 3.5 (Canonization Theorem for Polynomials). Suppose f (x1, . . . , xℓ) is a full nonidentity of A0. Then the T-ideal of f contains a critical nonidentity. REPRESENTABILITY OF AFFINE ALGEBRAS OVER AN ARBITRARY FIELD (VER. VI:34.0) 9 4. The proof of the Canonization Theorem for Polynomials The proof of the Canonization Theorem for Polynomials is done in several stages: (1) Eliminate unwanted semisimple substitutions. (2) Make sure that that the substitutions are in the "correct" semisimple compo- nents. (3) Provide a molecule inside the polynomial where we can compute the action of characteristic coefficients. 4.1. Unmixed case. First, following Kemer, we dispose of the following easy case. We say a substitution is unmixed if it does not involve bridges, i.e., all substitutions are in a single Peirce component. Here we need only multiply by a Capelli polynomial of the matrix degree, and then proceed directly to the method of §6. This aspect is crucial to our proof, since substitutions alone are not sufficient to take care of examples such as the non-finitely generated T-space of [33] (generated by {[x1, x2]xpk−1 , k ∈ N} in the Grassmann algebra with two generators; also see [22, 23]). xpk−1 2 1 4.2. The mixed case: the hiking procedure. To complete the proof of the Canonization Theorem for Polynomials, we must turn to the mixed case. The main feature in the proof of the Canonization Theorem for Polynomials is hiking. The notion of hiking passes from branches of quivers to com- binatorics of nonidentities, showing how to modify a non-identity of a T-ideal I to another non-identity in I whose algebraic operations leave us in the same quiver. In our combinatorics we need to cope with the danger that our substitutions are wrong, or the base field of the semisimple component is of the wrong size. To prevent this, we make substitutions of multilinear polynomials for indeterminates inside f , called hiking, which force the evaluations to become 0 in such situations. In other words, hiking replaces f by a more complicated polynomial in its T-ideal, which yields a zero valuation when we start with a wrong substitution in the original indeterminates of f . We have three kinds of variables: • Core variables, used for exclusive absorption inside the radical (such as variables which appear in commutators with central polynomial), • variables used for hiking, • variables inside Capelli polynomials used for computing the actions of charac- teristic coefficients. Example 4.1. An easy example of the underlying principle: If k = 2 with n1 > n2, then the quiver Γ consists of two blocks and an arrow connecting them, so we replace a variable y of f with a radical substitution by hn1,1[hn1,2, z]yhn2. The corresponding specialization remains in the radical. Then we are ready to utilize the techniques given below in §6 to compute characteristic coefficients, bypassing the complications of hiking. Suppose we have the polynomial f , with a radical evaluation. We replace it and have a hiked polynomial. If g belongs to the T-ideal generated by g, then one of the variables in h must have a radical evaluation. After making the substitution we get a 10 ALEXEI BELOV-KANEL, LOUIS ROWEN, AND UZI VISHNE new polynomial g ′ of the same form as g. This is like the Phoenix property described in characteristic 0. But in general we need a rather intricate analysis. Definition 4.2. Given a polynomial f (x1, . . . , xℓ) and another polynomial g, we write fxi 7→g to denote that g is substituted for xi. We say that f is hiked to f := fxi 7→g at xi if g is linear in xi. We call the replacement g of xi a molecule of the hiked polynomial. A complex molecule is the product of molecules. Definition 4.3. A polynomial is docked (of length d) if it can be written in the form gu,1hn(y)gu,2hn(y) · · · gu,dhn(y)gu,d+1 Xu for suitable polynomials gu,i (perhaps constant) in which the y indeterminates do not occur. (In other words the y indeterminates occur only in the hn(y).) Unfortunately, if the xi repeat then the molecules repeat, and thus the variables y repeat. A docked polynomial f (x1, . . . , xt; y, y ′, y ′′; z, z′) is critical if any nonzero substitu- tion of the yi is right. Thus the docks are attached to molecules. If f is hiked to various polynomials fj we also say it can be hiked to P fj. (Likewise for other indeterminates that appear once the hiking is initiated.) Remark 4.4. First suppose that the depth u = k, i.e., all nj = n, and there are no nonzero external radical substitutions. In other words, the only nonzero substitutions involve specializing all the xi to semisimple elements in blocks of degree n. Then we simply replace f by hf , which trivially is docked, and the theorem is proved. So in the continuation, we assume that u < k, which means there is some nonzero substitution f (x1, . . . , xk) in our dominant branch B, for which some xi is an n-bridge. We fix x1, . . . , xk in what follows, and call it our fundamental substitution, with bridge at this i. We prove a more technical version of the Canonization Theorem, to handle the mixed case. Theorem 4.5 (Hiking Theorem for Polynomials). Suppose f (x1, . . . , xℓ) is a full non- identity of A0, possibly with mixed or pure substitutions. Then f can be hiked to a critical nonidentity in which all of the substitutions of the xi are right. 5. Details of hiking The proof of Theorem 4.5 is through a succession of hiking steps in order both to eliminate "wrong" substitutions and then docking, i.e., insert hn into the polyno- mial. The latter is achieved by replacing zi by hnzi and z′ ihn; i.e., we pass to fzi 7→h nzi,z ′ i by z′ ih n. 7→z ′ i The hiking procedure requires three different stages. REPRESENTABILITY OF AFFINE ALGEBRAS OVER AN ARBITRARY FIELD (VER. VI:34.0) 11 5.1. Preliminary hiking. Our initial use of hiking is to resolve some technical issues. First, we want to elimi- nate the effect of (q1, q2)-Frobenius gluing for q1 6= q2, since it can complicate docking. Toward this end, we substitute zi′cnj (y)q1/q2 for zi′, for each instance of Frobenius gluing. It makes the Frobenius gluing identical on f . We also need the base fields of the components all to be the same. When B′ is another branch with the same degree vector, and the corresponding base fields for the i and replace xi by i-th vertex of B and B′ are ni and n′ (cti ni − cni)xi. This cuts off the specializations to matrices over finite fields of the wrong order. i respectively, we take ti = qn′ 5.2. First stage of hiking. zi1[xi1, hn−1]zi1+1cni1+1, (where as always the cni1 We have a quasi-linear nonidentity f of a Zariski closed algebra A0 which has a fun- damental substitution in some branch B, where xi1 in A0 is an n-bridge, corresponding to an edge in the full quiver whose initial vertex is labeled by (nℓ, ti1) and whose ter- minal vertex is labeled by (ni1+1, ti1+1) where n = max{ni1, ni1+1}. We replace xi1 by involve new indeterminates in v), cni1 and zi1, zi1+1 also are new indeterminates which we call "docking indeterminates"); this yields a quasi-linear polynomial in which any substitution of xi1 into a diagonal block of degree < n1 or a bridge which is not an n-bridge is 0. For each semisimple substitu- tion xi in a block of degree ni, taking [xi1, hni1 ] yields 0. This removes all semisimple component substitutions in h of such xi whose degree is too "small," i.e., less than ni. For the time being, we could still have radical substitutions, but first stage hiking does prepare for their elimination in the second stage. The number of extra n-bridges in a specialization of cni1 (v)]zi1+1cni1+1(v) is called its (first stage) bridge contribution. (In other words, one takes the total number of bridges, and subtracts 1 if xi is an n-bridge.) The term [xi1, hn′ (v)] is called i1 the core of the bridge contribution. (v)zi1[xi1, hn′ i1 Lemma 5.1. Any nonzero specialization of h is either n-semisimple, or its bridge contribution is positive. Proof. By definition, if the bridge contribution is 0 then every substitution has to be semisimple or a (j, j)-bridge for some j. If n does not appear then the graph would have t such bridges. (cid:3) Lemma 5.2. After the first stage of hiking, a wrong specialization of an n-semisimple element cannot be m-semisimple for m < n unless its bridge contribution is at least 2. Proof. When evaluating hn on semisimple elements of degree m we get 0 unless we pass away from the m-semisimple component, which requires two bridges. (cid:3) Lemma 5.3. After the first stage of hiking, a wrong specialization of an n-semisimple element is either n-semisimple or its bridge contribution is at least 1. Proof. When evaluating hn on semisimple elements of degree m we get 0 unless we pass away from the m-semisimple component, which requires two bridges. (cid:3) The first stage of hiking does not instantly zero out bridges xi, but does prepare for their elimination in the second stage. 12 ALEXEI BELOV-KANEL, LOUIS ROWEN, AND UZI VISHNE Appending the Capelli polynomials also sets the stage for eliminating other unwanted substitutions in the second stage. After repeated applications of first stage hiking, we wind up with a new polynomial f (x1, . . . , xℓ; v; z) where we still have our original indeterminates xi but have adjoined new indeterminates from v and z. 5.3. Second stage of hiking. Example 5.4. To introduce the underlying principle, here is a slightly more compli- cated example. Consider the quiver of three arrows, from degree 2 to degree 1, degree 1 to degree 1, and finally from degree 1 to degree 1. First we multiply on the left by c4[h1,1, z1]z2. The second substitution could have an unwanted position inside the first matrix block of degree 2, since c4[h1,1, z1]z2 can be evaluated in the larger component. We take fx1 7→c2y′yx1 − fx1 7→x1c2y′y, i.e., we multiply by a central polynomial h2 on the left and subtract it from a parallel evaluation of h2 on the right. The unwanted substitution then cancels out with the other substitution and leaves 0. In the second stage of hiking, in the blended case, we arrange for all nonzero substi- tutions to be pure radical. Suppose f (x, . . . , xℓ; y; z; z′) is already hiked after the first stage. Suppose in the branch B the indeterminate zi occurs of degree di and the indeterminate zu+1 occurs of degree d′, where 1 ≤ j ≤ u. Proposition 5.5. There are three cases to consider: (1) There is a string xi−1xi · · · xjxj+1 where xi, · · · , xj are all semisimple of the same degree xnj whereas xi−1, xi, xjxj+1 are both n-bridges. We take the polynomial fzi 7→h n(y′)di zi − fz ′ u+1 7→z ′ u+1hnu (y′)ti , (2) where the branch B has depth u and ti designates the maximal degree of xi in a monomial of B, where y ′ is a fresh new set of indeterminates (2) There is a string x1xi · · · xjxj+1 where x1 · · · , xj are all semisimple of the same degree xn1 whereas xjxj+1 is an n-bridge. We take the polynomial (3) There is a string xi−1xi · · · xkxk where xi, · · · , xk−1 are all semisimple of the same degree xn1 whereas xi−1xi is an n-bridge. We take the polynomial fz17→h n(y′)d1 z1. (3) fz17→h n(y′)d1 z1. (4) This hiking zeroes out semisimple evaluations of highest degree (n), but not a radical evaluation at the u block. Proof. (Note that (1) is the usual case, but we also need (2) and (3) to handle terms lying at the ends of the polynomial.) The expression (2) yields zero on a semisimple substitution, but not on a radical substitution, since exactly one of the two summands of (2) would be 0. (cid:3) Lemma 5.6. The second stage of hiking forces any nonzero specialization of an n- bridge also to be a n-bridge. REPRESENTABILITY OF AFFINE ALGEBRAS OVER AN ARBITRARY FIELD (VER. VI:34.0) 13 Proof. In order to provide a nonzero value, at least one of its vertices must be of degree n. But if both were n the evaluation would be 0, by Lemmas 5.1–5.3 and Remark 5.5. Thus we get an n-bridge. (cid:3) Lemma 5.7. After the first and second stages of hiking, the positions of semisimple substitutions of degree n are fixed; in other words, semisimple substitutions of degree n are n-right. Proof. Lemma 5.6 "uses up" all the places for n-bridges, since more n-bridges would yield a substitution contradicting the maximality of the number of n-bridges in B. If B has no semisimple substitutions of degree n then there is no room for any semisimple substitutions of degree n, and we are done. But if B has a semisimple substitution of degree n, that substitution must border an n-bridge, fixing the order of the pair of indices in the n-bridge, and thus fixing the positions of all the gaps of index n between n-bridges, so we are done. (cid:3) Remark 5.8. Although this is taken care of in the proof, we can remove finite compo- nents simply by substituting xm i for xi, for suitable ℓ, m. i − xℓ Proof of Theorem 4.5. Just iterate the hiking procedure down from n. (cid:3) Proof of Theorem 3.5. One obtains the dock by replacing f by fzu 7→c n(y′)tu zu, z ′ u+1 7→z ′ u+1cnu (y′)t1 . (cid:3) Example 5.9. Let us run through the hiking procedure, taking A0 =     ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ 0 0 0 ∗ ∗ 0 0 0 ∗ ∗ I1 → II 2, .     We have the full quiver and take the nonidentity f = x1[x2, x3]x4+x4[x2, x3]x2 1. We have nonzero specializations with x1 in the first matrix component, x2 an external radical specialization, and x3 in the second matrix component, which we denote as C = M2(K), but also we have a nonzero specialization of all variables into C. To avoid this situation, we replace f by f (c3(y)zx1z′, x2, x3, x4) = c3(y)zx1z′[x2, x3]x4 + x4[x2, x3]2c3(y)zx1z′c3(y)zx1z′. Now any specialization into C becomes 0, so we have eliminated some "wrong" special- izations. For stage 2 we take f (x; y; y ′; z; z′) := f (c3(y)c3(y ′)2zx1z′, x2, x3, x4) − f (c3(y)zx1z′c3(y ′), x2, x3, x4) = (c3(y)c3(y ′)2zx1z′[x2, x3]x4 + x4[x2, x3]2c3(y)c3(y ′)2zx1z′c3(y)zx1z′) − (c3(y)zx1z′c3(y ′)[x2, x3]x4 + x4[x2, x3]2c3(y)zx1z′c3(y ′)zx1z′c3(y ′)), where we see the specialization of highest degree to the first matrix component has been eliminated. We can eliminate the nonzero specializations of h(y ′′)3 of degree 1 by 14 ALEXEI BELOV-KANEL, LOUIS ROWEN, AND UZI VISHNE taking f (x; y; y ′; c3(y ′′)z; z′) − f (x; y; y ′; z; z′c3(y ′′)) which leaves us only with a radical specialization and a critical polynomial with a single dock c3(y ′′). Note how quickly the polynomial becomes complicated even though we have hiked only one indeterminate. Remark 5.10. Other examples of hiking are given in [16]. The main difference between the hiking procedure of this paper and that of stage 3 hiking of [16] is in the treatment of the Frobenius. Stage 4 hiking is analogous. 6. Characteristic coefficient-absorbing polynomials inside T-ideals We have already pinpointed the xi that must have substitutions into semisimple blocks of degree n, in order to utilize the well-understood properties of semisimple matrices (especially the coefficients of their characteristic polynomials, which we call characteristic coefficients). We follow the discussion of coefficient-absorbing poly- nomials from [16, Theorem 4.26] and [17, §6.3], although we can skip much of it because we already have a docked polynomial. Any matrix a ∈ Mn(K) can be viewed either as a linear transformation on the n-dimensional space V = K (n), and thus having Hamilton-Cayley polynomial fa of degree n, or (via left multiplication) as a linear transformation a on the n2-dimensional space V = Mn(K) with Hamilton-Cayley polynomial fa of degree n2. The matrix a can be identified with the matrix a ⊗ I ∈ Mn(K) ⊗ Mn(K) ∼= Mn2(K), so its eigenvalues have the form β ⊗ 1 = β for each eigenvalue β of a. From this, we conclude: Proposition 6.1 ([14, Proposition 2.4]). Suppose a ∈ Mn(F ). Then the characteris- tic coefficients of a are integral over the F -algebra C generated by the characteristic coefficients of a. Proof. The integral closure of C contains all the eigenvalues of a, which are the eigen- values of a, so the characteristic coefficients of a also belong to the integral closure. (cid:3) Next we use hiking also to force the characteristic coefficients of the matrices to com- mute with each other. Using Theorem 3.5, we work with quasi-linear polynomials and pinpoint semisimple substitutions of degree n, in order to utilize the well-understood properties of semisimple matrices (especially the characteristic coefficients). Having obtained semisimple substitutions of degree n, we have two ways of obtaining intrinsically the coefficients of the characteristic polynomial ga = λn + n−1 Xk=1 (−1)kαk(a)λn−k of a matrix a. Fixing k, we write αk for αk(a), which we call the k-characteristic coefficient of a. We want to extract these characteristic coefficients, by means of polynomials. REPRESENTABILITY OF AFFINE ALGEBRAS OVER AN ARBITRARY FIELD (VER. VI:34.0) 15 Definition 6.2. In any matrix ring Mn(W ), we define n αmat(a) := Xj=1 X ej,i1aei2,i2a · · · aeikikaei1,j, (5) the inner sum taken over all index vectors of length k. We can also define the characteristic coefficients via polynomials. Definition 6.3. Given a quasi-linear polynomial f (x; y) in indeterminates labeled xi, yi, we say f is characteristic coefficient-absorbing with respect to a full quiver Γ if f (A0(Γ))+ absorbs multiplication by any characteristic coefficient of any element in each docked (diagonal) matrix block of a molecule of A0(Γ). Lemma 6.4 (as in [15, Lemma 3.6]). Write the polynomial f of Theorem 3.5 as a sum of homogeneous components P fj. Each fj is characteristic coefficient absorbing in the blocks of degree n. Proof. The proof can be formulated in the language of [10, Theorem J, Equation 1.19, page 27] (with the same proof), as follows, writing Ta,j for the transformation given by left multiplication by a: a a1, . . . , T kt a at, r1, . . . , rm), (6) αkf (a1, . . . , at, r1, . . . , rm) = X f (T k1 summed over all vectors (k1, . . . , kt) with each ki ∈ {0, 1} and k1 + · · · + kt = k, where αk is the k-th characteristic coefficient of a linear transformation Ta : V → V. (cid:3) Since the purpose of Lemma 6.4 was to obtain the conclusion (6), we merely as- sume (6). Lemma 6.5. For any homogeneous polynomial f (x1, x2, . . . ) quasi-linear in x1 with respect to a matrix algebra Mn(F ), satisfying (6), there is a polynomial f in the T-ideal generated by f which is characteristic coefficient absorbing. Proof. Take the polynomial of Lemma 6.4. (cid:3) Remark 6.6. Notation as in (6), the Cayley-Hamilton identity for n × n matrices is 0 = = n Xk=0 Xk=0 n (−1)kαkf (a1, . . . , at, r1, . . . , rm)λn−k (−1)k Xk1+···+kt=k f (T k1 a a1, . . . , T kt a at, r1, . . . , rm)λn−k, which is thus an identity in the T-ideal generated by f . Definition 6.7. We call the identity n Xk=0 (−1)k Xk1+···+kt=k f (T k1 a a1, . . . , T kt a at, r1, . . . , rm)λn−k obtained in Remark 6.6, the Hamilton-Cayley identity induced by f . Definition 6.8. Fixing 0 ≤ k < n, we denote this implicit definition in Lemma 6.5 of αk, the k-th characteristic coefficient of a, as αpol ¯q(a). 16 ALEXEI BELOV-KANEL, LOUIS ROWEN, AND UZI VISHNE If the vertex corresponding to r has matrix degree ni, taking an ni × ni matrix w, we define αpol ¯q u(w) as in the action of Definition 6.8 and then the left action Likewise, for an nj × nj matrix w we define the right action au,v 7→ αpol ¯q u(w)au,v. au,v 7→ au,vαpol ¯q v(w). (7) (8) (However, we only need the action when the vertex is non-empty; we forego the action for empty vertices.) Remark 6.9 (For f homogeneous.). Take f of Lemma 6.5, and one more indetermi- nate y ′′. There is a Capelli polynomial cn2 (y ′′) and p-power ¯q such that i cn2 i (αky ′′)xicn′ i 2(y ′′) = α¯q k(y1)cn2 i (y ′′)xicn′ i 2(y ′′) (9) on any diagonal block. Since characteristic coefficients commute on any diagonal block, we see from this that cn2 i (y ′′)xicn′ i 2(y′′)cn2 i (z)xicn′ i 2(z) − cn2 i (z)xicn′ i 2(z)cn2 i (y ′′)xicn′ i 2(y′′) (10) vanishes identically on any diagonal block, where z = αky ′′. One concludes from this that substituting (10) for xi would hike f one step further. But there are only finitely many ways of performing this hiking procedure. Thus, after a finite number of hikes, we arrive at a polynomial in which we have complete control of the substitutions, and the characteristic coefficients defined via polynomials commute. 7. Resolving ambiguities for nonhomogeneous polynomials When f is homogeneous, we can skip this section. The non-homogeneous case is more delicate. Since we may be in nonzero characteristic, in the main situation our quasi-linear hiked polynomials are not homogeneous. In §6 we obtained actions on each monomial component separately, but we need to provide a uniform action on each of these components. 7.1. Removing ambiguity of matrix degree for nonhomogeneous polynomi- als. First we want to make sure that we are working in the same matrix degree for each monomial. Definition 7.1. A hiked polynomial is uniform if there is some indeterminate xi for which, in each of its monomials, the molecule obtained from hiking xi is semisimple of the same matrix degree. Our objective in this section is to hike to a uniform polynomial. First we use §4.1 to dispose of the easy case where each hiked monomial has a semisimple molecule (Definition 4.2). Definition 7.2. A radical element of a complex molecule is isolated if multiplication by any radical element on the left or right is zero. Remark 7.3. The product of two isolated elements is 0, by definition. Proposition 7.4. Any polynomial can be hiked to a uniform polynomial. REPRESENTABILITY OF AFFINE ALGEBRAS OVER AN ARBITRARY FIELD (VER. VI:34.0) 17 Proof. Multiply xi by a new indeterminate x′ i and hike that. We are done unless it yields a radical substitution. Since J t+1 = 0, we get an isolated element after at most t hikes. (cid:3) 7.2. Removing ambiguities for nonhomogeneous polynomials having molecules of the same matrix degree. We have just reduced to the case where all monomials have molecules of some xi of the same matrix degree n, but we still must contend with the possibility that xi has different degrees in different monomials, and then our characteristic coefficient arguments work differently for the different monomials. Take the finitely many matrix components Ri, 1 ≤ i ≤ k, each of which has degree n and multiplicity qi. Multiplication by characteristic coefficients αi is integral over the multiplication by αqi i . We introduce a commuting indeterminate λi for each of the finitely many character- istic coefficients αi, i ∈ I, define C ′ to be C[λi : i ∈ I]. Defining the action of λi on Ri to be the same as that of αi, we have (λi − αi)Ri = 0. Definition 7.5. Given matrices a1, . . . , at, the symmetrized (k; j) characteristic co- efficient is the j-elementary symmetric function applied to the k-characteristic coeffi- cients of a1, . . . , at. For example, taking k = 1, the symmetrized (1, j)-characteristic coefficients αt are t Xj=1 tr(aj), Xj1>j2 tr(aj1) tr(aj2), . . . , t Yj=1 tr(aj). Lemma 7.6. Any characteristic coefficient αk is integral over the ring with all the symmetrized characteristic coefficients adjoined. Proof. If αk,t denotes the (t; j)-characteristic coefficient, then αk satisfies the usual (cid:3) polynomial λn + (−1)j Pn j=1 αk,jλt−j. Remark 7.7. The reason that we need to introduce the symmetrized characteristic coefficient is that we might have several glued components and their molecules, which we cannot distinguish, so we need to find coefficients common to all of them. Proposition 7.8. Take f of Lemma 6.5. There is a uniform polynomial f hiked from f which is characteristic coefficient absorbing. Proof. f has no semisimple evaluations into a maximal component, and thus (induc- tively) all evaluations of f involving a maximal component have a "radical bridge," i.e., f has a monomial h where the indeterminates xi specialize to external radical substitutions. In particular ni = n = n1. We adjoin the characteristic values of all the complex molecules. Then we apply (cid:3) Proposition 7.4 to Lemmas 6.4 and 6.5. 8. Application of Shirshov's theorem Here is the connection to full quivers. 18 ALEXEI BELOV-KANEL, LOUIS ROWEN, AND UZI VISHNE Definition 8.1. For a Zariski closed algebra A ⊆ Mn(K) faithful over an integral do- main C, we denote by C the algebra obtained by adjoining to C the symmetrized char- acteristic coefficients of products of the Peirce components of the generic generators of A (of length up to the bound of Shirshov's Theorem [10, Chapter 2]). Characteristic Value Adjunction Theorem [16, Theorem 3.22]. Let A0 denote the algebra obtained by adjoining to A0 the characteristic matrix coefficients of prod- ucts of the sub-Peirce components of the generic generators of A0 (of length up to the bound of Shirshov's Theorem [10, Chapter 2]), and let C be the algebra obtained by adjoining to F these symmetrized characteristic coefficients. The T-ideal I generated by the polynomial f contains a nonzero T-ideal which is also an ideal of the algebra A0. Recall in view of Shirshov's theorem that we only need to adjoin a finite number of elements to obtain C. Lemma 8.2. The algebra A0 is a finite module over C, and in particular is Noetherian and representable. Proof. Let C ′ be the commutative algebra generated over C, by all the characteris- tic coefficients of (finitely many) products of the Peirce components of the generic generators of A0, as in Definition 8.1. Clearly C ⊆ C ′. Enlarge A0 to A′ 0 = C ′A0, which is a finite module over C ′ in view of Shirshov's Theorem. But C ′ is finite over C, in view of Lemma 7.6, implying A0 is finite over C. Thus A0 is Noetherian, and is representable by Anan'in's Theorem [3]. (cid:3) Also, for any characteristic coefficient-absorbing polynomial f with respect to the quiver of A0, the Hamilton-Cayley identity induced by f is an identity of A0, and thus of A0. 9. Conclusion of the proof of Theorem 1.1 Proof. Let I be the T-ideal generated by f , and I1 be the T-ideal of A0 generated by symmetrized ¯q-characteristic coefficient-absorbing polynomials of I in A := C ′A0. The ideal I0 is representable by Lemma 8.2, implying A0 ∩ I1 is representable, as desired. (cid:3) References [1] Aljadeff, E. and Kanel-Belov, A., Representability and Specht problem for G-graded algebras, Advances in Math., 225:5 (2010), 2391-2428. [2] Amitsur, S.A., On the characteristic polynomial of a sum of matrices, J. Linear and Multilinear Algebra 8 (1980), pp. 177–182. [3] Anan'in, A.Z., The representability of finitely generated algebras with chain condition, Arch. Math. 59 (1992), 275-277. [4] Belov, A., On non-Spechtian varieties. (Russian. English summary) Fundam. Prikl. Mat. 5, No. 1 (1999), 47–66. [5] Belov, A., Counterexamples to the Specht problem, Sb. Math. 191 (2000), pp. 329-340. [6] Belov, A., Local finite basis property and local representability of varieties of associative rings, Izvestia of Russian Academia of science, No 1, 2010, pp. 3–134. English transl.: Izvestiya: Math- ematics, vol. 74, No 1, pp. 1–126. REPRESENTABILITY OF AFFINE ALGEBRAS OVER AN ARBITRARY FIELD (VER. VI:34.0) 19 [7] Belov, A., On varieties generated by a ring which is finite-dimensional over its centroid, Comm. Moscow Math. Soc., Russian Math. Surveys 62:2, pp. 379-400. [8] Belov, A., The Gel'fand-Kirillov Dimension of relatively free associative algebras, Sbornik Mathe. 195 (2004), pp. 1703-1726. [9] Belov, A., On rings that are asymptotically close to associative ones. (Russian. Russian summary) Mat. Tr. 10 (2007), no. 1, 2996. [10] Belov, A. and Rowen, L.H. Computational Aspects of Polynomial Identities, Research Notes in Mathematics 9, AK Peters, 2005. [11] Belov, A., Karasik, Y., and Rowen, L.H. Computational Aspects of Polynomial Identities, Re- search Notes in Mathematics 9, swecond edition, CRC, 2016. [12] Belov, A., Rowen, L.H., and U. Vishne, Zariski closed algebras and their representations, Trans. Amer. Math. Soc. 362, no. 9 (2010), 4695–4734. [13] Belov, A., Rowen, L.H., and U. Vishne, Full quivers of representations of algebras, Trans. Amer. Math. Soc. 364, 5525–5569, (2012). [14] Belov, A., Rowen, L.H., and U. Vishne, PI-varieties associated to full quivers of representations of algebras, Trans. Amer. Math. Soc. 365 (2013), no. 5, 2681–2722. [15] Belov, A., Rowen, L.H., and Vishne, U., Application of full quivers to polynomial identities, Comm. in Alg. 39, (2011), 4535–4551. [16] Belov, A., Rowen, L.H., and U. Vishne, Specht's problem for affine algebras over arbitrary com- mutative Noetherian rings, Trans. Amer. Math. Soc. 367 (2015), 5553–5596. [17] Belov, A., Rowen, L.H., and U. Vishne, Full exposition of Specht's problem, Serdica Mathematical Journal 38 (3), (2012), pp. 313–370. [18] Bergman, G. and Dicks, W., On universal derivations, J. Algebra 36 (1975), 193–211. [19] Bernsteın, I. N., Gelfand, I. M. and Ponomarev, V. A. Coxeter functors, and Gabriel's theorem (Russian) Uspehi Mat. Nauk 28(2(170)), 19–33, (1973). [20] Braun, A., The nilpotence of the radical in a finitely generated PI-ring, J. Algebra 89 (1984), 375–396. [21] Drensky, V. S. Identities in Lie algebras. (Russian) Algebra i Logika 13 (1974), 265–290, 363–364. [22] Grishin,A.V., Examples of T -spaces and T-ideals in Characteristic 2 without the Finite Basis Property(in Russian), Fundam. Prikl. Mat. 5 (1), no. 6 (1999), 101–118. [23] Grishin,A.V., On the existence of a finite basis in a T-space of generalized polynomials and on representability Grishin, Russian Mathematical Surveys 56(4) (2001), 755. [24] Iltyakov, A.V., Finiteness of basis identities of a finitely generated alternative PI-algebra, Sibir. Mat. Zh. 31 (1991), no. 6, 87–99; English translation: Sib. Math. J. 31 (1991), 948–961. [25] Iltyakov, A.V., On finite basis identities of identities of Lie algebra representations, Nova J. Algebra Geom. 1 no. 3 (1992), 207–259. [26] Kemer, A.R., The representability of reduced-free algebras, Algebra i Logika 27(3), 274–294, (1988). [27] Kemer, A.R., Identities of finitely generated algebras over an infinite field, Math. USSR Izv. 37, 69–97 (1991). [28] Kemer, A.R., Identities of Associative Algebras, Transl. Math. Monogr., 87, Amer. Math. Soc. (1991). [29] Kambayashi, Tatsuji; Miyanishi, Masayoshi; Takeuchi, Mitsuhiro. Unipotent algebraic groups, Lecture Notes in Mathematics 414 Springer-Verlag, Berlin-New York, 1974. [30] Lewin, J., A matrix representation for associative algebras. I and II, Trans. Amer. Math. Soc. 188(2) (1974), 293–317. [31] Rowen, L.H., Ring theory, Vol. 2, Academic Press Pure and Applied Mathematics 128 (1988). [32] Rowen, L.H., Ring Theory I. Pure and Applied Mathematics 127, Academic Press, 1988. [33] Shchigolev, V.V., Examples of infinitely basable T -spaces, Mat. Sb. 191 no. 3 (2000), 143–160; translation: Sb. Math. 191 no. 3-4 (2000), 459–476. [34] Vais, A. Ja. and Zelmanov, E. I., Kemer's theorem for finitely generated Jordan algebras, Izv. Vyssh. Uchebn. Zved. Mat. (1989), no. 6, 42–51; translation: Soviet Math. (Iz. VUZ) 33 no. 6 (1989), 38–47. 20 ALEXEI BELOV-KANEL, LOUIS ROWEN, AND UZI VISHNE Department of Mathematics, Bar-Ilan University, Ramat-Gan 52900,Israel E-mail address: {belova, rowen, vishne}@math.biu.ac.il
1607.00906	1	1607	2016-07-04T14:35:04	Monoids over which products of indecomposable acts are indecomposable	[ "math.RA" ]	In this paper we prove that for a monoid $S$, products of indecomposable right $S$-acts are indecomposable if and only if $S$ contains a right zero. Besides, we prove that subacts of indecomposable right $S$-acts are indecomposable if and only if $S$ is left reversible. Ultimately, we prove that the one element right $S$-act $\Theta_S$ is product flat if and only if $S$ contains a left zero.	math.RA	math	Monoids over which products of indecomposable acts are indecomposable Mojtaba Sedaghatjoo ∗† and Ahmad Khaksari‡ Abstract In this paper we prove that for a monoid S, products of indecompos- able right S-acts are indecomposable if and only if S contains a right zero. Besides, we prove that subacts of indecomposable right S-acts are indecomposable if and only if S is left reversible. Ultimately, we prove that the one element right S-act ΘS is product flat if and only if S contains a left zero. Keywords: flat, super flat. Indecomposable act, left reversible monoid, Baer criterion, product 2000 AMS Classification: Primary: 20M30; Secondary: 20M50. 1. Introduction Throughout this paper, S stands for a monoid and 1 denotes its identity element. A nonempty set A together with a mapping A × S → A, (a, s) as, is called a right S-act or simply an act (and is denoted by AS) if a(st) = (as)t and a1 = a for all a ∈ A, s, t ∈ S. Left S-acts can be defined similarly. We mean by A ⊔ B the disjoint union of sets A and B. The one element act is called zero act and is denoted by ΘS. A right S-act AS is called decomposable provided that there exist subacts BS, CS ⊆ AS such that AS = BS ∪ CS and BS ∩ CS = ∅; in this case AS = BS ∪ CS is called a decomposition of AS. Otherwise AS is called S, endowed with the indecomposable. For a nonempty set I, SI denotes the setQI natural componentwise right S-action (si)i∈I s = (sis)i∈I . We refer the reader to [1, 6] for more details on the concepts mentioned in this paper. Since for a given monoid S any right S-act AS is uniquely the disjoint union of indecomposable acts called indecomposable components of AS, analogous to the bricks forming a wall, indecomposable acts deserve to be taken into consideration. A pioneering work in this account goes back to [7], where the collection of all indecomposable right S-acts is partitioned into equivalence classes corresponding to the components of the right S-act R formed by letting S act on its right congruences by translation. ∗ Department of Mathematics, College of Sciences, Persian Gulf University, Bushehr, Iran, Email: [email protected] †Corresponding Author. ‡ Department of Mathematics, Payame Noor University, Tehran, Iran, Email: [email protected] 2 As mentioned, every right S-act AS has a unique decomposition into inde- composable subacts, indeed, indecomposable components of AS are the equiv- alence classes of the relation ∼ on AS defined in [8] by a ∼ b if there exist s1, s2, . . . , sn, t1, t2, . . . , tn ∈ S, a1, a2, . . . , an ∈ AS such that a = a1s1, a1t1 = a2s2, a2t2 = a3s3, . . . , an−1tn−1 = ansn, antn = b. We shall call this sequence of equalities a scheme of length n. Therefore, elements a, b ∈ AS are in the same indecomposable component if and only if there exists a scheme of length n as above connecting a to b. Note that for a natural number m > n, the scheme length can be increased to m by adding the equality b.1 = b.1 to the end of scheme iteratively. The paper comprises three sections as follows. In the first section we present a short account of the needed notions. The second one concerns with indecomposable acts over left reversible monoids. We prove that in Baer criterion for acts, the condition of possessing a zero element can be abandoned in case that S is not left reversible. In third section we engage in the main results of this paper, that is, conditions under which properties of indecomposability, product flatness and super flatness are preserved under products. Furthermore, we prove that for the one element act ΘS, the tensoring functor ΘS ⊗ − preserves limits if and only if it preserves products, equivalently, products of indecomposable left S-acts are indecomposable. 2. Indecomposable acts over left reversible monoids In this section we investigate indecomposable acts over left reversible monoids (that are monoids satisfying nonempty intersection for any pair of right ideals) and give some characterizations for left reversible monoids regarding indecomposability property. In the next proposition we show that for left reversible monoids the length of the preceding scheme can be considered to be 2. 2.1. Proposition. For a monoid S the following are equivalent: i) S is a left reversible monoid, ii) a right S-act AS is indecomposable if and only if for any a, a′ ∈ AS there exist s, s′ ∈ S such that as = a′s′, iii) any indecomposable right S-act contains at most one zero element. i =⇒ ii. Let S be a left reversible monoid and suppose that a, a′ ∈ AS Proof. for an indecomposable right S-act AS. So there exists a scheme connecting a to a′, of the form a = a1s1, a1t1 = a2s2, a2t2 = a3s3, . . . , antn = a′, for ai ∈ AS, si, ti ∈ S, 1 6 i 6 n. Left reversibility of S provides u1, v1 ∈ S such that s1u1 = t1v1, and in consequence au1 = a1s1u1 = a1t1v1 = a2s2v1. Proceeding inductively, we get u, v ∈ S satisfying au = antnv = a′v as desired. ii =⇒ iii. On the contrary suppose that an indecomposable right S-act contains two different zero elements namely θ1 and θ2. By assumption there exist s, t ∈ S for which θ1 = θ1s = θ2t = θ2 a contradiction. iii =⇒ i. By way of contradiction suppose that I ∩ J = ∅ for two right ideals I and J of S. Now define a right congruence ρ on S by xρy if x = y or x, y ∈ I or 3 x, y ∈ J. Take a ∈ I and b ∈ J. Then S/ρ is a cyclic indecomposable right S-act with two different zero elements namely [a] and [b], a contradiction. Recall that Baer criterion for right S-acts asserts that a right S-act is injective if and only if it possesses a zero element and is injective relative to all inclusions into cyclic right S-acts. In what follows we prove that if S is not left reversible then the condition of possessing a zero element in Baer criterion could be omitted. Let S be a monoid that is not left reversible. A right S-act 2.2. Proposition. QS is injective if and only if it is injective relative to all inclusions into cyclic right S-acts. Necessity is clear. To prove sufficiency, let QS be a right S-act that is Proof. injective relative to all inclusions into cyclic right S-acts. Suppose that I ∩ J = ∅ for two right ideals I and J and ρ is the Rees congruence on SS defined by the right ideal J. Consider the homomorphism f : IS −→ QS given by f (i) = qi, i ∈ I for some q ∈ QS. Since I can be identified with a subact of S/ρ, our assumption yields a homomorphism ¯f : S/ρ −→ QS making the following diagram commutative. ⊆ S/ρ ¯f IS f QS Now S/ρ contains a zero element and so does QS which thanks to Baer criterion QS is injective. Here a question can be posed that whether a monoid S over which injective acts are precisely ones that are injective relative to all inclusions into cyclic acts, is not left reversible. In the next proposition we characterize monoids over which subacts of inde- composable acts are indecomposable. 2.3. Proposition. are indecomposable if and only if S is left reversible. For a monoid S all subacts of indecomposable right S-acts Proof. implies that aS ∪ bS is indecomposable and therefore aS ∩ bS 6= ∅. Necessity. Let a, b ∈ S. Since S is indecomposable, our assumption Sufficiency. This is a straightforward result of Proposition 2.1, part ii). Recall that for a nonempty set I, I S is an I-cofree right S-act where f s for f ∈ I S, s ∈ S is defined by (f s)(t) = f (st) for every t ∈ S. It should be mentioned that the 1-cofree objects or terminal objects in Act-S are precisely one element acts which are indecomposable. The next proposition characterizes monoids over which non-zero cofree acts are decomposable. 2.4. Proposition. For a monoid S the following are equivalent: i) all non-zero cofree S-acts are decomposable, ii) there exists a non-zero decomposable cofree right S-act, iii) S is left reversible. 4 i =⇒ ii is clear. ii =⇒ iii. By way of contradiction suppose that Proof. aS ∩ bS = ∅ for some a, b ∈ S. Let X S be a non-zero decomposable X-cofree act and f, g ∈ X S. Let h ∈ X S be given by h(x) =(f (x) g(x) if x ∈ aS, otherwise. So we get the scheme f = f.1, f a = ha, hb = gb, g.1 = g, which implies that f and g are in the same indecomposable component. Therefore X S is indecomposable a contradiction. iii =⇒ i. Let S be a left reversible monoid and X S be a non-zero cofree S-act. Take constant functions f = cx1 and g = cx2 in X S for different elements x1 and x2 in X. Then f and g are zero elements of X S (note that zero elements of X S are precisely constant functions). In light of Proposition 2.1, since X S contains two different zero elements, it is not indecomposable. Regarding the fact that any scheme in an arbitrary act can be translated into another one in its factor act, generally factor acts of indecomposable acts are indecomposable. It is clear that coproducts of indecomposable acts are not inde- composable, though the next proposition states that pushouts of indecomposable acts are indecomposable. It is worth pointing out that for a monoid S, since right S-acts are nonempty, the category of right S-acts is not complete nor cocomplete. Indeed, this category has products and coequalizers and has neither coproducts and equalizers. Note that in this category coproduct of nonempty families of objects exists. Hence, coproducts can not be considered as a sort of pushouts. 2.5. Proposition. For a monoid S consider a pushout situation Y1 f1xXS −−−−→ f2 Y2 in the category Act-S of right S-acts where Y1 and Y2 are indecomposable and suppose that ((q1, q2), QS) is the pushout of the pair (f1, f2). Then QS is indecom- posable. It is known that QS is isomorphic to (Y1 ⊔ Y2)/ν where ν is the con- Proof. gruence relation on (Y1 ⊔ Y2) generated by all pairs (f1(x), f2(x)), x ∈ X. Let [y1], [y2] ∈ (Y1 ⊔ Y2)/ν for some y1, y2 ∈ Y1 ⊔ Y2. Since Y1 and Y2 are indecom- posable, in view of the preceding argument we should just engage in the case that y1 ∈ Y1, y2 ∈ Y2 or vice versa. Without restriction of generality, we can consider only the first case. Take an element x0 ∈ X. Therefore there exist two schemes connecting y1 to f1(x0) and y2 to f2(x0). Thus we get two schemes in (Y1 ⊔Y2)/ν connecting [y1] to [f1(x0)] and [y2] to [f2(x0)] which, using the equality [f1(x0)] = [f2(x0)] provides the desired result. 5 As amalgamated coproduct of objects in a category is a sort of pushout, the next corollary follows. 2.6. Corollary. posable. Amalgamated coproducts of indecomposable acts are indecom- 3. Products of indecomposable acts There have been published several works on preservation of acts properties under products, for instance [2, 3, 4, 5, 9]. In this section we investigate another version of the problem for indecomposability property. Note that products of indecomposable acts are not indecomposable in general, for instance if S is a left zero semigroup with an identity element externally adjoined, then there is no scheme in S × S connecting (1, a) to (a, 1) for 1 6= a ∈ S. As a product of a family of right S-acts is a sort of pullback, then indecomposability property is not preserved under pullback and consequently coamalgamated product. It is easy to check that for non-empty sets I and J with J ≤ I, if SI is indecomposable, then so is SJ . Now suppose that SS×S is indecomposable. Let I be a non-empty set and (ai)I , (bi)I ∈ SI . Put J = {(ai, bi) i ∈ I}. We index J by a set K as J = {(uk, vk) k ∈ K}. Since SS×S is indecomposable and K = J ≤ S × S, by the preceding argument, SK is indecomposable and hence there exists a scheme in SJ connecting (uk)K to (vk)K. The corresponding scheme in SI is the one connecting (ai)I to (bi)I as desired. Thereby, the next corollary is obtained. 3.1. Corollary. indecomposable for each nonempty set I. For a monoid S, SS×S is indecomposable if and only if SI is A subject of interest in the study of tensor products is preservation of limits by tensoring functor AS ⊗ − for a right S-act AS which is investigated in [3]. Following terms used in this reference, a right S-act AS is called (finitely) super flat if the functor AS ⊗ − preserves all (finite) limits, and (finitely) product flat if it preserves all (finite) products. Now if finite products of indecomposable acts are indecomposable then S × S is indecomposable. In the next theorem we show that this is a sufficient condition for finite products of indecomposable acts to be indecomposable which is equivalent to the condition that the one-element left S-act SΘ is finitely product flat. Besides in the sequel we show that products of indecomposable acts are indecomposable if and only if the one element left S-act SΘ is product flat. 3.2. Theorem. For a monoid S the following are equivalent. i) finite products of indecomposable acts are indecomposable, ii) finite products of cyclic acts are indecomposable, iii) Sn is indecomposable for each n ∈ N, iv) Sn is indecomposable for some 1 6= n ∈ N, v) S × S is indecomposable, vi) the one element left S-act SΘ is finitely product flat. It is sufficient to prove the implication v =⇒ i and the term v ⇐⇒ vi is Proof. valid by [3, Corollary 2.5]. Suppose that S × S is indecomposable. We just need to prove that the product of two indecomposable right S-act is indecomposable and 6 then applying induction provides the desired result. Let AS and BS be indecom- posable right S-acts and let (a, b), (a′, b′) ∈ A × B for some a, a′ ∈ AS, b, b′ ∈ BS. In view of the last argument of Section 1, there exist two schemes as: a = a1s1, a1t1 = a2s2, a2t2 = a3s3, . . . , antn = a′ and b = b1u1, b1v1 = b2u2, b2v2 = b3u3, . . . , bnvn = b′ both of length n for some n ∈ N, ai ∈ AS , bi ∈ BS, 1 ≤ i ≤ n. By assumption, there exists a scheme connecting (s1, u1) to (t1, v1) in S × S of the form (s1, u1) = (x1, y1)w1, (x1, y1)z1 = (x2, y2)w2, (x2, y2)z2 = (x3, y3)w3, . . . , (xm, ym)zm = (t1, v1) which yields the scheme (a, b) = (a1s1, b1u1) = (a1x1, b1y1)w1, (a1x1, b1y1)z1 = (a1x2, b1y2)w2, (a1x2, b1y2)z2 = (a1x3, b1y3)w3, . . . , (a1xm, b1ym)zm = (a1t1, b1v1) and hence we can assert that (a, b) and (a1t1, b1v1) are in the same indecomposable component. Processing inductively, we conclude that (a, b) and (antn, bnvn) = (a′, b′) are in the same indecomposable component. Regarding Theorem 3.2 and Corollary 3.1 the next corollary is obtained. 3.3. Corollary. S × S is indecomposable. For a finite monoid S, SS×S is indecomposable if and only if If products of indecomposable acts are indecomposable, then SI is indecompos- able for each nonempty set I, though, in comparison with Theorem 3.2, this is a strict implication (see Example 3.11). Hereby, we need an additional condition on S to fill the gap namely Condition right(left)-FI under which there exists a fixed natural number n such that any pair of elements in any indecomposable right(left) S-act can be connected via a scheme of length n (see [3, Corollary 2.11]). In the next proposition we characterize monoids satisfying Condition right-FI. 3.4. Proposition. Monoids satisfying condition right-FI are precisely left re- versible monoids. Necessity. Suppose, by way of contradiction, that aS ∩ bS = ∅ for some Proof. a, b ∈ S. For each i ∈ N, set Si = {(i, s) s ∈ S} and endow it with the right S-action (i, s)t = (i, st) for s, t ∈ S. Let us denote the element (i, s) by s(i) for Si and ¯An = An/ρn where ρn i ∈ N, s ∈ S. For n ∈ N we define An = is the right congruence on An generated by the pairs (a(i), b(i+1)), 1 ≤ i ≤ n. Because of aS ∩ bS = ∅, for x, y ∈ An we have xρny only if x, y ∈ Si ∪ Si+1 for some 1 ≤ i ≤ n − 1. On the other hand, since a and b are not right invertible, n Si=1 [1(i)]ρn = {1(i)} for any 1 ≤ i ≤ n. Now we have the following scheme 7 [1(1)]ρn = [1(1)]ρn 1, [1(1)]ρn a = [1(2)]ρn b, a = [1(3)]ρn [1(2)]ρn b , . . . , [1(n−1)]ρn a = [1(n)]ρn b of length n connecting [1(1)]ρn 1 ≤ i ≤ n} is a generating set for ¯An and these generators are all in the same indecomposable component, ¯An is indecomposable. Let there exist another scheme . Since {[1(i)]ρn to [1(n)]ρn [1(1)]ρn = [a1]ρn s1, [a1]ρn s2, t1 = [a2]ρn [a2]ρn t2 = [a3]ρn s3, . . . , [am]ρn tm = [1(n)]ρn to [1(n)]ρn of length m connecting [1(1)]ρn for si, ti ∈ S, ai ∈ An, 1 ≤ i ≤ m. From 1(1)ρn a1s1 and amtm ρn 1(n), we get a1s1 = 1(1) and amtm = 1(n) which imply that a1 ∈ S1 and am ∈ Sn. Since a1t1 ρn a2s2, a2 ∈ S1 ∪ S2. Continuing inductively, am ∈ S1 ∪ S2 . . . ∪ Sm. Now, am ∈ Sn implies that n ≤ m and hence, is of length n. Considering ¯An, the shortest scheme connecting [1(1)]ρn for each n ∈ N, S doesn't satisfy condition right-FI, a contradiction. to [1(n)]ρn Sufficiency. If S is left reversible then, by Proposition 2.1, any pair of elements in any indecomposable right S-act is connected by a scheme of length 2. In the next proposition we characterize monoids for which products of indecom- posable acts are indecomposable. 3.5. Proposition. For a monoid S the following are equivalent: i) products of indecomposable right S-acts are indecomposable, ii) S is left reversible and SS×S is indecomposable, iii) S satisfies condition right-FI and SS×S is indecomposable, iv) non-zero cofree acts are decomposable and SS×S is indecomposable, v) all subacts of indecomposable right S-acts are indecomposable and SS×S is indecomposable. Proof. first two statements are equivalent. By virtue of Propositions 2.3, 2.4 and 3.4 it is enough to prove that the i =⇒ ii Suppose, contrary to our claim, that S is not left reversible. For each n ∈ N, let ¯An be the right S-act constructed in the proof of Proposition ¯An which is indecomposable by assumption. Therefore there 3.4. Set A = Qn∈N is a scheme of length m, connecting ([1(1)]ρn n∈N. Considering ) this scheme componentwise, for each n ∈ N there exists a scheme of length m in ¯An, connecting [1(1)]ρn . But according to the proof of Proposition 3.4, m ≥ n for each n ∈ N, a contradiction. n∈N to ([1(n)]ρn to [1(n)]ρn ) ii) =⇒ i) Let S be a left reversible monoid and let {Ai i ∈ I} be a family of in- Ai. In light of Proposition 2.1, decomposable right S-acts and (ai)i∈I , (bi)i∈I ∈ Qi∈I for each i ∈ I, there exist si, ti ∈ S such that aisi = biti. Let us denote a typical element of SI , with the same element x ∈ S in each component, by (x)i∈I . Since 8 SI is indecomposable by Corollary 3.1 and (si)i∈I , (1)i∈I ∈ SI , another applica- tion of Proposition 2.1, part ii, provides s, t ∈ S such that (si)i∈I s = (1)i∈I t. The same arguments provides existence of α, β ∈ S such that (tis)i∈I α = (t)i∈I β. Now for each i ∈ I we have aitα = aisisα = bitisα = bitβ, which yields (ai)i∈I tα = (bi)i∈I tβ as desired. For commutative monoids, the left reversibility condition in Proposition 3.5 is fulfilled and the following corollary is obtained. 3.6. Corollary. are indecomposable if and only if SS×S is indecomposable. For a commutative monoid S products of indecomposable acts Recall that a monoid S is called right collapsible if for any s, t ∈ S there exists u ∈ S such that su = tu. 3.7. Lemma. right S-acts are indecomposable if and only if S is right collapsible. For a left reversible monoid S, finite products of indecomposable Let S be a left reversible monoid, s, t ∈ S, and let finite products of Proof. indecomposable right S-acts be indecomposable. Since S × S is indecomposable, using Proposition 2.1 there exist u, v ∈ S such that (1, s)u = (1, t)v that is u = v and su = tu. Conversely, suppose that S is a right collapsible monoid and (a, b), (c, d) ∈ S×S. Under our assumption, there exist u1, u2 ∈ S such that au1 = cu1, bu2 = du2. Also u1u = u2u for some u ∈ S. Then (a, b)u1u = (c, d)u1u as desired. Considering the strict implication right collapsible =⇒ lef t reversible for monoids, Lemma 3.7 and Theorem 3.2 give the following result. 3.8. Corollary. A monoid S is right collapsible if and only if S is left reversible and S × S is indecomposable. 3.9. Theorem. indecomposable if and only if S has a right zero. For a monoid S products of indecomposable right S-acts are Necessity. Suppose that S is represented by an index set I as S = Proof. {si i ∈ I}. In view of Proposition 3.5 and Corollary 3.1, S is left reversible and SI is indecomposable. By Proposition 2.1, (si)i∈I s = (1)i∈I t for some s, t ∈ S. Thus for each x ∈ S, xs = t. Taking x = 1, gives s = t that is xt = t for every x ∈ S as desired. Sufficiency. Let z ∈ S be a right zero and let I be a nonempty set. Take (ai)i∈I , (bi)i∈I ∈ SI. Thus (ai)i∈I z = (bi)i∈I z and then SI is indecomposable. By assumption S is left reversible and regarding Proposition 3.5 the result follows. It is worth to mention that the two conditions in the sufficient 3.10. Remark. part of Corollary 3.8, regarding Example 3.11, are independent. Besides, analo- gously to the Corollary 3.8 and the strict implication monoid with right zero =⇒ left reversible monoid, Theorems 3.2 and 3.9 state that monoids with right zeros are precisely left reversible monoids for which SS×S is indecomposable. The next example shows that the conditions occurring in part ii) of Proposition 3.5 are independent. Let S be a left reversible monoid which doesn't have a right 3.11. Example. zero for instance a nontrivial finite group. Proposition 3.5 and 3.9 imply that SS×S is not indecomposable. On the other hand, let S = Tn consist of transformations of the set {1, 2, . . . , n} with mappings written on the left side for some 1 6= n ∈ N. 9 Set ci =(cid:18) 1 i 2 . . . n i . . . i (cid:19) for 1 ≤ i ≤ n. For (α, β) ∈ S × S, we have (α, β)c1 = (ci1 , ci2 ) where i1, i2 ∈ {1, 2, . . . , n} which states that (α, β) and (ci1 , ci2 ) are in the same indecomposable component. Now the following scheme (c1, c1) = (1, c1)c1 (1, c1)ci1 = (ci1 , c1)1 (ci1 , c1)1 = (ci1 , 1)c1 , (ci1 , 1)ci2 = (ci1 , ci2 ) , in S × S, implies that (c1, c1) and (ci1 , ci2 ) are in the same indecomposable component. From this S × S is indecomposable. Using Corollary 3.3 SI is inde- composable for each nonempty set I and since S contains two left zeros, S is not left reversible. In the above example we observed that for the monoid S = Tn, S × S is indecomposable. So a question can be posed that whether for the monoid of full transformations of a nonempty set X, SI is indecomposable for each nonempty set I. Note that in [3], Proposition 3.8 states that for a proper right ideal K of a monoid S if the Rees factor act S/K is finitely product flat then S/K is super flat. So a natural question that comes to the mind is the case that K = S. In the next proposition we show that in this case product flatness is equivalent to super flatness. Indeed in [3, Corollary 2.11] it is proved that the one element right S-act ΘS is product flat if and only if S satisfies condition left-F I and the left S-act SSI is indecomposable for each nonempty set I. Hereby we give the next proposition which is an improvement of this result. 3.12. Proposition. For a monoid S the following are equivalent: i) the one element right S-act ΘS is super flat, ii) the one element right S-act ΘS is product flat, iii) products of indecomposable left S-acts are indecomposable. iv) S contains a left zero. i =⇒ ii is trivial. ii =⇒ iii follows immediately by Proposition 3.5 and Proof. [3, Corollary 2.11]. Theorem 3.9 implies the equivalence of iii and iv and iv =⇒ i follows by [3, Corollary 3.6]. Acknowledgements The authors wish to express their appreciation to the anonymous referee for his/her constructive input into the paper. References [1] Adamek, J., Herrlich, H. and Strecker, G. Abstract and Concrete Categories The Joy of Cats (John Wiley and Sons, New York, 1990). 10 [2] Bulman-Fleming, S. Products of projective S-systems, Comm. Algebra 19 (3), 951 -- 964, 1991. [3] Bulman-Fleming, S. and Laan, V. Tensor products and preservation of limits, for acts over monoids, Semigroup Forum 63, 161 -- 179, 2001. [4] Bulman-Fleming, S and McDowell, K. Coherent monoids, in: Latices, Semigroups and Universal Algebra, ed. J. Almeida et al. (Plenum Press, New York, 1990). [5] Gould, V. Coherent monoids, J. Austral. Math. Soc. 53(Series A), 166 -- 182, 1992. [6] Kilp, M., Knauer, U. and Mikhalev, A. Monoids, Acts and Categories, (W. de gruyter, Berlin, 2000). [7] Nico, W.R. A classification of indecomposable S-sets, J. Algebra 54(1), 260 -- 272, 1978. [8] Renshaw, J. Monoids for which condition (P)acts are projective, Semigroup Forum 61(1), 46 -- 56, 1998. [9] Sedaghatjoo, M., Khosravi, R. and Ershad, M. Principally weakly and weakly coherent monoids, Comm. Algebra 37(12), 4281 -- 4295, 2009.
1210.1821	1	1210	2012-10-05T17:35:19	Nijenhuis algebras, NS algebras and N-dendriform algebras	[ "math.RA", "math.CO", "math.QA" ]	In this paper we study (associative) Nijenhuis algebras, with emphasis on the relationship between the category of Nijenhuis algebras and the categories of NS algebras. This is in analogy to the well-known theory of the adjoint functor from the category of Lie algebras to that of associative algebras, and the more recent results on the adjoint functor from the categories of dendriform and tridendriform algebras to that of Rota-Baxter algebras. We first give an explicit construction of free Nijenhuis algebras and then apply it to obtain the universal enveloping Nijenhuis algebra of an NS algebra. We further apply the construction to determine the binary quadratic nonsymmetric algebra, called the N-dendriform algebra, that is compatible with the Nijenhuis algebra. As it turns out, the N-dendriform algebra has more relations than the NS algebra.	math.RA	math	NIJENHUIS ALGEBRAS, NS ALGEBRAS AND N-DENDRIFORM ALGEBRAS PENG LEI AND LI GUO Abstract. In this paper we study (associative) Nijenhuis algebras, with emphasis on the relation- ship between the category of Nijenhuis algebras and the categories of NS algebras. This is in analogy to the well-known theory of the adjoint functor from the category of Lie algebras to that of associative algebras, and the more recent results on the adjoint functor from the categories of dendriform and tridendriform algebras to that of Rota-Baxter algebras. We first give an explicit construction of free Nijenhuis algebras and then apply it to obtain the universal enveloping Nijen- huis algebra of an NS algebra. We further apply the construction to determine the binary quadratic nonsymmetric algebra, called the N-dendriform algebra, that is compatible with the Nijenhuis al- gebra. As it turns out, the N-dendriform algebra has more relations than the NS algebra. Contents Introduction 1. 2. Free Nijenhuis algebra on an algebra 2.1. A basis of the free Nijenhuis algebra 2.2. The product in a free Nijenhuis algebra 2.3. The proof of Theorem 2.4 3. NS algebras and their universal enveloping algebras 4. From Nijenhuis algebras to N-dendriform algebras 4.1. Background and the statement of Theorem 4.2 4.2. The proof of Theorem 4.2 References 1 3 3 4 6 9 11 11 13 15 1. Introduction Through the antisymmetry bracket [x, y] := xy − yx, an associative algebra A defines a Lie algebra structure on A. The resulting functor from the category of associative algebras to that of Lie algebras and its adjoint functor have played a fundamental role in the study of these algebraic structures. A similar relationship holds for Rota-Baxter algebras and dendriform algebras. This paper studies a similar relationship between (associative) Nijenhuis algebras and NS al- gebras. A Nijenhuis algebra is a nonunitary associative algebra N with a linear endomorphism P satisfying the Nijenhuis equation: P(x)P(y) = P(P(x)y) + P(xP(y)) − P2(xy), (1) The concept of a Nijenhuis operator on a Lie algebra originated from the important concept of a Nijenhuis tensor that was introduced by Nijenhuis [25] in the study of pseudo-complex mani- folds in the 1950s and was related to the well-known concepts of Schouten-Nijenhuis bracket, the ∀x, y ∈ N. 1 2 PENG LEI AND LI GUO Frolicher-Nijenhuis bracket [13] and the Nijenhuis-Richardson bracket. Nijenhuis operator oper- ators on a Lie algebra appeared in [20] in a more general study of Poisson-Nijenhuis manifolds and then more recently in [14, 15] in the context of the classical Yang-Baxter equation. The Nijenhuis operator on an associative algebra was introduced by Carinena and coauthors [4] to study quantum bi-Hamiltonian systems. In [26], Nijenhuis operators are constructed by anal- ogy with Poisson-Nijenhuis geometry, from relative Rota-Baxter operators. Note the close analogue of the Nijenhuis operator with the more familiar Rota-Baxter opera- tor of weight λ (where λ is a constant) defined to be a linear endomorphism P on an associative algebra R satisfying P(x)P(y) = P(P(x)y) + P(xP(y)) + λP(xy), ∀x, y ∈ R. The latter originated from the probability study of G. Baxter [2], was studied by Cartier and Rota and is closely related to the operator form of the classical Yang-Baxter equation. Its study has experienced a quite remarkable renascence in the last decade with many applications in mathe- matics and physics, most notably the work of Connes and Kreimer on renormalization of quantum field theory [5, 10, 11]. See [16] for further details and references. The recent theoretic developments of Nijenhuis algebras have largely followed those of Rota- Baxter algebras. Commutative Nijenhuis algebras were constructed in [7, 12] following the con- struction of free commutative Rota-Baxter algebras [17]. Another development followed the relationship between Rota-Baxter algebras and dendriform algebras. Recall that a dendriform algebra, defined by Loday [22], is a vector space D with two binary operations ≺ and ≻ such that (x ≺ y) ≺ z = x ≺ (y ⋆ z), (x ≻ y) ≺ z = x ≻ (y ≺ z), (x ⋆ y) ≻ z = x ≻ (y ≻ z), x, y, z ∈ D, where ⋆ :=≺ + ≻. Similarly a tridendriform algebra, defined by Loday and Ronco [23], is a vector space T with three binary operations ≺, ≻ and · that satisfy seven relations. Aguiar [1] showed that for a Rota-Baxter algebra (R, P) of weight 0, the binary operations x ≺P y := xP(y), x ≻P y := P(x)y, ∀x, y ∈ R, define a dendriform algebra on R. Similarly, Ebrahimi-Fard [6] showed that, for a Rota-Baxter algebra (R, P) of non-zero weight, the binary operations x ≺P y := xP(y), x ≻P y := P(x)y, x ·P y := λxy, ∀x, y ∈ R, define a tridendriform algebra on R. As an analogue of the tridendriform algebra, the concept of an NS algebra was introduced by Leroux [21], to be a vector space M with three binary operations ≺, ≻ and • that satisfy four relations (see Eq. (16)). As an analogue of the Rota-Baxter algebra case, it was shown [21] that, for a Nijenhuis algebra (N, P), the binary operations x ≺P y := xP(y), x ≻P y := P(x)y, x •P y := −P(xy), ∀x, y ∈ R, defines an NS algebra on R. Considering the adjoint functor of the functor induced by the above mentioned map from Rota- Baxter algebras to (tri-)dendriform algebras, the Rota-Baxter universal enveloping algebra of a (tri-)dendriform algebra was constructed in [8]. For this purpose, free Rota-Baxter algebras was first constructed. In this paper we give a similar approach for Nijenhuis algebras, but we go beyond the case of Rota-Baxter algebras. Our first goal is to give an explicit construction of free Nijenhuis algebras NIJENHUIS ALGEBRAS, NS ALGEBRAS AND N-DENDRIFORM ALGEBRAS 3 in Section 2. We consider both the cases when the free Nijenhuis algebra is generated by a set and by another algebra. Other than its role in the theoretical study of Nijenhuis algebras, this construction allows us to construct the universal enveloping algebra of an NS algebra. We achieve this in Section 3. Knowing that a Nijenhuis algebra gives an NS algebra, it is natural to ask what other dendriform type algebras that Nijenhuis algebras can give in a similar way. As a second application of our construction of free Nijenhuis algebras, we determine all "quadratic nonsymmetric" relations that can be derived from Nijenhuis algebras and find that one can actually derive more relations than defined by the NS algebra in Eq. (16). This discussion is presented in Section 4. Notation: In this paper k is taken to be a field. A k-algebra is taken to be nonunitary associative unless otherwise stated. 2. Free Nijenhuis algebra on an algebra We start with the definition of free Nijenhuis algebras. Definition 2.1. Let A be a k-algebra. A free Nijenhuis algebra over A is a Nijenhuis algebra FN(A) with a Nijenhuis operator PA and an algebra homomorphism jA : A → FN(A) such that, for any Nijenhuis algebra N and any algebra homomorphism f : A → N, there is a unique Nijenhuis algebra homomorphism ¯f : FN(A) → N such that ¯f ◦ jA = f : A jA FN(A) f '❖❖❖❖❖❖❖❖❖❖❖❖❖❖❖ ¯f N For the construction of free Nijenhuis algebras, we follow the construction of free Rota-Baxter algebras [8, 16] by bracketed words. Alternatively, one can follow [9] to give the construction by rooted trees that is more in the spirit of operads [24]. One can also follow the approach of Grobner-Shirshov bases [3]. Because of the lack of a uniform approach (see [18, 19] for some recent attempts in this direction) and to be notationally self contained, we give some details. We first display a k-basis of the free Nijenhuis algebra in terms of bracketed words in § 2.1. The product on the free Nijenhuis algebra is given in § 2.2 and the universal property of the free Nijenhuis algebra is proved in § 2.3. 2.1. A basis of the free Nijenhuis algebra. Let A be a k-algebra with a k-basis X. We first display a k-basis X∞ of FN(A) in terms of bracketed words from the alphabet set X. Let ⌊ and ⌋ be symbols, called brackets, and let X′ = X ∪ {⌊, ⌋}. Let M(X′) denote the free semigroup generated by X′. Definition 2.2. ([8, 16]) Let Y, Z be two subsets of M(X′). Define the alternating product of Y and Z to be (2) Λ(Y, Z) = (cid:16)[r≥1 (cid:0)Y⌊Z⌋(cid:1)r(cid:17)[(cid:16)[r≥0 (cid:0)Y⌊Z⌋(cid:1)rY(cid:17)[(cid:16)[r≥1 (cid:0)⌊Z⌋Y(cid:1)r(cid:17)[(cid:16)[r≥0 (cid:0)⌊Z⌋Y(cid:1)r ⌊Z⌋(cid:17). We construct a sequence Xn of subsets of M(X′) by the following recursion. Let X0 = X and, for n ≥ 0, define Xn+1 = Λ(X, Xn). / / ' 4 Further, define (3) PENG LEI AND LI GUO X∞ = [n≥0 Xn = lim −→ Xn. Here the second equation in Eq. (3) follows since X1 ⊇ X0 and, assuming Xn ⊇ Xn−1, we have By [8, 16] we have the disjoint union Xn+1 = Λ(X, Xn) ⊇ Λ(X, Xn−1) ⊇ Xn. (4) X∞ = (cid:16)Gr≥1 (cid:0)X⌊X∞⌋(cid:1)r(cid:17)G(cid:16)Gr≥0 (cid:0)X⌊X∞⌋(cid:1)rX(cid:17) G(cid:16)Gr≥1 (cid:0)⌊X∞⌋X(cid:1)r(cid:17)G(cid:16)Gr≥0 (cid:0)⌊X∞⌋X(cid:1)r ⌊X∞⌋(cid:17). Further, every x ∈ X∞ has a unique decomposition (5) where xi, 1 ≤ i ≤ b, is alternatively in X or in ⌊X∞⌋. This decomposition will be called the standard decomposition of x. x = x1 · · · xb, For x in X∞ with standard decomposition x1 · · · xb, we define b to be the breadth b(x) of x, we define the head h(x) of x to be 0 (resp. 1) if x1 is in X (resp. in ⌊X∞⌋). Similarly define the tail t(x) of x to be 0 (resp. 1) if xb is in X (resp. in ⌊X∞⌋). 2.2. The product in a free Nijenhuis algebra. Let FN(A) = Mx∈X∞ kx. We now define a product ⋄ on FN(A) by defining x ⋄ x′ ∈ FN(A) for x, x′ ∈ X∞ and then extending bilinearly. Roughly speaking, the product of x and x′ is defined to be the concatenation whenever t(x) , h(x′). When t(x) = h(x′), the product is defined by the product in A or by the Nijenhuis relation in Eq. (1). To be precise, we use induction on the sum n := d(x) + d(x′) of the depths of x and x′. Then n ≥ 0. If n = 0, then x, x′ are in X and so are in A and we define x ⋄ x′ = x · x′ ∈ A ⊆ FN(A). Here · is the product in A. Suppose x ⋄ x′ have been defined for all x, x′ ∈ X∞ with n ≥ k ≥ 0 and let x, x′ ∈ X∞ with n = k + 1. First assume the breadth b(x) = b(x′) = 1. Then x and x′ are in X or ⌊X∞⌋. Since n = k + 1 is at least one, x and x′ cannot be both in X. We accordingly define (6) x ⋄ x′ =  xx′, xx′, ⌊⌊x⌋ ⋄ x′⌋ + ⌊x ⋄ ⌊x′⌋⌋ − ⌊⌊x ⋄ x′⌋⌋, if x ∈ X, x′ ∈ ⌊X∞⌋, if x ∈ ⌊X∞⌋, x′ ∈ X, if x = ⌊x⌋, x′ = ⌊x′⌋ ∈ ⌊X∞⌋. Here the product in the first and second case are by concatenation and in the third case is by the induction hypothesis since for the three products on the right hand side we have d(⌊x⌋) + d(x′) = d(⌊x⌋) + d(⌊x′⌋) − 1 = d(x) + d(x′) − 1, d(x) + d(⌊x′⌋) = d(⌊x⌋) + d(⌊x′⌋) − 1 = d(x) + d(x′) − 1, d(x) + d(x′) = d(⌊x⌋) − 1 + d(⌊x′⌋) − 1 = d(x) + d(x′) − 2 which are all less than or equal to k. NIJENHUIS ALGEBRAS, NS ALGEBRAS AND N-DENDRIFORM ALGEBRAS 5 Now assume b(x) > 1 or b(x′) > 1. Let x = x1 · · · xb and x′ = x′ 1 · · · x′ b′ be the standard decompositions from Eq. (5). We then define (7) where xb ⋄ x′ is well-defined since by Eq. (6), we have h(xb) = h(xb ⋄ x′ t(xb−1) , h(xb ⋄ x′ x ⋄ x′ = x1 · · · xb−1(xb ⋄ x′ 2) , t(xb ⋄ x′ 1) and h(x′ 1). 1) x′ 2 · · · x′ b′ 1 is defined by Eq. (6) and the rest is given by concatenation. The concatenation 1). Therefore, 1) = t(xb ⋄ x′ 1) and t(x′ We have the following simple properties of ⋄. Lemma 2.3. Let x, x′ ∈ X∞. We have the following statements. (a) h(x) = h(x ⋄ x′) and t(x′) = t(x ⋄ x′). (b) If t(x) , h(x′), then x ⋄ x′ = xx′ (concatenation). (c) If t(x) , h(x′), then for any x′′ ∈ X∞, (xx′) ⋄ x′′ = x(x′ ⋄ x′′), x′′ ⋄ (xx′) = (x′′ ⋄ x)x′. Extending ⋄ bilinearly, we obtain a binary operation FN(A) ⊗ FN(A) → FN(A). For x ∈ X∞, define (8) Obviously ⌊x⌋ is again in X∞. Thus NA extends to a linear operator NA on FN(A). Let NA(x) = ⌊x⌋. jX : X → X∞ → FN(A) be the natural injection which extends to an algebra injection (9) jA : A → FN(A). The following is our first main result which will be proved in the next subsection. Theorem 2.4. Let A be a k-algebra with a k-basis X. (a) The pair (FN(A), ⋄) is an algebra. (b) The triple (FN(A), ⋄, NA) is a Nijenhuis algebra. (c) The quadruple (FN(A), ⋄, NA, jA) is the free Nijenhuis algebra on the algebra A. The following corollary of the theorem will be used later in the paper. Corollary 2.5. Let M be a k-module and let T (M) = Ln≥1 M⊗n be the reduced tensor algebra jT (M) over M. Then FN(T (M)), together with the natural injection iM : M → T (M) −−−→ FN(T (M)), is a free Nijenhuis algebra over M, in the sense that, for any Nijenhuis algebra N and k-module map f : M → N there is a unique Nijenhuis algebra homomorphism f : FN(T (M)) → N such that f ◦ kM = f . Proof. This follows immediately from Theorem 2.4 and the fact that the construction of the free algebra on a module (resp. free Nijenhuis algebra on an algebra, resp. free Nijenhuis on a module) is the left adjoint functor of the forgetful functor from algebras to modules (resp. from Nijenhuis algebras to algebras, resp. from Nijenhuis algebras to modules), and the fact that the composition of two left adjoint functors is the left adjoint functor of the composition. (cid:3) 6 PENG LEI AND LI GUO 2.3. The proof of Theorem 2.4. Proof. (a). We just need to verify the associativity. For this we only need to verify (10) for x′, x′′, x′′′ ∈ X∞. We will do this by induction on the sum of the depths n := d(x′) + d(x′′) + d(x′′′). If n = 0, then all of x′, x′′, x′′′ have depth zero and so are in X. In this case the product ⋄ is given by the product · in A and so is associative. (x′ ⋄ x′′) ⋄ x′′′ = x′ ⋄ (x′′ ⋄ x′′′) Assume the associativity holds for n ≤ k and assume that x′, x′′, x′′′ ∈ X∞ have n = d(x′) + d(x′′) + d(x′′′) = k + 1. If t(x′) , h(x′′), then by Lemma 2.3, (x′ ⋄ x′′) ⋄ x′′′ = (x′x′′) ⋄ x′′′ = x′(x′′ ⋄ x′′′) = x′ ⋄ (x′′ ⋄ x′′′). A similar argument holds when t(x′′) , h(x′′′). Thus we only need to verify the associativity when t(x′) = h(x′′) and t(x′′) = h(x′′′). We next reduce the breadths of the words. Lemma 2.6. If the associativity (x′ ⋄ x′′) ⋄ x′′′ = x′ ⋄ (x′′ ⋄ x′′′) holds for all x′, x′′ and x′′′ in X∞ of breadth one, then it holds for all x′, x′′ and x′′′ in X∞. Proof. We use induction on the sum of breadths m := b(x′) + b(x′′) + b(x′′′). Then m ≥ 3. The case when m = 3 is the assumption of the lemma. Assume the associativity holds for 3 ≤ m ≤ j and take x′, x′′, x′′′ ∈ X∞ with m = j + 1. Then j + 1 ≥ 4. So at least one of x′, x′′, x′′′ have breadth greater than or equal to 2. First assume b(x′) ≥ 2. Then x′ = x′ 1x′ 2 with x′ 1, x′ 2 ∈ X∞ and t(x′ 1) , h(x′ 2). Thus by Lemma 2.3, we obtain (x′ ⋄ x′′) ⋄ x′′′ = ((x′ 1x′ 2) ⋄ x′′) ⋄ x′′′ = (x′ 1(x′ 2 ⋄ x′′)) ⋄ x′′′ = x′ 1((x′ 2 ⋄ x′′) ⋄ x′′′). Similarly, x′ ⋄ (x′′ ⋄ x′′′) = (x′ 1x′ 2) ⋄ (x′′ ⋄ x′′′) = x′ 1(x′ 2 ⋄ (x′′ ⋄ x′′′)). Thus (x′ ⋄ x′′) ⋄ x′′′ = x′ ⋄ (x′′ ⋄ x′′′) whenever (x′ 2 ⋄ x′′) ⋄ x′′′ = x′ from the induction hypothesis. A similar proof works if b(x′′′) ≥ 2. 2 ⋄ (x′′ ⋄ x′′′). The latter follows Finally if b(x′′) ≥ 2, then x′′ = x′′ 1 x′′ 2 with x′′ 1 , x′′ 2 ∈ X∞ and t(x′′ 1 ) , h(x′′ 2 ). By applying Lemma 2.3 repeatedly, we obtain (x′ ⋄ x′′) ⋄ x′′′ = (x′ ⋄ (x′′ 1 x′′ 2 )) ⋄ x′′′ = ((x′ ⋄ x′′ 1 )x′′ 2 ) ⋄ x′′′ = (x′ ⋄ x′′ 1 )(x′′ 2 ⋄ x′′′). In the same way, we have (x′ ⋄ x′′ 1 )(x′′ 2 ⋄ x′′′) = x′ ⋄ (x′′ ⋄ x′′′). This again proves the associativity. (cid:3) To summarize, our proof of the associativity has been reduced to the special case when x′, x′′, x′′′ ∈ X∞ are chosen so that (a) n := d(x′) + d(x′′) + d(x′′′) = k + 1 ≥ 1 with the assumption that the associativity holds when n ≤ k. (b) the elements have breadth one and (c) t(x′) = h(x′′) and t(x′′) = h(x′′′). NIJENHUIS ALGEBRAS, NS ALGEBRAS AND N-DENDRIFORM ALGEBRAS 7 By item (b), the head and tail of each of the elements are the same. Therefore by item (c), either all the three elements are in X or they are all in ⌊X∞⌋. If all of x′, x′′, x′′′ are in X, then as already shown, the associativity follows from the associativity in A. So it remains to consider the case when x′, x′′, x′′′ are all in ⌊X∞⌋. Then x′ = ⌊x′⌋, x′′ = ⌊x′′⌋, x′′′ = ⌊x′′′⌋ with x′, x′′, x′′′ ∈ X∞. Using Eq. (6) and bilinearity of the product ⋄, we have (x′ ⋄ x′′) ⋄ x′′′ = (cid:0)⌊⌊x′⌋ ⋄ x′′⌋ + ⌊x′ ⋄ ⌊x′′⌋⌋ − ⌊⌊x′ ⋄ x′′⌋⌋(cid:1) ⋄ ⌊x′′′⌋ = ⌊⌊x′⌋ ⋄ x′′⌋ ⋄ ⌊x′′′⌋ + ⌊x′ ⋄ ⌊x′′⌋⌋ ⋄ ⌊x′′′⌋ − ⌊⌊x′ ⋄ x′′⌋⌋ ⋄ ⌊x′′′⌋ = ⌊⌊⌊x′⌋ ⋄ x′′⌋ ⋄ x′′′⌋ + ⌊(cid:0)⌊x′⌋ ⋄ x′′(cid:1) ⋄ ⌊x′′′⌋⌋ − ⌊⌊(cid:0)⌊x′⌋ ⋄ x′′(cid:1) ⋄ x′′′⌋⌋ +⌊⌊x′ ⋄ ⌊x′′⌋⌋ ⋄ x′′′⌋ + ⌊(cid:0)x′ ⋄ ⌊x′′⌋(cid:1) ⋄ ⌊x′′′⌋⌋ − ⌊⌊(cid:0)x′ ⋄ ⌊x′′⌋(cid:1) ⋄ x′′′⌋⌋ −⌊⌊⌊x′ ⋄ x′′⌋⌋ ⋄ x′′′⌋ − ⌊⌊x′ ⋄ x′′⌋ ⋄ ⌊x′′′⌋⌋ + ⌊⌊⌊x′ ⋄ x′′⌋ ⋄ x′′′⌋⌋. Applying the induction hypothesis in n to the fifth term (cid:0)x′ ⋄ ⌊x′′⌋(cid:1) ⋄ ⌊x′′′⌋ and the eighth term, and then use Eq. (6) again, we obtain (x′ ⋄ x′′) ⋄ x′′′ = ⌊⌊⌊x′⌋ ⋄ x′′⌋ ⋄ x′′′⌋ + ⌊(cid:0)⌊x′⌋ ⋄ x′′(cid:1) ⋄ ⌊x′′′⌋⌋ − ⌊⌊(cid:0)⌊x′⌋ ⋄ x′′(cid:1) ⋄ x′′′⌋⌋ −⌊x′ ⋄ ⌊⌊x′′ ⋄ x′′′⌋⌋⌋ − ⌊⌊(cid:0)x′ ⋄ ⌊x′′⌋(cid:1) ⋄ x′′′⌋⌋ − ⌊⌊⌊x′ ⋄ x′′⌋⌋ ⋄ x′′′⌋ +⌊⌊x′ ⋄ ⌊x′′⌋⌋ ⋄ x′′′⌋ + ⌊x′ ⋄ ⌊⌊x′′⌋ ⋄ x′′′⌋⌋ + ⌊x′ ⋄ ⌊x′′ ⋄ ⌊x′′′⌋⌋⌋ −⌊⌊⌊x′ ⋄ x′′⌋ ⋄ x′′′⌋⌋ − ⌊⌊(x′ ⋄ x′′) ⋄ ⌊x′′′⌋⌋⌋ + ⌊⌊⌊(x′ ⋄ x′′) ⋄ x′′′⌋⌋⌋ +⌊⌊⌊x′ ⋄ x′′⌋ ⋄ x′′′⌋⌋ +⌊⌊x′ ⋄ ⌊x′′⌋⌋ ⋄ x′′′⌋ + ⌊x′ ⋄ ⌊⌊x′′⌋ ⋄ x′′′⌋⌋ + ⌊x′ ⋄ ⌊x′′ ⋄ ⌊x′′′⌋⌋⌋ = ⌊⌊⌊x′⌋ ⋄ x′′⌋ ⋄ x′′′⌋ + ⌊(cid:0)⌊x′⌋ ⋄ x′′(cid:1) ⋄ ⌊x′′′⌋⌋ − ⌊⌊(cid:0)⌊x′⌋ ⋄ x′′(cid:1) ⋄ x′′′⌋⌋ −⌊x′ ⋄ ⌊⌊x′′ ⋄ x′′′⌋⌋⌋ − ⌊⌊(cid:0)x′ ⋄ ⌊x′′⌋(cid:1) ⋄ x′′′⌋⌋ − ⌊⌊⌊x′ ⋄ x′′⌋⌋ ⋄ x′′′⌋ −⌊⌊(x′ ⋄ x′′) ⋄ ⌊x′′′⌋⌋⌋ + ⌊⌊⌊(x′ ⋄ x′′) ⋄ x′′′⌋⌋⌋. By a similar computation, we obtain x′ ⋄(cid:0)x′′ ⋄ x′′′(cid:1) = ⌊⌊⌊x′⌋ ⋄ x′′⌋ ⋄ x′′′⌋ + ⌊⌊x′ ⋄ ⌊x′′⌋⌋ ⋄ x′′′⌋ − ⌊⌊⌊x′ ⋄ x′′⌋⌋ ⋄ x′′′⌋ +⌊x′ ⋄ ⌊⌊x′′⌋ ⋄ x′′′⌋⌋ − ⌊⌊x′ ⋄(cid:0)⌊x′′⌋ ⋄ x′′′(cid:1)⌋ +⌊⌊x′⌋ ⋄(cid:0)x′′ ⋄ ⌊x′′′⌋(cid:1)⌋ + ⌊x′ ⋄ ⌊x′′ ⋄ ⌊x′′′⌋⌋⌋ − ⌊⌊x′ ⋄(cid:0)x′′ ⋄ ⌊x′′′⌋(cid:1)⌋⌋ −⌊⌊⌊x′⌋ ⋄ (x′′ ⋄ x′′′)⌋⌋ + ⌊⌊⌊x′ ⋄ (x′′ ⋄ x′′′)⌋⌋⌋ − ⌊x′ ⋄ ⌊⌊x′′ ⋄ x′′′⌋⌋⌋. Now by induction, the i-th term in the expansion of (x′ ⋄ x′′) ⋄ x′′′ matches with the σ(i)-th term in the expansion of x′ ⋄ (x′′ ⋄ x′′′). Here the permutation σ ∈ Σ11 is given by (11) σ = 1 2 3 4 5 6 1 6 9 2 4 7 11 5 3 7 8 9 10 11 10 ! . 8 This completes the proof of Theorem 2.4.(a). (b). The proof follows from the definition NA(x) = ⌊x⌋ and Eq. (6). (c). Let (N, ∗, P) be a Nijenhuis algebra with multiplication ∗. Let f : A → N be a k-algebra homomorphism. We will construct a k-linear map ¯f : FN(A) → N by defining ¯f (x) for x ∈ X∞. We achieve this by defining ¯f (x) for x ∈ Xn, n ≥ 0, inductively on n. For x ∈ X0 := X, define ¯f (x) = f (x). Suppose ¯f (x) has been defined for x ∈ Xn and consider x in Xn+1 which is, by definition and Eq. (4), Λ(X, Xn) = (cid:16)Gr≥1 (X⌊Xn⌋)r(cid:17)G(cid:16)Gr≥0 (X⌊Xn⌋)rX(cid:17) 8 PENG LEI AND LI GUO ⌊Xn⌋(X⌊Xn⌋)r(cid:17)G(cid:16)Gr≥0 Let x be in the first union component Fr≥1(X⌊Xn⌋)r above. Then G(cid:16)Gr≥0 r ⌊Xn⌋(X⌊Xn⌋)rX(cid:17). x = (x2i−1⌊x2i⌋) Yi=1 for x2i−1 ∈ X and x2i ∈ Xn, 1 ≤ i ≤ r. By the construction of the multiplication ⋄ and the Nijenhuis operator NA, we have x = ⋄r i=1(x2i−1 ⋄ ⌊x2i⌋) = ⋄r i=1(x2i−1 ⋄ NA(x2i)). ¯f (x) = ∗r i=1(cid:0) ¯f (x2i−1) ∗ N(cid:0) ¯f (x2i))(cid:1). Define (12) where the right hand side is well-defined by the induction hypothesis. Similarly define ¯f (x) if x is in the other union components. For any x ∈ X∞, we have PA(x) = ⌊x⌋ ∈ X∞, and by the definition of ¯f in (Eq. (12)), we have (13) So ¯f commutes with the Nijenhuis operators. Combining this equation with Eq. (12) we see that if x = x1 · · · xb is the standard decomposition of x, then (14) ¯f (x) = ¯f (x1) ∗ · · · ∗ ¯f (xb). ¯f (⌊x⌋) = P( ¯f (x)). Note that this is the only possible way to define ¯f (x) in order for ¯f to be a Nijenhuis algebra homomorphism extending f . It remains to prove that the map ¯f defined in Eq. (12) is indeed an algebra homomorphism. For this we only need to check the multiplicity (15) for all x, x′ ∈ X∞. For this we use induction on the sum of depths n := d(x) + d(x′). Then n ≥ 0. When n = 0, we have x, x′ ∈ X. Then Eq. (15) follows from the multiplicity of f . Assume the multiplicity holds for x, x′ ∈ X∞ with n ≥ k and take x, x′ ∈ X∞ with n = k + 1. Let x = x1 · · · xb and x′ = x′ b′ be the standard decompositions. Since n = k + 1 ≥ 1, at least one of xb and x′ b′ is in ⌊X∞⌋. Then by Eq. (6) we have, ¯f (x ⋄ x′) = ¯f (x) ∗ ¯f (x′) 1 · · · x′ 1⌋⌋ − ⌊⌊xb ⋄ x′ 1⌋⌋(cid:1), 1) by the definition of ¯f . In the third case, In the first two cases, the right hand side is ¯f (xb) ∗ ¯f (x′ we have, by Eq. (13), the induction hypothesis and the Nijenhuis relation of the operator P on N, 1⌋ + ⌊xb ⋄ ⌊x′ 1⌋ ∈ ⌊X∞⌋. if xb ∈ X, x′ if xb ∈ ⌊X∞⌋, x′ if xb = ⌊xb⌋, x′ 1 ∈ ⌊X∞⌋, 1 ∈ X, 1 = ⌊x′ ¯f (xb ⋄ x′ 1), 1), ¯f (xbx′ ¯f (xbx′ ¯f(cid:0)⌊⌊xb⌋ ⋄ x′ 1) =  ¯f(cid:0)⌊⌊xb⌋ ⋄ x′ = ¯f (⌊⌊xb⌋ ⋄ x′ =P( ¯f (⌊xb⌋ ⋄ x′ =P( ¯f (⌊xb⌋) ∗ ¯f (x′ =P(P( ¯f (xb)) ∗ ¯f (x′ =P( ¯f (xb)) ∗ P( ¯f (x′ 1⌋⌋ − ⌊⌊xb ⋄ x′ 1⌋ + ⌊xb ⋄ ⌊x′ 1⌋) + ¯f (⌊xb ⋄ ⌊x′ 1)) + P( ¯f (xb ⋄ ⌊x′ 1⌋⌋) − ¯f (⌊⌊xb ⋄ x′ 1⌋⌋) 1⌋)) − P( ¯f (⌊xb ⋄ x′ 1⌋)) 1⌋⌋(cid:1) 1)) + P( ¯f (xb) ∗ ¯f (⌊x′ 1)) + P( ¯f (xb) ∗ P( ¯f (x′ 1)) 1⌋)) − P(P( ¯f (xb) ∗ ¯f (x′ 1))) 1))) − P(P(( ¯f (xb) ∗ ¯f (x′ 1))) NIJENHUIS ALGEBRAS, NS ALGEBRAS AND N-DENDRIFORM ALGEBRAS 9 = ¯f (⌊xb⌋) ∗ ¯f (⌊x′ = ¯f (xb) ∗ ¯f (x′ 1). 1⌋) Therefore ¯f (xb ⋄ x′ 1). Then 1) = ¯f (xb) ∗ ¯f (x′ ¯f (x ⋄ x′) = ¯f(cid:0)x1 · · · xb−1(xb ⋄ x′ 1)x′ 2 · · · x′ b′(cid:1) = ¯f (x1) ∗ · · · ∗ ¯f (xb−1) ∗ ¯f (xb ⋄ x′ = ¯f (x1) ∗ · · · ∗ ¯f (xb−1) ∗ ¯f (xb) ∗ ¯f (x′ = ¯f (x) ∗ ¯f (x′). 1) ∗ ¯f (x′ 2) · · · ¯f (x′ b′) 2) · · · ¯f (x′ b′) 1) ∗ ¯f (x′ This is what we need. (cid:3) 3. NS algebras and their universal enveloping algebras The concept of an NS algebra was introduced by Leroux [21] as an analogue of the dendriform algebra of Loday [22] and the tridendriform algebra of Loday and Ronco [23]. Definition 3.1. An NS algebra is a module M with three binary operations ≺, ≻ and • that satisfy the following four relations (x ≺ y) ≺ z = x ≺ (y ⋆ z), (x ≻ y) ≺ z = x ≻ (y ≺ z), (16) (x ⋆ y) ≻ z = x ≻ (y ≻ z), (x ⋆ y) • z + (x • y) ≺ z = x ≻ (y • z) + x • (y ⋆ z). for x, y, z ∈ M. Here ⋆ denotes ≺ + ≻ + •. NS algebras share similar properties as dendriform algebras. For example, the operation ⋆ defines an associative operation. Another similarity is the following theorem which is an analogue of the results of Aguiar [1] and Ebrahimi-Fard [6] that a Rota-Baxter algebra gives a dendriform algebra or a tridendriform algebra. Theorem 3.2. ([21]) A Nijenhuis algebra (N, P) defines an NS algebra (N, ≺P, ≻P, •P), where (17) x ≺P y = xP(y), x ≻P y = P(x)y, x •P y = −P(xy). Let NA denote the category of Nijenhuis algebras and let NS denote the category of NS al- gebras. It is easy to see that the map from NA to NS in Theorem 3.2 is compatible with the morphisms in the two categories. Thus we obtain a functor (18) E : NA → NS. We will study its left adjoint functor. Motivated by the enveloping algebra of a Lie algebra and the Rota-Baxter enveloping algebra of a tridendriform algebra [8], we are naturally led to the following definition. Definition 3.3. Let M be an NS-algebra. A universal enveloping Nijenhuis algebra of M is a Nijenhuis algebra UN(M) ∈ NA with a homomorphism ρ : M → UN(M) in NS such that for any N ∈ NA and homomorphism f : M → N in NS, there is a unique f : UN(M) → N in NA such that f ◦ ρ = f . 10 PENG LEI AND LI GUO Let M := (M, ≺, ≻, •) ∈ NS. Let T (M) = Ln≥1 M⊗n be the tensor algebra. Then T (M) is the free algebra generated by the k-module M. By Corollary 2.5, FN(T (M)), with the natural injection iM : M → T (M) → FN(T (M)), is the free Nijenhuis algebra over the vector space M. Let JM be the Nijenhuis ideal of FN(T (M)) generated by the set (19) (cid:8)x ≺ y − xP(y), x ≻ y − P(x)y, x • y = P(x ⊗ y)(cid:12)(cid:12)(cid:12) x, y ∈ M(cid:9) Let π : FN(T (M)) → FN(T (M))/JM be the quotient map. Theorem 3.4. Let (M, ≺, ≻, •) be an NS algebra. The quotient Nijenhuis algebra FN(T (M))/JM, together with ρ := π ◦ iM, is the universal enveloping Nijenhuis algebra of M. Proof. The proof is similar to the case of tridendriform algebras and Rota-Baxter algebras [8]. So we skip some of the details. Let (N, P) be a Nijenhuis algebra and let f : M → N be a homomorphism in NS. More pre- P, •P). We will complete the following commutative cisely, we have f : (M, ≺, ≻, •) → (N, ≺′ P, ≻′ diagram, using notations from Corollary 2.5. (20) M f xN T (M) <② ② kM ② ② ② ② ② ② jT (M) '◆◆◆◆◆◆◆◆◆◆◆ iM FN(T (M)) f f f π FN(T (M))/JM By the universal property of the free algebra T (M) over M, there is a unique homomorphism f : T (M) → N such that f ◦ kM = f . So f (x1 ⊗ · · · ⊗ xn) = f (x1) ∗ · · · ∗ f (xn). Here ∗ is the product in N. Then by the universal property of the free Nijenhuis algebra FN(T (M)) over T (M), there is a unique morphism ¯f : FN(T (M)) → N in NA such that ¯f ◦ jT (M) = f . By Corollary 2.5, ¯f = f . Then (21) f ◦ iM = f ◦ jT (M) ◦ kM = f ◦ kM = f . So for any x, y ∈ M, we check that f (x ≺ y − xP(y)) = 0, f (x ≻ y − P(x)y) = 0, f (x • y − P(x ⊗ y)) = 0. Thus JM is in ker( f ) and there is a morphism f : FN(T (M))/JM → N in NA such that f = f ◦ π. Then by the definition of ρ = π ◦ iM in the theorem and Eq. (21), we have f ◦ ρ = f ◦ π ◦ iM = f ◦ iM = f . This proves the existence of f . Suppose f ′ : FN(T (M))/JM → N is also a homomorphism in NA such that f ′ ◦ ρ = f . Then ( f ′ ◦ π) ◦ iM = f = ( f ◦ π) ◦ iM. By Corollary 2.5, the free Nijenhuis algebra FN(T (M)) over the algebra T (M) is also the free Nijenhuis algebra over the vector space M with respect the natural injection iM. So we have f ′ ◦ π = f ◦ π in NA. Since π is surjective, we have f ′ = f . This proves the uniqueness of f . (cid:3) ' / / < x o o NIJENHUIS ALGEBRAS, NS ALGEBRAS AND N-DENDRIFORM ALGEBRAS 11 4. From Nijenhuis algebras to N-dendriform algebras In this section, we consider an inverse of Theorem 3.2 in the following sense. Suppose (N, P) is a Nijenhuis algebra and define binary operations x ≺P y = xP(y), x ≻P y = P(x)y, x •P y = −P(xy). By Theorem 3.2, the three operations satisfy the NS relations in Eq. (16). Our inverse question is, what other quadratic nonsymmetric relations could (N, ≺P, ≻P, •P) satisfy? We recall some background on binary quadratic nonsymmetric operads in order to make the question precise. We then determine all the quadratic nonsymmetric relations that are consistent with the Nijenhuis operator. 4.1. Background and the statement of Theorem 4.2. For details on binary quadratic nonsym- metric operads, see [16, 24]. Definition 4.1. Let k be a field. (a) A graded vector space is a sequence P := {Pn}n≥0 of k-vector spaces Pn, n ≥ 0. (b) A nonsymmetric (ns) operad is a graded vector space P = {Pn}n≥0 equipped with partial compositions: (22) ◦i := ◦m,n,i : Pm ⊗ Pn −→ Pm+n−1, 1 ≤ i ≤ m, such that, for λ ∈ Pℓ, µ ∈ Pm and ν ∈ Pn, the following relations hold. (i) (λ ◦i µ) ◦i−1+ j ν = λ ◦i (µ ◦ j ν), (ii) (λ ◦i µ) ◦k−1+m ν = (λ ◦k ν) ◦i µ, (iii) There is an element id ∈ P1 such that id ◦ µ = µ and µ ◦ id = µ for µ ∈ Pn, n ≥ 0. 1 ≤ i ≤ ℓ, 1 ≤ j ≤ m. 1 ≤ i < k ≤ ℓ. An ns operad P = {Pn} is called binary if P1 = k.id and Pn, n ≥ 3 are induced from P2 by composition. Then in particular, for the free operad, we have (23) P3 = (P2 ◦1 P2) ⊕ (P2 ◦2 P2), which can be identified with P⊗2 among the binary operations in P2 are derived from P3. 2 ⊕ P⊗2 2 . A binary ns operad P is called quadratic if all relations Thus a binary, quadratic, ns operad is determined by a pair (V, R) where V = P2, called the space of generators, and R is a subspace of V ⊗2 ⊕ V ⊗2, called the space of relations. So we can denote P = P(V)/(R). Note that a typical element of V ⊗2 is of the form Thus a typical element of V ⊗2 ⊕ V ⊗2 is of the form k Pi=1 ⊙ (1) i ⊗ ⊙ (1) (2) i with ⊙ i , ⊙ (2) i ∈ V, 1 ≤ i ≤ k. k Xi=1 ⊙ (1) (2) i ⊗ ⊙ i , m Xj=1  (4) j  ⊙ (3) j ⊗ ⊙ (1) , ⊙ i (2) , ⊙ i , ⊙ (3) j , ⊙ (4) j ∈ V, 1 ≤ i ≤ k, 1 ≤ j ≤ m, k, m ≥ 1. For a given binary quadratic ns operad P = P(V)/(R), a k-vector space A is called a P-algebra if A has binary operations (indexed by) V and if, for k Xi=1 (cid:0) ⊙ (1) (2) i ⊗ ⊙ i , m Xj=1 ⊙ (3) j ⊗ ⊙ (4) j (cid:1) ∈ R ⊆ V ⊗2 ⊕ V ⊗2 12 (1) with ⊙ i (2) , ⊙ i , ⊙ (3) j (24) PENG LEI AND LI GUO , ⊙ (4) j ∈ V, 1 ≤ i ≤ k, 1 ≤ j ≤ m, we have k m (x ⊙ (2) (1) i y) ⊙ i z = Xi=1 Xj=1 x ⊙ (3) j (y ⊙ (4) j z), ∀ x, y, z ∈ A. For example, from Eq. (16) the NS algebras are precisely the P-algebras where P = P(V)/(R) with R being the subspace of V ⊗2 ⊕ V ⊗2 spanned by the four elements (≺ ⊗ ≺, ≺ ⊗⋆), (≻ ⊗ ≺, ≻ ⊗ ≺), (⋆⊗ ≻, ≻ ⊗ ≻), (⋆ ⊗ • + • ⊗ ≺, ≻ ⊗ • + • ⊗⋆), where ⋆ =≺ + ≻ + •. Theorem 4.2. Let V = k{≺, ≻, •} be the vector space with basis {≺, ≻, •} and let P = P(V)/(R) be a binary quadratic ns operad. The following statements are equivalent. (a) For every Nijenhuis algebra (N, P), the quadruple (N, ≺P, ≻P, •P) is a P-algebra. (b) The relation space R of P is contained in the subspace of V ⊗2 ⊕ V ⊗2 spanned by (25) (26) (≺ ⊗ ≺, ≺ ⊗⋆), (≻ ⊗ ≺, ≻ ⊗ ≺), (≻ ⊗⋆, ≻ ⊗ ≻), (≺ ⊗•, • ⊗ ≻), (≻ ⊗ • + • ⊗ ≺ + • ⊗•, ≻ ⊗ • + • ⊗ ≺ + • ⊗•), where ⋆ =≺ + ≻ + • . More precisely, any P-algebra A satisfies the relations (x ≺ y) ≺ z = x ≺ (y ≺ z) + x ≺ (y ≺ z) + x ≺ (y • z), (x ≻ y) ≺ z = x ≻ (y ≺ z), (x ≺ y) ≻ z + (x ≻ y) ≻ z + (x • y) ≻ z = x ≻ (y ≻ z), (x ≺ y) • z = x • (y ≻ z), ∀x, y, z ∈ A (x ≻ y) • z + (x • y) ≺ z + (x • y) • z = x ≻ (y • z) + x • (y ≺ z) + x • (y • z). Note that the relations of the NS algebra in Eq. (16) is contained in the space spanned by the relations in Eq. (25). We call P defined by the relations in Eq. (25) the N-dendriform operad and call a quadruple (A, ≺, ≻, •) satisfying Eq. (26) an N-dendriform algebra. Let ND denote the category of N-dendriform algebras. Then we have the following immediate corollary of The- orem 4.2. Corollary 4.3. (a) There is a natural functor (27) F : NA → ND, (N, P) 7→ (N, ≺P, ≻P, •P). (b) There is a natural (inclusion) functor (28) G : ND → NS, (M, ≺, ≻, •) 7→ (M, ≺, ≻, •). NIJENHUIS ALGEBRAS, NS ALGEBRAS AND N-DENDRIFORM ALGEBRAS 13 (c) The functors F and G give a refinement of the functor E : NA → NS in Eq. (16) in the sense that the following diagram commutes (29) NA F E '❖❖❖❖❖❖❖❖❖❖❖❖❖ ND G NS 4.2. The proof of Theorem 4.2. With V = k{≺, ≻, •}, we have V ⊗2 ⊕ V ⊗2 = M⊙1,⊙2,⊙3,⊙4∈{≺,≻,•} k(⊙1 ⊗ ⊙2, ⊙3 ⊗ ⊙4). Thus any element r of V ⊗2 ⊕ V ⊗2 is of the form r := a1(≺ ⊗ ≺, 0) + a2(≺ ⊗ ≻, 0) + a3(≺ ⊗•, 0) +b1(≻ ⊗ ≺, 0) + b2(≻ ⊗ ≻, 0) + b3(≻ ⊗•, 0) +c1(• ⊗ ≺, 0) + c2(• ⊗ ≻, 0) + c3(• ⊗ •, 0) +d1(0, ≺ ⊗ ≺) + d2(0, ≺ ⊗ ≻) + d3(0, ≺ ⊗•) +e1(0, ≻ ⊗ ≺) + e3(0, ≻ ⊗ ≻) + e3(0, ≻ ⊗•) + f1(0, •⊗ ≺) + f2(0, •⊗ ≻) + f3(0, • ⊗ •) where the coefficients are in k. ((a) ⇒ (b)) Let P = P(V)/(R) be an operad satisfying the condition in Item (a). Let r be in R expressed in the above form. Then for any Nijenhuis algebra (N, P), the quadruple (N, ≺P, ≻P, •P) is a P-algebra. Thus a1(x ≺P y) ≺P z + a2(x ≺P y) ≻P z + a3(x ≺P y) •P z +b1(x ≻P y) ≺P z + b2(x ≻P y) ≻P z + b3(x ≻P y) •P z +c1(x •P y) ≺P z + c2(x •P y) ≻P z + c3(x •P y) •P z +d1x ≻P (y ≻P z) + d2x ≻P (y ≺P z) + d3x ≻P (y •P z) +e1x ≺P (y ≻P z) + e2x ≺P (y ≺P z) + e3x ≺P (y •P z) + f1x •P (y ≻P z) + f2x •P (y ≺P z) + f3x •P (y •P z) = 0, ∀x, y, z ∈ N. By the definitions of ≺P, ≻P, •P in Eq.(17), we have a1xP(y)P(z) + a2P(xP(y))z − a3P(xP(y)z) + b1P(x)yP(z) +b2P(P(x)y)z − b3P(P(x)yz) − c1P(xy)P(z) − c2P(P(xy))z +c3P(P(xy)z) + d1P(x)P(y)z + d2P(x)yP(z) −d3P(x)P(yz) + e1xP(P(y)z) + e2xP(yP(z)) − e3xP(P(yz)) − f1P(xP(y)z) − f2P(xyP(z)) + f3P(xP(yz)) = 0. Since P is a Nijenhuis operator, we further have a1xP(yP(z)) + a1xP(P(y)z) − a1xP2(yz) + a2P(xP(y))z − a3P(xP(y)z) +b1P(x)yP(z) + b2P(P(x)y)z − b3P(P(x)yz) / / ' 14 PENG LEI AND LI GUO −c1P(xyP(z)) − c1P(P(xy)z) + c1P2(xyz) − c2P(P(xy))z + c3P(P(xy)z) +d1P(x)P(y)z + d2P(x)yP(z) − d3P(xP(yz)) − d3P(P(x)yz)) + d3P2(xyz) +e1xP(P(y)z) + e2xP(yP(z)) − e3xP(P(yz)) − f1P(xP(y)z) − f2P(xyP(z)) + f3P(xP(yz)) = 0. Collecting similar terms, we obtain (a1 + e2)xP(yP(z)) + (a1 + e1)xP(P(y)z) − (a1 + e3)xP(P(yz) + (a2 + d1)P(xP(y))z −(a3 + f1)P(xP(y)z) + (b1 + d2)P(x)yP(z) + (b2 + d1)P(P(x)y)z − (b3 + d3)P(P(x)yz) −(c1 + f2)P(xyP(z)) + (c3 − c1)P(P(xy)z) + (c1 + d3)P2(xyz) −(c2 + d1)P(P(xy))z + ( f3 − d3)P(xP(yz)) = 0. Now we take the special case when (N, P) is the free Nijenhuis algebra (FN(T (M)), PT (M)) defined in Corollary 2.5 for our choice of M = k{x, y, z} and PT (M)(u) = ⌊u⌋. Then the above equation is just (a1 + e2)x⌊y⌊z⌋⌋ + (a1 + e1)x⌊⌊y⌋z⌋ − (a1 + e3)x⌊⌊yz⌋ + (a2 + d1)⌊x⌊y⌋⌋z −(a3 + f1)⌊x⌊y⌋z⌋ + (b1 + d2)⌊x⌋y⌊z⌋ + (b2 + d1)⌊⌊x⌋y⌋z − (b3 + d3)⌊⌊x⌋yz⌋ −(c1 + f2)⌊xy⌊z⌋⌋ + (c3 − c1)⌊⌊xy⌋z⌋ + (c1 + d3)⌊⌊xyz⌋⌋ −(c2 + d1)⌊⌊xy⌋⌋z + ( f3 − d3)⌊x⌊yz⌋⌋ = 0. Note that the set of elements x⌊y⌊z⌋⌋, x⌊⌊y⌋z⌋, x⌊⌊yz⌋, ⌊x⌊y⌋⌋z, ⌊x⌊y⌋z⌋, ⌊x⌋y⌊z⌋, ⌊⌊x⌋y⌋z, ⌊⌊x⌋yz⌋, ⌊xy⌊z⌋⌋, ⌊⌊xy⌋z⌋, ⌊⌊xyz⌋⌋, ⌊⌊xy⌋⌋z, ⌊x⌊yz⌋⌋ is a subset of the basis X∞ of the free Nijenhuis algebra FN(T (M)) and hence is linearly indepen- dent. Thus the coefficients must be zero, that is, a1 = −e1 = −e2 = −e3, a2 = b2 = c2 = −d1, a3 = − f1, b1 = −d2, b3 = c1 = c3 = − f2 = − f3 = −d3. Substituting these equations into the general relation r, we find that the any relation r that can be satisfied by ≺P, ≻P, •P for all Nijenhuis algebras (N, P) is of the form r = a1(cid:16)(x ≺ y) ≺ z − x ≺ (y ≺ z) − x ≺ (y ≺ z) − x ≺ (y • z)(cid:17) +b1(cid:16)(x ≻ y) ≺ z − x ≻ (y ≺ z)(cid:17) +d1(cid:16)x ≻ (y ≻ z) − (x ≺ y) ≻ z − (x ≻ y) ≻ z − (x • y) ≻ z(cid:17) +a3(cid:16)(x ≺ y) • z − x • (y ≻ z)(cid:17) +b3(cid:16)(x ≻ y) • z + (x • y) ≺ z + (x • y) • z − x ≻ (y • z) − x • (y ≺ z) − x • (y • z)(cid:17), NIJENHUIS ALGEBRAS, NS ALGEBRAS AND N-DENDRIFORM ALGEBRAS 15 where a1, b1, d1, a3, b3 ∈ k can be arbitrary. Thus r is in the subspace prescribed in Item (b), as needed. ((b) ⇒ (a)) We check directly that all the relations in Eq. (26) are satisfied by (N, ≺P, ≻P, •P) for every Nijenhuis algebra (N, P). First of all (x ≺P y) ≺P z = xP(y)P(z) = xP(yP(z)) + xP(P(y)z) − xP2(yz) = x ≺P (y ≺P z) + x ≺P (y ≻P z) + x ≺P (y •P z), proving the first equation in Eq. (26). The proofs of the second and third equations are similar. For the fourth equation, we have (x ≺P y) •P z = −P((xP(y))z) = −P(x(P(y)z)) = x •P (y ≻P z). Finally for the last equation, we verify and (x ≻P y) •P z + (x •P y) ≺P z + (x •P y) •P z = −P((P(x)y)z) − P(xy)P(z) + P(P(xy)z) = −P(P(x)yz) − P(xyP(z)) − P(P(xy)z) + P2(xyz) + P(P(xy)z) = −P(P(x)yz) − P(xyP(z)) + P2(xyz), x ≻P (y •P z) + x •P (y ≺P z) + x •P (y •P z) = −P(x)P(yz) − P(x(yP(z))) + P(xP(yz)) = −P(xP(yz)) − P(P(x)yz) + P2(xyz) − P(xyP(z)) + P(xP(yz)) = −P(P(x)yz) + P2(xyz) − P(xyP(z)). So the two sides of the last equation agree. Thus if the relation space R of an operad P = P(V)/(R) is contained in the subspace spanned by the vectors in Eq. (25), then the corresponding relations are linear combinations of the equations in Eq. (26) and hence are satisfied by (N, ≺P, ≻P, •P) for each Nijenhuis algebra (N, P). Therefore (N, ≺P, ≻P, •P) is a P-algebra. This completes the proof of Theorem 4.2. References [1] M. Aguiar, On the associative analog of Lie bialgebras, Journal of Algebra, 244, (2001), 492-532. 2, 9 [2] G. Baxter, An analytic problem whose solution follows from a simple algebraic identity, Pacific J. Math., 10, (1960), 731-742. 2 [3] L. A. Bokut, Y. Chen and J. Qiu, Grobner-Shirshov bases for associative algebras with multiple operators and free Rota-Baxter algebras, J. Pure Appl. Algebra 214 (2010) 89-110. 3 [4] J. Carinena, J. Grabowski and G. Marmo, Quantum bi-Hamiltonian systems, Internat. J. Modern Phys. A, 15, (2000), 4797 -- 4810. 2 [5] A. Connes and D. Kreimer, Renormalization in quantum field theory and the Riemann-Hilbert problem. I. The Hopf algebra structure of graphs and the main theorem., Comm. Math. Phys., 210, (2000), no. 1, 249-273. 2 [6] K. Ebrahimi-Fard, Loday-type algebras and the Rota-Baxter relation, Lett. Math. Phys. 61 (2002) 139- 147. 2, 9 [7] K. Ebrahimi-Fard, On the associative Nijenhuis relation, The Electronic Journal of Combinatorics, Vol- ume 11(1), R38, (2004). 2 [8] K. Ebramihi-Fard and L. Guo, Rota-Baxter algebras and dendriform algebras, Jour. Pure Appl. Algebra 212 (2008), 320-339. 2, 3, 4, 9, 10 16 PENG LEI AND LI GUO [9] K. Ebramihi-Fard and L. Guo, Free Rota-Baxter algebras and rooted trees, J. Algebra and Its Applications 7 (2008), 167-194. 3 [10] K. Ebrahimi-Fard, L. Guo and D. Kreimer, Spitzer's Identity and the Algebraic Birkhoff Decomposition in pQFT, J. Phys. A: Math. Gen., 37, (2004), 11037-11052. 2 [11] K. Ebrahimi-Fard, L. Guo and D. Manchon, Birkhoff type decompositions and the Baker-Campbell- Hausdorff recursion, Comm. Math. Phys., 267, (2006), 821-845. arXiv:math-ph/0602004. 2 [12] K. Ebrahimi-Fard and P. Leroux, Generalized shuffles related to Nijenhuis and TD-algebras, Comm. Algebra 37 (2009) 3065-3094. 2 [13] A. Frolicher and A. Nijenhuis, Theory of vector valued differential forms. Part I, Indag. Math. 18 (1956) 338-360. 2 [14] I. Z. Golubchik and V.V. Sokolov, One more type of classical Yang-Baxter equation, Funct. Anal. Appl. 34 (2000), 296-298. 2 [15] I. Z. Golubchik and V.V. Sokolov, Generalized Operator Yang-Baxter Equations, Integrable ODEs and Nonassociative Algebras, J. of Nonlinear Math. Phys. 7 (2000), 184-197. 2 [16] L. Guo, An Introduction to Rota-Baxter Algebras, to be published by Higher Education Press (China) and International Press (US). 2, 3, 4, 11 [17] L. Guo and W. Keigher, Baxter algebras and shuffle products, Adv. Math., 150, (2000), 117-149. 2 [18] L. Guo, W. Sit and R. Zhang, On Rota's problem for linear operators in associative algebras, Proc. ISSAC 2011, 147-154. 3 [19] L. Guo, W. Sit and R. Zhang, Differential type operators and Grobner-Shirshov bases, preprint. 3 [20] Y. Kosmann-Schwarzbach and F. Magri, Poisson-Nijenhuis structures, Ann. Inst. Henri Poincar´e 53 (1990), 35-81. 2 [21] P. Leroux, Construction of Nijenhuis operators and dendriform trialgebras, Int. J. Math. Math. Sci. (2004), no. 40-52, 2595-2615, arXiv:math.QA/0311132 2, 9 [22] J.-L. Loday, Dialgebras, in Dialgebras and related operads, Lecture Notes in Math., 1763, (2001), 7-66. 2, 9 [23] J.-L. Loday and M. Ronco, Trialgebras and families of polytopes, in "Homotopy Theory: Relations with Algebraic Geometry, Group Cohomology, and Algebraic K-theory" Contemp. Math. 346 (2004) 369-398. 2, 9 [24] J.-L. Loday and B. Vallette, Algebraic Operads, Grundlehren Math. Wiss. 346, Springer, Heidelberg, 2012. 3, 11 [25] A. Nijenhuis, Xn−1-forming sets of eigenvectors. Indag. Math. 13 (1951) 200-212. 1 [26] K. Uchino, Twisting on associative algebras and Rota-Baxter type operators, J. Noncommut. Geom. 4 (2010) 349-379. 2 Department of Mathematics, Lanzhou University, Lanzhou, Gansu 730000, China E-mail address: [email protected] Department of Mathematics and Computer Science, Rutgers University, Newark, NJ 07102, USA E-mail address: [email protected]
1808.10756	2	1808	2018-09-18T03:53:08	Leavitt path algebras with bounded index of nilpotence	[ "math.RA" ]	In this paper we completely describe graphically Leavitt path algebras with bounded index of nilpotence. We show that the Leavitt path algebra $L_{K}(E)$ has index of nilpotence at most $n$ if and only if no cycle in the graph $E$ has an exit and there is a fixed positive integer $n$ such that the number of distinct paths that end at any given vertex $v$ (including $v$, but not including the entire cycle $c$ in case $v$ lies on $c$) is less than or equal to $n$. Interestingly, the Leavitt path algebras having bounded index of nilpotence turn out to be precisely those that satisfy a polynomial identity. Furthermore, Leavitt path algebras with bounded index of nilpotence are shown to be directly-finite and to be $\mathbb{Z}$-graded $\Sigma$-$V$ rings. As an application of our results, we answer an open question raised in \cite{JST} whether an exchange $\Sigma$-$V$ ring has bounded index of nilpotence.	math.RA	math	LEAVITT PATH ALGEBRAS WITH BOUNDED INDEX OF NILPOTENCE KULUMANI M. RANGASWAMY AND ASHISH K. SRIVASTAVA p e S 8 1 ] . A R h t a m [ 2 v 6 5 7 0 1 . 8 0 8 1 : v i X r a Abstract. In this paper we completely describe graphically Leavitt path alge- bras with bounded index of nilpotence. We show that the Leavitt path algebra LK (E) has index of nilpotence at most n if and only if no cycle in the graph E has an exit and there is a fixed positive integer n such that the number of distinct paths that end at any given vertex v (including v, but not including the entire cycle c in case v lies on c) is less than or equal to n. Interestingly, the Leavitt path algebras having bounded index of nilpotence turn out to be precisely those that satisfy a polynomial identity. Furthermore, Leavitt path algebras with bounded index of nilpotence are shown to be directly-finite and to be Z-graded Σ-V rings. As an application of our results, we answer an open question raised in [10] whether an exchange Σ-V ring has bounded index of nilpotence. 1. Introduction The objective of this paper is to characterize Leavitt path algebras with bounded index of nilpotence. A ring R is said to have bounded index of nilpotence if there is a positive integer n such that xn = 0 for all nilpotent elements x in R. If n is the least such integer then R is said to have index of nilpotence n. We show that the Leavitt path algebra L := LK(E) of a directed graph E over a field K has index of nilpotence at most n if and only if no cycle in the graph E has an exit and there is a fixed positive integer n such that the number of distinct paths that end at any given vertex v (including v, but not including the cycle c in case v lies on c) is less than or equal to n. In this case, L becomes a subdirect product of matrix rings Mt(K) or Mt(K[x, x−1]) of finite order t ≤ n. Examples are constructed showing that L need not decompose as a direct sum of these matrix rings Mt(K) or Mt(K[x, x−1]), though the decomposition is possible when E is row-finite. We show that a Leavitt path algebra L with bounded index of nilpotence is always directly-finite and that L is a Z-graded Σ-V ring, that is, each graded simple left/right L-module is graded Σ-injective. Examples show that the converse of these statements do not hold. Interestingly, it turns out that the graphical conditions on E that ensure L has a bounded index of nilpotence are exactly the same graphical conditions on E that were shown in [4] to imply that L satisfies a polynomial identity. When E is a finite graph, these graphical conditions also imply that L has GK-dimension ≤ 1. Such 2010 Mathematics Subject Classification. 16D50, 16D60. Key words and phrases. Leavitt path algebras, bounded index of nilpotence, direct-finiteness, simple modules, injective modules. The work of the second author is partially supported by a grant from Simons Foundation (grant number 426367). 1 2 KULUMANI M. RANGASWAMY AND ASHISH K. SRIVASTAVA statements illustrate a unique phenomenon in the study of Leavitt path algebras where a single graph property of E often implies different ring-theoretic properties for L and these ring-theoretic properties are usually independent of each other for general rings (see [15] for several illustrations of this phenomenon of Leavitt path algebras). This feature of Leavitt path algebras makes them really useful tools in constructing examples of rings of various desired ring-theoretic properties. Finally, as an application of our results, we answer a question raised in [10], whether an exchange Σ-V ring has bounded index of nilpotence. For the general notation, terminology and results in Leavitt path algebras, we refer the reader to [1]. We give below an outline of some of the needed basic concepts and results. A (directed) graph E = (E0, E1, r, s) consists of two sets E0 and E1 together with maps r, s : E1 → E0. The elements of E0 are called vertices and the elements of E1 edges. A vertex v is called a sink if it emits no edges and a vertex v is called a regular vertex if it emits a non-empty finite set of edges. An infinite emitter is a vertex which emits infinitely many edges. For each e ∈ E1, we call e∗ a ghost edge. We let r(e∗) denote s(e), and we let s(e∗) denote r(e). A path µ of length n > 0 is a finite sequence of edges µ = e1e2 · · · en with r(ei) = s(ei+1) for all i = 1, · · ·, n − 1. In this case µ∗ = e∗ 1 is the corresponding ghost path. A vertex is considered a path of length 0. n · · · e∗ 2e∗ A path µ = e1 . . . en in E is closed if r(en) = s(e1), in which case µ is said to be based at the vertex s(e1). A closed path µ as above is called simple provided it does not pass through its base more than once, i.e., s(ei) 6= s(e1) for all i = 2, ..., n. The closed path µ is called a cycle if it does not pass through any of its vertices twice, that is, if s(ei) 6= s(ej) for every i 6= j. An exit for a path µ = e1 . . . en is an edge e such that s(e) = s(ei) for some i and e 6= ei. An infinite rational path p is an infinite path of the form p = x1x2 . . . xn . . . where there is an m ≥ 1 such that xk = g, a fixed closed path for all k ≥ m and that xk is an edge ek if k < m. Thus p will be of the form p = e1e2 · · · em−1gggg · ·· where g is a closed path and the ei are edges. An infinite path which is not rational is called an irrational path. A graph E is said to satisfy Condition (K), if every vertex v on a closed path c is also the base of a another closed path c′different from c. A graph E is said to satisfy Condition (L) if every cycle has an exit. If there is a path from vertex u to a vertex v, we write u ≥ v. A subset D of vertices is said to be downward directed if for any u, v ∈ D, there exists a w ∈ D such that u ≥ w and v ≥ w. A subset H of E0 is called hereditary if, whenever v ∈ H and w ∈ E0 satisfy v ≥ w, then w ∈ H. A hereditary set is saturated if, for any regular vertex v, r(s−1(v)) ⊆ H implies v ∈ H. Given an arbitrary graph E and a field K, the Leavitt path algebra LK (E) is defined to be the K-algebra generated by a set {v : v ∈ E0} of pair-wise orthogonal idempotents together with a set of variables {e, e∗ : e ∈ E1} which satisfy the following conditions: (1) s(e)e = e = er(e) for all e ∈ E1. (2) r(e)e∗ = e∗ = e∗s(e) for all e ∈ E1. (3) (The "CK-1 relations") For all e, f ∈ E1, e∗e = r(e) and e∗f = 0 if e 6= f . LEAVITT PATH ALGEBRAS WITH BOUNDED INDEX OF NILPOTENCE 3 (4) (The "CK-2 relations") For every regular vertex v ∈ E0, v = X ee∗. e∈E 1,s(e)=v Every Leavitt path algebra LK(E) is a Z-graded algebra, namely, LK(E) = M n∈Z induced by defining, for all v ∈ E0 and e ∈ E1, deg(v) = 0, deg(e) = 1, deg(e∗) = −1. Here the Ln are abelian subgroups satisfying LmLn ⊆ Lm+n for all m, n ∈ Z. Further, for each n ∈ Z, the homogeneous component Ln is given by Ln Ln = {Pkiαiβ∗ i ∈ L : αi − βi = n}. Elements of Ln are called homogeneous elements. An ideal I of LK(E) is said to be a graded ideal if I = M (I ∩ Ln). If A, B are graded modules over a graded ring R, n∈Z we write A ∼=gr B if A and B are graded isomorphic and we write A ⊕gr B to denote a graded direct sum. We will also be using the usual grading of a matrix of finite order. For this and for the various properties of graded rings and graded modules, we refer to [6], [8] and [12]. A breaking vertex of a hereditary saturated subset H is an infinite emitter w ∈ E0\H with the property that 0 < s−1(w) ∩ r−1(E0\H) < ∞. The set of all breaking vertices of H is denoted by BH. For any v ∈ BH, vH denotes the element v − Ps(e)=v,r(e) /∈H ee∗. Given a hereditary saturated subset H and a subset S ⊆ BH, (H, S) is called an admissible pair. Given an admissible pair (H, S), the ideal generated by H ∪ {vH : v ∈ S} is denoted by I(H, S). It was shown in [16] that the graded ideals of LK(E) are precisely the ideals of the form I(H, S) for some admissible pair (H, S). Moreover, LK(E)/I(H, S) ∼= LK(E\(H, S)). Here E\(H, S) is a Quotient graph of E where (E\(H, S))0 = (E0\H) ∪ {v′ : v ∈ BH\S} and (E\(H, S))1 = {e ∈ E1 : r(e) /∈ H} ∪ {e′ : e ∈ E1 with r(e) ∈ BH\S} and r, s are extended to (E\(H, S))0 by setting s(e′) = s(e) and r(e′) = r(e)′. It is known (see [14]) that if P is a prime ideal of L with P ∩ E0 = H, then E0\H is downward directed. Let Λ be an infinite set and R, a ring. Then MΛ(R) denotes the ring of Λ × Λ matrices in which all except at most finitely many entries are non-zero. 2. Results In this section, we characterize Leavitt path algebras having bounded index of nilpo- tence. We begin with the following useful proposition some part of which might be implicit in earlier works on Leavitt path algebras. Proposition 2.1. Let E be an arbitrary graph and let L := LK (E). (a) Let v be a vertex in E which does not lie on a closed path. If, for some n ≥ 1, there are n distinct paths p1, · · ·, pn in E that end at v, then the set Tn = { kijpip∗ j : kij ∈ K} is a subring of L isomorphic to the matrix ring n X i=1 n X j=1 Mn(K). 4 KULUMANI M. RANGASWAMY AND ASHISH K. SRIVASTAVA (b) Let v be a vertex in E lying on a cycle c and let f be an exit for c at v. Then, for every integer n ≥ 1, the subset Sn = { kijcif f ∗(c∗)j : kij ∈ K} is a subring of L isomorphic to the n X i=1 n X j=1 matrix ring Mn(K). (c) Let v be a vertex lying on a cycle c without exits in E. If, for some n ≥ 1, there are n distinct paths p1, · · ·, pn in E that end at v and do not go through the entire cycle c, then again the set Tn = { kijpip∗ j : kij ∈ K} is a subring of n X i=1 n X j=1 L isomorphic to the matrix ring Mn(K). j pk 6= 0 if and only if pj = pk. Because, if p∗ Proof. (a) First observe that p∗ j pk 6= 0, then either pj = pkp′ or pk = pjq′ for some paths p′, q′. Since r(pj) = r(pk) = v, we get s(p′) = v = r(p′) and s(q′) = v = r(q′). Since v does not lie on a closed path, we conclude that p′ = v = q′. So pj = pk. Conversely, if pj = pk, then clearly p∗ j pk = j . Clearly, (εii)2 = εii and εijεkl = εil or 0 p∗ j pj = v 6= 0. For all i, j, let εij = pip∗ according as j = k or not. Thus the εij form a set of matrix units and it is readily seen that the set Tn = { kijpip∗ j : kij ∈ K} is a subring of L isomorphic to the n X i=1 n X j=1 matrix ring Mn(K). (b) Suppose c is a cycle in E with an exit f at a vertex v. Consider the set {εij = cif f ∗(c∗)j : 1 ≤ i, j ≤ n}. Clearly, the εij form a set of matrix units as (εii)2 = εii and εijεkl = εil or 0 according as j = k or not. It is then easy to check kijcif f ∗(c∗)j : kij ∈ K} is a subring of L isomorphic to that the set Sn = { n X i=1 the matrix ring Mn(K). n X j=1 (c) Let v be the base of a cycle c without exits and p1, · · ·, pn be n distinct paths that end at v and not go through the entire cycle c. Using the fact that c is a cycle without exits and repeating the arguments as in (a), it follows that p∗ j pk 6= 0 if and j with 1 ≤ i, j ≤ n. Clearly, (εii)2 = εii only if pj = pk. As before let εij = pip∗ and εijεkl = εil or 0 according as j = k or not. Thus the εij form a set of matrix units and it is readily seen that Tn = { kijpip∗ j : kij ∈ K} is a subring of L n X i=1 n X j=1 isomorphic to the matrix ring Mn(K). (cid:3) We are now ready to describe all the Leavitt path algebras with bounded index of nilpotence. It is interesting to note that the Leavitt path algebras having bounded index of nilpotence are precisely those that satisfy a polynomial identity. Recall that an algebra A over a field K is said to satisfy a polynomial identity if there exists a non-zero element f in K[x1, . . . , xn] such that f (a1, . . . , an) = 0 for all ai in A. Clearly every commutative ring satisfies a polynomial identity but there are many interesting classes of noncommutative rings too that satisfy a polynomial identity. For instance, the Amitsur-Levitzky theorem (see [13]) states that, for any n ≥ 1, the matrix ring Mn(R) over a commutative ring R satisfies a polynomial identity of degree 2n. In [4] it is shown that the Leavitt path algebra LK (E) of an LEAVITT PATH ALGEBRAS WITH BOUNDED INDEX OF NILPOTENCE 5 arbitrary graph E over a field K satisfies a polynomial identity if and only if no cycle in E has an exit and there is a fixed positive integer d such that the number of distinct paths that end at any given vertex v (including v, but not including the entire cycle c in case v lies on c) is less than or equal to d. When E is a finite graph, then the Leavitt path algebra LK (E) satisfying a polynomial identity is known to be equivalent to the Gelfand-Kirillov dimension of LK(E) being at most one [4]. Theorem 2.2. Let E be an arbitrary graph. Then the following properties are equivalent for L := LK(E): (1) L has index of nilpotence less than or equal to n; (2) No cycle in E has an exit and there is a fixed positive integer n such that the number of distinct paths that end at any given vertex v (including v, but not including the entire cycle c in case v lies on c) is less than or equal to n; (3) For any graded prime ideal P of L, L/P ∼=gr Mt(K) or Mt(K[x, x−1]) where t ≤ n with appropriate matrix gradings; (4) L is a graded subdirect product of graded rings {Ai : i ∈ I} where, for each i, Ai ∼=gr Mti (K) or Mti(K[x, x−1]) with appropriate matrix gradings where, for each i, ti ≤ n, a fixed positive integer. (5) L satisfies a polynomial identity. Proof. Assume (i), that is, assume that the index of nilpotence of L is ≤ n. We claim that no cycle in E can have an exit. Because, otherwise, by Proposition 2.1 (b), L will contain subrings of matrices of arbitrary finite size and this will give rise to unbounded index of nilpotence for L. Thus every vertex v in E either does not lie on a closed path or lies on a cycle without exits. If there are more than n distinct paths ending at v, then again, by Proposition 2.1 (a) and (c), L will contain a copy of a matrix ring of order greater than n over K which will imply that the index of nilpotence of L is greater than n, a contradiction. This proves (ii). Assume (ii). Let P = I(H, S) be a graded prime ideal of L. Our hypothesis implies that no cycle in E\(H, S) has an exit and that n is also the upper bound for the number of distinct paths ending at any vertex in E\(H, S). So E\(H, S) contains no infinite irrational paths. This means that every path ends at a sink or at a cycle without exits. Also, as I(H, S) is a graded prime ideal, Theorem 3.12 of [14] implies that (E\(H, S))0 is downward directed. Consequently, E\(H, S) contains either (a) exactly one sink w or (b) exactly one cycle c without exits based at a vertex v. Now in case (a), there are no more than n distinct paths ending at w and, in case (b), there are no more than n paths which end at v and do not go through the cycle c. We then appeal to Corollary 2.6.5 and Lemma 2.7.1 of [1] to conclude that L/P ∼= LK(E\(H, S)) ∼= Mt(K) or Mt(K[x, x−1]) according as E\(H, S) contains a sink or a cycle without exits. This proves (iii). Assume (iii). Now, for any graded prime ideal P , L/P ∼=gr Mt(K) or Mt(K[x, x−1]) with appropriate matrix gradings where t ≤ n, a fixed positive integer. It is known that the intersection I of all graded prime ideals of L is zero. For the sake of com- pleteness, we shall outline the argument. If I 6= 0, being a graded ideal, I the will contain vertices. But, given any vertex v, a graded ideal M maximal among graded ideals with respect to v /∈ M is a graded prime ideal, because, for any two homogeneous elements a, b, if a /∈ M and b /∈ M , then v ∈ M + LaL and v ∈ M + LbL. Consequently, v = v2 ∈ (M + LaL)(M + LbL) = M + LaLbL. Since 6 KULUMANI M. RANGASWAMY AND ASHISH K. SRIVASTAVA v /∈ M , aLb " M . Thus M is a graded prime ideal. But then v /∈ M implies v /∈ I, a contradiction. Thus ∩{P : P graded prime ideal} = 0. Consequently, L is a graded subdirect product of the graded rings L/P graded isomorphic to Mt(K) or Mt(K[x, x−1]) under appropriate matrix gradings, where t ≤ n, a fixed positive integer. This proves (iv). Assume (iv). If t ≤ n, then the matrix rings Mt(K) and Mt(K[x, x−1]) will each have index of nilpotence ≤ n. Consequently, a subdirect product of such rings will also have nilpotence index ≤ n. This proves (i). Assume (iv). By the Amitsur-Levitzky theorem [13], both Mt(K) and Mt(K[x, x−1]) with t ≤ n are polynomial identity rings satisfying a polynomial identity of degree ≤ 2n. From this, it is clear that the subdirect product L also satisfies a polynomial identity of degree ≤ n. This proves (v). The implication (v) =⇒ (ii) has been established in [4]. (cid:3) The Leavitt path algebra in Theorem 2.2 need not decompose as a direct sum of matrix rings, as the following example shows. Example 2.3. Consider the following "infinite clock" graph E: •w1 •w2 ↑ ր •v −→ •w3 ... . . . . . . · · · ւ wn Thus E0 = {v}∪{w1, w2, ···, wn, ···} where the wi are all sinks. For each n ≥ 1, let en denote the single edge connecting v to wn. The graph E is acyclic and so every ideal of L is graded ([8]). The number of distinct paths ending at any given sink (including the sink) is ≤ 2. For each n ≥ 1, Hn = {wi : i 6= n} is a hereditary saturated set, BHn = {v} and (E\(Hn, BHn))0 = {v, wn} is downward directed. Also E\(Hn,BHn) trivially satisfies Condition (L). Hence the ideal Pn generated by Hn ∪ {v − ene∗ n} is a graded primitive ideal by Theorem 4.3(iii) of [14] and LK(E)/P ∼= M2(K). Moreover, every graded primitive (equivalently, prime) ideal P of LK(E) is equal to Pn for some n. By [9, Theorem 4.12], LK(E) is a graded Σ-V ring. But LK (E) cannot decompose as a direct sum of the matrix rings M2(K). Be- cause, otherwise, v would lie in a direct sum of finitely many copies of M2(K). Since the ideal generated by v is LK(E), LK(E) will then be a direct sum of finitely many copies of M2(K). This is impossible since LK(E) contains an infinite set of orthogonal idempotents {ene∗ n : n ≥ 1}. We can also describe the internal structure of this ring LK (E). The socle S M2(K) and : i ≥ 1}, S ∼= L of LK(E) is the ideal generated by the sinks {wi LK(E)/S ∼= K. ℵ0 But the decomposition is possible if the graph is row-finite, as shown in the following theorem. Theorem 2.4. Let E be a row-finite graph. Then the following properties are equiv- alent for L := LK (E): (i) L has bounded index of nilpotence ≤ n; LEAVITT PATH ALGEBRAS WITH BOUNDED INDEX OF NILPOTENCE 7 (ii) There is a fixed positive integer n and a graded isomorphism L ∼=gr M i∈I Mni(K) ⊕ M j∈J Mnj (K[x, x−1]) where I, J are arbitrary index sets and, for all i ∈ I and j ∈ J, ni ≤ n and nj ≤ n . In particular, L is graded semi-simple (that is, a direct sum of graded simple left/right ideals). Proof. Assume (i). By Theorem 2.2, no cycle in E has an exit and the number of distinct paths that end at any vertex v is ≤ n, with the proviso that if v sits on a cycle c, then these paths do not include the entire cycle c. If A is the graded ideal generated by all the sinks in E and all the vertices on cycles without exits, then, by Corollary 2.6.5 and Lemma 2.7.1 of [1], A ∼= M Mnj (K[x, x−1]). By giving Mni(K) ⊕ M j∈J i∈I appropriate matrix gradings, this isomorphism becomes a graded isomorphism. We claim that L = A. Let H ⊆ A be the set consisting of all the sinks and all the vertices on cycles in E. By hypothesis, every path in E that does not include an entire cycle has length ≤ n and ends at a vertex in H. So if u is any vertex in E, using the fact that all the vertices in E are regular and by a simple induction on the length of the longest path from u, we can conclude that u belongs to the saturated closure of H. This implies that L = A ∼=gr M Mnj (K[x, x−1]). This Mni(K) ⊕ M j∈J proves (ii). i∈I (ii) =⇒ (i) follows from the fact that the matrix rings Mni(K) and Mnj (K[x, x−1]) (cid:3) with ni, nj ≤ n have index of nilpotence ≤ n. One consequence of Theorem 2.2 is the following. Proposition 2.5. Let E be an arbitrary graph. If L := LK (E) has bounded index of nilpotence n, then L is a graded Σ-V ring. Proof. If L has bounded index of nilpotence n, then for any graded primitive ideal P of L (since it is also graded prime), we have, by Theorem 2.2(iii), L/P ∼=gr Mt(K) or Mt(K[x, x−1]) with appropriate matrix gradings, where t ≤ n. By [9, Theorem 4.12], we then conclude that L is a graded Σ-V ring. (cid:3) The converse of the above result does not hold, as can be seen in the two examples below. Example 2.6. Let E be the "inverse infinite clock" graph consisting of a sink w and countably infinite edges {en : n ≥ 1} with r(en) = w and s(en) = wn for all n. •w1 •w2 ↓ ւ •w ←− •w3 ... . . . . . . · · · ր •wn Then LK(E) ∼= M∞(K), the infinite ω × ω matrix with finitely many non-zero en- tries. Now, under appropriate matrix grading, M∞(K) is graded semisimple (that is, 8 KULUMANI M. RANGASWAMY AND ASHISH K. SRIVASTAVA a graded direct sum of graded simple modules) and so all graded left/right M∞(K)- modules are graded injective and hence L is a graded Σ-V ring. But L does not have bounded index of nilpotence by Theorem 2.2(ii). Example 2.7. Consider the following graph F consisting of two cycles g and c connected by an edge: • −→ • • −→ • ↑ ↑ ↓ ↓ • ←− • −→ •v ←− • g c Now F is downward directed, c is a cycle without exits and the various powers of the cycle g give rise to infinitely many distinct paths that end at the base v of the cycle c. Hence LK (F ) ∼= M∞(K[x, x−1]) (by Lemma 2.7.1 of [1]) and is graded semisimple. Hence each graded simple module over LK(F ) is graded Σ-injective. But LK(F ) does not have bounded index of nilpotence, as M∞(K[x, x−1]) contains subrings isomorphic to Mn(K[x, x−1]) for every positive integer n. In the monograph [10], the following open question (6.33: Problem 3, Chapter 6) was raised: Question 2.8. Does every exchange right Σ-V ring have bounded index of nilpo- tence? We answer this question in the negative in the following remark. Remark 2.9. Consider the graph E of Example 2.6. As noted there, LK(E) ∼= M∞(K) which is semisimple and hence is a Σ-V ring. Since E is acyclic, LK (E) is von Neumann regular [2] and hence is an exchange ring. But LK(E) does not have bounded index of nilpotence by Theorem 2.2(ii). Remark 2.10. It was shown in [9] that a Leavitt path algebra LK(E) is directly- finite (equivalently, graded directly-finite with respect to vertices) if and only if no cycle in E has an exit. In view of Theorem 2.2, it is clear that a Leavitt path algebra of bounded index is always directly-finite. But, for arbitrary rings with bounded index ∼= Z(pn), the ring of nilpotence, this is not the case. Let S = Q Rk, where each Rk of integers modulo a fixed integer n ≥ 2. Now (pS)n = 0 and if a ∈ S is nilpotent, then a ∈ pS and an = 0. Consequently, S has bounded index of nilpotence. In fact, ∼= the index of nilpotence of S is n. But S is not directly-finite, since S ∼= Q Q k≥3 ∼= · · ·. Rk Rk k≥2 Conversely, a directly-finite Leavitt path algebra need not have bounded index of nilpotence. If E is the graph consisting of an infinite line segment • −→ • −→ • −→ · · ·• −→ · · · then clearly LK(E) is directly-finite, but, by Theorem 2.2 (ii), LK(E) ∼= M∞(K) does not have bounded index of nilpotence. LEAVITT PATH ALGEBRAS WITH BOUNDED INDEX OF NILPOTENCE 9 References [1] G. Abrams, P. Ara and M. Siles Molina, Leavitt path algebras, Lecture Notes in Mathematics, vol. 2191, Spinger (2017). [2] G. Abrams and K. M. Rangaswamy, Regularity conditions for arbitrary Leavitt path algebras, Algebr. Represent. Theory, 13 (2010), 319 - 334. [3] P. Ara and K. M. Rangaswamy, Finitely presented simple modules over Leavitt path algebras, J. Algebra 417 (2014), 333-352. [4] J. P. Bell, T. H. Lenagan and K. M. Rangaswamy, Leavitt path algebras satisfying a polynomial identity, J. Algebra and Applications, vol. 15 (2016), 1650084 (13 pages). [5] K. R. Goodearl, von Neumann regular rings, Krieger Publishing Co., Malabar, 1991. [6] R. Hazrat, The graded structure of Leavitt path algebras, Israel J. Math. 195 (2013), 833-895. [7] R. Hazrat, Leavitt path algebras are graded von Neumann regular rings, J. Algebra 401 (2014), 220-233. [8] R. Hazrat and K. M. Rangaswamy, On graded irreducible representations of Leavitt path algebras, J. Algebra 450 (2016), 458-486. [9] R. Hazrat, K. M. Rangaswamy and A. K. Srivastava, Leavitt path algebras: Graded direct- finiteness and graded Σ-injective simple modules, J. Algebra, 503 (2018), 299-328. [10] S. K. Jain, A. K. Srivastava and A. A. Tuganbaev, Cyclic modules and the structure of rings, Oxford Math. Monographs, Oxford University Press, (2012). [11] T. Y. Lam, Lectures on Modules and Rings, Grad. Texts in Math., vol. 189, Springer, New York, 1999. [12] C. Nastasescu and F. Van Oystaeyen, Graded Rings Theory, North Holland Publ. Co., 1987. [13] C. Procesi, Rings with polynomial identities, Pure and Applied Math series, vol. 17 (Marcel Dekker, New York 1973). [14] K. M. Rangaswamy, The theory of prime ideals of Leavitt path algebras over arbitrary graphs, J. Algebra, 375 (2013), 73-96. [15] K. M. Rangaswamy, A survey of some of the recent developments in Leavitt path algebras, arXiv:1808.04466v1 [math RA] 13 August 2018. [16] M. Tomforde, Uniqueness theorems and ideal structure of Leavitt path algebras, J. Algebra 318 (2007), 270-299. Departament of Mathematics, University of Colorado at Colorado Springs, Colorado- 80918, USA E-mail address: [email protected] Department of Mathematics and Statistics, St. Louis University, St. Louis, MO- 63103, USA E-mail address: [email protected]
1408.6075	3	1408	2015-09-17T13:33:33	A Sylow theorem for the integral group ring of PSL(2,q)	[ "math.RA", "math.GR" ]	For G = PSL(2,p^f) denote by ZG the integral group ring, by V(ZG) the group of normalized units of ZG and let r be a prime different from p. Using the so called HeLP-method we prove, that units of r-power order in V(ZG) are rationally conjugate to elements of G. As a consequence we prove, that subgroups of prime power order in V(ZG) are rationally conjugate to subgroups of G, provided p = 2 or f =1.	math.RA	math	A Sylow theorem for the integral group ring of PSL(2, q) Leo Margolis March 1, 2018 Abstract: For G = PSL(2, pf ) denote by ZG the integral group ring over G and by V (ZG) the group of units of augmentation 1 in ZG. Let r be a prime different from p. Using the so called HeLP-method we prove that units of r-power order in V (ZG) are rationally conjugate to elements of G. As a consequence we prove that subgroups of prime power order in V (ZG) are rationally conjugate to subgroups of G, if p = 2 or f = 1. Let G be a finite group and ZG the integral group ring over G. Denote by V (ZG) the group of units of augmentation 1 in ZG, i.e. those units whose coefficients sum up to 1. We say that a finite subgroup U of V (ZG) is rationally conjugate to a sub- group W of G, if there exists a unit x ∈ QG such that x−1Ux = W. The question if some, or even all, finite subgroups of V (ZG) are rationally conjugate to subgroups of G was proposed by H. J. Zassenhaus in the '60s and published in [Zas74]. This so called Zassenhaus Conjectures motivated a lot of research. E.g. A. Weiss proved the strongest version, that all finite subgroups of V (ZG) are rationally conjugate to subgroups of G, provided G is nilpotent [Wei88] [Wei91]. K. W. Roggenkamp and L. L. Scott obtained a counterexample [Rog91] to this strong conjecture. The version, which asks whether all finite cyclic subgroups of V (ZG) are rationally conjugate to subgroups of G, the so called First Zassenhaus Conjecture, is however still open, see e.g. [Her08a], [CMdR13]. Though mostly solvable groups were considered when studying such questions, there are some results available for series of non-solvable groups. E.g. a work on the symmetric groups [Pet76] or for Lie-groups of small rank [Ble99]. The groups PSL(2, q), which are also the object of study in this paper, found also some special attention in [Wag95], 1 [Her07], [HHK09], [BK11] or in [BM15b]. In this paper we will limit our attention to "Sylow-like" results, i.e. to finite p-subgroups of V (ZG). We say that a weak Sylow theorem holds for V (ZG), if every finite p-subgroup of V (ZG) is isomorphic to some subgroup of G, and that a strong Sylow theorem holds for V (ZG), if every finite p-subgroup of V (ZG) is even rationally conjugate to a sub- group of G. First Sylow-like results for integral group rings were obtained in [KR93]. Later M. A. Dokuchaev and S. O. Juriaans proved a strong Sylow theorem for special classes of solvable groups [DJ96] and M. Hertweck, C. Höfert and W. Kimmerle proved a weak Sylow theorem for PSL(2, pf ), where p = 2 or f ≤ 2. The results of this article are as follows: Theorem 1: Let G = PSL(2, pf ), let r be a prime different from p and let u be a torsion unit in V (ZG) of r-power order. Then u is rationally conjugate to a group element. Theorem 2: Let G = PSL(2, pf ) such that f = 1 or p = 2. Then any subgroup of prime power order of V (ZG) is rationally conjugate to a subgroup of G, i.e. a strong Sylow theorem holds in V (ZG). 1 HeLP-method and known results Let G be a finite group. A very useful notion to study rational conjugacy of torsion agg ∈ ZG and xG be the conjugacy class ag is called the partial augmentation units are partial augmentations: Let u = Pg∈G of the element x ∈ G in G. Then εx(u) = Pg∈xG of u at x. This relates to rational conjugacy via: Lemma 1.1 ([MRSW87, Th. 2.5]). Let u ∈ V(ZG) be a torsion unit. Then u is rationally conjugate to a group element if and only if εx(uk) ≥ 0 for all x ∈ G and all powers uk of u. It is well known that if u 6= 1 is a torsion unit in V(ZG), then ε1(u) = 0 by the so called Berman-Higman Theorem [Seh93, Prop. 1.4]. If εx(u) 6= 0, then the order of x divides the order of u [MRSW87, Th. 2.7], [Her06, Prop. 3.1]. Moreover the exponent of G and of V(ZG) coincide [CL65, Cor. 4.1] and if U is a finite subgroup of V(ZG) the order of U divides the order of G [ŽK67] (or [Seh93, Lemma 37.3]). We will use these 2 facts in the following without further mention. Let u be a torsion unit in V (ZG) of order n and ζ a primitive complex n-th root of unity and let K be some field, whose characteristic p does not divide n. Let ξ be a (not necessarily primitive) complex n-th root of unity and let D be a K-representation of G with character ϕ. Here ϕ is understood as an ordinary or Brauer character. It was first obtained by Luthar and Passi for K having characteristic 0 [LP89] and later generalized by Hertweck for positive characteristic [Her07] that the multipilicity of ξ as an eigenvalue of D(u), which we denote by µ(ξ, u, ϕ) and which is of course a non-negative integer, may be computed as µ(ξ, u, ϕ) = 1 nXdn d6=1 TrQ(ζ d)/Q(ϕ(ud)ξ−d) + 1 n XxG εx(u)TrQ(ζ)/Q(ϕ(x)ξ−1), x p−regular where as usual TrQ(ζ)/Q(x) = Xσ∈Gal(Q(ζ)/Q) σ(x). The multiplicity of a complex root of unity as an eigenvalue of a matrix over a field of positive characteristic should be understood, here and in the rest of the paper, in the sense of Brauer. If u is of prime power order rk for the first sum in the expression above we obtain TrQ(ζ d)/Q(ϕ(ud)ξ−d) = 1 r µ(ξr, ur, ϕ). 1 nXdn d6=1 Using these formulas to find possible partial augmentations for torsion units in integral group rings of finite groups is today called HeLP-method. For a diagonalizable matrix A we will write A ∼ (a1, ..., an), if the eigenvalues of A, with multiplicities, are a1, ..., an. d d and pf −1 All subgroups of G = PSL(2, pf ) were first known to Dickson [Dic01, Theorem 260]. Let d = gcd(2, p − 1). Up to conjugation there is exactly one cyclic group of order p, pf +1 respectively in G and every element of G lies in a conjugate of such a group. In particular there is only one conjugacy class of involutions in G. The Sylow p-subgroups are elementary-abelian, the Sylow subgroups for all other primes, which are odd, are cyclic and if p 6= 2 the Sylow 2-subgroup is dihedral or a Kleinian four-group. There are d conjugacy classes of elements of order p. If g ∈ G is not of order p or 2 its only distinct conjugate in hgi is g−1. We denote by a a fixed element of order pf −1 and d 3 by b a fixed element of order pf +1 d. The modular representation theory of PSL(2, pf ) in defining characteristic is well known. All irreducible representations were first given by R. Brauer and C. Nesbitt [BN41]. The explicit Brauer table of SL(2, pf ), which contains the Brauer table of PSL(2, pf ), may be found in [Sri64]. In particular all characters are real valued since any p-regular element is conjugate to its inverse. However, I was not able to find the following Lemma in the literature, except, without proof, in Hertwecks preprint [Her07], so a short proof is included. Lemma 1.2. Let G = PSL(2, pf ) and d = gcd(2, p − 1). There are p-modular represen- tations of G given by Θ0, Θ1, Θ2, ... such that there is a primitive pf −1 -th root of unity α and a primitive pf +1 -th root of unity β satisfying d d Θk(b) ∼ (1, β, β−1, β2, β−2, ..., βk, β−k), Θk(a) ∼ (1, α, α−1, α2, α−2, ..., αk, α−k) for every k ∈ N0. Proof: The group SL(2, pf ) acts on the vector space spanned by the homogenous poly- nomials in two commuting variables x, y of some fixed degree e extending the natural operation on the 2-dimensional vector space spanned by x, y, see e.g. [Alp86, p. 14-16]. Since (cid:0) −1 0 0 −1(cid:1) xiyj = (−1)i+jxiyj this action affords a PSL(2, pf )-representation if and 2 . Let γ be an eigenvalue of an element in SL(2, q) mapping 2 (a) has the only if e is even and p is odd or p = 2. So let from now on e be even. Call this representation Θ e onto a under the natural projection from SL(2, pf ) to PSL(2, pf ). Then Θ e same eigenvalues as Θ e {γi−j 0 ≤ i, j ≤ e, i + j = e} = {(γ2t −e first part of the claim. Now let δ be an eigenvalue of an element in SL(2, pf ) mapping onto b under the nat- ural projection from SL(2, pf ) to PSL(2, pf ). The action of SL(2, pf ) may of course be extended to SL(2, p2f ). So Θ e matrix may be seen as an element in SL(2, p2f ). Then doing the same calculations as above and setting β = δ2 proves the Lemma. 0 γ−1(cid:17) xiyj = γi−jxiyj, so the eigenvalues are 2 }. Thus setting α = γ2 proves the 0 δ−1(cid:1)(cid:1) , where the 2 (cid:16)(cid:16) γ 0 γ−1(cid:17)(cid:17) . Now(cid:16) γ 2 (b) has the same eigenvalues as Θ e 0 2 (cid:0)(cid:0) δ 0 0 2 ≤ t ≤ e Notation: The Brauer-character belonging to a representation Θi from the Lemma above will be denoted by ϕi. 4 Using the HeLP-method R. Wagner [Wag95] and Hertweck [Her07] obtained already some results about rational conjugacy of torsion units of prime power order in PSL(2, pf ). Part of Wagners result was published in [BHK04]. Lemma 1.3. [Wag95] (also [Her07, Prop. 6.1]) Let G = PSL(2, pf ) and f ≤ 2. If u is a unit of order p in V (ZG), then u is rationally conjugate to a group element. Remark: The HeLP-method does not suffice to prove rational conjugacy to a group element of a unit of order p in V (Z PSL(2, pf )), if p is odd and f ≥ 3. There is also no other method or idea around how one could e.g. obtain whether a unit of order 3 in V (Z PSL(2, 27)) is rationally conjugate to a group element or not. Lemma 1.4. [Her07, Prop. 6.4] Let G = PSL(2, pf ) and let r be a prime different from p. If u is a unit of order r in V (ZG), then u is rationally conjugate to an element of G. Lemma 1.5. [Her07, Prop. 6.5] Let G = PSL(2, pf ), let r be a prime different from p and u a torsion unit in V (ZG) of order rn. Let m < n and denote by S a set of εx(u) = 0. representatives of conjugacy classes of elements of order rm in G. Then Px∈S If moreover g is an element of order rn in G, then µ(1, u, ϕ) = µ(1, g, ϕ) for every p-modular Brauer character ϕ of G. If one is interested not only in cyclic groups the following result is very useful. It may be found e.g. in [Seh93, Lemma 37.6] or in [Val94, Lemma 4]. Lemma 1.6. Let G be a finite group, U a finite subgroup of V (ZG) and H a subgroup of G isomorphic to U. If σ : U → H is an isomorphism such that χ(u) = χ(σ(u)) for all u ∈ U and all irreducible complex characters χ of G, then U is rationally conjugate to H. 2 Proof of the results We will first sum up some elementary number theoretical facts. The notatian a ≡ b (c) will mean, that a is congruent b modulo c. Proposition 2.1. Let t and s be natural numbers such that s divides t and denote by ζt and ζs a primitive complex t-th root of unity and s-th root of unity respectively. Then TrQ(ζt)/Q(ζs) = µ(s) ϕ(t) ϕ(s) , 5 where µ denotes the Möbius function and ϕ Euler's totient function. So for a prime r and natural numbers n, m with m ≤ n we have TrQ(ζrn )/Q(ζrm) =  rn−1(r − 1), m = 0 −rn−1, m = 1 0, m > 1 Let moreover i and j be integers prime to r, then TrQ(ζrn )/Q(ζ i rmζ −j rm ) =  rn−1(r − 1), −rn−1, 0, i ≡ j (rm) i 6≡ j (rm), i 6≡ j (rm−1) i ≡ j (rm−1) Proof of Proposition 2.1: Let s = pf1 k be the prime factorisation of s. For a natural number l let I(l) = {i ∈ N 1 ≤ i ≤ l, gcd(i, l) = 1}. As is well known, t i ∈ I(t)}. From this the case s = 1 follows immediately. Gal(Q(ζt)/Q) = {σi : ζt 7→ ζ i Otherwise we have 1 · ... · pfk TrQ(ζt)/Q(ζs) = Xi∈I(t) = ( −1, fj = 1 fj > 1 0, fj j ζ i p ζ i s = ϕ(t) ϕ(s) Xi∈I(s) ζ i s = ϕ(t) ϕ(s) k Yj=1 Xi∈I(p fj j ) ζ i p fj j . and this gives the first formula. The other formulas Now Pi∈I(p fj j ) are special cases of this general formula since ϕ(rn) = (r − 1)(rn−1). Proof of Theorem 1: Let G = PSL(2, pf ), let r be a prime different from p and let u be a torsion unit in V (ZG) of order rn. Let ζ be an primitive complex rn-th root of unity and set TrQ(ζ)/Q = Tr. If n = 1, then by Lemma 1.4 u is rationally conjugate to an element in G, so assume n ≥ 2. Assume further that by induction ur is rationally conjugate to an element in G. We will proceed by induction on m, where 1 ≤ m ≤ n. For a fixed m let l = rm−1 if r is odd and l = rm−2 if r = 2. Let {xi 1 ≤ i ≤ l, gcd(i, r) = 1} be a full set of 1 = xi (this representatives of conjugacy classes of elements of order rm in G such that xi is possible by the group theoretical properties of G given above) and let S be a set of representatives of conjugacy classes of elements of G of r-power order not greater than rn containing x1, ..., xl. The proof will be divided in several steps: 2 2 6 a) For m < n we will show εxi(u) = 0 for every 1 ≤ i ≤ l, i.e. the partial augmenta- tions of u at elements of order rm vanish. This will be proved by an induction on k ≤ m showing (i) εxi(u) = εxj (u) for i ≡ ±j (rm−k). If r is even it suffices to prove this for k = m − 1, if r is odd we will prove it for k = m. It then follows from Lemma 1.5 that εxi(u) = 0 for all xi. b) For m = n after reordering the x1, ..., xl we will show εx1(u) = 1 and εxi(u) = 0 for i ≥ 2. Together with a) this proves the Theorem. This will be achieved by an induction on k to show several facts: (i) εx1(u) = 1 and εxi(u) = 0 for i ≡ ±1 (rn−k), i 6= 1. (ii) εxi(u) = εxj (u) for i ≡ ±j (rn−k) and i 6≡ ±1 (rn−k). These statements will be proved for k = n − 1. If r is even this is already enough. εxi(u) = 0 for α 6≡ ±1 (r), which In case r is odd we will moreover prove that Pi≡α(r) then also implies the Theorem. So let m be a natural number such that m < n. If m = 0 statement a) is the Berman- Higman Theorem and if r = 2 and m = 1 it follows from Lemma 1.5 and the fact that there is only one conjugacy class of involutions in G. So assume we know εx(u) = 0 for ◦(x) < rm. The representations and the corresponding characters of G from Lemma 1.2 will be used freely. Statement (i) in a) is certainly true for k = 0, so assume εxi(u) = εxj (u) for i ≡ ±j (rm−k) for some k. Since ur is rationally conjugate to a group element, there exists a primitive rn−1-th root of unity ζrn−1 such that Θrk(ur) ∼ (1, ζrn−1, ζ −1 rn−1, ..., ζ rk rn−1, ζ −rk rn−1). rn−1, ζ 2 rn−1, ζ −2 Now all p-modular Brauer characters of G are real valued and thus using the last state- rk ), where for ment in Lemma 1.5 we obtain that Θrk(u) ∼ (1, a1, a−1 every i we have ai a root of unity such that arm−k 6= 1. So for every primitive rm−k-th root of unity ζrm−k we have µ(ζrm−k, u, ϕrk) = 0. Let ζrm be a primitive rm-th root of unity rm, ζ −rk rm. Let moreover α such that we have Θrk(x1) ∼ (1, ζrm, ζ −1 be a natural number prime to r such that 1 ≤ α ≤ l. Thus µ(ξα, u, ϕrk) = 0 and εx(u) = 0 for ◦(x) < rm. From here on a sum over i, if not specified, will always mean a sum over all i satisfying 1 ≤ i ≤ l and r ∤ i. Then using rm ) and set ξ = ζ rk rm , ..., ζ rk 1 , a2, a−1 2 , ..., ark, a−1 i 7 the HeLP-method we get 0 = µ(ξα, u, ϕrk) = 1 r µ(ξαr, ur, ϕrk) + εx(u)Tr(ϕrk(x)ξ−α) 1 rn Xx∈S εx(u)Tr(ϕrk(x)ξ−α) + εxi(u)Tr(ϕrk(xi)ξ−α) 1 rn Xi εx(u)Tr(ξ−α) + εxi(u)Tr((ξi + ξ−i)ξ−α) 1 rn Xi εxi(u)Tr((ξi + ξ−i)ξ−α). (1) = = = 1 r 1 r 1 r µ(ξαr, ur, ϕrk) + µ(ξαr, ur, ϕrk) + µ(ξαr, ur, ϕrk) + ◦(x)>rm 1 rn Xx∈S rn Xx∈S Tr(ξ−α) 1 rn + 1 rn Xi In the third line we used that if ζ is a root of unity of r-power order such that ζ rm−k 6= 1, then ζξ has the same order as ζ and so Tr(ζξ) = 0 by Proposition 2.1. Thus for an element x ∈ S of order at least rm we get Tr(ϕrk(x)ξ−α) =( Tr(ξ−α), Tr(ξ−α + (ξi + ξ−i)ξ−α), x = xi ◦(x) > rm Note that as i is prime to r the congruence i ≡ α (rm−k) implies −i 6≡ α (rm−k) for rm−k /∈ {1, 2} and these exceptions don't have to be considered by our assumptions on m and k. There are now two cases to consider. First assume k < m − 1, so ξ is at least of order r2 and thus Tr(ξ) = 0. Moreover µ(ξαr, ur, ϕrk) = 0 and using Proposition 2.1 in (1) we obtain 0 = εxi(u)Tr((ξi + ξ−i)ξ−α) 1 1 = rn Xi rn Xi≡±α(rm−k) = Xi≡±α(rm−k) εxi(u) − εxi(u)(rn−1(r − 1)) + 1 rn Xi≡±α(rm−k−1) i6≡±α(rm−k) εxi(u)(−rn−1) 1 εxi(u). r Xi≡±α(rm−k−1) (2) So rXi≡±α(rm−k) εxi(u) = Xi≡±α(rm−k−1) εxi(u). 8 But since by induction εxi(u) = εxj (u) for i ≡ ±j (rm−k) the summands on the left hand side are all equal and since changing α by rm−k−1 does not change the right hand side of the equation we get εxi(u) = εxj (u) for i ≡ ±j (rm−k−1). Now consider k = m − 1, then ξ is a primitive r-th root of unity and thus Tr(ξ) = −rn−1 and µ(ξαr, ur, ϕrk) = µ(1, ur, ϕrk) = 1. So using Proposition 2.1 in (1) we get 1 rn X±i≡α(r) εxi(u)(rn−1(r − 1) − rn−1) (3) εxi(u). 0 = 1 r + −rn−1 rn + εxi(u) − = X±i≡α(r) So εxi(u)(−2rn−1) + 1 rn X±i6≡α(r) r Xi 2 εxi(u). r X±i≡α(r) εxi(u) = 2Xi Now by Lemma 1.5 the right side of this equation is zero and by induction all summands on the left side are equal. Hence varying α gives εxi(u) = 0 for every xi and part a) is proved. So assume m = n. As in the computation above we have Θrk(ur) ∼ (1, ζrn−1, ζ −1 rn−1, ..., ζ rk rn−1, ζ −rk rn−1) rk ), for some primitive rn−1-th root of unity ζrn−1 and Θrk(u) ∼ (1, a1, a−1 2 , ..., ark, a−1 where ai are roots of unity such that arn−k 6= 1 for 1 ≤ i ≤ rk−1 and ark is some primitive rn−k-th root of unity. Set ξ = ark and reorder the x1, ..., xl such that Θ1(x1) ∼ Θ1(u), but still xi 1 is rationally conjugate to ur. We will proceed by induction on k. Let α be a natural number prime to r with 1 ≤ α ≤ l. Using the HeLP-method and εx(u) = 0 for ◦(x) < rn we obtain, doing the same calculations as in (1): 1 = xi. Then xr 1 , a2, a−1 i µ(ξα, u, ϕrk) = 1 r µ(ξαr, ur, ϕrk) + Tr(ξ−α) rn + 1 rn Xi εxi(u)Tr((ξi + ξ−i)ξ−α). (4) As ur is rationally conjugate to xr get 1 we know that ξ±r are eigenvalues of Θrk(ur). So we µ(ξα, u, ϕrk) =( 1, α ≡ ±1 (rn−k) 0, else and µ(ξαr, ur, ϕrk) =( 1, α ≡ ±1 (rn−k−1) 0, else 9 There are now several cases to consider: Statement (ii) from b) is clear for k = 0. So let k < n − 1. For i 6≡ ±1 (rn−k−1), set α = i. Then µ(ξα, u, ϕrk) = µ(ξrα, ur.ϕrk) = 0 and since ξ has order at least r2, also Tr(ξ) = 0. Thus we can do the same computations as in (2) to obtain (ii) for k + 1. So (ii) holds for k = n − 1. To obtain the base case for (i) set k = 0. Then ξ is at least of order r2, i.e. Tr(ξ) = 0, and from (4) we obtain (similar to the computation in (2)): and 1 = 0 = 1 r 1 r + εx1(u) − + εxα(u) − 1 r Xi≡±1(rn−1) εxi(u) 1 r Xi≡±1(rn−1) εxi(u) for α ≡ ±1 (rn−1) and α 6= 1. Substracting (6) from (5) gives 1 = εx1(u) − εxα(u) (5) (6) (7) for every α ≡ ±1 (rn−1) and α 6= 1. Let t = {i ∈ Ni ≤ l, i ≡ ±1 (rn−1)}. Then summing up (5) with the equations (6) for all α ≡ ±1 (rn−1) gives 1 = t r εxi(u) − + Xi≡±1(rn−1) t r Xi≡±1(rn−1) εxi(u) = t r + (1 − t r εxi(u). ) Xi≡±1(rn−1) εxi(u) = 1 and the base case of (i) follows from (7). So Xi≡±1(rn−1) Xi≡±1 (rn−k) So assume 0 ≤ k < n − 1. Then ξ is at least of order r2, i.e. Tr(ξ) = 0. By induction εxi(u) = 1 and for α ≡ ±1 (rn−k) from (4) computing as in (2) we obtain 1 = 1 r εxi(u) − + Xi≡±1(rn−k) 1 r Xi≡±1(rn−k−1) εxi(u) = 1 r + 1 − 1 εxi(u). r Xi≡±1(rn−k−1) For α 6≡ ±1 (rn−k) and α ≡ ±1 (rn−k−1) we obtain the same way 1 εxi(u). r Xi≡±1(rn−k−1) 0 = 1 r εxi(u) − + Xi≡±α(rn−k) 10 Thus subtracting the last equation from the one before gives 1 = 1 − Xi≡±α(rn−k) εxi(u). The summands on the right hand side are all equal by (ii), so εxα(u) = 0, as claimed. Finally let r be odd, k = n − 1 and α 6≡ ±1 (r). Then ξ is of order r, i.e. Tr(ξ) = −rn−1, and µ(ξαr, ur, ϕrk) = µ(1, ur, ϕrk) = 3. So from (4) computing as in (5) we obtain 0 = 3 r + −rn−1 rn − 2 r Xi εxi(u) + Xi≡±α(r) εxi(u) = Xi≡±α(r) εxi(u). As by (ii) all summands in the last sum are equal, we get εxα(u) = 0 and the Theorem is finally proved. Proof of Theorem 2: Let G = PSL(2, pf ) such that f = 1 or p = 2. Assume first that r is an odd prime, which is not p, and R is an r-subgroup of V (ZG). As every r-subgroup of G is cyclic so is R by [Her08b, Theorem A] and thus R is rationally conjugate to a subgroup of G by Theorem 1. If p 6= 2 and R is a 2-subgroup of V (ZG), then R is either cyclic or dihedral or a Kleinian four group by [HHK09, Theorem 2.1]. If R is cyclic, then it is rationally conjugate to a subgroup of G by Theorem 1. If R is dihedral or a Kleinian four group let S = hsi be a maximal cyclic subgroup of R. Then any element of R outside of S is an involution and moreover s is rationally conjugate to an element g ∈ G by Theorem 1. R is isomorphic to some subgroup H of G, such that the maximal cyclic subgroup of H is generated by g. As there is only one conjugacy class of invoultions in G every isomorphism σ between R and H mapping s to g satisfies χ(σ(u)) = χ(u) for every irreducible complex character of G and every u ∈ R. Thus R is rationally conjugate to H by Lemma 1.6. If p = 2 and P is a 2-subgroup of V (ZG) then all non-trivial elements of P are involu- tions, so P is elementary abelian and the order of P divides the order of G. As there is again only one conjugacy class of involutions in G every isomorphism σ between P and a subgroup of G isomorphic with P satisfies χ(σ(u)) = χ(u) for every irreducible complex character of G and every u ∈ P. So P is rationally conjugate to a subgroup of G by Lemma 1.6. Finally if p is odd and P is a p-subgroup of V (ZG), then P is cyclic of order p and thus rationally conjugate to a subgroup of G by Lemma 1.3. 11 Remark: Let G = PSL(2, pf ) and let n be a number prime to p. The structure of the Brauer table of G in defining characteristic yields immidiately, that if we can prove that a unit u ∈ V (ZG) of order n is rationally conjugate to an element in G applying the HeLP-method to the Brauer table, then these calculations will hold over any PSL(2, q), if n and q are coprime. In this sense it would be interesting, and seems actually achiev- able, to determine a subset Apf of N such that we can say: The HeLP-method proves that a unit u ∈ V (ZG) of order n is rationally conjugate to an element in G if and only if n ∈ Apf . Test computations yield the conjecture that Apf actually contains all odd numbers prime to p. If this turned out to be true this would yield, using the results in [Her07], the First Zassenhaus Conjecture for the groups PSL(2, p), where p is a Fermat- or Mersenne prime. Other interesting questions concerning torsion units of the integral group ring of G = PSL(2, pf ) were mentioned at the end of [HHK09] and are still open today: If the order of u ∈ V (ZG) is divisible by p, is u of order p? Are units of order p rationally conjugate to elements of G? Are p-subgroups in V (ZG) necessarily abelian? For the last question a positive answer in case f = 3 is given in [BM15b]. Acknowledgement: The computations given above were all done by hand, but some motivating computations were done using a GAP-implementation of the HeLP-algorithm written by the author in collaboration with Andreas Bächle [BM15a]. I also thank the referee for many suggestions improving the readability of the paper. References [Alp86] J. L. Alperin, Local representation theory, Cambridge Studies in Advanced Mathematics, vol. 11, Cambridge University Press, Cambridge, 1986, Mod- ular representations as an introduction to the local representation theory of finite groups. [BHK04] V. Bovdi, C. Höfert, and W. Kimmerle, On the first Zassenhaus conjecture for integral group rings, Publ. Math. Debrecen 65 (2004), no. 3-4, 291–303. [BK11] [Ble99] A. Bächle and W. Kimmerle, On torsion subgroups in integral group rings of finite groups, J. Algebra 326 (2011), 34–46. Frauke M. Bleher, Finite groups of Lie type of small rank, Pacific J. Math. 187 (1999), no. 2, 215–239. 12 [BM15a] [BM15b] [BN41] [CL65] A. Bächle and L. Margolis, HeLP – Hertweck-Luthar-Passi method, GAP package, Version 1.0, http://homepages.vub.ac.be/abachle/help/, 2015. Andreas Bächle and Leo Margolis, Torsion subgroups in the units of the integral group ring of PSL(2, p3), Arch. Math. (Basel) 105 (2015), no. 1, 1–11. R. Brauer and C. Nesbitt, On the modular characters of groups, Ann. of Math. (2) 42 (1941), 556–590. James A. Cohn and Donald Livingstone, On the structure of group algebras. I, Canad. J. Math. 17 (1965), 583–593. [CMdR13] Mauricio Caicedo, Leo Margolis, and Ángel del Río, Zassenhaus conjecture for cyclic-by-abelian groups, J. Lond. Math. Soc. (2) 88 (2013), no. 1, 65–78. [Dic01] [DJ96] [Her06] Leonard Eugene Dickson, Linear groups: With an exposition of the Galois field theory, Teubner, Leipzig, 1901. Michael A. Dokuchaev and Stanley O. Juriaans, Finite subgroups in integral group rings, Canad. J. Math. 48 (1996), no. 6, 1170–1179. Martin Hertweck, On the torsion units of some integral group rings, Algebra Colloq. 13 (2006), no. 2, 329–348. [Her07] , Partial augmentations and Brauer character values of torsion units in group rings, arXiv:math.RA/0612429v2, 2004 - 2007. [Her08a] , Torsion units in integral group rings of certain metabelian groups, Proc. Edinb. Math. Soc. (2) 51 (2008), no. 2, 363–385. [Her08b] , Unit groups of integral finite group rings with no noncyclic abelian finite p-subgroups, Comm. Algebra 36 (2008), no. 9, 3224–3229. [HHK09] Martin Hertweck, Christian R. Höfert, and Wolfgang Kimmerle, Finite groups of units and their composition factors in the integral group rings of the group PSL(2, q), J. Group Theory 12 (2009), no. 6, 873–882. [KR93] W. Kimmerle and K. W. Roggenkamp, A Sylow-like theorem for integral group rings of finite solvable groups, Arch. Math. (Basel) 60 (1993), no. 1, 1–6. 13 [LP89] I. S. Luthar and I. B. S. Passi, Zassenhaus conjecture for A5, Proc. Indian Acad. Sci. Math. Sci. 99 (1989), no. 1, 1–5. [MRSW87] Z. Marciniak, J. Ritter, S. K. Sehgal, and A. Weiss, Torsion units in integral group rings of some metabelian groups. II, J. Number Theory 25 (1987), no. 3, 340–352. [Pet76] [Rog91] [Seh93] [Sri64] [Val94] [Wag95] [Wei88] Gary L. Peterson, Automorphisms of the integral group ring of Sn, Proc. Amer. Math. Soc. 59 (1976), no. 1, 14–18. Klaus W. Roggenkamp, Observations on a conjecture of Hans Zassenhaus, Groups-St. Andrews 1989, Vol. 2, London Math. Soc. Lecture Note Ser., vol. 160, Cambridge Univ. Press, Cambridge, 1991, pp. 427–444. S. K. Sehgal, Units in integral group rings, Pitman Monographs and Surveys in Pure and Applied Mathematics, vol. 69, Longman Scientific & Technical, Harlow; copublished in the United States with John Wiley & Sons, Inc., New York, 1993, With an appendix by Al Weiss. Bhama Srinivasan, On the modular characters of the special linear group SL(2, pn), Proc. London Math. Soc. (3) 14 (1964). Angela Valenti, Torsion units in integral group rings, Proc. Amer. Math. Soc. 120 (1994), no. 1, 1–4. R. Wagner, Zassenhausvermutung über die Gruppen PSL(2, p), Master's the- sis, Universität Stuttgart, Mai 1995. Alfred Weiss, Rigidity of p-adic p-torsion, Ann. of Math. (2) 127 (1988), no. 2, 317–332. [Wei91] , Torsion units in integral group rings, J. Reine Angew. Math. 415 (1991), 175–187. [Zas74] Hans Zassenhaus, On the torsion units of finite group rings, Studies in math- ematics (in honor of A. Almeida Costa) (Portuguese), Instituto de Alta Cultura, Lisbon, 1974, pp. 119–126. [ŽK67] È. M. Žmud and G. Č. Kurennoı, The finite groups of units of an integral group ring, Vestnik Har'kov. Gos. Univ. 1967 (1967), no. 26, 20–26. 14 Leo Margolis, Fachbereich Mathematik, Universität Stuttgart, Pfaffenwaldring 57, 70569 Stuttgart, Germany. [email protected] 15
1806.10386	2	1806	2018-11-05T09:04:29	Closure operations, Continuous valuations on monoids and Spectral spaces	[ "math.RA", "math.AC" ]	We present several naturally occurring classes of spectral spaces using commutative algebra on pointed monoids. For this purpose, our main tools are finite type closure operations and continuous valuations on monoids which we introduce in this work. In the process, we make a detailed study of different closure operations on monoids. We prove that the collection of continuous valuations on a topological monoid with topology determined by any finitely generated ideal is a spectral space.	math.RA	math	Closure operations, Continuous valuations on monoids and Spectral spaces Samarpita Ray∗ Abstract We present several naturally occurring classes of spectral spaces using commutative algebra on pointed monoids. For this purpose, our main tools are finite type closure operations and continuous valuations on monoids which we introduce in this work. In the process, we make a detailed study of different closure operations on monoids. We prove that the collection of continuous valuations on a topological monoid with topology determined by any finitely generated ideal is a spectral space. 1 Introduction In basic commutative algebra, topological spaces arise essentially in two contexts. First, as the Zariski spec- trum Spec(A) of a commutative ring A and secondly while topologizing a ring with the I-adic topology, where I is an ideal of the ring (see, for instance, [1]). A celebrated result by Hochster ([13]) shows that a topological space X is homeomorphic to Spec(A) for some commutative ring A if and only if X has the following properties : (1) X is T0, (2) X is quasi-compact (i.e., any open cover of X admits a finite subcover), (3) X admits a basis of quasi-compact open subspaces that is closed under finite intersections (4) and every irreducible closed subspace Y of X has a unique generic point. The striking thing about the above description is the fact that it is purely topological in nature. Such a topological space X is called spectral. Unlike the Zariski spectrum, the relation between the I-adic topology and spectral spaces is not so apparent. However, one of the most important tools to understand the I-adic topology is through valuations (see, for instance, [4]) and several spectral spaces naturally emerge on study- ing valuations. For example, given any field F , the set of all valuation rings having the quotient field F is a spectral space (see, for instance, [20, Example 2.2 (8)]). The valuation spectrum of any ring is a spectral space [15]. Moreover, for any ring A with the I-adic topology having a finitely generated ideal of definition, the space of continuous valuations Cont(A) forms a spectral space [15]. On the one hand, the Zariski spectra are the building blocks of schemes in algebraic geometry and on the other, the space Cont(A) is the starting point of non-archimedean geometry. This shows that spectral spaces help in linking algebra, topology and geometry. ∗Department of Mathematics, Indian Institute of Science, Bangalore, India. email: [email protected]. Keywords: abelian monoids; Spectral space and spectral map; Zariski, hull-kernel, patch, inverse and ultrafilter topologies; closure operation; continuous valuation. MSC(2010): 13A15, 13A18, 13B22, 54D80 1 Recently, Finocchiaro [8] developed a new criterion involving ultrafilters to characterize spectral spaces. This helped them to show that certain familiar objects appearing in commutative algebra can be realized as the spectrum of a ring (see, for instance, [9], [10]). For example, Finocchiaro, Fontana and Spirito proved that the collection of submodules of a module over a ring has the structure of a spectral space [10]. They further used closure operations from classical commutative ring theory to bring more such spectral spaces in light. Moreover, the methods of [10] were used in [2], [3] to uncover many spectral spaces coming from objects in an abelian category and from modules over tensor triangulated categories. In this paper, we use both Hochster's characterization and Finocchiaro's criterion to present natural classes of spectral spaces involving pointed monoids. Our interest in monoids began by looking at the paper [12] by Weibel and Flores which studies certain geometric structures involving monoids. This interest in the geometry over monoids lies in its natural as- sociation to toric geometry, which was pointed out in the work of Cortias, Haesemeyer, Walker and Weibel [6]. A detailed work on the commutative and homological algebra on monoids was presented by Flores in [11]. As such, in this work we look into the topological aspects of monoids via spectral spaces. We begin by showing that the collection of all prime ideals of a monoid or, in other words, the spectrum of a monoid, endowed with the Zariski topology is homeomorphic to the spectrum of a ring, i.e., it is a spectral space. We further prove that the collection of all ideals as well as the collection of all proper ideals of a monoid are also spectral spaces (Corollary 3.4). As in [11], the notion of A-sets over a monoid A is the analogue of the notion of modules over a ring. We introduce closure operations on monoids and obtain natural classes of spectral spaces using finite type closure operations on A-sets. In the process, different notions of closure operations like integral, saturation, Frobenius and tight closures are introduced for monoids inspired by the corresponding closure operations on rings from classical commutative algebra. We discuss their persistence and localization properties in detail. Other than rings, valuation has also been studied on ring-like objects like semirings (see, for instance [18], [19]). Valuation monoids were studied in [6]. In this work, we prove that the collection of all valuation monoids having the same group completion forms a spectral space (Proposition 5.3) and that the valuation spectrum of any monoid gives a spectral space (Theorem 5.18). Finally, we prove that the collection of continuous valuations on a topological monoid whose topology is determined by any finitely generated ideal (in the sense of Definition 5.27) also gives a spectral space (Theorem 5.30). The paper is organized as follows. In §2, we recall some background materials on commutative algebra on monoids and spectral spaces. In §3, we obtain classes of spectral spaces using A-sets over a monoid A and in particular using ideals of a monoid A. For instance, we prove that the collection of A-subsets of an A-set forms a spectral space in a manner similar to [10]. Following this, in §4, we introduce different closure operations on monoids and study their persistence and localization properties as well. This is done with a view towards obtaining spectral spaces involving finite type closure operations. Finally, in §5, we study valuations and continuous valuations on monoids and several classes of spectral spaces appear. Acknowledgements: I am grateful to my Ph.D advisor Professor Abhishek Banerjee for many helpful discussions on this work. I would like to thank Professor Sudesh Kaur Khanduja and Professor Pooja Singla for their helpful comments and suggestions. I would also like to express my gratitude to the anonymous referee for a very careful reading of the paper and for several useful comments and suggestions. 2 Background 2.1 Monoids Throughout this work, by a monoid we shall always mean a commutative pointed (unique basepoint 0) monoid with identity element 1. A morphism of monoids is a basepoint preserving map defined in the 2 obvious way. Let us now briefly recall the commutative algebra on monoids from [11]. A monoid A is called cancellative if for all a, b, c ∈ A with a 6= 0, ab = ac implies b = c. An ideal I of a monoid A is a pointed subset such that IA ⊆ A. An ideal I ⊆ A is generated by a subset Y of I if every x ∈ I can be written as ay for some a ∈ A and some y ∈ Y . If this Y can be chosen to be finite, I is called a finitely generated ideal. An ideal p 6= A is called prime if the complement A\ p is closed under multiplication, or equivalently if A/p is torsion free (see, for instance, [11, § 2.1.3]). The complement of the set of units A× of a monoid A forms an ideal and is therefore the unique maximal ideal of A, written as mA. In this sense, every monoid is local. A morphism f : A −→ B is local if f (mA) ⊆ mB. Let I ⊆ A be an ideal. The subset I × {0} ⊆ A × A generates a congruence, i.e., an equivalence relation compatible with the monoid operation, whose associated quotient monoid is written as A/I and identifies all elements of I with 0 leaving A \ I untouched. An A-set is a commutative pointed set M with a binary operation · : A × M −→ M satisfying, (i) 1.x = x, (ii) 0A.x = 0X ; a.0X = 0A and (iii) (a.b).x = a.(b.x) for every a, b ∈ A and x ∈ M , where 0A denotes the basepoint of A and 0M denotes the basepoint of M . Morphisms of A-sets are defined in the obvious way. A subset N of an A-set M is called an A-subset of M if ay ∈ N for every a ∈ A and y ∈ N . An A-set M is said to be finitely generated if there exists a finite subset S ⊆ M such that every x ∈ M can be written as x = az for some a ∈ A and z ∈ S. An A-set is Noetherian if every A-subset is finitely generated. Localization of a monoid A or of an A-set are defined in the obvious way. When S is the multiplicatively closed subset of non-zero divisors of A, the localization S−1A is called the total monoid of fractions. When A is torsion free, i.e., 0 generates a prime ideal, the localization A+ := A(0) of A at (0) is called the group completion of A. Note that it is A+ \ 0 which is a group but not A+. The canonical map A → A+ is an inclusion only when A is cancellative. 2.2 Spectral spaces A topological space X is called a spectral space (after Hochster [13]) if it is homeomorphic to the spectrum Spec(R) of a ring R with the Zariski topology. As stated in the introduction, spectral spaces can be described in purely topological terms. A map f : X −→ Y of spectral spaces is called a spectral map if for any open and quasi-compact subspace Y ′ of Y , the set f −1(Y ′) is open and quasi-compact. In particular, any spectral map of spectral spaces is continuous. A spectral space X need not be Hausdorff unless X is zero dimensional. However, there is a natural way to refine the topology of X in order to make X an Hausdorff space without losing compactess. The patch topology (or constructible topology) on X is the topology which has as a sub-basis for its closed sets the closed sets and quasi-compact open sets of the original space, or in other words, which has the quasi-compact open sets and their complements as an open sub-basis. Endowing X with the patch topology makes it a Hausdorff spectral space (see, [13, § 2]) and it is denoted as Xpatch. By a patch in X we mean a subset of X closed in Xpatch. If Y is a patch in a spectral space X, then Y is also a spectral space with the subspace topology (see, [13, Proposition 9] or [20, Proposition 2.1] ). We shall use this property in Section 5 to show certain topological spaces are spectral. Another way of understanding spectral spaces is via ultrafilters. Let us recall the definition of ultrafilters (say from [8]) before we state the criterion. Definition 2.1. A nonempty collection F of subsets of a given set X is called a filter on X if the following properties hold: (i) ∅ /∈ F . (ii) If Y, Z ∈ F , then Y ∩ Z ∈ F . (iii) If Z ∈ F and Z ⊂ Y ⊂ X, then Y ∈ F . 3 A filter F on X is an ultrafilter if and only if it is maximal. Equivalently, a filter F is an ultrafilter on X if for each subset Y of X, either Y ∈ F or X \ Y ∈ F . We shall denote an ultrafilter by U. For further details and examples of filters, see, for instance, [17]. Theorem 2.2. (see,[8, Corollary 3.3]) Let X be a topological space. (1) The following conditions are equivalent: (i) X is a spectral space. (ii) X satisfies the T0-axiom and there is a subbasis S of X such that XS(U) := {x ∈ X [∀S ∈ S, x ∈ S ⇐⇒ S ∈ U]} 6= ∅ for any ultrafilter U on X. (2) If the previous equivalent conditions hold and S is as in (ii), then a subset Y of X is closed, with respect to the patch topology, if and only if YS(V) := {x ∈ X [∀S ∈ S, x ∈ S ⇐⇒ S ∩ Y ∈ V]} is a subset of Y for any ultrafilter V on X. Corollary 2.3. (see,[10, Corollary 1.2]) Let X be a topological space satisfying the equivalent conditions of Theorem 2.2 and let S be as in Theorem 2.2(ii). Then S is a subbasis of quasi-compact open subspaces of X. 3 A-subsets and Spectral spaces We begin this section by showing that the spectrum of a monoid, topologically speaking, is same (homeo- morphic) to the spectrum of a ring, i.e., it is a spectral space. Proposition 3.1. Let M Spec(A) denote the set of all prime ideals of a monoid A. It is a topological space when equipped with the Zariski topology whose closed sets are given by V (I) := {p ∈ M Spec(A) I ⊆ p} for an ideal I of A. Then, M Spec(A) is a spectral space. Proof. Clearly, a subbasis of open sets for this topology is given by {D(x) x ∈ A} where D(x) := {p ∈ M Spec(A) x /∈ p}. Further, since D(x) ∩ D(y) = D(xy) for any x, y ∈ A, the collection {D(x) x ∈ A} gives a basis for the Zariski topology on M Spec(A). For any D(x) containing the unique maximal ideal mA of A, we have D(x) = M Spec(A) and this shows that M Spec(A) is quasi-compact. Since this implies that M Spec(A) is quasi-compact for any monoid A, we have D(x), which is same as M Spec(Ax), is quasi- compact for all x ∈ A. Hence, {D(x) x ∈ A} gives a basis of quasi-compact open subspaces that is clearly closed under finite intersections. It is also easy to see that M Spec(A) with this topology is T0. Let Y be an irreducible closed subspace of M Spec(A). Then, Y can be partially ordered by inclusion, i.e., p ≤ q ⇐⇒ p ⊆ q ⇐⇒ q ∈ {p}. Consider any arbitrary non-empty chain of prime ideals {pα}α∈Λ in Y . Since Y is a closed subset, Y = V (I) for some ideal I of A. Thus, I ⊆ pα for all α ∈ Λ. Hence, I ⊆ ∩α∈Λpα and therefore ∩α∈Λpα, which is a prime ideal, is in Y . Thus, by Zorn's lemma, Y has a minimal element and let us denote it by pY . Then, {pY } = Y because otherwise it will contradict the irreducibility of Y . Hence, pY is a generic point of Y and it is unique because of being minimal. Thus, M Spec(A) is a spectral spaces by Hochster's characterization. Let S be any set. We denote the power set of S as 2S. The topology on 2S given by an open sub-basis of sets of the form, D(F ) := {R ⊆ S F * R} where F is a finite subset of S, is called the hull-kernel topology on 2S. The complement of the set D(F ) is denoted as V (F ) := {R ⊆ S F ⊆ R}. Given any A-set M , let SSet(MA) denote the set of all A-subsets of M . By the hull-kernel topology on SSet(MA), we will denote the subspace topology induced by the hull-kernel topology on 2M . We will show that SSet(MA) is a spectral space. This is analogous to the result of Finocchiaro, Fontana and Spirito [10, Proposition 2.1] for the case of submodules of a module over a ring. 4 Proposition 3.2. For any monoid A and an A-set M , SSet(MA) is a spectral space. Moreover, the collection of sets S := {D(x1, . . . , xn) x1, . . . , xn ∈ M} is a subbasis of quasi-compact open subspaces of SSet(MA). Proof. Let X denotes SSet(MA). Let N, N ′ ∈ X such that N 6= N ′. Suppose, without loss of generality, that there exists some y ∈ N such that y /∈ N ′. Then SSet(MA) \ V (y) is an open neighborhood of N ′ not containing N . This shows that the hull-kernel topology on X is T0. By Theorem 2.2, it is now enough to show that there is a subbasis S of X such that XS(U) 6= ∅ for any ultrafilter U on X. For a given ultrafilter U of X, set NU := {y ∈ M V (y) ∈ U}. For a ∈ A and y ∈ NU , we have V (y) ⊆ V (ay) and hence ay ∈ NU (Definition 2.1(iii)). Hence, NU is an A-subset of M . We claim that NU ∈ XS(U) where XS(U) is as in Theorem 2.2. Indeed, if NU ∈ S = D(x1, . . . , xn) for some S ∈ S, then there exists at least one xi /∈ NU for i = 1, . . . , n. Therefore, by the definition of NU , V (xi) /∈ U and this implies D(xi) ∈ U. Since D(xi) ⊆ D(x1, . . . , xn), we have S = D(x1, . . . , xn) ∈ U (Definition 2.1(iii)). Conversely, let S = D(x1, . . . , xn) ∈ U. If possible, let NU /∈ D(x1, . . . , xn). Then NU ∈ V (x1, . . . , xn) i.e., V (xi) ∈ U for all i. Since U is a filter, Tn i=1 V (xi) = V (x1, . . . , xn) ∈ U. Hence, both D(x1, . . . , xn) and V (x1, . . . , xn) belongs to U and so does their intersection, i.e., ∅ ∈ U. This contradicts the definition of a filter. Thus NU ∈ D(x1, . . . , xn) and our claim is proved. Thus, X is a spectral space and the sets in S gives a subbasis of quasi-compact open subspaces of X by Corrolary 2.3. The next proposition is an analogue of the result [10, Proposition 2.4] for the case of modules over a ring. Proposition 3.3. Let M be an A-set. Then, SSet•(MA) := SSet(MA) \ M endowed with the hull-kernel subspace topology is a spectral space if and only if M is finitely generated. Proof. Assume that M is a finitely generated A-set with a finite set of generators, which we denote by F . A subbasis S for X := SSet•(MA) is given by the open sets {D(x1, ..., xn) x1, ...., xn ∈ M}. If U is an ultrafilter on X, the subset NU := {y ∈ M V (y) ∩ X ∈ U} is an A-subset of M following the proof of Proposition 3.2. If only we can show that NU is a proper subset of M , it follows immediately that NU ∈ XS(U) proving that X is spectral by Theorem 2.2. If NU = M , then V (F ) ∩ X ∈ U and as the empty set is not a member of any filter, we must have V (F ) ∩ X 6= ∅. So, pick any A-subset N ∈ V (F ) ∩ X. But then N contains F , the generating set of M . This shows that M = N ∈ U giving a contradiction. Thus, we have NU 6= M . Conversely, assume that M is not finitely generated. Then the family of open subsets {D(x) x ∈ M} gives a cover of X such that for any finite subset F of M , the collection of open sets {D(x)x ∈ F} is not a subcover of X. Indeed, consider the A-subset N := hFi of M . Since M is not finitely generated, clearly N 6= M . Thus, N ∈ X \S{D(x) x ∈ F}. This shows that X is not quasi-compact and hence not a spectral space if M is not finitely generated. This completes the proof. As a particular case of Propositions 3.2 and 3.3, we have the following corollary. Corollary 3.4. The set of all ideals of A, Id(A) := SSet(AA) and the set of all proper ideals of A, Id(A)• := SSet(AA) \ {A}, are spectral spaces. 4 Closure operations on monoids and spectral spaces In commutative ring theory, closure operations are defined on ideals of a ring or more generally on submodules of a module over a ring (see, for instance, [7]). We begin this section by defining closure operations for monoids likewise. 5 Definition 4.1. Given a monoid A and an A-set M , a closure operation on SSet(MA) is a map cl : SSet(MA) −→ SSet(MA) N 7→ N cl satisfying the following conditions: (i) (Extension) N ⊆ N cl for all N ∈ SSet(MA). (ii) (Order-preservation) If N1 ⊆ N2, then N cl (iii) (Idempotence) N cl = (N cl)cl for all N ∈ SSet(MA). A closure operator cl is said to be finite type if N cl := S{Lcl L ⊆ N, L ∈ SSet(MA), L is finitely generated }. 2 for all N1, N2 ∈ SSet(MA). 1 ⊆ N cl Given any closure operation on submodules of a module over a ring, a closure operation of finite type can be constructed from it (see, for instance [7, Construction 3.1.6]). We prove the same for A-subsets of an A-set over a monoid A in the following proposition. Proposition 4.2. Let cl : SSet(MA) −→ SSet(MA) be a closure operation on SSet(MA). We set, N clf := [{Lcl L ⊆ N, L ∈ SSet(MA), L is finitely generated } Then clf is a closure operator of finite type. Further, cl is of finite type if and only if cl = clf . Proof. Take any x ∈ N and consider the finitely generated A-subset hxi of N . Since hxi ⊆ hxicl, we have x ∈ N clf . This shows extension. Order-preservation is obvious. As for idempotence, suppose x ∈ (N clf )clf . Then there exists some finitely generated A-subset L ⊆ N clf such that x ∈ Lcl. Let {l1, l2, ..., ln} be a finite cl. We generating set of L. Since each li ∈ N clf , there exists finitely generated Ki ⊆ N such that li ∈ Ki set, K := Sn i=1 Ki. Clearly, K is finitely generated. Any element of L is of the form λli for some λ ∈ A In other words, L ⊆ K cl. and li belonging to the generating set of L and therefore λli ∈ Ki Hence, x ∈ Lcl ⊆ (K cl)cl = K cl and this implies x ∈ N clf . So we have, (N clf )clf ⊆ N clf . The other way, N clf ⊆ (N clf )clf , follows from the extension property of clf . Thus, we have shown idempotence. Hence, clf is a closure operation. The last statement of the lemma follows from the definition of finite type closure operation. cl ⊆ K cl. The next proposition is an analogue of the result, [10, Proposition 3.4] on finite type closure operations on modules over rings. Proposition 4.3. Let M be an A-set and cl be a closure operation of finite type on SSet(MA). Then, the set SSetcl(MA) := {N ∈ SSet(MA) N = N cl} is a spectral space. Moreover, SSetcl(MA) is closed in SSet(MA), endowed with the patch topology. Proof. Let U be an ultrafilter on SSetcl(MA). It is enough to show that NU (as in Proposition 3.2) is in SSet(MA)cl to prove that SSetcl(MA) is spectral (by Theorem 2.2). Let x ∈ (NU )cl. Since cl is a closure operation of finite type, there exists a finitely generated A-set hl1, . . . , lni ⊆ NU such that x ∈ hl1, . . . , lnicl. Hence, x ∈ K cl for all K ⊇ hl1, . . . , lni. In other words, x ∈ K cl for all K ∈ V (l1, . . . , ln). Therefore, V (l1, . . . , ln) ∩ SSetcl(MA) ⊆ V (x) ∩ SSetcl(MA). Since V (li) ∩ SSetcl(MA) ∈ U for all li i=1(cid:0)V (li) ∩ SSetcl(MA)(cid:1) = V (l1, . . . , ln) ∩ SSetcl(MA) ∈ U. Hence, (by the definition of NU ), we have ∩n V (x) ∩ SSetcl(MA) ∈ U and so x ∈ NU . This shows (NU )cl ⊆ NU and thus NU = (NU )cl. Hence, SSetcl(MA) is a spectral space. Finally, it follows from Theorem 2.2(2) that SSetcl(MA) is closed in SSet(MA), endowed with the patch topology. Traditionally closure operations for rings are defined on a whole class of rings rather than one ring at a time (see, for instance, [7]). Similarly, we can define closure operations on a class of monoids and define an important property of closure operations called persistence : 6 Definition 4.4. Let M be a subcategory of the category of monoids. Let cl be a closure operation defined on the monoids of M. We say that cl is persistent if for any morphism of monoids φ : A −→ B in M and any ideal I of A, we have φ(I cl)B ⊆ (φ(I)B)cl. For a persistent closure operation cl on a subcategory M of monoids, we say that cl commutes with localization in M if for any monoid A ∈ M and any multiplicative set S ⊆ A, such that the localization map A −→ S−1A is in M, we have S−1(I cl) = (S−1I)cl for every ideal I of A. Examples 4.5. Simple examples of closure operations on I, the collection of all ideals of a monoid A, are: (1) (Identity closure) I cl := I for all I ∈ I. (2) (Indiscrete closure) I cl := A for all I ∈ I. (3) We define the radical closure of an ideal I ∈ I as I cl := √I for all I ∈ I where √I is the intersection of all the prime ideals of A containing I or equivalently it is the set {x ∈ A ∃ n ∈ N such that xn ∈ I}. It is easy to show that identity, indiscrete and radical closures are persistent on the category of all monoids and commutes with localization. We now introduce some interesting closure operations on monoids like Frobenius, tight, integral and satu- ration closure operations, which are inspired by such similar operations for rings in classical commutative algebra. We begin by introducing Frobenius closure of ideals over monoids. See, for instance, [7] for details on Frobenius closure of ideals over rings. Given an ideal I of a monoid A, let I n := {Qn Proposition 4.6. Let I ∈ I. We set, i=1 xi xi ∈ I} and I [n] := {axn a ∈ A, x ∈ I}. I F rob := {x ∈ A ∃ n ∈ N such that xn ∈ I [n]} (1) This defines a closure operation on I and we call this the Frobenius closure1 on I. (2) Frobenius closure is persistent on the category of all monoids and commute with localization. Proof. Since extension and order-preservation are obvious, it is enough to prove the idempotence condition for (1). Take x ∈ (I F rob)F rob. Then, xn ∈ (I F rob)[n] for some n ∈ N, i.e., xn = ayn for some a ∈ A and y ∈ I F rob. Again this implies yk ∈ I [k] for some k ∈ N, i.e., yk = a′zk for some a′ ∈ A and z ∈ I. Hence, xnk = (xn)k = ak(yk)n = ak(a′zk)n = aka′nznk. Therefore, xnk ∈ I [nk] or in other words x ∈ I F rob. The other inclusion is obvious. Let φ and S be as in Definition 4.4. Consider any φ(x) for x ∈ I F rob. Then, xn ∈ I [n] for some n ∈ N. Thus, xn = ayn for some y ∈ I and a ∈ A. Hence, (φ(x))n = φ(xn) = φ(ayn) = φ(a)φ(y)n ∈ (φ(I)B)[n]. This shows persistence. Now, pick any x/s ∈ S−1(I F rob) where x ∈ I F rob and s ∈ S. Since x ∈ I F rob, there exists some n ∈ N such that xn ∈ I [n]. Thus, xn = ayn for some a ∈ A and y ∈ I. Since, (x/s)n = xn/sn = ayn/sn = (a/1)(y/s)n ∈ (S−1I)[n], we have x/s ∈ (S−1I)F rob. This proves (2). Tight closure of ideals over rings was introduced by M. Hochster and C. Huneke in 1986 [14]. Inspired by the same, we introduce tight closure of ideals over monoids. Proposition 4.7. Let A be a Noetherian cancellative monoid and I ∈ I. Then, I must be finitely generated and let {y1, . . . , yr} denote a finite set of generators of I. We set, I tight := {x ∈ A ∃ 0 6= a ∈ A such that axn ∈ I [n] ∀ n ≫ 0}. 1We call this the "Frobenius closure" closure because it is motivated by the usual Frobenius closure for rings of characteristic p > 0. If R is a ring of characteristic p > 0, the association a 7→ ape (e ∈ N) defines a ring endomorphism. Clearly, this does not hold in general for arbitrary n ∈ N. However, since the additive structure does not arise in the case of monoids, we can consider here a Frobenius closure for any n. 7 (1) This defines a closure operation on I and we call this the tight closure operation. (2) Tight closure is persistent on the subcategory M of Noetherian cancellative monoids whose morphisms have zero kernel. Proof. Note that any element of I [n] is of the form ayn where a ∈ A and y ∈ I. Again, y must be of the form a′yi for some a′ ∈ A and yi ∈ {y1, . . . , yr}. Henceforth, I [n] = hyn r i. Let {z1, . . . , zk} be a finite set of generators of the ideal I tight. Pick any x ∈ (I tight)tight. Then, there exists a ∈ A such for some a′ ∈ A and for some zi ∈ {z1, . . . , zk}. that axn ∈ hzn Since zi ∈ I tight, there exists some 0 6= a′′ ∈ A such that a′′zn r i for all n ≫ 0. Therefore, r i for all n ≫ 0. Note that a′′a 6= 0 since both a′′, a 6= 0 and A is a cancellative a′′axn = a′′a′zn monoid. Thus, we have x ∈ I tight. Since I tight ⊆ (I tight)tight is obvious, we have I tight = (I tight)tight. This shows idempotence. Let φ ∈ M. Consider any φ(x) for x ∈ I tight. Then, there exists 0 6= a ∈ A such that axn ∈ I [n] forall n ≫ 0. This implies φ(a)φ(x)n = φ(axn) ∈ φ(I [n]) ⊆ φ(I [n])B = (φ(I)B)[n] for all n ≫ 0. But, φ(a) 6= 0 by assumption. This shows persistence. ki for all n ≫ 0, i.e, axn = a′zn 1 , ..., yn i ∈ hyn i ∈ hyn 1 , . . . , zn 1 , . . . , yn 1 , ..., yn i A classical study of integral closure of ideals over rings can be found, for instance, in [16]. We now introduce integral closure of ideals over monoids. Proposition 4.8. Let I ∈ I. We set, I int := {x ∈ A ∃ n ∈ N such that xn ∈ I n} (1) This defines a closure operation on I and we call this the integral closure. (2) Integral closure is persistent on the category of all monoids and commutes with localization. i=1 ki, then clearly yk Proof. Since extension and order-preservation are obvious, we only show idempotence for (1). The inclusion I int ⊆ (I int)int follows immediately from extension. To show the other way, pick any x ∈ (I int)int. Then, there exists some n ∈ N such that xn ∈ (I int)n. Thus, xn = Qn i=1 yi for some yi ∈ I int. Now, yi ∈ i ∈ I ki for some ki ∈ N. If we set, k = Qn I cl =⇒ yki i ∈ I k for all yi. Therefore, xnk = Qn i ∈ I nk and so x ∈ I int. i=1 yk Let φ and S be as in Definition 4.4. Consider any φ(x) for x ∈ I int. Then, xn ∈ I n for some n ∈ N. Hence, (φ(x))n = φ(xn) ∈ φ(I n) ⊆ φ(I n)B = (φ(I)B)n. This shows persistence. Now, pick any x/s ∈ S−1(I int) where x ∈ I int and s ∈ S. Since x ∈ I int, there exists some n ∈ N such that xn ∈ I n ⊆ I. Thus, (x/s)n = xn/sn ∈ S−1I which implies x/s ∈ (S−1I)int. Hence, S−1(I int) ⊆ (S−1I)int. For the other way, let y/t ∈ (S−1I)int for y ∈ A and t ∈ S. Then, (y/t)n ∈ (S−1I)n for some n ∈ N. Thus, (y/t)n = Qn i=1 yi/ti i=1 yi(cid:17). i=1 ti(cid:17) = utn(cid:16)Qn for some yi ∈ I and ti ∈ S. Hence, there exists some u ∈ S such that uyn(cid:16)Qn (cid:17) ∈ I n as yi ∈ I. Thus, Therefore, (cid:16)uy(cid:16)Qn y/t = uy(cid:16)Qn i (cid:17) = untn(cid:16)Qn i=1 ti(cid:17) ∈ S−1(I int). This proves (2). i=1 ti(cid:17)(cid:17)n i=1 ti(cid:17)/ut(cid:16)Qn = unyn(cid:16)Qn i=1 yi(cid:17)(cid:16)Qn i=1 tn−1 i=1 tn i See, for instance, [7] for details on saturation closure of ideals over rings. Here we introduce saturation closure of ideals over monoids. Proposition 4.9. Fix a finitely generated ideal a ∈ I. Then for any I ∈ I, we set I a−sat := [n∈N (I : an) = {x ∈ A ∃ n ∈ N such that anx ⊆ I} 8 (1) This defines a closure operation on I and we call this the a-saturation closure operation. (2) Consider the subcategory M of Noetherian monoids whose morphisms are surjective2 local morphisms. Saturation closure with respect to the unique maximal ideal of each monoid is persistent on this category and commutes with localization. Proof. We only show idempotence for (1). Take any x ∈ (I a−sat)a−sat. Then, there exists some nx ∈ N such that anxx ⊆ I a−sat = Sn∈N (I : an). Since a is finitely generated, so is anxx. Let S denote a finite generating set of anxx. Then, for each generator s ∈ S ⊆ anxx, there exists ns ∈ N such that anss ⊆ I. We set NS := Qs∈S ns ∈ N. Then, for every s ∈ S, aNS s ⊆ anS s ⊆ I. Consequently, anx+NS x ⊆ I. Hence, x ∈ I a−sat. The other inclusion I a−sat ⊆ (I a−sat)a−sat is obvious. Consider any x ∈ I a−sat. Then, there exists some n ∈ N such that (mA)nx ⊆ I. For any morphism φ : A −→ B in M, we have, φ(mA) = mB where mB denote the maximal ideal of B. Hence, (mB)nφ(x) = φ(mA)nφ(x) = φ((mA)nx) ⊆ φ(I). This shows persistence. Let mA denote the maximal ideal of a monoid A ∈ M. For an ideal I of A, the mA-saturation closure of I is defined as {x ∈ A ∃ n ∈ N such that (mA)nx ⊆ I}. Consider any multiplicative set S ⊆ A such that the localization A −→ S−1A is in M. This implies that S−1(mA) is the unique maximal ideal of S−1A. Pick any x/s ∈ S−1(I a−sat) for x ∈ I a−sat and s ∈ S. Then, there exists some n ∈ N such that (mA)nx ⊆ I. Clearly, this im- plies (S−1(mA))n(x/s) ⊆ S−1I and hence, S−1(I a−sat) ⊆ (S−1I)a−sat. To show the other way, take any y/t ∈ (S−1I)a−sat for y ∈ A and t ∈ S. Then, there exists some n ∈ N such that (S−1(mA))n(y/t) ⊆ S−1I. Since A is Noetherian, (mA)n is finitely generated and let {a1, . . . , ak} be a finite set of generators of (mA)n. Clearly, ai/1 ∈ (S−1(mA))n and hence (ai/1)(y/t) ∈ S−1I. This implies that there exists some si ∈ S such that aiysi ∈ I. Now, set y′ := y(Qk i=1 si). It is easy to see that (mA)ny′ ⊆ I. If s := Q si, then we have, y/t = y′/st ∈ S−1(I a−sat). 5 Continuous valuation and Spectral spaces Other than rings, valuation has also been studied on ring-like objects like semirings (see, for instance [18], [19]). Here, we study valuation on monoids. Given any field F , the set of all valuation rings having the quotient field F is a spectral space (see, for instance, [20, Example 2.2 (8)]). As shown by Huber in [15], the valuation spectrum of any ring gives a spectral space and for an f -adic (or Huber) ring, the collection of continuous valuations gives a spectral space too (see also, [22] and [5]). In this section, we prove similar results for monoids. We begin by briefly recalling totally ordered groups (resp. monoids). A totally ordered group (resp. totally ordered monoid) is a commutative group (resp. commutative monoid) Γ together with a total order ≤ on Γ which is compatible with the group operation. A homomorphism of totally ordered groups (resp. totally ordered monoids) is an order preserving homomorphism of groups (resp. monoids). Equivalently, a group homomorpism f : Γ1 −→ Γ2 is a homomorphism of totally ordered groups Γ1 and Γ2 if and only if for all a ∈ Γ1 with a ≤ 1, we have f (a) ≤ 1. Given a totally ordered abelian group (written multiplicatively) Γ, we can obtain a totally ordered monoid Γ∗ := Γ ∪ {0} by adjoining a basepoint 0 to Γ and extending the multiplication and the ordering of Γ to Γ∗ by setting γ · 0 = 0 · γ = 0 and γ ≥ 0 for all γ ∈ Γ. From now onwards, Γ∗ will always denote this monoid corresponding to a given group Γ. 2Note that the saturation closure with respect to the unique maximal ideal of each monoid is not persistent on the subcategory i of Noetherian monoids and local morphisms without the assumption of surjection. For example, consider the inclusion {0, 1} −→ A where A is any Noetherian cancellative monoid. Then the saturation closure of the ideal h0i in {0, 1} is {0, 1} and i({0, 1}) = {0, 1}. But, the saturation closure of the ideal hi(0)i is just {0} since A is cancellative. 9 Lemma 5.1. Let Γ be a totally ordered group with γ, δ,∈ Γ. Then, (1) γ < 1 ⇐⇒ γ−1 > 1 (2) γ, δ ≤ 1 =⇒ γδ ≤ 1 ; γ < 1, δ ≤ 1 =⇒ γδ < 1 ; γ, δ ≥ 1 =⇒ γδ ≥ 1 ; γ > 1, δ ≥ 1 =⇒ γδ > 1 If Γ′ is a subgroup of Γ, then for all γ, δ, η ∈ Γ, we have the following equivalent conditions: (i) γ ≤ δ ≤ 1 and γ ∈ Γ′ imply δ ∈ Γ′. (ii) γ, δ ≤ 1 and γδ ∈ Γ′ imply γ, δ ∈ Γ′. (iii) γ ≤ δ ≤ η and γ, η ∈ Γ′ imply δ ∈ Γ′. A subgroup Γ′ of Γ satisfying one of these equivalent conditions is called convex . Proof. Omitted. See, for instance, [22, Remark 1.6 and Remark 1.7 ]. Let Γ be a totally ordered group and let Γ′ be a subgroup of Γ. An element γ of Γ ∪ {0} is called cofinal for Γ′ if for all γ′ ∈ Γ′ there exists n ∈ N such that γn < γ′. Lemma 5.2. Let Γ be a totally ordered group and Γ′ ⊆ Γ be a convex subgroup. Let γ ∈ Γ be cofinal for Γ′ and let ∆ ( Γ′ be a proper convex subgroup. Then δγ is cofinal for Γ′ for all δ ∈ ∆. Proof. Omitted. See, for instance, [22, Proposition 1.20]. We now briefly recall valuation monoids from [6]. A monoid A is called a valuation monoid if A is cancellative and for every nonzero element α ∈ A+ (the group completion of A), at least one of α or 1/α belongs to A. For a valuation ring (A, +,·), the underlying multiplicative monoid (A,·) is a valuation monoid. Another simple example of a valuation monoid is the free pointed monoid on one generator. The set of units A× of a given valuation monoid A is a subgroup of A+\0 and the quotient group (A+\0)/A× is a totally ordered abelian group with the total ordering defined by x ≥ y if and only if x/y belongs to the image of A \ 0. We call the totally ordered pointed monoid Γ := ((A+ \ 0)/A×)∗ the value monoid of the valuation monoid A . The canonical surjection ord : A+ ։ Γ is called the valuation map of A . The monoid A can be identified with the set of x in A+ such that ord(x) ≤ 1 where 1 represents the identity of Γ. Note that ord(x) = 1 only if x is in A×. Thus, the maximal ideal mA of A is {x ∈ A+ ord(x) < 1}. Clearly, for all x, y ∈ A+, ord(x) ≥ 0 ; ord(xy) = ord(x)ord(y) ; ord(x) = 0 if and only if x = 0 (5.1) Conversely, given an abelian group G, consider the monoid G∗ obtained by adjoining a basepoint 0. Then, for a surjective morphism ord : G∗ ։ Γ onto a totally ordered monoid (Γ,·, 1, 0) that satisfies all the conditions in (5.1), we get a valuation monoid given by {g ∈ G ord(g) ≤ 1} whose group completion is G∗ and whose associated valuation map is ord. Proposition 5.3. Let A be a monoid contained in a monoid G∗, where G∗ is a monoid obtained by adjoining a basepoint 0 to an abelian group G. The set of all valuation monoids containing A and having the group completion G∗, is a spectral space. Proof. For any set S, the power set 2S endowed with the hull-kernel topology, which has as an open sub-basis the sets of the form D(F ) = {R ⊆ S F * R} where F is a finite subset of S, is a spectral space (see,[13, Theorem 8 and Proposition 9]). The sets D(F ) are quasi-compact and the complement of D(F ) is denoted as V(F ) = {R ⊆ S F ⊆ R}. Taking S = G∗, we see that 2G∗ is a spectral space. The set of all valuation monoids containing A and having the group completion G∗, is clearly a subset of 2G∗ and it is given by (cid:16)(cid:0) \a∈A V(a)(cid:1) ∩(cid:0) \g1,g2∈G∗ D(g1, g2) ∩ V(g1g2)(cid:1)(cid:17) \06=h∈G∗(cid:0)V(h) ∪ V(h−1(cid:1). 10 It is easy to see that this is a closed set in the patch topology on 2G∗. Since any patch closed set of a spectral space is also spectral in the subspace topology (see, [13, Proposition 9] or [20, Proposition 2.1] ), we have the required result. Note that by taking A = {0, 1} in Proposition 5.3, we have that the set of all valuation monoids having the same group completion G∗ forms a spectral space. Remark 5.4. It is worth mentioning that, like in Proposition 5.3, using the property that a patch in a spectral space is also spectral, the results obtained in Section 3 can be given shorter proofs. For example, given a monoid A, the collection of all proper ideals of A is a spectral space since it is given by the patch closed set, X = D(1) ∩ (cid:16)Ta,b∈A D(a) ∪ V (ba)(cid:17) of 2A and the collection of all prime ideals, M Spec(A), is a spectral space since it is given by the patch closed subset, X ∩(cid:16)Ta,b∈A D(ab) ∪ V (a) ∪ V (b)(cid:17) of 2A. This method of obtaining spectral spaces by expressing them as patches in the power set of some set, endowed with the hull-kernel topology, was used in [20, Example 2.2] for proving the ring theoretic versions of similar results. This method does not appear to extend beyond Section 3 of this paper. 5.1 Valuation spectrum of a monoid as a spectral space We begin this section by defining valuation on a monoid. Definition 5.5. Let A be a monoid and Γ be a totally ordered abelian group written multiplicatively. Let Γ∗ = Γ∪{0} denote the corresponding totally ordered monoid. We call a mapping v : A −→ Γ∗ is a valuation of A with values in Γ∗ if the following conditions are satisfied: (i) v(xy) = v(x)v(y) for all x, y ∈ A (ii) v(0) = 0, v(1) = 1 We call the subgroup of Γ generated by im(v) \ {0} the value group of v and we denote it by Γv. Note that for a valuation monoid A, the corresponding valuation map ord as defined by the equation (5.1) does satisfy the definition of a valuation as in Definition 5.5. Examples 5.6. (1) Let A be a monoid and let p be a prime ideal of A. Then, a 7→ (cid:26) 1, when a /∈ p 0, when a ∈ p gives a valuation with value group 1. We call every valuation of this type a trivial valuation. (2) Restricting any valuation on a ring (A, +,·) to the underlying multiplicative monoid (A,·) gives a valu- ation on the monoid. (3) Let φ : G −→ Γ be a group homomorphism from an abelian group G to a totally ordered abelian group Γ. Adjoining a basepoint 0 to both the groups, we obtain a valuation v : G∗ −→ Γ∗ on the monoid G∗ given by v(g) := φ(g) if g 6= 0 and v(0) := 0. Definition 5.7. Let v be a valuation on a monoid A and let Γv be its value group. Then the characteristic subgroup of Γv is the convex subgroup (convex with respect to the total ordering of Γ) generated by the set {v(x) ∈ Γv v(x) ≥ 1}. We denote it as cΓv. It is easy to check that cΓv is the set of all γ ∈ Γ such that v(x)−1 ≤ γ ≤ v(x) for some x ∈ A with v(x) ≥ 1. 11 Examples 5.8. (1) Let A := G∗ be a monoid obtained from a group G. For any valuation v on A, it is easy to check that cΓv = Γv. (2) Let A be a valuation monoid and v be the corresponding valuation map. Then, {v(x) ∈ Γv v(x) ≥ 1} = {1} and hence cΓv = 1. For a trivial valuation on a monoid, we also have cΓv = 1. Lemma 5.9. Let v be a valuation on a monoid A. Then, supp(v) := v−1(0) is a prime ideal. Proof. For any x ∈ v−1(0) and any a ∈ A, we have v(ax) = v(a)v(x) = v(a)0 = 0. Hence ax ∈ v−1(0) and this shows that v−1(0) is an ideal. To further show that this ideal is prime, take x, y ∈ A such that xy ∈ v−1(0). Then v(xy) = v(x)v(y) = 0. If v(y) 6= 0, then v(x) = v(x)v(y)v(y)−1 = 0v(y)−1 = 0. Hence, supp(v) is a prime ideal. Since supp(v) is a prime ideal, A/supp(v) is torsion free and so we can consider the group completion of A/supp(v). Clearly, the valuation v on A induces a valuation on its quotient A/supp(v) and the inverse image of 0 under this induced valuation reduces to 0. This again induces a valuation ¯v : (A/supp(v))+ −→ Γ∗ on the group completion (A/supp(v))+ defined by ¯v(¯a/¯b) := v(a)v(b)−1 where ¯a, ¯b ∈ A/supp(v) are the images of elements a, b ∈ A. It can be easily checked that ¯v is well-defined by using the multiplicativity of v and the fact that ¯a = {a} for every non-zero ¯a ∈ A/supp(v). Note that Γv = Γ¯v. Lemma 5.10. Let v be a valuation on a monoid A. For any x, y ∈ (A/supp(v))+, ¯v satisfies the following conditions: (i) ¯v(x) ≥ 0 (ii) ¯v(xy) = ¯v(x)¯v(y) (iii) ¯v(x) = 0 if and only if x = 0 Proof. Let us only check that x 6= 0 implies ¯v(x) 6= 0. If possible, let there exists some 0 6= x ∈ (A/supp(v))+ such that ¯v(x) = 0. Then, 1 = ¯v(1) = ¯v(xx−1) = ¯v(x)¯v(x−1) = 0. This gives the required contradiction. As discussed in the beginning of § 5, we have a valuation monoid A(v) := {x ∈ (A/supp(v))+ ¯v(x) ≤ 1} corresponding to the valuation ¯v. Note that ¯v(x) = 1 if and only if x is invertible in A(v). Hence, mA(v) := {x ∈ (A/supp(v))+ ¯v(x) < 1} is the maximal ideal of A(v). Proposition 5.11. Two valuations v and w on a monoid A are called equivalent if the following equivalent conditions are satisfied: (i) There exists an isomorphism of totally ordered monoids, f : (Γv)∗ −→ (Γw)∗, such that w = f ◦ v. (ii) supp(v) = supp(w) and A(v) = A(w). (iii) v(x) ≤ v(y) if and only if w(x) ≤ w(y) for all x, y ∈ A. Note that f in (i) restricts to an isomorphism Γv −→ Γw of totally ordered groups. Proof. Since homomorphism of totally ordered monoids preserves order, (i) implies (iii) obviously. It is also easy to see that (iii) implies (ii). Let us now assume (ii). Let p := supp(v) = supp(w). Then, we have induced valuations ¯v and ¯w on (A/p)+. We claim that {x ∈ (A/p)+ ¯v(x) ≥ 1} = {x ∈ (A/p)+ ¯w(x) ≥ 1}. Indeed, let x ∈ (A/p)+ such that ¯v(x) ≥ 1. If ¯v(x) > 1, then x /∈ A(v) = A(w) and so ¯w(x) > 1. Now, if ¯v(x) = 1, we claim that ¯w(x) must also be 1. This is because if ¯w(x) 6= 1, then either ¯w(x) > 1 or ¯w(x) < 1. If ¯w(x) > 1, then x /∈ A(w) = A(v) and so ¯v(x) > 1 which gives a contradiction. If ¯w(x) < 1, then x ∈ mA(w) = mA(v), i.e., ¯v(x) < 1 which also gives a contradiction. Hence, ¯w(x) = 1 and this proves our claim. This also shows that ¯v and ¯w restrict respectively to the surjective group homomorphisms ¯v : (A/p)+ \ {0} −→ Γv and ¯w : (A/p)+ \ {0} −→ Γw 12 with the same kernel. Thus, there exists a unique group homomorphism f : Γv −→ Γw such that f ◦ ¯v = ¯w. It can be easily checked that f is an isomorphism. Note that f maps elements ≤ 1 to elements ≤ 1 and hence it is a homomorphism of totally ordered groups. We can extend f to (Γv)∗ by setting f (0) = 0. Finally, it is easy to see that f ◦ ¯v = ¯w implies f ◦ v = w. This proves (i). From now onwards, we will not differentiate between a valuation and its equivalence class. We are now all set to define the valuation spectrum of a monoid. Definition 5.12. Let A be a monoid. The valuation spectrum Spv(A) is the set of all equivalence classes of valuations on A equipped with the topology generated by the subsets Spv(A)(cid:16) x y(cid:17) := {v ∈ Spv(A) v(x) ≤ v(y) 6= 0} where x, y ∈ A. We now proceed to show that Spv(A) is a spectral space for which we first discuss some preparatory lemmas. Lemma 5.13. Given two valuations v 6= w in Spv(A) there exists some open set U = Spv(A)( x x, y ∈ A such that either v ∈ U but w /∈ U , or v /∈ U but w ∈ U . In other words, Spv(A) is a T0 space. Proof. Since v and w are two non-equivalent valuations on A, there exists x, y ∈ A such that v(x) ≤ v(y) but w(x) (cid:2) w(y) (or conversely) by Proposition 5.11(iii). If v(y) 6= 0, then there exists the open set y(cid:17) such that v ∈ U but w /∈ U . Now, if v(y) = 0, then we also have v(x) = 0. Since U = Spv(A)(cid:16) x x(cid:17) such that w ∈ W but w(y) (cid:12) w(x), clearly w(x) 6= 0. Thus, there exists the open set V = Spv(A)(cid:16) y v /∈ V . This finishes the proof. To the equivalence class v of a valuation on A we attach a binary relation v on A by defining xvy if and only if v(x) ≤ v(y). This gives us a map, y ) with φ : Spv(A) −→ P (A × A) v 7→ v into the power set P (A × A) of A × A. We identify P (A × A) with {0, 1}A×A and endow it with the product topology, where {0, 1} carries the discrete topology. Thus, P (A× A) is compact (by Tychonoff) and Hausdorff, since {0, 1} is compact Hausdorff. Remark 5.14. Note that in the definition of v, we do not require v(y) 6= 0. Lemma 5.15. The map φ : Spv(A) −→ P (A × A), v 7→ v is an injection and a binary relation belongs to im(φ) if and only if satisfies the following axioms for all a, b, c ∈ A: (1) ab or ba (2) If ab and bc then ac (3) If ab then acbc (4) If acbc and 0 ∤ c then ab (5) 0 ∤ 1 Proof. Let w and w′ be two non-equivalent valuations on A. Then, there exists x, y ∈ A such that w(x) ≤ w(y) but w′(x) (cid:2) w′(y) (or conversely) by Proposition 5.11 (iii). Thus, w 6= w′. This shows that φ is injective. Obviously, the axioms stated in this lemma are satisfied by v for any valuation v on A. Let us now verify that the above axioms encode exactly the image of φ. Fix such a binary relation satisfying all the axioms 13 and declare a ∼ b if ab and ba. Let [a] denote the equivalence class of a. Multiplication of equivalence classes can be defined via multiplication of representatives and can be verified to be well defined easily. If ab ∼ 0 , then 0ab. Since 0 = 0a, if 0 ∤ a, then by axiom (4) we get 0b. Hence, ab ∼ 0 if and only if either a ∼ 0 or b ∼ 0. Thus, multiplication of non zero equivalence classes gives a non zero equivalence class and [a] 6= 0}. It is a commutative monoid with identity ( therefore we get the unpointed monoid M := {[a] the class of 1 which is distinct from that of 0 by axiom (5)). By axiom (4), it is also cancellative. Take distinct [a] and [b] from M . Then either ab or ba by axiom (1). If ab, then we set [a] ≤ [b]. This ordering is independent of the choice of representatives of the equivalence classes by axiom (2). Thus, M is a totally ordered cancellative commutative monoid and we can create an abelian group corresponding to this monoid. Declare two points (m1, m2) (think of it as m1/m2) and (m′ 1m2 where m1, m2, m′ 2 ∈ M . Clearly this relation is reflective and symmetric. Transitivity follows by axioms (3) and (4). Therefore, this gives an equivalence relation and we denote each equivalence class (m1, m2) as a fraction m1/m2. We multiply two classes in the obvious way and it is well-defined due to the cancellative property. If Γ denote the set of equivalance classes, then it is clearly an abelian group with identity 1/1 and (m1/m2)−1 = m2/m1. Since M is cancellative, the canonical morphism M −→ Γ is an injection. We extend the total ordering from M to Γ in the usual way, i.e., m1/m2 ≤ m′ 2 ≤ m′ 1m2. This is well defined by axiom (3) and the cancellative property of M . We define, v : A −→ Γ ∪ {0} by a 7→ (cid:26) [a]/1, when a ≁ 0 [0]/1, when a ∼ 0 2) equivalent if m1m′ 2 = m′ 1/m′ 2 when m1m′ 1, m′ 1, m′ It is easy to check that v is a valuation and that = v. Hence, the axioms listed in the lemma entirely encodes the image of φ. Lemma 5.16. The image of φ : Spv(A) −→ P (A × A), v 7→ v is a closed compact subset of P (A × A). Proof. Each of the axioms (for fixed a, b, c) as stated in Lemma 5.15 defines a closed subset of P (A × A). For example, let us see that axiom (2) defines a closed condition. Recall that P (A × A) can be identified with {0, 1}A×A. Let πa,b : {0, 1}A×A −→ {0, 1} be the projection onto the (a, b)-th component. Then the collection of elements of {0, 1}A×A satisfying axiom (2) is the union of π−1 a,c(1) and the complement of a,b(1)∩π−1 π−1 b,c (1). Clearly, this collection gives a closed set since {0, 1} has the discrete topology and {0, 1}A×A has the product topology, which makes the projection maps continuous. Hence the image of φ, being the set of all binary relations on A which satisfies the axioms in the Lemma 5.15, is closed. Since P (A × A) is compact, the closed set im(φ) of P (A × A) is also compact. Theorem 5.17. Let X ′ be a quasi-compact topological space and let U be a collection of clopen sets of X ′. Endow the set underlying X ′ with the topology which has U as an open subbasis and call the resulting topological space X. Assume that X is T0. Then X is a spectral space and the basis obtained from the subbasis U gives a basis of open quasi-compact subsets of X. Moreover, Xpatch = X ′. Conversely, every spectral space arises from its patch space in this way. Proof. See for instance [13, Proposition 7] and [22, Proposition 3.31]. We define U (a, b) := {v ∈ Spv(A) avb} = {v ∈ Spv(A) v(a) ≤ v(b)}. It follows immediately from the definition of v that bv0 if and only if v(b) = 0, i.e., U (b, 0) = {v ∈ Spv(A) v(b) = 0}. Thus clearly by definition, Spv(A)( a b ) = U (a, b) ∩ (Spv(A) \ U (b, 0)). 14 Theorem 5.18. Let A be a monoid. (i) The valuation spectrum Spv(A) is a spectral space with a basis of quasi-compact open subsets given by the subbasis consisting of sets of the form Spv(A)(cid:0) a sets Spv(A)(cid:0) a b(cid:1) gives a basis for the patch topology on Spv(A). (ii) Any monoid morphism f : A −→ B induces a spectral map Spv(f ) : Spv(B) −→ Spv(A) v 7→ v ◦ f b(cid:1) for a, b ∈ A. The Boolean algebra generated by the We obtain a contravariant functor from the category of monoids to the category of spectral spaces and spectral maps. πa,b−−→ {0, 1} to the (a, b)-th a,b{1} for the projection P (A×A) ∼= {0, 1}A×A Proof. (i) Consider the preimage π−1 component. It is easy to see that π−1 a,b{1} meets the image of φ : Spv(A) ֒→ P (A × A) in exactly U (a, b). Thus, in other words, we have U (a, b) = π−1 a,b{1} ∩ im(φ). The projection πa,b is a continuous map to the discrete space {0, 1}. Thus, U (a, b), which is the preimage of the clopen set {1}, must be clopen in im(φ). This means U (a, b) is clopen in im(φ) for any (a, b) ∈ A×A. Hence, Spv(A)(cid:0) a b(cid:1) = U (a, b)∩(Spv(A)\U (a, 0)) b(cid:17) a, b ∈ A}. Endowing the set underlying im(φ) with is also clopen in im(φ). We set, U := {Spv(A)(cid:16) a the topology generated by U, we obtain the topological space Spv(A). Also, Spv(A) is T0 by Lemma 5.13. Hence, our claim follows by Theorem 5.17. (ii) Let f : A −→ B be a monoid morphism which induces the map Spv(f ) : Spv(B) −→ Spv(A) defined by v 7→ v ◦ f . Since the sets of the form Spv(A)(cid:0) a b(cid:1) give a basis of open quasi-compact subsets of Spv(A) φ(b)(cid:17) is an open quasicompact subset of Spv(B), we have, Spv(f ) is and Spv(f )−1(Spv(A)(cid:0) a b(cid:1)) = Spv(B)(cid:16) φ(a) a spectral map. 5.2 Subspace of continuous valuations as a spectral space Let x and y be two points of a topological space X. Then, x is called a generization of y or that y is called a specialization of x if y ∈ {x}. Lemma 5.19. Let v : A −→ Γ ∪ {0} be a valuation on a monoid A. Let H be a subgroup of Γv. We can define a map, vH : A −→ H ∪ {0} by setting vH (a) := v(a) if v(a) ∈ H and vH (a) := 0 if v(a) /∈ H. If H is convex and cΓv ⊆ H, then vH is a valuation. Moreover, vH is a specialization of v in Spv(A). Proof. Clearly, vH (0) = 0 and vH (1) = 1. We only need to check that vH (ab) = vH (a)vH (b) for any a, b ∈ A. Suppose vH (ab) = 0 or in other words v(ab) /∈ H. But since v(ab) = v(a)v(b), at least one of v(a) and v(b) must not belong to H. Thus, at least one of vH (a) and vH (b) must be 0 which gives vH (ab) = vH (a)vH (b). Now, let vH (ab) = v(ab) or in other words v(ab) ∈ H. Then, to show vH (ab) = vH (a)vH (b), we must show that both v(a), v(b) ∈ H. We have the following possible cases: (1) If v(a) ≥ 1, v(b) ≥ 1, then v(a), v(b) ∈ H since cΓv ⊆ H. (2) If v(a) < 1, v(b) < 1, then v(a), v(b) ∈ H since v(ab) ∈ H (by Lemma 5.1). (3) If v(a) ≥ 1, v(b) < 1, then v(a) ∈ H since cΓv ⊆ H and hence v(b) = v(ab)v(a)−1 ∈ H. To show that vH is a specialization of v, it is enough to check that vH (a) ≤ vH (b) 6= 0 implies v(a) ≤ v(b) 6= 0 for a, b ∈ A. Clearly, vH (b) 6= 0 shows that v(b) ∈ H and in particular v(b) 6= 0. If possible, assume v(a) > v(b). Then, v(a) /∈ H and hence v(a) < 1 as H contains cΓv. Thus, v(b) < v(a) < 1 and hence v(a) ∈ H as H is convex. This gives the required contradiction. 15 Lemma 5.20. Let v : A −→ Γ ∪ {0} be a valuation on a monoid A. Let H ⊆ Γv be a convex subgroup such that cΓv ( H. Then, I := {x ∈ A v(x) is cofinal for H} is an ideal of A and √I = I. Proof. Let a ∈ A and x ∈ I. If v(a) ≤ 1, then v(ax) ≤ v(x) and hence ax ∈ I. If v(a) > 1, then v(a) ∈ cΓv ⊆ H. Thus, v(ax) = v(a)v(x) is cofinal for H because cΓv ( H (by Lemma 5.2). It is easy to see that √I = I. We call an ideal I of A rad-finite if there exists a finitely generated ideal J of A such that √I = √J. If v(I) ∩ cΓv = ∅, then there exists a greatest convex Lemma 5.21. Let I be a rad-finite ideal of A. subgroup H of Γv such that v(x) is cofinal for H for all x ∈ I. Moreover, if v(I) 6= {0}, then cΓv ⊆ H and v(I) ∩ H 6= ∅. Proof. If v(I) = {0} we can choose H = Γv. So let us assume that v(I) 6= {0} and hence cΓv 6= Γv. Note that for any subgroup H of Γv, we have v(I)∩ H = ∅ ⇐⇒ v(√I)∩ H = ∅. This together with Lemma 5.20, shows that it is enough to prove the lemma for a finitely generated ideal I. Let T be a finite set of generators of I and let H be the convex subgroup of Γv generated by h := max{v(t) t ∈ T}. Since v(I) ∩ cΓv = ∅, we must have, h < 1 and thus h is cofinal for H. Consequently, v(t) is cofinal for H for every t ∈ T . Moreover, v(I) ∩ cΓv = ∅ implies v(x) < cΓv for all x ∈ I. For if v(x) ≥ 1, then v(x) ∈ cΓv, giving a contradiction and if γ ≤ v(x) ≤ 1 for some γ ∈ cΓv, then v(x) ∈ cΓv (as cΓv is convex), giving a contradiction again. Hence, v(x) < cΓv for all x ∈ I and so in particular we have, h < cΓv < h−1. Since H is a convex subgroup generated by h, this implies cΓv ( H. Since T generates I and we have shown that v(t) is cofinal for H for all t ∈ T , it follows from Lemma 5.20 that v(x) is cofinal for H for all x ∈ I. Let H ′ be another convex subgroup of Γv such that v(x) is cofinal for H ′ for all x ∈ I. Then, in particular, h is cofinal for H ′ which implies that H ′ ⊆ H. Lemma 5.21 allows us to define the following subgroup of Γv, for every valuation v of a monoid A and every let cΓv(I) be the group cΓv if v(I) ∩ cΓv 6= 0 and otherwise let cΓv(I) be the greatest rad-finite ideal I: convex subgroup H of Γv such that v(x) is cofinal for H for all x ∈ I. Note that by Lemma 5.21, cΓv(I) is a convex subgroup of Γv which always contains cΓv. Hence, by Lemma 5.19, vcΓv (I) is a valuation. Lemma 5.22. Let v be a valuation of A and I be a rad-finite ideal of A. Then, the following conditions are equivalent: (i) Γv = cΓv(I). (ii) Γv = cΓv or v(x) is cofinal in Γv for every x ∈ I. (iii) Γv = cΓv or v(x) is cofinal in Γv for every element x of a set of generators of I. Proof. The equivalence of (i) and (ii) follows from the definition of cΓv(I) and the equivalence of (ii) and (iii) follows from Lemma 5.20. Let I be a rad-finite ideal of A. We set Spv(A, I) := {v ∈ Spv(A) Γv = cΓv(I)} and equip it with the subspace topology of Spv(A). Note that we get the same subspace if we replace I by any ideal J such that √I = √J. We have the following map: r : Spv(A) −→ Spv(A, I) v 7→ vcΓv (I) (5.2) Clearly, r(v) = v for every v ∈ Spv(A, I) i.e., r is a retraction. Lemma 5.23. Let R be the set of all subsets U of Spv(A, I) of the form: U = {v ∈ Spv(A, I) v(x1) ≤ v(y) 6= 0, . . . , v(xn) ≤ v(y) 6= 0} where y, x1, . . . , xn ∈ A with I ⊆ phx1, . . . , xni. Then R is a basis of the topology of Spv(A, I) that is closed under finite intersections. 16 Proof. Clearly, each set in R is open in Spv(A, I) since it has the induced topology of Spv(A). Let us now show that R is closed under finite intersections. Let x0, . . . , xn and y0, . . . , ym be elements of A. Then, n \i=1 {v ∈ Spv(A) v(xi) ≤ v(x0) 6= 0}(cid:17)\(cid:16) m \j=1 {v ∈ Spv(A) v(yi) ≤ v(x0) 6= 0}(cid:17) (cid:16) = \ i=1,...,n ; j=1,...,m {v ∈ Spv(A) v(xiyj) ≤ v(x0y0) 6= 0} and, if I ⊆ ph{xi i = 0, ..., n}i and I ⊆ ph{yj j = 0, ..., m}i, then, I ⊆ qh{xiyj i = 0, ..., n ; j = 0, ..., m}i Hence, R is closed under finite intersections. Now let us show that R gives a basis for the topology on Spv(A, I). Let v ∈ Spv(A, I) and let U be an open neighborhood of v in Spv(A). We choose x0, . . . , xn ∈ A such that v ∈ W := {w ∈ Spv(A) w(xi) ≤ w(x0) 6= 0 for all i = 1, . . . , n} ⊆ U We can make this choice because the sets like W form a basis for the topology of Spv(A). Now we have two possible cases, Γv = cΓv or Γv 6= cΓv. First assume Γv = cΓv. Since v(x0) ∈ cΓv, there exists some d ∈ A such that v(x0d) ≥ 1. Hence, v ∈ W ′ := {w ∈ Spv(A) w(xid) ≤ w(x0d) 6= 0 for all i = 1, . . . , n and w(1) ≤ w(x0d) 6= 0} ⊆ W Thus, we have, W ′ ∩ Spv(A, I) ∈ R. Next assume Γv 6= cΓv. Let {s1, s2, . . . , sm} be a finite set of generators of a finitely generated ideal J for which √I = √J. By Lemma 5.22, v(sj ) is cofinal in Γv for all j = 1, . . . , m. Thus we can choose a k ∈ N such that v(sk j ) ≤ v(x0) for all j = 1, . . . , m. Hence, v ∈ W ′ := {w ∈ Spv(A) w(xi) ≤ w(x0) 6= 0 and w(sk j ) ≤ w(x0) 6= 0 for all i, j} ⊆ W where i = 1, . . . , n and j = 1, . . . , m. Thus, we have W ′ ∩ Spv(A, I) ∈ R. This completes the proof. Lemma 5.24. Let I be an ideal of a monoid A and y, x1, . . . , xn ∈ A such that I ⊆ phx1, . . . , xni. For U = {v ∈ Spv(A, I) v(x1) ≤ v(y) 6= 0, . . . , v(xn) ≤ v(y) 6= 0} W = {v ∈ Spv(A) v(x1) ≤ v(y) 6= 0, . . . , v(xn) ≤ v(y) 6= 0} we have, r−1(U ) = W where r is as in (5.2). In particular, for any v ∈ Spv(A) if v(I) 6= 0, then r(v)(I) 6= 0. Proof. Since U ⊆ W and as every point v ∈ r−1(U ) specializes to a point r(v) of U , we have r−1(U ) ⊆ W because W is open in Spv(A). Conversely, take any w ∈ W . We want to show that r(w) ∈ U . If w(I) = 0, then cΓw(I) = Γw by definition and thus r ∈ Spv(A, I). Hence, r(w) = w. Now assume w(I) 6= 0. We claim that r(w)(xi) ≤ r(w)(y) for all i = 1, . . . , n. The inequality is obvious if r(w)(xi) = 0 i.e., when w(xi) /∈ cΓw(I). If w(xi) ∈ cΓv(I), then w(y) also belongs to cΓv(I). This is because either w(y) ≥ 1 in which case w(y) ∈ cΓv(I) since cΓv ⊆ cΓv(I), or w(xi) ≤ w(y) ≤ 1 in which case w(y) ∈ cΓv(I) since cΓv(I) is convex. Thus, we have r(w)(xi) = w(xi) ≤ w(y) = r(w)(y). It remains to show that r(w)(y) 6= 0. So, if possible, let r(w)(y) = 0. Then r(w)(xi) = 0 for all i = 1, . . . , n. Since I ⊆ phx1, . . . , xni, this implies r(w)(x) = 0 for all x ∈ I. But by Lemma 5.21, w(I) 6= 0 implies w(I) ∩ cΓw(I) 6= ∅, i.e., there exists some x ∈ I such that r(w)(x) 6= 0. This gives the required contradiction. Note that we have also proved that for any v ∈ Spv(A) if v(I) 6= 0, then r(v)(I) 6= 0. 17 Theorem 5.25. The topological space Spv(A, I) is spectral and the set R (as in Lemma 5.23) forms a basis of quasi-compact open subsets of the topology, which is stable under finite intersections. Moreover, the retraction r : Spv(A) −→ Spv(A, I) (as in (5.2)) is a spectral map. Proof. Let R denote the Boolean algebra of subsets of Spv(A, I) generated by R. Let X denote the set underlying Spv(A, I). Endow X with the topology generated by R. By Lemma 5.24 and Theorem 5.18(i), we see that r : (Spv(A))patch −→ X is continuous. Since (Spv(A))patch is quasi-compact (see, for instance,[13, Theorem 1]) and r is surjective, X is quasi-compact. Since X has the topology generated by R, every set of R is clopen in X. By Lemma 5.23, endowing X with the topology generated by R gives the topological space Spv(A, I). Since Spv(A) is T0, so is Spv(A, I). Hence by Theorem 5.17, Spv(A, I) is a spectral space and R is a basis of open quasi-compact subsets of Spv(A, I). By Lemma 5.23, R is stable under finite intersections and by Lemma 5.24, r : Spv(A) −→ Spv(A, I) is a spectral map. A topological monoid is a monoid A endowed with a topology such that the monoid operation is continuous. Monoids endowed with the discrete and indiscrete topologies or the underlying multiplicative monoid of a topological ring are some simple examples of topological monoids. Some interesting examples are: Examples 5.26. (1) M Spec(A) is a topological monoid for any commutative monoid A (see, [21, § 2]). (2) Given any topological space X, we can construct a topological monoid A := X S{1, 0} whose underlying space is X adjoined with two disjoint points 1 and 0. The point 1 represents the identity of the monoid and 0 represents the basepoint. The monoid operation is defined as : a.b = 0 for all a, b 6= 1 and a.1 = 1.a = a. We declare a set U ⊆ A is open if and only if U ∩ X is open. It is easy to check that this gives a topology on A and also makes it a topological monoid. (3) Given a topological group (G,∗) we can obtain a monoid A := G ∪ {0} by adjoining a basepoint 0 to G. The monoid operation is defined as : a.b = a ∗ b for all a, b 6= 0 and a.0 = 0.a = 0. We declare a set U ⊆ A is open if and only if U ∩ G is open. This makes A a topological monoid. We now introduce the notion of I-topology on a monoid induced by an ideal I. This is inspired by the classical I-adic topology on a ring induced by an ideal I of the ring (see, for instance, [1, Chapter 10]). Definition 5.27. Given an ideal I of a monoid A, we can define a topology on the monoid by declaring U ⊆ A is open if either 0 /∈ U or I n ⊆ U for some n ∈ N ∪ {0} where I 0 := A. It can be easily checked that the monoid multiplication is continuous with respect to this topology and hence this gives a topological monoid. We call such a topology on a monoid defined by an ideal I as I-topology. If an ideal I of a monoid A is such that I n = {0}, then it is easy to check that the I-topology is the same as the topology induced by the following metric: ∞ Tn=0 d(a, b) = (cid:26) 1 2n , 0, for a 6= b and for the largest non-negative integer n such that both a, b ∈ I n for a = b It may be easily checked that this gives a complete metric space! Definition 5.28. Let v be a valuation on a topological monoid A with value group Γv. We call v continuous if the map v : A −→ Γv∪{0} is continuous where Γv ∪{0} is endowed with the following topology: U ⊆ Γv ∪{0} is open if 0 /∈ U or if there exists γ ∈ Γv such that the set {v(x) ∈ Γv v(x) < γ} ⊆ U . We denote by Cont(A) the subspace of Spv(A) consisting of the continuous valuations on A. We now discuss examples of continuous and non-continuous valuations. 18 Examples 5.29. For any abelian group (G,∗), consider the cancellative monoid A := G ∪ {0} obtained by adjoining a basepoint 0 to G. Explicitly, the monoid operation is defined as : a.b = a ∗ b for all a, b 6= 0 and a.0 = 0.a = 0. Take any non-trivial ideal I of A, say, for instance, I = hgi where g is a non-identity element of the group G. Endow A with the I-topology. Now, consider the following valuation v : A −→ {0, 1} of A with the value group Γv = {1}: v(a) = (cid:26) 1, when a 6= 0 0, when a = 0 By Definition 5.28, the topology on {0, 1} is discrete. But, v−1({0}) = {0} is not open in the I-topology of A. This is because I n 6= 0 for all n ∈ N, as A is cancellative. Thus, this gives an example of non-continuous valuation. If we take I = {0}, then the I-topology on A will make the valuation v continuous. In fact, any valuation on a monoid having discrete topology is continuous. Let A be a topological monoid. We call an element x ∈ A topologically nilpotent if xn −→ 0 as n −→ ∞. Note that for a topological monoid A having the I-topology for some ideal I, √I gives the complete collection of topologically nilpotent elements of A. Theorem 5.30. Let A be a monoid and I be a finitely generated ideal of A. Endowing A with the I-topology we obtain a topological monoid. Then, (i) For a valuation v : A −→ Γv ∪ {0}, v ∈ Cont(A) ⇐⇒ {x ∈ A v(x) < γ} is open for all γ ∈ Γv (ii) Cont(A) = {v ∈ Spv(A, I) v(x) < 1 for all x ∈ I}. (iii) Cont(A) is a spectral space. Proof. (i) If v ∈ Cont(A), then clearly {x ∈ A v(x) < γ} is open for all γ ∈ Γv. To show the other way, consider any open set U of Γv ∪ {0}. If U is such that 0 /∈ U , then 0 /∈ v−1(U ) and hence v−1(U ) is open in A. If U is such that there exists some γ ∈ Γv for which {v(x) ∈ Γv v(x) < γ} ⊆ U , then the open set {x ∈ A v(x) < γ} is contained in v−1(U ). Since {x ∈ A v(x) < γ} is an open set of A containing 0, there exists some n ∈ N such that I n ⊆ {x ∈ A v(x) < γ} ⊆ v−1(U ). Hence, v−1(U ) is open in A. (ii) Let v ∈ Cont(A). Then, as we have just shown, {x ∈ A v(x) < γ} is an open neighborhood of 0 for all γ ∈ Γv. Since every element a ∈ I is topologically nilpotent, for every γ ∈ Γ there exists some n ∈ N such that an ∈ {x ∈ A v(x) < γ} i.e., v(a)n < γ. Thus, v(a) is cofinal in Γv for every a ∈ I. This implies that v(a) < 1 as 1 ∈ Γv. By Lemma 5.22, we also have Γv = cΓv(I). Hence, v ∈ {v ∈ Spv(A, I) v(x) < 1 for all x ∈ I}. Conversely, let v ∈ Spv(A, I) with v(a) < 1 for all a ∈ I. We want to show that v(a) is cofinal in Γv for all a ∈ I. If cΓv 6= Γv, we are done by Lemma 5.22. Now let cΓv = Γv and let γ ∈ Γv be any given element. For γ ≥ 1, we are done by our hypothesis that v(a) < 1. Otherwise, if γ < 1, then by the definition of the characteristic subgroup there exists t ∈ A such that v(t) 6= 0 and v(t)−1 ≤ γ < 1. Note that the map A −→ A defined by x 7→ tx is continuous since A is a topological monoid. Thus the preimage of the open set I under this map is open and is given by {x ∈ A tx ∈ I}. Since this is an open neighborhood of 0 in A, there exists some n ∈ N such that I n ⊆ {x ∈ A tx ∈ I}. For any a ∈ I, we therefore have tan ∈ I. Hence, by assumption v(tan) < 1 and so v(a)n < γ. This proves our claim. Let S = {a1, a2, . . . am} be a finite set of generators of I. Set δ := max{v(ai)}. Then, there exists n ∈ N such that δn < γ. Now, let x be any element in I n+1. Then, x = aax1 . . . axn+1 for some a ∈ A and axi ∈ S for i = 1, . . . , n + 1. Hence, v(x) = v(aax1 . . . axn+1) = v(aax1 )v(ax2 ) . . . v(axn+1) < v(aax1)δn < γ since aax1 ∈ I. So by assumption v(aax1 ) < 1. Thus, I n+1 ⊆ {x ∈ A v(x) < γ}. This shows {x ∈ A v(x) < γ} is open for any γ ∈ Γv and hence by (i), v is continuous. This completes the proof of (ii). (iii) Note that Spv(A, I)\ Cont(A) = Sx∈I{v ∈ Spv(A, I) 1 ≤ v(x)} is open in Spv(A, I) since each set on 19 the right-hand side is open by Lemma 5.23. Thus, Cont(A) is a closed subspace of Spv(A, I). Recall that Spv(A, I) is a spectral space by Theorem 5.25. Since any closed subspace of a spectral space is spectral [22, Remark 3.13], we have that Cont(A) is spectral. References [1] M. F. Atiyah and I. G. Macdonald. Introduction to commutative algebra. Addison-Wesley Ser. Math. Reading, MA: Addison-Wesley, 1969. [2] A. Banerjee, Closure operators in abelian categories and spectral spaces, Theory Appl. Categ. 32 (2017), Paper No. 20, 719 -- 735. [3] A. Banerjee, On some spectral spaces associated to tensor triangulated categories, Arch. Math. (Basel) 108 (2017), no. 6, 581 -- 591. [4] N. Bourbaki. Elements of Mathematics. Commutative algebra. Chapters 1 -- 7. Translated from the French, Reprint of the 1989 English translation. Berlin: Springer-Verlag, 1998. [5] B. Conrad, Spectral spaces and consrtuctible sets. url: math.stanford.edu/ conrad/Perfseminar/Notes/L3.pdf. [6] G. Cortinas, C. Haesemeyer, M.E. Walker and C. Weibel, Toric varieties, monoid schemes and cdh descent, J. Reine Angew. Math. 698 (2015), 1 -- 54. [7] N. Epstein, A guide to closure operations in commutative algebra, Progress in commutative algebra 2, 1 -- 37, Walter de Gruyter, Berlin, 2012. [8] C. A. Finocchiaro, Spectral spaces and ultrafilters, Comm. Algebra 42 (2014), no. 4, 1496 -- 1508. [9] C. A. Finocchiaro, Marco Fontana, Dario Spirito, New Distinguished Classes of Spectral Spaces: A Survey. Multiplicative ideal theory and factorization theory, 117 -- 143, Springer Proc. Math. Stat.,170, Springer, [Cham], 2016. [10] C. A. Finocchiaro, Marco Fontana, Dario Spirito, A topological version of Hilbert's Nullstellensatz, J. Algebra 461 (2016), 25 -- 41. [11] J. Flores, Homological Algebra for Commutative Monoids, PhD thesis, New Brunswick Rutgers, The State University of New Jersey, 2015. [12] J. Flores, C. Weibel, Picard groups and class groups of monoid schemes, J. Algebra 415 (2014), 247 -- 263. [13] M. Hochster, Prime ideal structure in commutative rings, Trans. Amer. Math. Soc. 142 (1969), 43 -- 60. [14] M. Hochster and C. Huneke, Tight closure, invariant theory, and the Briancon-Skoda theorem, J. Amer. Math. Soc. 3 (1990), no. 1, 31 -- 116. [15] R. Huber, Continuous valuations, Math. Zeitschrift 212 (1993), 45 -- 477. [16] C. Huneke and I. Swanson, Integral Closure of Ideals, Rings, and Modules, volume 336 of London Mathematical Society Lecture Note Series. Cambridge University Press, Cambridge, 2006. [17] T. Jech, Set Theory, Springer, New York, 1997 (First Edition, Academic Press, 1978). [18] J. Jun,Valuation of Semirings, J. Pure Appl. Algebra 222 (2018), no. 8, 2063 -- 2088. [19] P. Nasehpour,Valuation semirings, J. Algebra Appl. 17 (2018), no.4, 1850073, 23 pp. [20] B. Olberding, Topological aspects of irredundant intersections of ideals and valuation rings, In Multiplicative Ideal Theory and Factorization Theory: Commutative and Non-Commutative Perspectives, Springer Proc. Math. Stat., 170 (2016), 277 -- 307, Springer, [Cham]. [21] I. Pirashvili, On the spectrum of monoids and semilattices, J. Pure Appl. Algebra 217 (2013), no. 5, 901 -- 906. [22] T. Wedhorn, Adic spaces, June 19, 2012. url: http : / / www2 . math . uni - paderborn . de / fileadmin / Mathematik/People/wedhorn/Lehre/AdicSpaces.pdf. 20
1610.02232	1	1610	2016-10-07T11:51:26	Filtered K-theory for graph algebras	[ "math.RA", "math.OA" ]	We introduce filtered algebraic $K$-theory of a ring $R$ relative to a sublattice of ideals. This is done in such a way that filtered algebraic $K$-theory of a Leavitt path algebra relative to the graded ideals is parallel to the gauge invariant filtered $K$-theory for graph $C^*$-algebras. We apply this to verify the Abrams-Tomforde conjecture for a large class of finite graphs.	math.RA	math	FILTERED K-THEORY FOR GRAPH ALGEBRAS SØREN EILERS, GUNNAR RESTORFF, EFREN RUIZ, AND ADAM P. W. SØRENSEN Abstract. We introduce filtered algebraic K-theory of a ring R relative to a sublattice of ideals. This is done in such a way that filtered algebraic K-theory of a Leavitt path algebra relative to the graded ideals is parallel to the gauge invariant filtered K-theory for graph C ∗-algebras. We apply this to verify the Abrams-Tomforde conjecture for a large class of finite graphs. 1. Introduction Since the inception of Leavitt path algebras in [AAP05, AMP07] it has been known that there is a strong connection between Leavitt path algebras and graph C∗-algebras. In particular many results for both graph C∗-algebras and Leavitt path algebras have the same hypotheses when framed in terms of the underlying graph and the conclusions about the structure of the algebras are analogous. For instance, by [JP02, Theorem 4.1] and [APPSM06, Theorem 4.5] the following are equivalent for a graph E. (1) E satisfies Condition (K) (no vertex is the base point of exactly one return path). (2) C∗(E) has real rank 0. (3) LC(E) is an exchange ring. That real rank 0 is the analytic analogue of the algebraic property of being an exchange ring is justified in [AGOP98, Theorem 7.2]. One of the most direct connections we could possibly have between Leavitt path algebras and graph C∗-algebras would be: If E, F are graphs then LC(E) ∼= LC(F ) ⇐⇒ C∗(E) ∼= C∗(F ). This is called the isomorphism question and it is unknown if it is true. As currently stated the question is very imprecise, while it is clear what is meant by isomorphism of C∗-algebras, we could consider isomorphisms of Leavitt path algebras both as rings, algebras, and ∗-algebras. In the last case the forward implication of the isomorphism question holds. In [AT11] Abrams and Tomforde take a systematic look at the isomorphism question and many related questions, for instance whether or not the above holds with Morita equivalence in place of isomorphism. They provide evidence in favor of a positive answer to the Morita equivalence question and elevate one direction to a conjecture. Conjecture 1.1 (The Abrams-Tomforde Conjecture). Let E and F be graphs. If LC(E) is Morita equivalent to LC(F ), then C∗(E) is (strongly) Morita equivalent to C∗(F ). In [RT13] the third named author and Tomforde use ideal related algebraic K- theory to verify the Abrams-Tomforde conjecture of large classes of graphs. They Date: September 10, 2018. 2010 Mathematics Subject Classification. 16B99, 46L05, 46L55. Key words and phrases. Leavitt path algebra, Graph C ∗-algebras. 1 2 SØREN EILERS, GUNNAR RESTORFF, EFREN RUIZ, AND ADAM P. W. SØRENSEN introduce ideal related algebraic K-theory as a Leavitt path algebra analogue for fil- tered K-theory for graph C∗-algebras. This then allows them to prove the Abrams- Tomforde conjecture for all classes of graphs where the associated C∗-algebras are classified by filtered K-theory. The authors have shown in [ERRS16] that when classifying graph C∗-algebras that do not have real rank 0, it can be useful to replace the full filtered K-theory with a version that only looks at gauge invariant ideals. Motivated by this, we develop a version of ideal related algebraic K-theory relative to a sublattice of ideals. Our goal is to get an ideal related K-theory for Leavitt path algebras that only considers graded ideals, but we try to state our result in greater generality. We look at a sublattice S of ideals in some ring R and consider the spectrum of these ideals, that is the set of S-prime ideals. This set is equipped with the Jacobson (or hull-kernel) topology. In nice cases there exists a lattice isomorphism from the open sets in the spectrum to the ideals in S. Specializing to the case of a Leavitt path algebra Lk(E), we show that the spectrum associated to the graded ideals is homeomorphic to the spectrum of gauge invariant ideals in C∗(E). Using this we define filtered algebraic K-theory of Lk(E) relative to the graded ideals in complete analogy to the C∗-algebra definition. We then follow the work of [RT13] and establish the Abrams-Tomforde conjecture for all graphs where the C∗- algebras are classified by filtered K-theory of gauge invariant ideals. By [ERRS16] this includes a large class of finite graphs. 2. Preliminaries In this section we set up the notation we will use throughout the paper and we recall the needed definitions. We begin with the definitions of graphs, graph C∗-algebras and Leavitt path algebras. Definition 2.1. A graph E is a quadruple E = (E0, E1, r, s) where E0 is the set of vertices, E1 is the set of edges, and r and s are maps from E1 to E0 giving the range and source of an edge. Standing Assumption 2.2. Unless explicitly stated otherwise, all graphs are assumed to be countable, i.e., the set of vertices and the set of edges are countable sets. We follow the notation and definition for graph C∗-algebras in [FLR00] and warn the reader that this is not the convention used in the monograph by Raeburn ([Rae05]). Definition 2.3. Let E = (E0, E1, r, s) be a graph. The graph C∗-algebra C∗(E) is defined to be the universal C∗-algebra generated by mutually orthogonal projections (cid:8)pv : v ∈ E0(cid:9) and partial isometries(cid:8)se : e ∈ E1(cid:9) satisfying the relations We get our definition of Leavitt path algebras from [AAP05, AMP07]. Definition 2.4. Let k be a field and let E be a graph. The Leavitt path algebra Lk(E) is the universal k-algebra generated by pairwise orthogonal idempotents {v v ∈ E0} and elements {e, e∗ e ∈ E1} satisfying • e∗f = 0, if e 6= f , • e∗e = r(e), • s(e)e = e = er(e), esf = 0 if e, f ∈ E1 and e 6= f , ese = pr(e) for all e ∈ E1, e ≤ ps(e) for all e ∈ E1, and, • s∗ • s∗ • ses∗ • pv =Pe∈s−1(v) ses∗ e for all v ∈ E0 with 0 < s−1(v) < ∞. FILTERED K-THEORY FOR GRAPH ALGEBRAS 3 • e∗s(e) = e∗ = r(e)e∗, and, • v =Pe∈s−1(v) ee∗, if s−1(v) is finite and nonempty. Recall that graph C∗-algebras come with a natural gauge action and that Leavitt path algebras come with a natural grading. We now turn to the ideal structure of Leavitt path algebras and graph C∗-algebras, where we are particularly interested in graded ideals and gauge invariant ideals. Standing Assumption 2.5. Unless explicitly stated otherwise, all ideals in rings are two-sided ideals and all ideals in a C∗-algebra are closed two-sided ideals. Definition 2.6. For any ring R we denote by I(R) the lattice of ideals in R. As per usual we write v ≥ w if there is a path from the vertex v to the vertex w. We call a subset H ⊆ E0 hereditary if v ∈ H and v ≥ w imply that w ∈ H, and we say that H is saturated if for every v ∈ E0 with 0 < s−1(v) < ∞ and r(s−1(v)) ⊆ H we have v ∈ H. If H is saturated and hereditary we define BH =(cid:8)v ∈ E0 \ H : s−1(v) = ∞ and 0 < s−1(v) ∩ r−1(E0 \ H) < ∞(cid:9) . In other words, BH consists of infinite emitters that are not in H and emit a non- zero finite number of edges to vertices not in H. We say that those vertices are breaking for H. Definition 2.7 ([Tom07, Definition 5.4]). An admissible pair (H, S) consists of a saturated hereditary subset H and a subset S of BH . We put an order on the set of admissible pairs by letting (H, S) ≤ (H ′, S′) if and only if H ⊆ H ′ and S ⊆ H ′ ∪ S′. This is in fact a lattice order. Theorem 2.8 ([BHRS02, Theorem 3.6] and [Tom07, Theorem 5.7]). Let E be a graph and let k be a field. • There is a canonical lattice isomorphism from the set of admissible pairs to (H,S) for the image the set of gauge invariant ideals of C∗(E). We write I top of an admissible pair. • There is a canonical lattice isomorphism from the set of admissible pairs (H,S) for the image of an to the set of graded ideals of Lk(E). We write I alg admissible pair. One of the main reasons the sublattice of graded ideals can be used to study the Morita equivalence classes of Leavitt path algebras is that the graded ideals are preserved by (not necessarily graded) ring isomorphisms. Lemma 2.9. Let E be a graph and let k be a field. Suppose I is an ideal in Lk(E). Then I is graded if and only if I is generated by idempotents. Proof. Suppose I is graded. Then I = I alg definition (see for instance [Tom07, Definition 5.5]) I alg H} and (H,S) for some admissible pair (H, S). By (H,S) is generated by {v : v ∈ {v − Xs(e)=v r(e) /∈H ee∗ : v ∈ S}. Hence I is generated by idempotents. Suppose instead I is generated by idempotents. Let e ∈ I be an idempotent in the generating set S of idempotents for I. By [HLM+14, Theorem 3.4], e is equivalent in M∞(Lk(G)) to a finite sum of the idempotents of the form v ∈ E0 i where s(e) = w ∈ E0, s−1(w) = ∞, and each ei is an element of s−1(w). Then Se where e is replaced by these new idempotents in the generating set S will generate the ideal I. Thus, I is generated by idempotents in the vertex and w −Pn i=1 eie∗ 4 SØREN EILERS, GUNNAR RESTORFF, EFREN RUIZ, AND ADAM P. W. SØRENSEN set and idempotents of the form v −Pn and each ei is an element of s−1(v). Therefore, I is a graded ideal. i=1 eie∗ i , where s(e) = v ∈ E0, s−1(v) = ∞, (cid:3) Finally we recall from [ERRS16, Section 3] the definition of Primeγ(C∗(E)) and FKtop,+(Specγ(C∗(E)); C∗(E)). Definition 2.10. Let E = (E0, E1, r, s) be a graph. Let Primeγ(C∗(E)) denote the set of all proper ideals that are prime within the set of proper gauge invariant ideals. We give Primeγ(C∗(E)) the Jacobson topology and can then show that C∗(E) has a canonical structure as a Primeγ(C∗(E))-algebra. So when E has finitely many vertices -- or, more generally, Primeγ(C∗(E)) is finite -- we can consider the reduced filtered ordered K-theory of C∗(E): FKtop,+(Specγ(C∗(E)); C∗(E)). Loosely speaking this is the collection of the K-groups associated to certain sub- quotients I/J of gauge invariant ideals I, J in C∗(E) together with certain maps of the associated six-term exact sequences. 3. S-Prime spectrum for a ring We will now introduce the Prime-spectrum of a ring relative to a sublattice of ideals. Our primary motivation is to look at prime graded ideals in Leavitt path algebras. Definition 3.1. Let R be a ring and let S be a sublattice of I(R) containing the trivial ideals {0} and R. An ideal P ∈ S is called S-prime if P 6= R and for any ideals I, J ∈ S, IJ ⊆ P =⇒ I ⊆ P or J ⊆ P. We denote by SpecS(R) the set of all S-prime ideals of R. We note that if P is S-prime and I, J are in S then IJ ⊆ I ∩ J so we have I ∩ J ⊆ P =⇒ I ⊆ P or J ⊆ P. We will equip SpecS(R) with the Jacobson (or hull-kernel) topology. For each subset T ⊆ SpecS(R) we define the kernel of T as ker(T ) = \p∈T p and the closure of T as (3.1) T = {p ∈ SpecS(R) : p ⊇ ker(T )} . Note that if R is a commutative ring and S = I(R), then SpecS(R) is the spectrum of R with the Zariski topology. Lemma 3.2. Let R be a ring and let S be a sublattice of I(R) closed under arbitrary intersections and containing the trivial ideals {0} and R. The closure operation defined in (3.1) satisfies the Kuratowski closure axioms, that is (1) ∅ = ∅, (2) T ⊆ T , for all T ⊆ SpecS(R), (3) T = T , for all T ⊆ SpecS(R), and, (4) T1 ∪ T2 = T1 ∪ T2, for all T1, T2 ⊆ SpecS(R). Proof. Once we recall that by definition ker(∅) = R it is clear that 1. holds and since we have p ⊇ ker(T ) for all p ∈ T , 2. also holds. For 3. we observe that ker(T ) = ker(T ), and then clearly T = T . FILTERED K-THEORY FOR GRAPH ALGEBRAS 5 Finally suppose that T1, T2 ⊆ SpecS(R). Since ker(T1 ∪ T2) = ker(T1) ∩ ker(T2) we have that T1 ∪ T2 = {p ∈ SpecS(R) : p ⊇ ker(T1 ∪ T2)} = {p ∈ SpecS(R) : p ⊇ ker(T1) ∩ ker(T2)} = {p ∈ SpecS(R) : p ⊇ ker(T1) or p ⊇ ker(T2)} = T1 ∪ T2. So 4. holds. (cid:3) We now describe the open sets in the Jacobson topology. To this end we define for each I ∈ S the set W (I) = {p ∈ SpecS(R) : p + I} . Lemma 3.3. Let R be a ring and let S be a sublattice of I(R) closed under ar- bitrary intersections and containing the trivial ideals {0} and R. Then for all U ⊆ SpecS(R), U is open if and only if Furthermore, if I ∈ S is such that U = W (ker(U c)). I = ker({p ∈ SpecS (R) : p ⊇ I}), then W (I) is open. Proof. Let U be a subset of SpecS(R). Then U is open if and only if U c = U c if and only if U c = {p ∈ SpecS(R) : p ⊇ ker(U c)} if and only if U = {p ∈ SpecS(R) : p + ker(U c)} = W (ker(U c)). Let now I ∈ S be such that I = ker({p ∈ SpecS (R) : p ⊇ I}). To ease notation we let H = {p ∈ SpecS (R) : p ⊇ I}, so that I = ker(H). Then W (I)c = {p ∈ SpecS(R) : p + I}c = {p ∈ SpecS (R) : p + ker(H)}c = {p ∈ SpecS(R) : p ⊇ ker(H)} = H. Hence W (I) is open. (cid:3) We now define a lattice isomorphism between the open sets of SpecS(R) and the elements of S. Theorem 3.4. Let R be a ring and let S be a sublattice of I(R) closed under arbitrary intersections and containing the trivial ideals {0} and R. Suppose that for each I ∈ S we have that I = ker({p ∈ SpecS (R) : p ⊇ I}). Define φ : O(SpecS(R)) → S by φ(U ) = ker(U c). Then φ is a lattice isomorphism. 6 SØREN EILERS, GUNNAR RESTORFF, EFREN RUIZ, AND ADAM P. W. SØRENSEN Proof. To show that φ is bijective we define γ : S → O(SpecS(R)) by γ(I) = W (I) and check that it is an inverse. Note that by Lemma 3.3 the set W (I) is in fact open. For each I ∈ S we have φ(γ(I)) = φ(W (I)) = ker(W (I)c) = ker({p ∈ SpecS(R) : p ⊇ I}) = I, by the assumption on I. On the other hand, if U ⊆ SpecS(R) is open we can use Lemma 3.3 to get γ(φ(U )) = γ(ker(U c)) = W (ker(U c)) = U. Hence φ is bijective. To show that φ is a lattice isomorphism it only remains to verify that both φ and γ preserves order. Let U, V be open subsets of SpecS(R) with U ⊆ V . Then V c ⊆ U c so φ(U ) = ker(U c) ⊆ ker(V c) = φ(V ), and hence φ is order preserving. Let now I, J ∈ S be such that I ⊆ J. Then W (I)c = {p ∈ SpecS(R) : p ⊇ I} ⊇ {p ∈ SpecS(R) : p ⊇ J} = W (J)c, which implies that γ(I) = W (I) ⊆ W (J) = γ(J), i.e., γ is order preserving. (cid:3) In keeping with the notation from C∗-algebras we define R[U ] = φ(U ) for every U ∈ O(PrimeS(R)). Whenever we have open sets V ⊆ U we can form the quotient R[U ]/R[V ]. The next lemma shows that the quotient R[U ]/R[V ] only depends on the set difference U \ V up to canonical isomorphism. Lemma 3.5. Let R be a ring and let S be a sublattice of I(R) closed under arbitrary intersections and containing the trivial ideals {0} and R. Suppose that for each I ∈ S we have that I = ker({p ∈ SpecS (R) : p ⊇ I}). Then for all U, V ∈ O(SpecS(R)) we have R[U ∪ V ] = R[U ] + R[V ] and R[U ∩ V ] = R[U ] ∩ R[V ]. Consequently, if V1, V2, U1, U2 ∈ O(SpecS(R)) are such that V1 ⊆ U1, V2 ⊆ U2, and U1 \V1 = U2 \V2, then there exits an isomorphism from R[U1]/R[V1] to R[U2]/R[V2] and this isomorphism is natural, i.e., if also V3, U3 ∈ O(SpecS(R)) with V3 ⊆ U3 and U3 \ V3 = U1 \ V1, then the composition of the isomorphisms from R[U1]/R[V1] to R[U2]/R[V2] and from R[U2]/R[V2] to R[U3]/R[V3] is equal to the isomorphism from R[U1]/R[V1] to R[U3]/R[V3]. Proof. The first part of the theorem follows from the fact that φ is a lattice iso- morphism (Theorem 3.4) and that S is a sublattice. Suppose now V1, V2, U1, U2 ∈ O(X) are as in the statement of the Lemma. Then V1 ∪ U2 = U1 ∪ U2 = U1 ∪ V2 and therefore R[U2] + R[V1] = R[V1 ∪ U2] = R[U1 ∪ V2] = R[U1] + R[V2]. Since U2 ∩ (V1 ∪ V2) = V2 we get (R[U2] + R[V1])/(R[V1] + R[V2]) ∼= R[U2]/(R[U2] ∩ R[V1 ∪ V2]) = R[U2]/R[U2 ∩ (V1 ∪ V2)] = R[U2]/R[V2]. Similarly (R[U1] + R[V2])/(R[V1] + R[V2]) ∼= R[U1]/R[V1]. FILTERED K-THEORY FOR GRAPH ALGEBRAS 7 Hence R[U1]/R[V1] ∼= (R[U1] + R[V2])/(R[V1] + R[V2]) = (R[U2] + R[V1])/(R[V1] + R[V2]) ∼= R[U2]/R[V2]. Suppose that we also have V3, U3 ∈ O(SpecS(R)) with V3 ⊆ U3 and U3 \ V3 = U1 \ V1. Then V1 ∪ U2 = U1 ∪ U2 = U1 ∪ V2, V2 ∪ U3 = U2 ∪ U3 = U2 ∪ V3, V1 ∪ U3 = U1 ∪ U3 = U1 ∪ V3, V1 = U1 ∩ (V1 ∪ V2) = U1 ∩ (V1 ∪ V3) = U1 ∩ (V1 ∪ V2 ∪ V3), V2 = U2 ∩ (V1 ∪ V2) = U2 ∩ (V2 ∪ V3) = U2 ∩ (V1 ∪ V2 ∪ V3), and V3 = U3 ∩ (V1 ∪ V3) = U3 ∩ (V2 ∪ V3) = U3 ∩ (V1 ∪ V2 ∪ V3). Now, by considering the isomorphism constructed above, one then gets that the isomorphism is natural from Noether's isomorphism theorem. (cid:3) Definition 3.6. Let X be a topological space and let Y be a subset of X. We call Y locally closed if Y = U \ V where U, V ∈ O(X) with V ⊆ U . We let LC(X) be the set of locally closed subsets of X. Definition 3.7. Let R be a ring and let S be a sublattice of I(R) closed under arbitrary intersections and containing the trivial ideals {0} and R. Suppose that for each I ∈ S we have that I = ker({p ∈ SpecS (R) : p ⊇ I}). For Y = U \ V ∈ LC(SpecS(R)), define R[Y ] := R[U ]/R[V ]. By Lemma 3.5, R[Y ] does not depend on U and V up to a canonical choice of isomorphism. 4. Specγ(Lk(E)) and Primeγ(C∗(E)) Having set up our notion of primitive ideal spectrum relative to a sublattice, we will now apply it to the graded ideals of Leavitt path algebras. Definition 4.1. Let E be a graph and let k be a field. We denote by Iγ(Lk(E)) the sublattice of I(Lk(E)) consisting of all graded ideals of Lk(E) and for brevity we let Specγ(Lk(E)) = SpecIγ (Lk(E))(Lk(E)). Similarly we let Iγ(C∗(E)) be the sublattice of I(C∗(E)) consisting of all gauge invariant ideals of C∗(E). Recall from [ERRS16, Section 3] that Primeγ(C∗(E)) denotes the collection of prime gauge invariant ideals of C∗(E). We first prove that the lattice of graded ideals and the lattice of gauge invariant ideals are isomorphic in a canonical way. Lemma 4.2. Let E be a graph. The map β : Iγ(Lk(E)) → Iγ(C∗(E)) that is given by β(I alg (H,S) is a lattice isomorphism. Furthermore β maps Specγ(Lk(E)) bijectively onto Primeγ(C∗(E)). (H,S)) = I top 8 SØREN EILERS, GUNNAR RESTORFF, EFREN RUIZ, AND ADAM P. W. SØRENSEN Proof. By Theorem 2.8 there is a lattice isomorphism βalg from the set of admis- sible pairs to Iγ(Lk(E)) given by βalg((H, S)) = I alg (H,S), and a lattice isomorphism βtop from the set of admissible pairs to Iγ(C∗(E)) given by βtop((H, S)) = I top (H,S). Consequently, β = βtop ◦ β−1 alg is a lattice isomorphism. Let S = Iγ(Lk(E)). It follows from [NvO82, Proposition II.1.4] that a graded ideal I of Lk(E) is S-prime if and only if I is a prime ideal of Lk(E). Thus, by [Ran13, Theorem 3.12], every S-prime ideal I of Lk(E) is of the form • I = I alg • I = I alg (H,S), where E0 \ H is a maximal tail and S = BH , or (H,S) where E0 \ H = M (u) and S = BH \ {u} for some breaking vertex, and that these ideals are distinct. In [ERRS16, Section 3] it is shown that every ideal I in Primeγ(C∗(E)) is of the form • I = I top • I = I top (H,S), where E0 \ H is a maximal tail and S = BH , or (H,S) where E0 \ H = M (u) and S = BH \ {u} for some breaking vertex, and that these ideals are distinct. Hence I top if I alg Primeγ(C∗(E)). (H,S) is in Primeγ(C∗(E)) if and only (H,S) is in Specγ(Lk(E)). In other words β maps Specγ(Lk(E)) bijectively onto (cid:3) We can now prove that the collection of graded ideals satisfies the kernel as- sumption we used in Section 3. Proposition 4.3. Let E be a graph. If I is a proper graded ideal of Lk(E), then I = ker(cid:0)(cid:8)p ∈ Specγ(Lk(E)) : p ⊇ I(cid:9)(cid:1) . Proof. Let β be the lattice isomorphism from Lemma 4.2 and let I ∈ Iγ(Lk(E)) be a proper ideal. By [ERRS16, Lemma 3.5] we have that β(I) = \ q∈Primeγ (C ∗(E)) q⊇β(I) q. Since I is a graded ideal I = I alg intersection of graded ideals is again graded we also have (H,S) for some admissible pair (H, S). As the \p∈Specγ (Lk(E)) p⊇I p = I alg (H′,S′), for some admissible pair (H ′, S′). We will now show that I top (H′,S′) is an intersection of ideals that all contain I alg Since I alg (H,S) = I top (H,S), I alg (H′,S′). (H,S) ⊆ I alg (H′,S′) (H′,S′) as β is order preserving. If q ∈ Primeγ(C∗(E)) (H,S) ⊆ β−1(q). Therefore β−1(q) is one the ideals which implies that I top is such that I top whose intersection define I alg (H,S) ⊆ I top (H,S) ⊆ q, then I alg (H′,S′) so (H′,S′) = β(I alg I top We now have the following inclusions (H′ ,S′)) ⊆ β(β−1(q)) = q. I top (H,S) ⊆ I top (H′,S′) ⊆ q = β(I) = I top (H,S). \ q∈Primeγ (C ∗(E)) q⊇I top (H,S) FILTERED K-THEORY FOR GRAPH ALGEBRAS 9 Therefore, I top (H,S) = I top (H′,S′). Hence (H, S) = (H ′, S′) so I = I alg (H,S) = I alg (H′,S′) = \p∈Specγ (Lk(E)) p⊇I p = ker(cid:0)(cid:8)p ∈ Specγ(Lk(E)) : p ⊇ I(cid:9)(cid:1) (cid:3) Lemma 4.4. Let k be a field and let {Iα}α∈S be a subset of Iγ(Lk(E)). Then the ideal I =Tα∈S Iα is an element of Iγ(Lk(E)). Proof. Recall that J ∈ Iγ(Lk(E)) if and only if J =Tn∈Z J ∩ Lk(E)n, where Lk(E)n =( lXk=1 k : λk ∈ k, µk, νk are finite paths, and µk − νk = n) . \n∈Z Iα ∩ Lk(E)n = \α∈S \n∈Z Iα ∩ Lk(E)n λkµkν∗ Then I ∩ Lk(E)n = \n∈Z \α∈S = \α∈S Iα = I. Hence, I ∈ Iγ(Lk(E)). Corollary 4.5. The map (cid:3) U 7→ \p∈Specγ (R)\U p is a lattice isomorphism from O(Specγ(Lk(E))) to Iγ(Lk(E)). Proof. This follows from Theorem 3.4 which is applicable by Proposition 4.3 and Lemma 4.4. (cid:3) As the final result in this section we prove that β restricts to a homeomorphism between the graded prime ideals and the gauge prime ideals. Theorem 4.6. Let E be a graph. Then φ = βSpecγ (Lk(E)) is a homeomorphism from Specγ(Lk(E)) to Primeγ(C∗(E)), where β is the lattice isomorphism from Lemma 4.2. Proof. We first observe that Lemma 3.3 and Proposition 4.3 combine to show that the open sets of Specγ(Lk(E)) are precisely the sets of the form W (I) for some proper ideal I ∈ Iγ(Lk(E)). Let a proper ideal I ∈ Iγ(Lk(E)) be given. Then β(W (I)) = β(cid:0)(cid:8)p ∈ Specγ(Lk(E)) : p + I(cid:9)(cid:1) =(cid:8)β(p) : p ∈ Specγ(Lk(E)) and p + I(cid:9) =(cid:8)β(p) : p ∈ Specγ(Lk(E)) and β(p) + β(I)(cid:9) = {q ∈ Primeγ(C∗(E)) : q + β(I)} . By [ERRS16, Lemma 3.6] the last set is open, and hence φ−1 is continuous. The above computation used that β was a lattice isomorphism and that we had complete, and similar looking, descriptions of the open sets in Specγ(Lk(E)) and Primeγ(C∗(E)). Hence a completely parallel computation will show that φ is also continuous. Therefore φ is a homeomorphism. (cid:3) 10 SØREN EILERS, GUNNAR RESTORFF, EFREN RUIZ, AND ADAM P. W. SØRENSEN 5. Filtered algebraic K-theory In this section we define filtered algebraic K-theory for rings and show that if two Leavitt path algebras over C have isomorphic filtered algebraic K-theory then the associated graph C∗-algebras have isomorphic filtered K-theory. We then use this result to answer the Abrams-Tomforde conjecture for a large class of finite graphs. Suppose R is a ring and S is a sublattice of ideals. Moreover, assume that every I ∈ S has a countable approximate unit consisting of idempotents, i.e., for every I ∈ S, there exists a sequence {en}∞ n=1 in I such that • en is an idempotent for all n ∈ N, • enen+1 = en for all n ∈ N, and • for all r ∈ I, there exists n ∈ N such that ren = enr = r. Then for any locally closed subset Y = U \ V of SpecS(R), we have the algebraic K-groups {K alg n (R[Y ])}n∈Z. Moreover, for all U1, U2, U3 ∈ O(SpecS(R)) with U1 ⊆ U2 ⊆ U3, by [RT13, Lemma 3.10], we have a long exact sequence in algebraic K-theory K alg n (R[U2 \ U1]) ι∗ / K alg n (R[U3 \ U1]) π∗ / K alg n (R[U3 \ U2]) ∂∗ / / K alg n−1(R[U2 \ U1]). Definition 5.1. Let R be a ring and let S be a sublattice of I(R) closed under arbitrary intersections and containing the trivial ideals {0} and R. Suppose that for each I ∈ S we have that I = ker({p ∈ SpecS (R) : p ⊇ I}). Moreover, assume that every I ∈ S has a countable approximate unit consisting of idempotents. (1) We define FKalg(SpecS(R); R) to be the collection n (R[Y ])}n∈Z,Y ∈LC(PrimeS (R)), {K alg equipped with the natural transformations {ι∗, π∗, ∂∗}. (2) We define FKalg,+(SpecS (R), R) to be the collection FKalg(SpecS (R); R) 0 (R[Y ]) for all Y ∈ LC(SpecS(R)). together with the positive cone of K alg Definition 5.2. Let R, R′ be rings, let S be a sublattice of I(R) closed under arbitrary intersections and containing the trivial ideals {0} and R, and let S′ be a sublattice of I(R′) closed under arbitrary intersections and containing the trivial ideals {0} and R′. Suppose that for each I ∈ S we have that and that for each I ′ ∈ S′ we have that I = ker({p ∈ SpecS (R) : p ⊇ I}), I ′ = ker({p ∈ SpecS ′(R′) : p ⊇ I ′}). Moreover, assume that every I ∈ S and every I ′ ∈ S′ have a countable approximate unit consisting of idempotents. An isomorphism from FKalg(SpecS(R); R) to FKalg(SpecS ′ (R′); R′) consists of a homeomorphism φ : SpecS(R) → SpecS ′ (R′) and isomorphisms αY,∗ from K∗(R[Y ]) to K∗(R′[φ(Y )]) for each Y ∈ LC(SpecS(R)) such the diagrams involving the nat- ural transformations commute. If the isomorphism from FKalg(SpecS(R); R) to FKalg(SpecS ′ (R′); R′) restricts to an order isomorphism on K0(R[Y ]) for all Y ∈ LC(SpecS(R)), we write FKalg,+(SpecS(R); R) ∼= FKalg,+(SpecS ′ (R′); R′). / / FILTERED K-THEORY FOR GRAPH ALGEBRAS 11 Lemma 5.3. Let E be a graph and let k be a field. Then every graded-ideal of Lk(E) has a countable approximate unit consisting of idempotents. Consequently, we may define FKalg,+(Specγ(Lk(E)); Lk(E)). Proof. Let F be a graph and set F 0 = {v1, v2, . . . }. Then {Pn n=1 is a count- able approximate unit consisting of idempotents for Lk(F ). Thus, every Leavitt path algebra has a countable approximate unit consisting of idempotents. The lemma now follows since by [RT14, Corollary 6.2] every graded-ideal of Lk(E) is isomorphic to a Leavitt path algebra. (cid:3) k=1 vk}∞ Lemma 5.4. Let E be a directed graph and let φ : Specγ(LC(E)) → Specγ(C∗(E)) be the homeomorphism given in Theorem 4.6. The for all U ∈ O(Specγ(LC(E))), there exists an admissible pair (H, S) such that LC(E)[U ] = I alg (H,S) and C∗(E)[φ(U )] = I top (H,S). Proof. This follows from the construction of φ in Theorem 4.6 as the restriction of the lattice isomorphism β that sends I alg (cid:3) (H,S) to I top (H,S). Let A be a C∗-algebra and let A be a ∗-algebra. Suppose ιA is a ∗-homomorphism from A to A. Denote the composition K alg n (A) Kn(ιA) / K alg n (A) / K top n (A) by γn,A, where K top Theorem 5.5. Let E be a directed graph and let n (A) is the (usual) topological K-theory of the C∗-algebra A. φ : Specγ(LC(E)) → Specγ(C∗(E)) be the homeomorphism given in Theorem 4.6. For all U1, U2, U3 ∈ O(Specγ(LC(E))) with U1 ⊆ U2 ⊆ U3, for all n ∈ Z, the diagram K alg n (LC(E)[U2 \ U1]) K alg n (LC(E)[U3 \ U1]) K alg n (LC(E)[U3 \ U2]) K alg n−1(LC(E)[U2 \ U1]) γn,C∗(E)[V2 \V1 ] γn,C∗(E)[V3 \V1 ] γn,C∗ (E)[V3 \V2] γn−1,C∗(E)[V2 \V1] K top n (C∗(E)[V2 \ V1]) / K top n (C∗(E)[V3 \ V1]) / K top n (C∗(E)[V3 \ V2]) / K top n−1(C∗(E)[V2 \ V1]) is commutative, where Vi = φ(Ui) Proof. This follows Lemma 5.4 and from [Cor11, Theorems 2.4.1 and 3.1.9] . (cid:3) Lemma 5.6. Let E be a graph. Then for all (H1, S1), (H2, S2) admissible pairs with (H1, S1) ≤ (H2, S2), we have that γ0,I top (H2 ,S2)/I top (H1 ,S1) : K alg 0 (I alg (H2,S2)/I alg (H1,S1)) → K top 0 (I top (H2,S2)/I top (H1,S1)) is an order isomorphism and γ1,I top (H2 ,S2)/I top (H1 ,S1) : K alg 1 (I alg (H2,S2)/I alg (H1,S1)) → K top 1 (I(H2,S2)/I(H1,S1)) is surjective with kernel a divisible group. Suppose F is a graph and suppose there exists an order isomorphism α0 : K alg 0 (I alg (H2,S2)/I alg (H1,S1)) → K alg 0 (I alg (H′ 2)/I alg (H′ 1)) 1,S′ 2,S′ and there exists an isomorphism 1)), i) is an admissible 2). Then α0 2, S′ 1 (I alg α1 : K alg (H2,S2)/I alg (H1,S1)) → K alg 2)/I alg (H′ 1,S′ where (Hi, Si) is an admissible pair of E for i = 1, 2 and (H ′ i, S′ pair of F for i = 1, 2 with (H1, S1) ≤ (H2, S2) and (H ′ and α1 induce isomorphisms (I top 1 (I alg (H′ (H2,S2)/I top (H1,S1)) → K top (I top (H′ 1, S′ 2)/I top 2,S′ 2,S′ (H′ 0 1) ≤ (H ′ 1)) 1,S′ 0 eα0 : K top / / / / / / / / / / / 12 SØREN EILERS, GUNNAR RESTORFF, EFREN RUIZ, AND ADAM P. W. SØRENSEN and (I top (H2,S2)/I top (H1,S1)) → K top 1 (I top (H′ 2)/I top (H′ 1)) 1,S′ 2,S′ 1 eα1 : K top such that eα0 is an order isomorphism and γi,I top (H′ /I top (H′ 1 ,S′ 1) 2 ,S′ 2) ◦ αi =eαi ◦ γi,I top (H2 ,S2)/I top (H1 ,S1) . (H2 ,S2)/I top (H2,S2)/I alg (H,S)) ⊆ I top (H1,S1) to I top Proof. Let ιE : LC(E) → C∗(E) be the ∗-homomorphism sending v to pv and e to se. Note that for all admissible pairs (H, S), ιE(I alg (H,S). Therefore, for all admissible pairs (H1, S1), (H2, S2) with (H1, S1) ≤ (H2, S2), ιE induces a ∗-homomorphism from I alg (H1,S1). We denote this map . Thus, the composition of this induced map in K-theory by ιE,I top with the homomorphism from K alg γn,I top (H2 ,S2)/I top We will show that it is enough to prove the first part of the lemma for the case (H2, S2) = (∅, ∅) and (H1, S1) = (E0, ∅). Let (H, S) be an admissible pair. Let E(H,S) be the graph given in [RT14, Definition 4.1]. By the proofs of [RT14, Theorems 5.1 and 6.1], there exist ∗-isomorphisms (H1,S1)) to K top (H2,S2)/I top (H2,S2)/I top (H2,S2)/I top (H1,S1)) is n (I top n (I top (H1 ,S1) (H1 ,S1) . β(H,S) : LC(E(H,S)) → I alg (H,S) and λ(H,S) : C∗(E(H,S)) → I top (H,S) given by β(H,S)(v) := β(H,S)(e) := β(H,S)(e∗) := λ(H,S)(qv) := λ(H,S)(te) := v vH αα∗ αr(α)H α∗ e α αr(α)H e∗ α∗ r(α)H α∗ if v ∈ H if v ∈ S if v = α ∈ F1(H, S) if v = α ∈ F2(H, S) if e ∈ E1 if e = α ∈ F 1(H, S) if e = α ∈ F 2(H, S) if e ∈ E1 if e = α ∈ F 1(H, S) if e = α ∈ F 2(H, S) pv pH v sαs∗ α sαpH r(α)s∗ α se sα sαpH r(α) if v ∈ H if v ∈ S if v = α ∈ F1(H, S) if v = α ∈ F2(H, S) if e ∈ E1 if e = α ∈ F 1(H, S) if e = α ∈ F 2(H, S). and Note that the diagram LC(E(H,S)) ιE(H,S) β(H,S) I alg (H,S) ι E,I top (H,S) /0 C∗(E(H,S)) λ(H,S) / I top (H,S) / / / FILTERED K-THEORY FOR GRAPH ALGEBRAS 13 commutes. Therefore, for two admissible pairs (H1, S1), (H2, S2) with (H1, S1) ≤ (H2, S2), the diagram LC(E(H2,S2))/β−1 (H2,S2)(I alg (H1,S1)) ιE(H2 ,S2) C∗(E(H2,S2))/λ−1 (H2,S2)(I top (H1,S1)) β(H2 ,S2) I alg (H2,S2)/I alg (H1,S1) λ(H2 ,S2) / I top (H2,S2)/I top (H1,S1) ι E,I top (H2 ,S2) /I top (H1 ,S1) where β(H2,S2) and λ(H2,S2) are the induced ∗-isomorphisms on the quotient, com- mutes. Therefore, it is enough to prove the lemma for the graph E(H2,S2). Hence, we may assume that (H2, S2) = (E0, ∅). Set (H1, S1) = (H, S) to simplify the notation. Let E \ (H, S) be the graph defined in [Tom07, Theorem 5.7(2)]. Then by the proof of [Tom07, Theorem 5.7(2)] and the discussion before [BHRS02, Corollary 5.7], there are ∗-isomorphisms and δ(H,S) : LC(E \ (H, S)) → LC(E)/I(H,S) η(H,S) : C∗(E \ (H, S)) → C∗(E)/I(H,S) such that the diagram LC(E \ (H, S)) δ(H,S) LC(E)/I(H,S) ιE\(H,S) ιE,C∗ (E)/I(H,S) C∗(E \ (H, S)) η(H,S) / C∗(E)/I(H,S) commutes. Hence, it is enough to prove the lemma for the graph E \ (H, S). Hence, we may assume that (H, S) = (∅, ∅). Thus, proving the claim. The fact that γ0,C ∗(E)/0 is an isomorphism follows from [HLM+14, Corollary 3.5]. To prove that γ1,C ∗(E)/0 is surjective and its kernel is a divisible group we reduce to the case that E is row-finite. Let F be a Drinen-Tomforde desingularization of E defined in [DT05]. Then there are embeddings ω : LC(E) → LC(F ) and ρ : C∗(E) → C∗(F ) such that the diagram LC(E) ω / LC(F ) ιE ιF C∗(E) / C∗(F ) ρ commutes, ω(LC(E)) is a full corner of LC(F ), and ρ(C∗(E)) is a full corner of C∗(F ). Hence, ω and ρ induce isomorphisms in K-theory. Therefore, it is enough to prove γ1,C ∗(E),0 is surjective with kernel a divisible group for the case that E is row-finite. The row-finite case follows from [RT13, Lemma 4.7]. The first part of the lemma now follows. For the last part of the lemma, since K0(ιE,I top (H2 ,S2)/I top (H1 ,S1) ) is an order isomor- /I top (H′ γ0,I top (H′ phism, it is clear that α0 induces an order isomorphism eα0 such that The fact that α1 induces an isomorphism eα1 such that γ1,I top eα1 ◦γ1,I top ◦ α0 =eα0 ◦ γ0,I top is the result of the kernel of γ1,I top (H′ 2 (H2 ,S2)/I top (H2 ,S2)/I top (H2 ,S2)/I top (H1 ,S1) (H1 ,S1) 2 ,S′ 2) 1 ,S′ 1) (H1 ,S1) ,S′ 2 ) ◦ α1 = /I top (H′ 1 being a divisible ,S′ 1 ) / / / / / / / / 14 SØREN EILERS, GUNNAR RESTORFF, EFREN RUIZ, AND ADAM P. W. SØRENSEN group and K1(I top (H′ 2)/I top (H′ 2,S′ 1)) being torsion free, thus [RT13, Lemma 4.8] applies. 1,S′ (cid:3) Theorem 5.7. Let E and F be graphs. (1) Suppose FKalg,+(Specγ(LC(E)); LC(E)) ∼= FKalg,+(Specγ(LC(F )); LC(F )). Then FKtop,+(Specγ(C∗(E)); C∗(E)) ∼= FKtop,+(Specγ(C∗(F )); C∗(F )). (2) Suppose E0, F 0 < ∞. If θ : FKalg,+(Specγ(LC(E)); LC(E)) → FKalg,+(Specγ(LC(F )); LC(F )) is an isomorphism such that θ0 sends [1LC(E)]0 ∈ K alg K alg 0 (LC(F )), then there exists an isomorphism 0 (LC(E)) to [1LC(F )]0 ∈ Θ : FKtop,+(Primeγ(C∗(E)); C∗(E)) → FKtop,+(Primeγ(C∗(F )); C∗(F )) such that Θ0 sends [1C ∗(E)]0 ∈ K top 0 (C∗(E)) to [1C ∗(F )]0 ∈ K top 0 (C∗(F )). Proof. The theorem follows from Lemmas 5.4 and 5.6, and Theorem 5.5. (cid:3) Corollary 5.8. Let E and F be graphs. (1) If LC(E) and LC(F ) are isomorphic as rings, then FKtop,+(Primeγ(C∗(E)); C∗(E)) ∼= FKtop,+(Primeγ(C∗(F )); C∗(F )). If, in addition, E0, F 0 < ∞, then there exists an isomorphism Θ : FKtop,+(Primeγ(C∗(E)); C∗(E)) → FKtop,+(Primeγ(C∗(F )); C∗(F )) such that Θ0 sends [1C ∗(E)]0 ∈ K top 0 (C∗(E)) to [1C ∗(F )]0 ∈ K top 0 (C∗(F )). (2) If LC(E) and LC(F ) are Morita equivalent, then FKtop,+(Primeγ(C∗(E)); C∗(E)) ∼= FKtop,+(Primeγ(C∗(F )); C∗(F )). Proof. 1. follows from Lemma 2.9 and Theorem 5.7. Suppose LC(E) and LC(F ) are Morita equivalent. Then by [AT11, Corol- lary 9.11], M∞(LC(E)) ∼= M∞(LC(F )) as rings. By [AT11, Proposition 9.8(2)], M∞(LC(E)) ∼= LC(SE) and M∞(LC(F )) ∼= LC(SF ) as C-algebras, where SE and SF are the stabilized graphs of E and F respectively (see [AT11, Definition 9.4]). Note that every graded ideal LC(SE) is of the from M∞(I) for a unique graded ideal of I of LC(E) and every graded ideal of LC(SF ) is of the from M∞(J) for a unique graded ideal J of LC(F ). We also have that FKalg,+(Specγ(LC(E)); LC(E)) ∼= FKalg,+(Specγ(LC(SE)); LC(SE)) ∼= FKalg,+(Specγ(LC(SF )); LC(SF )) ∼= FKalg,+(Specγ(LC(F )); LC(F )). Therefore, by Theorem 5.7, FKtop,+(Primeγ(C∗(E)); C∗(E)) ∼= FKtop,+(Primeγ(C∗(F )); C∗(F )). (cid:3) Corollary 5.9. The Abrams-Tomforde conjecture holds for the class of finite graphs that satisfy Condition (H) of [ERRS16, Definition 4.19]. In particular the Abrams- Tomforde conjecture holds for the class of finite graphs that satisfy Condition (K). Proof. The first part is just a combination of Corollary 5.8 and [ERRS16, Theorem 6.1]. Finally, all graphs that satisfy Condition (K) satisfy Condition (H). (cid:3) FILTERED K-THEORY FOR GRAPH ALGEBRAS 15 Acknowledgements This work was partially supported by the Danish National Research Foundation through the Centre for Symmetry and Deformation (DNRF92), by the VILLUM FONDEN through the network for Experimental Mathematics in Number Theory, Operator Algebras, and Topology, by a grant from the Simons Foundation (# 279369 to Efren Ruiz), and by the Danish Council for Independent Research -- Natural Sciences. References [AAP05] Gene Abrams and Gonzalo Aranda Pino. The Leavitt path algebra of a graph. J. Algebra, 293(2):319 -- 334, 2005. [AGOP98] P. Ara, K. R. Goodearl, K. C. O'Meara, and E. Pardo. Separative cancellation for [AMP07] projective modules over exchange rings. Israel J. Math., 105:105 -- 137, 1998. P. Ara, M. A. Moreno, and E. Pardo. Nonstable K-theory for graph algebras. Algebr. Represent. Theory, 10(2):157 -- 178, 2007. [APPSM06] G. Aranda Pino, E. Pardo, and M. Siles Molina. Exchange Leavitt path algebras and [AT11] stable rank. J. Algebra, 305(2):912 -- 936, 2006. Gene Abrams and Mark Tomforde. Isomorphism and Morita equivalence of graph algebras. Trans. Amer. Math. Soc., 363(7):3733 -- 3767, 2011. [Cor11] [BHRS02] Teresa Bates, Jeong Hee Hong, Iain Raeburn, and Wojciech Szyma´nski. The ideal structure of the C ∗-algebras of infinite graphs. Illinois J. Math., 46(4):1159 -- 1176, 2002. Guillermo Cortinas. Algebraic v. topological K-theory: a friendly match. In Topics in algebraic and topological K-theory, volume 2008 of Lecture Notes in Math., pages 103 -- 165. Springer, Berlin, 2011. D. Drinen and M. Tomforde. The C ∗-algebras of arbitrary graphs. Rocky Mountain J. Math., 35(1):105 -- 135, 2005. Søren Eilers, Gunnar Restorff, Efren Ruiz, and Adam P. W. Sørensen. Geometric classification of graph C ∗-algebras over finite graphs. ArXiv e-prints, April 2016. Neal J. Fowler, Marcelo Laca, and Iain Raeburn. The C ∗-algebras of infinite graphs. Proc. Amer. Math. Soc., 128(8):2319 -- 2327, 2000. [ERRS16] [FLR00] [DT05] [Rae05] [NvO82] [HLM+14] Damon Hay, Marissa Loving, Martin Montgomery, Efren Ruiz, and Katherine Todd. Non-stable K-theory for Leavitt path algebras. Rocky Mountain J. Math., 44(6):1817 -- 1850, 2014. Ja A. Jeong and Gi Hyun Park. Graph C ∗-algebras with real rank zero. J. Funct. Anal., 188(1):216 -- 226, 2002. C. Nastasescu and F. van Oystaeyen. Graded ring theory, volume 28 of North-Holland Mathematical Library. North-Holland Publishing Co., Amsterdam-New York, 1982. Iain Raeburn. Graph algebras, volume 103 of CBMS Regional Conference Series in Mathematics. Published for the Conference Board of the Mathematical Sciences, Washington, DC; by the American Mathematical Society, Providence, RI, 2005. Kulumani M. Rangaswamy. The theory of prime ideals of Leavitt path algebras over arbitrary graphs. J. Algebra, 375:73 -- 96, 2013. Efren Ruiz and Mark Tomforde. Ideal-related K-theory for Leavitt path algebras and graph C ∗-algebras. Indiana Univ. Math. J., 62(5):1587 -- 1620, 2013. Efren Ruiz and Mark Tomforde. Ideals in graph algebras. Algebr. Represent. Theory, 17(3):849 -- 861, 2014. Mark Tomforde. Uniqueness theorems and ideal structure for Leavitt path algebras. J. Algebra, 318(1):270 -- 299, 2007. [Tom07] [Ran13] [JP02] [RT13] [RT14] 16 SØREN EILERS, GUNNAR RESTORFF, EFREN RUIZ, AND ADAM P. W. SØRENSEN Department of Mathematical Sciences, University of Copenhagen, Universitetspark- en 5, DK-2100 Copenhagen, Denmark E-mail address: [email protected] Department of Science and Technology, University of the Faroe Islands, N´oat´un 3, FO-100 T´orshavn, the Faroe Islands E-mail address: [email protected] Department of Mathematics, University of Hawaii, Hilo, 200 W. Kawili St., Hilo, Hawaii, 96720-4091 USA E-mail address: [email protected] Department of Mathematics, University of Oslo, PO BOX 1053 Blindern, N-0316 Oslo, Norway E-mail address: [email protected]
1607.03942	2	1607	2016-08-08T17:03:37	Primeness property for graded central polynomials of verbally prime algebras	[ "math.RA" ]	Let $F$ be an infinite field. The primeness property for central polynomials of $M_n(F)$ was proved by A. Regev, i.e., if the product of two polynomials in distinct variables is central then each factor is also central. In this paper we consider the analogous property for $M_n(F)$ and determine, within the elementary gradings with commutative neutral component, the ones that satisfy this property, namely the crossed product gradings. Next we consider $M_n(R)$, where $R$ admits a regular grading, with a grading such that $M_n(F)$ is a homogeneous subalgebra and provide sufficient conditions - satisfied by $M_n(E)$ with the trivial grading - to prove that $M_n(R)$ has the primeness property if $M_n(F)$ does. We also prove that the algebras $M_{a,b}(E)$ satisfy this property for ordinary central polynomials. Hence over a field of characteristic zero every verbally prime algebra as the primeness property.	math.RA	math	PRIMENESS PROPERTY FOR GRADED CENTRAL POLYNOMIALS OF VERBALLY PRIME ALGEBRAS DIOGO DINIZ AND CLAUDEMIR FIDELIS BEZERRA JÚNIOR Abstract. Let F be an infinite field. The primeness property for central polynomials of Mn(F ) was established by A. Regev, i.e., if the product of two polynomials in distinct variables is central then each factor is also central. In this paper we consider the analogous property for Mn(F ) and determine, within the elementary gradings with commutative neutral component, the ones that satisfy this property, namely the crossed product gradings. Next we con- sider Mn(R), where R admits a regular grading, with a grading such that Mn(F ) is a homogeneous subalgebra and provide sufficient conditions - satis- fied by Mn(E) with the trivial grading - to prove that Mn(R) has the primeness property if Mn(F ) does. We also prove that the algebras Ma,b(E) satisfy this property for ordinary central polynomials. Hence we conclude that, over a field of characteristic zero, every verbally prime algebra has the primeness property. 1. Introduction The study of central polynomials is an important part of the theory of algebras with polynomial identites. Verbally prime algebras were introduced by A. Kemer [12] in his solution of the Specht problem and are of fundamental importance in the theory of p.i. algebras. The existence of central polynomials for verbally prime algebras was proved by Yu. P. Razmyslov [15] and earlier for matrix algebras, in- dependently, by Formanek [11] and Razmyslov [14]. A p.i. algebra A is verbally prime if whenever f (x1, . . . , xr) and g(xr+1, . . . , xs) are two polynomials in distinct variables and f · g is an identity for A then either f or g is an identity for A. In his structure theory of T -ideals Kemer proves that over a field of characteristic zero every non-trivial verbally prime p.i. algebra is equivalent to one of the alge- bras Mn(F ), Mn(E) (here E denotes the Grassmann algebra of a vector space of countable dimension) and certain subalgebras Ma,b(E) of Ma+b(E) (see Section 2). Amitsur proved in [3] that if the field F is infinite the ideal of identities of Mn(F ) is a prime ideal. As an analog of Amitsur's result for central polynomials Regev proved in [16] the following primeness property on the central polynomials: if f (x1, . . . , xr) and g(xr+1, . . . , xs) are two polynomials in distinct variables and f · g is a central polynomial for Mn(F ) then both f and g are central. Graded polynomial identities and graded central polynomials play an important role in the study of p.i. algebras, they were used in the theory developed by Kemer. Such identities are easier to describe in many important cases and they are related to the ordinary ones, for example, coincidence of graded identities implies the coincidence of the ordinary ones. In [9] regular gradings on the algebras Mn(E) 2010 Mathematics Subject Classification. 16R20 16W50 16R99. Key words and phrases. verbally prime algebras, central polynomials, graded algebras. The first author was partially supported by CNPq. 1 2 SILVA AND BEZERRA and Ma,a(E) (see [5]) were used to prove that, if the field F is of characteristic zero, these algebras satisfy the primeness property. In this paper the canonical Z2-grading on Ma,b(E) is used to prove (see Theorem 5.8) that these algebras also satisfy the primeness property. This proves that, over a field of characteristc zero, every verbally prime algebra satisfies the primeness property. We remark that this property arises implicitly in the description of (graded) central polynomials of some verbally prime algebras. For example, over an infinite field of characteristic p > 2 the central polynomials of E together with its identities form a so called limit T -space and the primeness property is useful to describe these central polynomials (see [8]). Graded simple algebras, over an algebraically closed field, have been described in [6], for a group G these are essentially matrix algebras over a twisted group algebra F αH, where α is a 2-cocicle with entries in F ×. If H is abelian then the canonical H-grading on F αH is a regular grading. Classification of abelian gradings on matrix algebras plays an important role in the cassification of gradings on Lie algebras (see [10]). A grading on Mn(F ) is elementary if every elementary matrix is homogeneous, these gradings play an important role in the description of graded simple algebras. The graded identities and central polynomials of matrix algebras have been described for various gradings. For example, with the canonical Z and Zn, central polynomials were described in [7] and more generally in [1] with crossed product gradings. It is then a natural question to consider the analogous primeness property for the graded polynomials of Mn(F ). Graded identities for the algebra Mn(F ) with an elementary grading by the group G such that the component associated to the neutral element of G is a commutative subalgebra were described in [4] and [9]. In Theorem 3.10 we prove that with such a grading the algebra Mn(F ) satisfies the primeness property if and only if the grading is a crossed product grading (i.e. G has order n) and there exists no nontrivial homomorphism G → F ×. In Section 4 we consider matrix algebras Mn(R) with entries in an algebra R that has a regular grading by an abelian group H. The gradings considered on Mn(R) are ones for which R and Mn(F ) are homogeneous subalgebras and R lies in the neutral component. Under natural hypothesis on the regular grading on R we prove that Mn(R) satisfies the property whenever Mn(F ) does. As a corollary we prove that Mn(E) (see Corollary 4.14) satisfies the primeness property. Finally in Section 5 the canonical Z2-grading on Ma,b is used to we prove the property for the ordinary central polynomials of these algebras. 2. Definitions and Preliminary Results Let F be an infinite field of characteristic different from 2, G a group with identity element ǫ and A an F -algebra. A grading by the group G on A is a vector space decomposition A = ⊕g∈GAg such that AgAh ⊆ Agh for every g, h in G. The subspaces Ag are called the homogeneous components of A and a non-zero element of Ag is said to be homogeneous of degree g. A subalgebra B is said to be a homogeneous subalgebra if B = ⊕g∈GAg ∩ B. We denote by F hXGi the free G-graded algebra, freely generated by the set XG = {xi,gi ∈ N, g ∈ G}. This algebra has a natural grading by G where the homogeneous component (F hXGi)g is the subspace generated by the monomials xi1,g1 · · · xim,gm such that g1 · · · gm = g. Let f (x1,g1 , . . . , xm,gm ) be a polynomial in F hXGi, an m-tuple (a1, . . . , am) such PRIMENESS PROPERTY FOR CENTRAL POLYNOMIALS OF VERBALLY PRIME ALGEBRAS3 that ai ∈ Agi for i = 1, . . . , m is called an f -admissible substitution (or simply an admissible substitution). If f (a1, . . . , am) = 0 for every admissible substitution (a1, . . . , am) we say that the polynomial f is a graded polynomial identity for A. We denote by Z(A) the centre of A and we say that f is a graded central polynomial for A if f is not a graded polynomial identity and f (a1, . . . , am) ∈ Z(A) for every admissible substitution (a1, . . . , am). Definition 2.1. Let A = ⊕g∈GAg be an algebra with a grading by the group G. We say that A satisfies the primeness property for (graded) central polynomials if, whenever the product f (x1,g1 , · · · , xr,gr ) · g(xr+1,gr+1 , · · · , xs,gs ) of two polynomials in distinct variables is a graded central polynomial, then both f (x1,g1 , · · · , xr,gr ) and g(xr+1,gr+1, · · · , xs,gs ) are graded central polynomials for A. The set of all graded central polynomials and all graded identities for A is a subspace of F hXGi invariant under all endomorphisms of F hXGi, such subspaces are called TG-spaces. The intersection of TG-spaces is also a TG-space. Given a subset S ⊂ F hXGi we denote by hSiTG the intersection of all TG-spaces that contain S, this is the TG-space generated by S. If S = {f } we use the notation hf iTG for the TG-space generated by {f }. We remark that hf iTG is generated as a vector space by the set {f (p1, . . . , pn)pi ∈ (F hXGi)gi }. If f is a graded central polynomial for A then every element of hf iTG either a graded identity or a graded central polynomial for A. If G = {ǫ} we recover the definition of ordinary polynomial identities and central polynomials, in this case we use the notation F hXi for the free associative algebra and xi for the variables. The Grassmann algebra E of a countable dimensional vector space with basis {e1, e2, . . . } has the presentation: h1, e1, e2, . . . eiej = −ejei, for all i,j ≥ 1i. It is well known that BE = {1, ei1ei2 . . . eik i1 < i2 < · · · < ik} is a basis for E. Hence we have E = E0 ⊕ E1 where E0 is the subspace generated by 1 and all monomials with even k and E1 is the subspace generated by the monomials with odd k. This decomposition is a Z2-grading on E. The algebra Ma+b(E) has a subalgebra denoted Ma,b(E) which consists of matrices of the form (cid:18) A B C D (cid:19) , where A ∈ Ma(E0), D ∈ Mb(E0) and B, C are blocks with entries in E1. This algebra has a canonical Z2-grading with homogeneous components (Ma,b(E))0 and (Ma,b(E))1 that consists of matrices of the form (cid:18) A 0 0 D (cid:19) and (cid:18) 0 B C 0 (cid:19) , respectively. Proposition 2.2. Let f (x1, . . . , xr) be a multihomogeneous polynomial of degree di in xi and hi(x1, . . . , xr, zi) the multihomogeneous component of f (x1 + z1, . . . , xr + zr) of degree one in zi. Then (1) [f (x1, . . . , xr), xr+1] =Xi hi(x1, . . . , xr, [xi, xr+1]). 4 SILVA AND BEZERRA Proof. The equality follows from the fact that the map a 7→ [a, xr+1] is a deriva- (cid:3) tion of F hXi (see[pg. 9][18]). Proposition 2.3. Let A be a G-graded algebra with basis B that consists of ho- mogeneous elements, V a subspace of A and f (x1,g1 , . . . , xn,gn ) a polynomial. The following assertions are equivalent: (i) For every admissible substitution a1, . . . , an by elements of A, the element f (a1, . . . , an) lies in V ; (ii) For every polynomial f ′(x1,g1 , . . . , xm,gm ) in hf iTG and every admissible substitution a1, . . . , am by elements of A, the element f ′(a1, . . . , am) lies in V . (iii) For every multihomogeneous polynomial f ′(x1,g1 , . . . , xm,gm ) in hf iTG and every admissible substitution b1, . . . , bm by elements of the basis B, the ele- ment f ′(b1, . . . , bm) lies in V . Proof The set S = {f (p1, . . . , pn)pi ∈ (F hXGi)gi } generates hf iTG as a vector space. If the assertion (i) holds for f then it holds for every element of S and therefore for every f ′ in hf iTG . Hence we conclude (ii). It is clear that (ii) implies (iii). Now we assume that (iii) holds. Let a1, . . . , an an admissible substitution j=1 λijbj, i = 1, . . . , n, where bj ∈ B. Let s1 = 0 and si+1 = si + ni for i = 2, . . . , m − 1. Denote qi(xsi+1,gi , . . . , xsi+ni,gi) = by elements of A. We write ai = Pni Pni j=1 λij xsi+j,gi and write f (q1, . . . qm) = f1 + · · · + fl, as a sum of multihomogeneous components. The field F is infinite and therefore the polynomials f1, . . . , fl lie in hf iTG . Let S be the substitution such that xsi+j,gi is replaced by bj. The result qjS of this substitution is qj(b1, . . . , bni) = ai and therefore f (a1, . . . , an) = f1S + · · ·+ flS. From (iii) we conclude that the elements f1S, . . . , flS lie in V , hence f (a1, . . . , an) also lies in V . (cid:3) Corollary 2.4. Let A be an algebra graded by the group G, V a subspace of A and let f (x1,g1 , . . . , xm,gm) be a polynomial such that for every admissible substitution a1, . . . , am by elements of A, the element f (a1, . . . , am) lies in V . Then for every b in Aǫ and every admissible substitution a1, . . . , am the commutator [f (a1, . . . , am), b] also lies in V . Proof. Proposition 2.2 implies that (2) [f (x1,g1 , . . . , xm,gm ), xm+1,ǫ] =Xi hi(x1,g1 , . . . , xm,gm , [xi,gi , xm+1,ǫ]). The variable xm+1,ǫ is homogeneous of degree ǫ, therefore [xi,gi , xm+1,ǫ] has de- gree gi and the polynomial hi(x1,g1 , . . . , xm,gm, [xi,gi , xm+1,ǫ]) lies in the TG-space generated by f . The substitution (a1, . . . , am, b) is hi-admissible, hence the previ- ous proposition implies that hi(a1, . . . , am, [ai, b]) lies in V . The result now follows directly from (2). (cid:3) Proposition 2.5. Let f (x1,g1 , . . . , xr,gr ), g(xr+1,gr+1, . . . , xs,gs ) be polynomials in In distinct variables. particular if f · g is a central polynomial for A then f ′g′ is also a central polynomial for A. If f ′ ∈ hf iTG and g′ ∈ hgiTG then f ′g′ lies in hf · giTG. PRIMENESS PROPERTY FOR CENTRAL POLYNOMIALS OF VERBALLY PRIME ALGEBRAS5 Proof. Note that f ′ and g′ are linear combinations of polynomials of the form f (p1, . . . , pr) and g(q1, . . . , qs) respectively, where p1, . . . , pr and q1, . . . , qs admis- sible substitutions by polynomials in F hXGi. Hence the product f ′ · g′ is a linear combination polynomials in the set S = {f (p1, . . . , pr) · g(q1, . . . , qr)p1, . . . , pr, q1, . . . , qs ∈ F hXi}. Since f and g are polynomials in distinct variables each element in S lies in hf · giT . Therefore f ′ · g′ lies in hf · giT . The second assertion is an obvious consequence of the first one. (cid:3) 3. Elementary gradings on Mn(F ) and the primeness property In this section A denotes the algebra Mn(F ). We denote Eij the elementary matrix with 1 in the (i, j)-th entry and 0 in the other entries. Definition 3.1. Let g = (g1, . . . , gn) be an n-tuple of elements in the group G. The decomposition A = ⊕g∈GAg, where Ag is the subspace generated by the elementary matrices Eij such that g−1 i gj = g, is a grading by G on A called the elementary grading induced by g. We recall that an elementary grading by an n-tuple of parwise distinct elements by a group G of order n is called a crossed product grading (see [1]). Our main result in this section characterizes the elementary gradings by an n-tuple of pairwise distinct elements on A for which this algebra satisfies the primeness property for graded central polynomials. We prove in Theorem 3.10 that these gradings are crossed product gradings by a group G that has no non-trivial homomorphism G → F ×. Proposition 3.2. Let A be the matrix algebra Mn(F ) with an elementary grading and f (x1,g1 , . . . , xr,gr ), g(xr+1,gr+1 , . . . , xs,gs) polynomials in distinct variables such that the product f · g is a graded central polynomial for A. Then there exists an invertible diagonal matrix P such that the result of every f -admissible (resp. g- admissible) substitution is a scalar multiple of P (resp. P −1). Proof. Let f (x1,g1 , . . . , xr,gr ) and g(xr+1,gr+1 , . . . , xs,gs ) be polynomials in dis- tinct variables such that f · g is a graded central polynomial for A. There ex- ists an admissible substitution A1, . . . , As such that f (A1, . . . , Ar) · g(Ar+1, . . . , As) is a non-zero scalar matrix. Hence g(Ar+1, . . . , As) is an invertible matrix, let P denote its inverse. If B1, . . . , Bs is a f -admissible substitution the product f (B1, . . . , Br) · g(Ar+1, . . . , As) is a scalar matrix. Therefore f (B1, . . . , Br) is a scalar multiple of P for every f -admissible substitution. Analogously, the result of every g-admissible substitution is a scalar multiple of P −1. We now prove that P is a diagonal matrix. It follows from Corollary 2.4 that if M is any homogeneous matrix of degree ǫ and B1, . . . , Br is an f -admissible substitution the commutator [f (B1, . . . , Br), M ] is a scalar multiple of P . Therefore [P, M ] = λM P for some scalar λM . The grading is elementary, hence the matrices Eii, i = 1, . . . , n have degree ǫ and for each i there exists a scalar λi such that (3) [P, Eii] = λiP. We write P = Pi,j pijEij . If λi = 0 for every i then an easy computation yields that P is a diagonal matrix. We prove that this must be the case. Assume that 6 SILVA AND BEZERRA a contradition since this matrix is not invertible. If pil 6= 0 for some l 6= i then we conclude that pii = 0 and pkl = 0 if k 6= i and l 6= i. Moreover pki = λipki If pki 6= 0 for some k 6= i for every k 6= i and −pil = λipil for every l 6= i. λi 6= 0 for some i. The left side of (3) is Pk6=i pkiEki −Pl6=i pilEil. Since λi 6= 0 then λi = 1 and pil = 0 for every l 6= i. In this case P = Pk6=i pkiEki which is λi = −1 and P =Pl6=i pilEil, this last matrix is not invertible and again we have a contradiction. The remaining possibility is that pki = 0 for every k 6= i and pli = 0 for every l 6= i which implies P = 0. This is also a contradiction. (cid:3) Our main objective now is to prove that for elementary gradings determined by tuples of pairwise distinct elements the primeness property holds only for crossed product gradings by suitable groups. The precise statement is in Theorem 3.10. Definition 3.3. Let A be the matrix algebra Mn(F ) with an elementary grading. We denote H the set of permutations σ in Sn for which the corresponding automor- phism Λσ, given by Λσ(Eij ) = Eσ(i)σ(j), is a graded automorphism of A. The map σ 7→ Λσ is a homomorphism and the set H in the definition above is a subgroup of Sn. Next we prove some results that involve this subgroup. Proposition 3.4. If f is a non-identity such that the result of every f -admissible substitution is a scalar multiple of P , it follows that for every σ ∈ H there exists a non-zero scalar λ(σ) such that The map λ : H → F × is a homomorphism. Λσ(P ) = λ(σ)−1P. Proof. There exists an admissible substitution f (A1, . . . , Am) = λP for some scalar λ 6= 0. The automorphisms Λσ are graded and therefore Λσ(λP ) = Λσ(f (A1, . . . , Am)) = f (Λσ(A1), . . . , Λσ(Am)) = λ′P, σ P . for some non-zero λ′ depending on σ. Hence there exists a non-zero scalar λσ It is clear that λ : H → F × given by σ 7→ λσ is a such that Λσ(P ) = λ−1 homomorphism. (cid:3) In the next lemma we describe the matrices P that satisfy the equality Λσ(P ) = λ(σ)−1P in terms of the homomorphism λ and of the canonical action of H on {1, 2, . . . , n}. Lemma 3.5. Let λ : H → F × be a homomorphism and for each i denote by P (i) =Pα∈H λ(α)Eα(i),α(i). The following assertions hold: (i) For every σ ∈ H we have Λσ(P (i)) = λ(σ)−1P (i); (ii) The matrices P (i) and P (j) have a non-zero entry in the same position if and only if i and j lie in the same H-orbit; (iii) Let i1, . . . , id be a system of representatives of the orbits of the canonical action of H on {1, 2, . . . , n}. The diagonal matrix P satisfies (4) Λσ(P ) = λ(σ)−1P for every σ ∈ H if and only if P is a linear combination of the matrices P (i1), . . . , P (id). Proof. We have Λσ(P (i)) = Xα∈H λ(α)Eσα(i),σα(i) = λ(σ)−1 Xα∈H λ(σα)Eσα(i),σα(i) = λ(σ)−1P (i), PRIMENESS PROPERTY FOR CENTRAL POLYNOMIALS OF VERBALLY PRIME ALGEBRAS7 hence assertion (i) holds. Assertion (ii) is clear since the non-zero entries of P (i) are in the positions corresponding to the orbit of i. Next we prove (iii). Let P = Pi piEii be a matrix that satisfies (4) and λ1, . . . , λd be scalars such that the is-th entry of the matrix λsP (is) is pis . Let P ′ = Ps λsP (is) and write P ′ = Pi p′ iEii. It follows from (ii) that pis = p′ is for s = 1, . . . , d. Assertion (i) implies that P ′ also satisfies (4), hence pσ(i) = p′ σ(i) whenever pi = p′ i and σ ∈ H. Since i1, . . . , id is a system of representatives this implies that P = P ′. (cid:3) For elementary gradings by tuples of pairwise distinct elements it is possible to provide for the matrices in the above lemma a non-identity f such that the result of every admissible substitution is a scalar multiple of P . This is done in the next proposition. Lemma 3.6. Let x1,g1 · · · xn,gn be a multilinear monomial of degree ǫ such that the n-tuple (g1, g1g2, . . . , g1 · · · gn) consists of pairwise distinct elements. If the n- tuples (Ei1,j1 , . . . , Ein,jn ) and (Ek1 ,l1, . . . , Ekn ,ln) are admissible substitutions and the result of each substitution is not zero then there exists σ ∈ H such that (Ek1 ,l1, . . . , Ekn ,ln) = (Eσ(i1),σ(j1), . . . , Eσ(in),σ(jn)). Proof. Since Ei1,j1 · · · Ein ,jn 6= 0 and the total degree of the polynomial is ǫ and we conclude that j1 = i2, . . . , jn−1 = in, jn = i1. Moreover the fact that (g1, g1g2, . . . , g1 · · · gn) consists of pairwise distinct elements implies that (i1, . . . , in) is a permutation of the elements 1, . . . , n. Analogously (k1, . . . , kn) is a permutation. We denote by σ the permutation in Sn such that σ(is) = ks, for s = 1, . . . , n. The substitutions considered are admissible, therefore the automorphism of A such that Eij 7→ Eσ(i),σ(j) maps the matrices Ei1,i2 , . . . , Ein−1,in , Ein,i1 onto elementary matrices of the same degree. This implies that this automorphism is a graded automorphism. (cid:3) Proposition 3.7. Let A be algebra Mn(F ) with an elementary grading induced by the n-tuple (h1, . . . , hn) of pairwise distinct elements. If there exists a non-zero diagonal matrix P and a non-identity f (x1,g1 , . . . , xm,gm ) such that the result of every admissible substitution is a scalar multiple of P then there exists a homo- morphism λ : H → F × such that P equals a linear combination of the matrices P (i1), . . . , P (id) in the previous lemma. Conversely let λ : H → F × be a homo- morphism and P = diag(p1, . . . , pn) a non-zero linear combination of the matrices P (i1), . . . , P (id), then the polynomial (5) f (x1,g1 , . . . , xn,gn ) =Xi pixσi(1),gσi (1) , . . . , xσi(n),gσi (n) , where σ is the n-cicle (12 . . . n), gi = h−1 n h1; is not an identity and has the property that every the result of every admissible substitution is a scalar multiple of P . i hi+1 for i = 1, . . . , n − 1 and gn = h−1 Proof. The first claim about P follows form Proposition 3.4 and (iii) of the Lemma 3.5. Now we prove the second claim. The polynomial f is multilinear, hence it suffices to prove that the result of every admissible substitution by elementary matrices is a scalar multiple of P . It is clear that f (E12, . . . , En1) = P , in particular f is not a graded identity. Lemma 3.6 implies that every admissible substitution such that the result is not zero is of the form (Λσ(E12), . . . , Λσ(En1)). Now (iii) of the Lemma 3.5 implies f (Λσ(E12), . . . , Λσ(En1)) = Λσ(P ) = λ(σ)−1P, 8 SILVA AND BEZERRA and the second claim follows. (cid:3) Lemma 3.8. If A has an elementary grading induced by an n-tuple of pairwise distinct elements of g then the orbit of every element in the set {1, 2, . . . , n} under the canonical action has H elements. In particular H divides n. Proof. If σ(i) = i for some σ ∈ H and some i then for every j the matrices Ei,j and Ei,σ(j) have the same degree. Since the n-tuple that determines the grading consists of pairwise distinct elements this implies that σ(j) = j. Hence the stabilizer of every element is trivial and the result follows. (cid:3) Proposition 3.9. Let A be algebra Mn(F ) with an elementary grading induced by the n-tuple (g1, . . . , gn) of pairwise distinct elements of the group G. Then A has a crossed product grading by the support if and only if H = n, moreover in this case the groups H and G are isomorphic. Proof. We assume first that A has a crossed product grading. For each g ∈ G let σg ∈ Sn be the permutation such that (gg1, . . . , ggn) = (gσg (1), . . . , gσg (n)). The map g 7→ σg is a group homomorphism. To verify this let g, h be elements in G, we have gσg (j) = ggj for every j and therefore gσg (σh(i)) = ggσh(i) = ghgi = gσgh(i). Hence σgσh = σgh. If σg(j) = j for some j we have gj = ggj, hence g = ǫ. This proves that the homomorphism is injective. Now we prove that its image lies in H. The degree of Λσg (Eij ) = Eσg (i)σg (j) is σg (i)gσg (j) = (ggi)−1(ggj) = g−1 g−1 i gj, hence Λσg is a graded automorphism. Lemma 3.8 implies that H has order at most n, therefore g 7→ σg is also surjective. Hence H and G are isomorphic and H = n. Now we assume that H = n. Lemma 3.8 implies that there is only one orbit, i.e., the action of H on {1, . . . , n} is transitive. Let g be in the support of the grading and Ekl be of degree g. For every i there exists σ ∈ H such that σ(k) = i, hence Eij has degree g for j = σ(l). Now if h also lies in the support of the grading then there exists a matrix Ejt of degree h and Eit = Eij Ejt has degree gh. Therefore gh lies in the support, it is clear that h−1 lies in the support if h does, hence the support is a subgroup of G of order n. Since the elements (ǫ, g−1 1 gn) are pairwise distinct elements in the support we conclude that the grading is a crossed product grading by the support. (cid:3) 1 g2, . . . , g−1 Theorem 3.10. Let A be the matrix algebra Mn(F ) with an elementary grading by an n-tuple of pairwise distinct elements of G. The algebra A has the primeness property for central polynomials if and only if the grading is a crossed product grading and the group G has no non-trivial homomorphism G → F ×. Proof. If A does not have a crossed product grading then it follows from Propo- sition 3.9 that H < n. In this case Lemma 3.8 implies that the number d of orbits in the H action on {1, 2, . . . , n} is greater than 1. Let i1, . . . , id be a system of rep- resentatives for the orbits, we apply Proposition 3.7 with the trivial homomorphism to obtain a polynomial f such that the result of every admissible substitution is a scalar multiple of the matrix P = P (i1) + · · · + P (id−1) − P (id). Since d > 1, the PRIMENESS PROPERTY FOR CENTRAL POLYNOMIALS OF VERBALLY PRIME ALGEBRAS9 polynomial f is not graded central. Clearly P 2 = I, therefore the product of two copies of f in distinct variables is central. Hence A does not have the primeness property for central polynomials. Now assume that A has a crossed product grad- ing and that G has a non-trivial homomorphism G → F ×. Proposition 3.9 implies that H is isomorphic to G. In this case there exists a non-trivial homomorphism λ : H → F ×. The action of H determines only one orbit, we apply Lemma 3.5 with this homomorphism and Proposition 3.7 to obtain a matrix P = P (i1) and a poly- nomial f such that the image of every admissible substitution is a scalar multiple of P . The entries in the diagonal of P are in the image of λ and therefore are n-th roots of the unit. This implies that P n = I, hence the product of n copies of f in distinct variables is central. The homomorphism λ is not trivial, hence P is not a scalar matrix and f is not a graded central polynomial. Therefore A does not have the primeness property for central polynomials. Finally assume that A has a crossed product grading and that the only ho- momorphism G → F × is the trivial one. In this case we fix two polynomials f (x1,g1 , . . . , xr,gr ) and g(xr+1,gr+1 , . . . , xs,gs ) in distinct variables such that the product f · g is a graded central polynomial for A. If we prove that the matrix P in Proposition 3.2 is a scalar matrix it follows that f and g are central polyno- mials. Proposition 3.4 implies that there exits a homomorphism λ : H → F × such that Λσ(P ) = λ(σ)−1P. Since the grading is a crossed product grading H is isomorphic to G. The only one dimensional homomorphism of G onto F × is the trivial one, hence λ is the trivial homomorphism. The order of H is n, hence Lemma 3.8 implies that the action of H on {1, . . . , n} determines only one orbit. In this case Lemma 3.5 implies that P is a scalar matrix. (cid:3) Let A = ⊕g∈GAg and A = ⊕h∈HAh be gradings on A by the groups G and H respectively. We say that the H-grading on A is a coarsening of the G-grading on A if for every g ∈ G there exists h ∈ H such that Ag ⊆ Ah. The next result proves that the primeness property is inherited by coarsenings of a grading with this property. Theorem 3.11. Let A = ⊕g∈GAg be a grading on A = Mn(F ) by the group G and A = ⊕h∈H Ah an H-grading that is a coarsening of the G-grading. If A has the primeness property for G-graded central polynomials then it has the primeness property for H-graded central polynomials. Proof. Let f (x1,h1 , . . . , xr,hr ) and g(xr+1,hr+1, . . . , xs,hs) be polynomials in dis- tinct variables such that the product f · g is an H-graded central polynomial for A. As in the proof of Proposition 3.2 we obtain a g-admissible substitution such that g(Ar+1, . . . , As) is an invertible matrix. Let (A1, . . . , Ar) be an f -admissible substitution. Since the grading by H is a coarsening the grading by G each matrix Ai is a sum of homogeneous matrices in the G-grading. Therefore we may write Ai = B1,i + · · · + Bki,i as a sum of elements of degree g1,i, . . . , gki,i, here gk,i 6= gl,i l=1 xi,gl,i and let f = f (p1, . . . , pr), g = g(pr+1, . . . , ps). Notice that pi, pj are in distinct variables if i 6= j. Hence f and g are in dis- tinct variables. It is clear that if we replace xi,gl,i by Bl,i in pi the result is Ai and therefore the results of this substitution in f and g are f (A1, . . . , Ar) If the product f · g is a graded identity then and g(Ar+1, . . . , As) respectively. if k 6= l. We write pi = Pki 10 SILVA AND BEZERRA f (A1, . . . , Ar) · g(Ar+1, . . . , As) = 0 and the hypothesis that g(Ar+1, . . . , As) is in- vertible implies that f (A1, . . . , Ar) = 0. Otherwise f · g is a central polynomial because f · g is central. The primeness property for G-graded central polynomials implies that f is central and therefore f (A1, . . . , Ar) is a central element. This proves that f is an H-graded central polynomial. Analogously one proves that g is central. (cid:3) 4. Matrix algebras over algebras with a regular grading Let R be an algebra (with unit) and A the matrix algebra Mn(R). We identify the scalar matrix diag (r, . . . , r) with r and consider R as a subalgebra of A. Notice that the matrix algebra B = Mn(F ) is also a subalgebra of A. In this section we consider gradings on A such that R and B are homogeneous subalgebras and R has the trivial grading. In this case it is clear that the grading A = ⊕g∈GAg on A is such that Ag = {rbr ∈ R, b ∈ Bg}. Our main result of this section (Theorem 4.12) is that if R has a suitable regular grading then A has the primeness property whenever B does. As a corollary we obtain the primeness property for Mn(E) over infinite fields. Next we recall the definition of a regular grading (see [2], [17]). Definition 4.1. Let R be an algebra and R = ⊕h∈HRh, a grading by the abelian group H. The H-grading above on R is said to be regular if there exists a commutation function β : H × H → F × such that (P1) For every n and every n-tuple (h1, . . . , hn) of elements of H, there exist r1, . . . , rn such that rj ∈ Rhj , and r1 · · · rn 6= 0. (P2) For every h1, h2 ∈ H and for every ah1 ∈ Ah1 , ah2 ∈ Ah2 , we have ah1ah2 = β(h1, h2)ah2 ah1. The regular H-grading on R is said to be minimal if for any h 6= ǫ there exists h′ such that β(h, h′) 6= 1. Remark 4.2. In the definition above the commutation function β : H × H → F × is a skew-symmetric bicharacter (see [2, Remark 13]), i.e., for every h0 ∈ H the maps h 7→ β(h0, h) and h 7→ β(h, h0) are characters of the group H and β(h2, h1) = β(h1, h2)−1 for every h1, h2 ∈ H. In most of the known examples of a minimal regular grading by an abelian group (see [5]) it turns out that the centre of the algebra coincides with the neutral com- ponent. In the next proposition we prove this coincidence under a minor condition on the regular grading. Proposition 4.3. Let R be an algebra with a regular grading by the group H. If the regular grading is minimal and for every h, h′ ∈ H and every 0 6= rh ∈ Rh there exists sh′ ∈ Rh′ such that rhsh′ 6= 0 then Z(R) = Rǫ. Proof. Since β is a bicharacter it is clear that β(ǫ, h) = 1 for every h ∈ H. Condition (P2) of the above definition now implies that Rǫ ⊆ Z(R). The center Z(R) is a homogeneous subalgebra, hence if Rǫ $ Z(R) there exists h 6= ǫ and 0 6= zh ∈ Rh such that zh ∈ Z(R). For h′ ∈ H let wh′ be such that zhwh′ 6= 0. We have wh′ zh = zhwh′ = β(h, h′)wh′ zh. Hence β(h, h′) = 1 for every h′ ∈ H. This is a contradiction because the regular grading is minimal. (cid:3) PRIMENESS PROPERTY FOR CENTRAL POLYNOMIALS OF VERBALLY PRIME ALGEBRAS11 Remark 4.4. None of the two hypotheses of the proposition previous can be re- moved. Indeed, if R is an algebra with regular grading is not minimal, then there exist ǫH 6= h ∈ H such that β(h, g) 6= 1, for all g ∈ H and so Z(R) contains properly Rǫ. We consider the algebra R with presentation: h1, f, e1, e2, . . . f ei = eif = 0, ejej = −ejeii. The subalgebra of R generated by the ei is the Grassmann algebra E. The decom- position R0 = E0 and R1 = E1 ⊕ hf i is a minimal regular Z2-grading on R. contains properly Rǫ. It is clear that f ∈ Z(A), hence Z(R) Lemma 4.5. Let R be an algebra with a regular grading by an abelian group H such that Z(R) = Rǫ and let A be the algebra Mn(R) with a G-grading. If f = f (x1,g1 , . . . , xr,gr ) is a polynomial such that the result of every admissible sub- stitution lies in Rh, for some h 6= ǫH , then f is a graded identity for A. Proof. Let (a1, . . . , ar) be an admissible substitution, Corollary 2.4 implies that for every h′ ∈ H and every m′ ∈ Rh′ the commutator [f (a1, . . . , ar), m′] lies in Rh. The group H is abelian, hence it is clear that f (a1, . . . , ar)·m′ and m′ ·f (a1, . . . , ar) lie in Rhh′ . Hence [f (a1, . . . , ar), m′] also lies in Vhh′. If h′ 6= ǫH then h 6= hh′ and this implies Vhh′ ∩ Vh = {0}. Therefore we conclude that [f (a1, . . . , ar), b] = 0. Hence f (a1, . . . , ar) commutes with every m′ in Rh′ if h′ 6= ǫ. Since β(ǫ, h) = 1 we conclude that f (a1, . . . , ar) commutes with every element in Rǫ. This implies that f (a1, . . . , ar) lies in Z(R) = RǫH . In this case f (a1, . . . , ar) lies in Rh ∩ RǫH = {0}, therefore f (a1, . . . , ar) = 0. Hence f is a graded identity for A. (cid:3) Let R be an algebra with a regular grading by the abelian group H and let β : H × H → F × denote its commutation function. If m(x1, . . . , xs) is a monomial of degree di in xi and (r1, . . . , rs) is an r-tuple of homogeneous elements of R then it follows from the condition (P2) of Definition 4.1 that m(r1, . . . , rs) = λrd1 1 · · · rds s , for some non-zero scalar λ. Remark 4.6. The scalar λ in the previous equality depends only on the monomial m, the commutation function β : H × H → F × and the s-tuple h = (h1, . . . , hs) such that ai ∈ Rhi. This scalar will be denoted ǫR h,m hence we have m(r1, . . . , rs) = ǫR h,mrd1 1 · · · rds s . Definition 4.7. Let f (x1,g1 , . . . , xn,gn ) be a multihomogeneous polynomial in the algebra F hXGi and h = (h1, . . . , hn) ∈ H n. We write f = Pi αimi as a linear combination of monomials and define the polynomial fh as fh(x1,g1 , . . . , xn,gn ) =Xi ǫR h,miαimi. Remark 4.8. If f (x1,g1 , . . . , xr,gr ) is a multihomogeneous polynomial of degree di in xi,gi , ai ∈ Rhi and Bi ∈ Mn(F ) then f (a1B1, . . . , arBr) = ad1 1 · · · adr r fh(B1, . . . , Br), where h = (h1, . . . , hr). 12 SILVA AND BEZERRA Next we consider an algebra R an algebra with a regular H-grading that satisfies the following condition: (R1) For any two monomials m1, m2 ∈ F hXH i, in distinct variables, that are not H-graded identities for R the product m1 · m2 is not a graded identity for R. The regular gradings in [2], [5] satisfy this property. An algebra R satisfying (R1) satisfies the graded analogue of the definition of a verbally prime algebra. Proposition 4.9. Let R be an algebra with a regular H-grading that satisfies (R1). If f, g ∈ F hXH i are non-identities in distinct variables for R then f · g is a non- identity for R. Proof. Since the field is infinite we may assume without loss of generality that f, g are multihomogeneous. It follows from Remark 4.6 that f, g are, modulo IdH (R), scalar multiples of monomials in distinct variables. Hence (R1) implies that f · g is a non-identity for R. (cid:3) Proof. Let h, h′ ∈ H and r′ h′ 6= 0, where s′ h = f + IdH (R) a non-zero element in R′ Corollary 4.10. Let R be an algebra with a minimal regular H-grading that satis- fies (R1). If R′ = F hXH i/IdH (R), with its canonical H-grading, then Z(R′) = R′ ǫ. h. Let m be such that xm,h′ and f are in distinct variables. Proposition 4.9 implies that f ·xm,h′ does not lie in IdH . Therefore r′ hs′ h′ = xm,h′ + IdH (R). The result now follows from Proposition 4.3. (cid:3) In the next propositon we remark that the algebra R above is verbally prime. Over an algebraically closed field of characteristic zero there is a description of algebras that have a nondegenerate regular grading by a finite group in [2, Corollary 42]. Proposition 4.11. An algebra R with a regular H-grading that satisfies (R1) must be verbally prime. Proof. Let f (x1, . . . , xt), g(xt+1, . . . , xs) be two non-identities, in distinct vari- ables, for R. Proposition 2.3 implies that there exists multihomogeneous polyno- mials f ′(x1, . . . , xt′ ) ∈ hf iT , g′(xt′+1, . . . , xs′ ) ∈ hf iT and homogeneous elements r1, . . . , rt′ , . . . , rs′ such that f ′(r1, . . . , rt′ ) 6= 0 and g(rt′+1, . . . , rs′ ) 6= 0. Remark 4.6 and (R1) imply that f ′(x1, . . . , xt′ ) · g′(xt′+1, . . . , xs′ ) is not an identity for R. Since f ′ · g′ lies in hf · giT we conclude that f · g is not an ordinary identity for R. (cid:3) In the next theorem we consider an algebra A = Mn(R) with a G-grading such that: (Gr1) R is a homogeneous subalgebra and R ⊆ Aǫ; (Gr2) Mn(F ) is a homogeneous subalgebra. Now we state our main result of the section. Theorem 4.12. Let A be the algebra Mn(R) with a G-grading that satisfies (Gr1), (Gr2). If R has a regular grading, by an abelian group H, satisfying (R1) and the algebra B = Mn(F ) (with the inherited G-grading) satisfies the primeness property for graded central polynomials then A also satisfies the primeness property. Proof. We first prove that we may assume without loss of generality that the regular grading on R satisfies (R1) and Z(R) = Rǫ. Let R with a regular H- grading that satisfies (R1), then every coarsening of this grading satisfies (R1). PRIMENESS PROPERTY FOR CENTRAL POLYNOMIALS OF VERBALLY PRIME ALGEBRAS13 Every regular grading on R admits a coarsening by a homomorphic image of H that is minimal, hence we may assume that R has a minimal regular grading that satisfies (R1). For simplicity we still denote the grading group by H. The canonical H-grading on the algebra R′ = F hXHi/IdH (R) is a minimal regular grading that satisfies (R1). Corollary 4.10 implies that Z(R′) = R′ ǫ. The algebra A′ = Mn(R′) with the G-grading such that A′ g = {r′M r′ ∈ R′, M ∈ Bg} satisfies the same G-graded identities as A. Hence A has the primeness property if and only if A′ does. Henceforth we assume that R has a regular grading by H such that (R1) and Z(R) = Rǫ hold. Let f (x1,g1 , . . . , xr,gr ) such that the product f ·g is a central polynomial for A for some g(xr+1,gr+1, . . . , xs,gs ) in distinct variables than f . We will prove that f is a graded central polynomial for A. We first claim that there exists g′(x1,g1 , . . . , xm,gm ) in hgiTG , an m-tuple h = (h1, . . . , hm) in H m and an g′ h-admissible substitution (A1, . . . , Am) such that g′ h(A1, . . . , Am) is an invertible matrix. Let BR be a basis of R of homogeneous elements in the regular H-grading and BB a basis of B of homogeneous matrices in the G-grading. It is clear that BA = {mM m ∈ BR, M ∈ BB} is a basis for A. The product f · g is not an identity, hence Proposition 2.3 implies that there exists f ′(x1,g1 , . . . , xm,gm) in hf iTG and g′(x1,g1 , . . . , xm,gm ) in hgiTG and an admissible substitution (a1A1, . . . , amAm) by elements of BA such that (6) f ′(a1A1, . . . , amAm) · g′(a1A1, . . . , amAm) 6= 0. Proposition 2.5 implies that the element above lies in the centre of A. Let hi denote the degree of ai in the regular grading of R and h = (h1, . . . , hm). Using Remark 4.8 we rewrite the above element as 1 · · · aem h(A1, . . . , Am) · g′ h(A1, . . . , Am), 1 · · · adm m ) · (ae1 m )f ′ (ad1 where di and ei denote the degree of xi,gi in f ′ and g′, respectively. Since this element is central the product f ′ h(A1, . . . , Am) is a scalar matrix, different from 0 because of (6). Hence g′ h(A1, . . . , Am) is an invertible matrix and the claim is proved. h(A1, . . . , Am) · g′ 1 · · · aem Let h−1 be the degree of (ae1 m ) in the regular grading of R. We claim that the result of every admissible substitution in the polynomial f lies Rh. It is clear that f is not an identity, hence this claim and Lemma 4.5 imply that f is central. Let f ′′(y1,g1 , . . . , ys,gs) be a polynomial in hf iTG and (b1B1, . . . , bsBs) an admissible substitution by elements of BA. Denote k = (k1, . . . , ks) the s-tuple where ki is the homogeneous degree of bi in the regular grading of R. The product f ′′(y1,g1, . . . , ys,gs) · g′(x1,g1 , . . . , xm,gm ) is a central polynomial. Hence the element (bf1 1 · · · bfs (7) where fi is the degree of yi,gi in f ′′, is central. We will prove that f ′′(b1B1, . . . , bsBs) lies in Rh, this and Lemma 2.3 imply the claim. We may assume that h(A1, . . . , Am), k (B1, . . . , Bk) · g′ 1 · · · aem s ) · (ae1 m )f ′′ f ′′(b1B1, . . . , bsBs) = (bf1 1 · · · bfs s )f ′′ k (B1, . . . , Bk) 6= 0. Since the product of two monomials m1, m2 ∈ F hXH i that are not identities for R is also not an identity, using (R1) we may assume without loss of generality that (bf1 1 · · · bfs s ) · (ae1 1 · · · aem m ) 6= 0. In this case f ′′ h(A1, . . . , Am) is a scalar matrix. The previous argument proves that the result of every admissible substitution in the polynomial k (B1, . . . , Bm) · g′ 14 SILVA AND BEZERRA h is a scalar matrix. The matrix f ′′ k · g′ f ′′ k (B1, . . . , Bm) is not 0 and the matrix k · g′ h(A1, . . . , Am) is invertible, hence f ′′ g′ h is not an identity. Since B satisfies the primeness property we conclude that f ′′ k is a central polynomial for B. Therefore f ′′ k (B1, . . . , Bk) is a scalar matrix. Since Z(R) = Rǫ it is clear that the centre of A is Rǫ. The element in (7) is central, hence it follows that (bf1 m ) lies in Rǫ. Therefore (bf1 s ) lies in Rh. Hence conclude that f ′′(b1B1, . . . , bsBs) lies in Rh. (cid:3) 1 · · · aem 1 · · · bfs 1 · · · bfs s ) · (ae1 Remark 4.13. Over a field of characteristic zero one may reduce the above ar- gument to multilinear polynomials instead of multihomogeneous ones. In this case the no extra hypothesis is needed on the regular grading on R to prove the previous theorem. Indeed condition (R1) for multilinear monomials is a direct consequence of (P1) in Definition 4.1. As an immediate consequence of the previous theorem we have the following corollaries. Corollary 4.14. The algebra Mn(E) (with the trivial grading) satisfies the prime- ness property for central polynomials. Proof. The canonical Z2-grading on E is regular and satisfies the hypothesis of Theorem 4.12. Moreover, the algebra Mn(F ) with the trivial grading satisfies the primeness property (see [16]). (cid:3) The previous corollary generalizes to infinite fields the main result of [9] which is valid for a field of characteristic zero. Corollary 4.15. Let G be a group with no non-trivial homomorphism G → F ×. If the algebra Mn(E) has a G-grading such that E is a homogeneous subalgebra with the trivial grading and Mn(F ) a homogeneous subalgebra with the crossed product grading then Mn(E) satisfies the primeness property for graded central polynomials. Theorem 4.12. The result now follows from Theorems 4.12 and 3.10. Proof. The canonical Z2-grading on E is regular and satisfies the hypothesis of (cid:3) The algebra R = Mm(F ) has a regular grading that satisfies the hypothesis of Theorem 4.12 (see [5]). Since Mn(R) ∼= Mmn(F ) one may produce other examples of gradings on matrix algebras with entries in the field F that satisfy the primeness property for graded central polynomials using the main result of this section. 5. Primeness property for the algebras Ma,b(E) The verbally prime algebras Mn(F ) and Mn(E) have regular gradings satisfying (R1), (R2). Therefore the primeness property for these algebras may be obtained as a consequence of Theorem 4.12 (with n=1). The remaining type of verbally prime algebra introduced by A. Kemer is Ma,b(E). If a 6= b no regular grading is known for this algebra. In this section we use the canonical Z2-grading and the results in Section 2 to prove that Ma,b(E) satisfies the primeness property for ordinary central polynomials. Definition 5.1. Let A, R be two H-graded algebras. We denote by Ab⊗R the H- graded algebra such that (Ab⊗R)h = Ah ⊗ Rh. The algebra Ma,b(E) is obtained as Ma+b(F )b⊗E where Ma+b(F ) has the ele- mentary Z2-grading induced by the tuple (0, . . . , 0, 1, . . . , 1) with a (resp. b) entries PRIMENESS PROPERTY FOR CENTRAL POLYNOMIALS OF VERBALLY PRIME ALGEBRAS15 equal to 0 (resp. 1). If R has a regular grading there is a natural relation between Given a monomial m(x1,h1 , · · · , xk,hk ) we denote ǫR Definition 5.2. Let f (x1,h1 , . . . , xn,hn ) be a polynomial in F hXH i and write f = the graded identities of A and Ab⊗R (see [2], [4]). P λimi as a linear combination of monomials m1, . . . , mn. We denote f ∗ the poli- nomial f ∗ =P ǫR Over a field of characteristic zero we have the following result. m the scalar in Remark 4.6. miλimi. Proposition 5.3. [2, Lemma 27] Let F be a field of characteristic zero, A be an H-graded algebra and R an algebra with a regular H-grading. If f is a multilinear polynomial then f is a graded identity for A if and only if f ∗ is a graded identity for Ab⊗R. precisely if H is finite and bR = b⊗ IdH ((Ab⊗R)b⊗bR) = IdH (A).This is proved in [2, Theorem 6]. In this case we obtain the following theorem. H−1 Remark 5.4. The "envelope operation" in Definition 5.1 is involutive. More R then bR has a regular H-grading and Theorem 5.5. Let F be a field of characteristic zero, A be a graded algebra graded by a finite abelian group H and R an algebra with a regular H-grading. The algebra A has the primeness property for H-graded central polynomial if and only if Ab⊗R has the primeness property. Proof. Let f (x1,h1, . . . , xr,hr ), g(xr+1,hr+1, . . . , xs,hs ) be multilinear polynomials in distinct variables. It is clear that (f · g)∗ = f ∗ · g∗. Note also that f 7→ f ∗ is an invertible linear map in the subspace of multilinear polynomials in a fixed set of variables. Hence every multilinear polynomial is of the form h∗ for some multilinear (f · g)∗ = f ∗ · g∗ we conclude that f · g is central for A. This implies that f, g are to consider multilinear polynomials only, hence suffices to prove that if f ∗ · g∗ is central then f ∗, g∗ are central. The previous proposition implies that f is central h. We assume that A has the primeness property. To prove that Ab⊗R we may for A if and only if f ∗ is central for Ab⊗R. Since f ∗ · g∗ is central for Ab⊗R and central and therefore f ∗, g∗ are central. Hence Ab⊗R has the primeness property. Now assume that Ab⊗R has the primeness property, by Remark 5.4 the H- envelope operation is involutive. Hence we conclude by the above implication that A has the primeness property. (cid:3) Remark 5.6. It follows from Theorem 3.10 that M2(F ) with its canonical elemen- tary Z2-grading does not satisfy the primeness property, hence by the above result M1,1(E) with its canonical Z2-grading does not satisfy the property. We may verify this directly: x2 2,1 are not graded central polynomials. 2,1 is a graded central polynomial, however x2 1,1 · x2 1,1, x2 We conclude this section by proving that Ma,b(E) satisfies the primeness property for ordinary central polynomials. Lemma 5.7. Let A = A0 ⊕ A1 denote the algebra Ma,b(E) with its canonical Z2- grading and let f (x1, . . . , xr) be a polynomial such that for some g(xr+1, . . . , xs), in distinct variables than f , the product f · g is a central polynomial for Ma,b(E). Then there exists an i in Z2 such that f (a1, . . . , ar) lies in Ai for every a1, . . . , ar in Ma,b(E). 16 SILVA AND BEZERRA Proof Let B denote the canonical basis of A. We prove first that there exists multihomogeneous polynomials f ′(x1, . . . , xm) in hf iT and g′(x1, . . . , xm) in hgiT and elements b1, . . . , bm in B such that f ′(b1, . . . , bm)·g′(b1, . . . , bm) 6= 0. Otherwise the polynomials of the form f ′ · g′ satisfy assertion (iii) of Proposition 2.3 with V = 0. Since these polynomials generate hf · giT as a vector space Proposition 2.3 implies that f · g is an identity for A, which is a contradiction. The elements f ′(b1, . . . , bm) and g′(b1, . . . , bm) have the form mM and rR, where M and R are matrices in Mn(F ) and m and r are monomials in E. Proposition 2.5 implies that the non-zero product (mM )(rR) lies in Z(A). Hence M R is a non-zero scalar matrix and therefore R is an invertible matrix. Let i be the element of Z2 such that g′(b1, . . . , bm) = rR lies in Ai. For a multihomogeneous polynomial f ′′(xi1 , . . . , xis ) in hf iT and bi1 , . . . , bis in B we write f ′′(bi1 , . . . , bis) = qQ, where q is a monomial in E and Q a matrix in Mn(F ). If qQ = 0 then it lies in Ai. Now we suppose that qQ 6= 0, we may assume without loss of generality that qr 6= 0. Since R is invertible and Q 6= 0 the product QR is not 0 and therefore (qQ)(rR) 6= 0. Proposition 2.5 implies that f ′′(bi1 , . . . , bis) · g′(b1, . . . , bm) is a central polynomial for A. Therefore (qQ)(rR) lies in Z(A) ⊆ A0. Note that qQ lies in Aj for some j in Z2. Hence (qQ)(rR) also lies in Aj+i. Since (qQ)(rR) 6= 0 this implies that j + i = 0. Therefore qQ lies in Ai. Proposition 2.3 with V = Ai yields the result. (cid:3) Theorem 5.8. Let F be an infinite field of characteristic different from 2. The F -algebra Ma,b(E) has the primeness property for central polynomials. Proof Let f and g be polynomials in distinct variables such that f · g is a central polynomial for A = Ma,b(E). It follows from the previous proposition that there ex- ists an i in Z2 such that f (a1, . . . , ar) lies in Ai for every a1, . . . , ar in Ma,b(E). The previous proposition and Corollary 2.4 imply that for every b in A the commutator [f (a1, . . . , ar), b] also lies in Ai. If b lies in A1 then the products f (a1, . . . , ar) · b and b · f (a1, . . . , ar) lie on Ai+1. Hence the commutator [f (a1, . . . , ar), b] lies on Ai+1. We conclude that [f (a1, . . . , ar), b] lies on Ai ∩ Ai+1, hence [f (a1, . . . , ar), b] = 0. To prove that f (a1, . . . , ar) commutes with every element in A0 it suffices to prove that it commutes with elements b = ei1 · · · ei2k Eii, for k > 0. We choose j such that b1 = ei1 Eij and b2 = e2 · · · ei2k Eji lie in A1. Since b = b1 · b2 we conclude that f (a1, . . . , ar) commutes with b = ei1 · · · ei2k Eii. Therefore the element f (a1, . . . , ar) is central. The proof that g is a central polynomial is analogous. (cid:3) Remark 5.9. Over a field of characteristic zero the only verbally prime algebras are Mn(F ), Mn(E) and Ma,b(E) (see [13]) and we conclude that every verbally prime algebra satisfies the primeness property for central polynomials. References [1] E. Aljadeff, Y. Karasik, Crossed products and their central polynomials, J. Pure Appl. Algebra 217 (2013), no. 9, 1634 -- 1641. [2] E. Aljadeff, D. Ofir, On regular G-gradings, Trans. Amer. Math. Soc. 367 (2015), no. 6, 4207 -- 4233. [3] S. Amitsur, The T-ideals of the free ring, J. London Math. Soc. 30 (1955), 470 -- 475. [4] Yu. Bahturin, V. Drensky, Graded polynomial identities of matrices, Linear Algebra and its Applications 357 (2002), 15 -- 34. [5] Yu. Bahturin, A. Regev, Graded tensor products, J. Pure Appl. Algebra 213 (2009), 1643 -- 1650. PRIMENESS PROPERTY FOR CENTRAL POLYNOMIALS OF VERBALLY PRIME ALGEBRAS17 [6] Yu. Bahturin, S.K. Sehgal, M.V. Zaicev, Finite-dimensional simple graded algebras, Sb. Math. 199 (2008), no. 7, 965 -- 983. [7] A. Brandão, Graded central polynomials for the algebra Mn(K), Rend. Circ. Mat. Palermo 57 (2008), no. 2, 265 -- 278. [8] A. Brandão, P. Koshlukov, A. Krasilnikov, E. da Silva, The central polynomials for the Grassmann algebra, Israel J. of Math. 179 (2010), 127 -- 144. [9] D. Diniz, A primeness property for central polynomials of verbally prime P.I. algebras, Linear and Multilinear Algebra 63 (2015), no. 11, 2151 -- 2158. [10] A. Elduque, M. Kochetov, Gradings on Simple Lie Algebras, AMS Math. Surv. Monographs 189 (2013), 336p. [11] E. Formanek, Central polynomials for matrix rings, J. Algebra 23 (1972), 129 -- 133. [12] A. R. Kemer, Varieties and Z2-graded algebras, Izv. Akad. Nauk SSSR Ser. Mat. 48 (1984), no. 5, 1042 -- 1059. [13] A. R. Kemer, On the multilinear components of the regular prime varieties, Methods in ring theory (Levico Terme, 1997), 171 -- 183, Lecture Notes in Pure and Appl. Math., 198, Dekker, New York, 1998. [14] Yu. P. Razmyslov, On a problem of Kaplansky (Russian), Izv. Akad. Nauk SSSR Ser. Mat. 37 (1973), 483 -- 501. Translation: Math. USSR, Izv. 7 (1973), 479 -- 496. [15] Yu. P. Razmyslov, Trace identities and central polynomials in the matrix superalgebra Mk,l, Math. USSR-Sb. 56 1 (1987), 187 -- 206. [16] A. Regev, A primeness property for central polynomials, Pacific J. Math. 83 (1979), 269 -- 271. [17] A. Regev, T. Seeman, Z2-graded tensor product of p.i. algebras. J. Algebra. 291 (2005), no. 1, 274 -- 296. [18] A. Giambruno, M. Zaicev, Polynomial Identities and Asymptotic Methods, Mathematical Surveys and Monographs, Volume 122, American Mathematical Society, (2005). Departamento de Matemática, UAME/CCT-UFCG, Avenida Aprígio Veloso 882, 58109-970 Campina Grande-PB, Brazil E-mail address: [email protected] Departamento de Matemática, UAME/CCT-UFCG, Avenida Aprígio Veloso 882, 58109-970 Campina Grande-PB, Brazil E-mail address: [email protected]
1705.03840	3	1705	2018-08-10T15:10:58	Torsion pairs over $n$-Hereditary rings	[ "math.RA", "math.CT" ]	We study the notions of $n$-hereditary rings and its connection to the classes of finitely $n$-presented modules, FP$_n$-injective modules, FP$_n$-flat modules and $n$-coherent rings. We give characterizations of $n$-hereditary rings in terms of quotients of injective modules and submodules of flat modules, and a characterization of $n$-coherent using an injective cogenerator of the category of modules. We show two torsion pairs with respect to the FP$_n$-injective modules and the FP$_n$-flat modules over $n$-hereditary rings. We also provide an example of a B\'ezout ring which is 2-hereditary, but not 1-hereditary, such that the torsion pairs over this ring are not trivial.	math.RA	math	TORSION PAIRS OVER n-HEREDITARY RINGS DANIEL BRAVO AND CARLOS E. PARRA Abstract. We study the notions of n-hereditary rings and its connection to the classes of finitely n-presented modules, FPn-injective modules, FPn-flat modules and n-coherent rings. We give characterizations of n-hereditary rings in terms of quotients of injective modules and submodules of flat modules, and a characterization of n-coherent using an injective cogenerator of the category of modules. We show two torsion pairs with respect to the FPn-injective modules and the FPn-flat modules over n-hereditary rings. We also provide an example of a B´ezout ring which is 2-hereditary, but not 1-hereditary, such that the torsion pairs over this ring are not trivial. Contents Introduction 1. Finitely n-presented modules 2. n-Hereditary rings and n-coherent rings 3. Relative homological algebra over n-hereditary rings 4. Torsion pairs and n-hereditary rings Appendix A. Modules over the ring Z ⊕ Li≥1 Z/2Z Appendix B. A categorical result applied to finitely n-presented modules References 1 3 4 7 11 13 15 17 Introduction The notion of torsion pair was introduced in the sixties by S. Dickson in the setting of abelian categories, generalizing the classical notion of torsion pairs for abelian groups; see [Dic66]. Since then, the theory of torsion pairs has been greatly developed and many applications have been given to areas such as representation theory of Artin algebras, homological algebra, non commutative localization theory, and tilting theory to mention a few; see [AS05], [BH09], [CT95], [Col99], [GT06], [HRS96], [Ste75], . The importance of torsion pairs is highlighted by the theorem of Popescu and Gabriel [Ste75, X, §4] which reduces the theory of Grothendieck categories to the study of categories of modules of quotients by a hereditary torsion Date: October 16, 2018. 2010 Mathematics Subject Classification. Primary 18E40; Secondary 16E60. Key words and phrases. n-hereditary, n-coherent, finitely n-presented, FPn-injective, FPn- flat, torsion pairs, B´ezout ring, injective cogenerator. The first author was partially supported by CONICYT + FONDECYT/Regular + 1180888. The second author was partially supported by CONICYT + FONDECYT/Iniciaci´on + 11160078. 1 2 DANIEL BRAVO AND CARLOS E. PARRA pairs. All this have made the theory of torsion pairs a valuable toolkit and an active research area on its own; see [BP16], [CGM07], [Hrb16], [PS15]. Recently the classes of FPn-injective modules and FPn-flat modules have been studied in detail, generalized to chain complexes and applications have been given to cotorsion pairs, duality pairs, and model categories; see [BGH14], [BP17], [ZP17]. In particular, some of those results showed that over certain generalization of coherent rings, namely n-coherent rings, the cotorsion pairs are well behaved. In this sense, it seems natural to investigate whether these classes of modules also fit in the theory of torsion pairs. Alternatively, one could ask if there are any conditions required on the ring such that any such torsion pair exist. Following the known facts that the classes of FPn-injective and FPn-flat modules are closed under products, summands and extensions, when n > 1, it remains as the main obstacle for these classes of modules to form part of a torsion pair, to be also closed by either quotients or submodules (indeed, we already know these classes are closed under pure submodules and pure quotients). A classical result from H. Cartan and S. Eilenberg shows that over hereditary rings the class of injective modules is closed under quotients; see [CE99]. C. Meg- giben observed a slightly more general result, namely, that over semi-herditary rings the class of FP-injective modules is also closed under quotients; see [Meg70]. In this way, we are motivated to investigate such closure properties for FPn-injective modules over a generalized version of semi-hereditary rings. Thus, we introduce the concept of n-hereditary rings to reach this goal and to investigate how far the classes of FPn-injective modules and FPn-flat modules are from being part of a torsion pair. This motivation naturally leads to investigate any relevant properties of these rings and its connections to these classes of modules. In fact, we show that over n-hereditary rings, the class of FPn-injective modules is the torsion class of a torsion pair, and that the class of FPn-flat modules is the torsion-free class of a torsion pair. We are also able to provide an example of 2-hereditary ring, that is not 1-hereditary (or semi-hereditary), which shows that the torsion pairs in question are in fact non trivial. This article is organized as follows. Section 1 describes the class of finitely n- presented modules, the basic object of our study; collect some of its properties and describe a relevant property for the sections to follow that also doesn't seem to be available previously in the literature. Section 2 introduces the notion of n- hereditary rings and investigate its relation with the class of the finitely n-presented modules and n-coherent rings. The notion of FPn-injective and FPn-flat modules is recalled in Section 3, where the key relations of these two classes of modules with n-hereditary rings are also investigated. We also show new a characterization of n-coherent rings in terms of closure properties of a certain collection of injective modules and the class of FPn-injective modules. These results are used in Sec- tion 4 to establish the two torsion pairs over n-hereditary rings, and its connection with (co)tilting classes; as an application of these results, we get from these tor- sion pairs a far from obvious results about the class of FPn-flat modules. Finally, two appendixes sections are added; in the first one, we show that the mentioned 2-hereditary ring example is a 2-coherent, B´ezout ring, but not 1-hereditary, and the necessary properties which give that the torsion pairs from the previous section TORSION PAIRS OVER n-HEREDITARY RINGS 3 are not trivial. The second appendix is about a property of Grothendieck cate- gories that when applied to the setting of modules, gives the key result for the new characterization of n-coherent rings. Throughout this paper, R denotes an associative ring with unit, R-Mod denotes category of left R-modules, and unless otherwise noted, the expression R-module mean left R-module. 1. Finitely n-presented modules Let n ≥ 0 be an integer. An R-module M is said to be finitely n-presented, if there is an exact sequence Fn → Fn−1 → · · · → F1 → F0 → M → 0, where the modules Fi are finitely generated and free (or projective) modules, for every 0 ≤ i ≤ n. Denote by F P n the class of all finitely n-presented modules. Thus F P 0 is the class of finitely generated modules, and F P 1 is the class of finitely presented modules. For convenience, we let F P −1 be the whole class of R-modules. Also we consider the class, F P∞, of the finitely ∞-presented modules, formed by modules that posses a resolution by finitely generated free (or projective) modules. Note that the class F P ∞ is not empty, since any finitely generated projective module is finitely ∞-presented. We immediately observe the following descending chain of inclusions: F P 0 ⊇ F P 1 ⊇ · · · ⊇ F P n ⊇ F P n+1 ⊇ · · · ⊇ F P ∞. (1.1) We include two examples of rings to show how the chain (1.1) behaves; for more details about these two examples we refer the reader to [BP17]. Example 1. Let k be a field and R be the following polynomial ring: R := k[x1, x2, x3, . . .]/(xixj )i,j≥1. In this ring every finitely 2-presented module is finitely generated free. Also we can quickly check that R/(x1) ∈ F P 1 \ F P 2 and that (x1) ∈ F P 0 \ F P 1. Thus we have that the chain of inclusions in (1.1) collapses at 2: F P 0 ⊃ F P 1 ⊃ F P 2 = F P n = F P∞. The next example shows that (1.1) may never collapse. Example 2. Let k be a field and consider the following ring: R := k[. . . , x3, x2, x1, y1, y2, y3, . . .]/(xj+1xj , x1y1, y1yi)i,j≥1 Then (y1) ∈ F P 0 \ F P 1, and (xi) ∈ F P i \ F P i+1 for i ≥ 1. Hence in this case, the chain in (1.1) is strict at every level. Several results about F P n and F P∞ are collected in [BP17]. We include here the following results. Proposition 3. Let n ≥ 0. F P n is closed under cokernels of monomorphisms, extensions, and direct summands. The following results also appear in [Bro82], and [Bie76]. Theorem 4. The following conditions are equivalent for every left R-module M and every n ≥ 0: (1) M ∈ F P n+1. 4 DANIEL BRAVO AND CARLOS E. PARRA (2) Exti (3) TorR R(M, −) commutes with direct limits for all 0 ≤ i ≤ n. i (−, M ) commutes with direct products for all 0 ≤ i ≤ n. The class F P∞ has all the properties from the previous proposition and one more, as indicated in the following result. Theorem 5. F P∞ is closed under kernels of epimorphisms. Remark 6. For any finitely generated module M , we have that Card(M ) ≤ max{ℵ0, Card(R)}. Hence we can choose Sn, a set of representatives of finitely generated modules in F P n, such that every module in F P n is isomorphic to a module in Sn. In the next two sections, we will use the class of finitely n-presented modules to describe two types of rings. 2. n-Hereditary rings and n-coherent rings As defined in [CE99], recall that a ring is said to be left hereditary if every left ideal is a projective module. This is also equivalent to saying that every submodule of a projective left module is also a projective module, or that every quotient (ho- momorphic image) of an injective left module is injective. A bit more general are left semi-hereditary rings; that is, rings such that every finitely generated left ideal is projective. This is equivalent to saying that every finitely generated submodule of a projective left module is also a projective module, or that every quotient (ho- momorphic image) of a FP-injective left module is FP-injective [Meg70, Theorem 2]. From these observations we define the following: Definition 7. A ring is said to be left n-hereditary if every finitely (n−1)-presented submodule of a finitely generated projective left module is also a projective module. This way a left 1-hereditary ring is the same as a left semi-hereditary ring, and if we allow for finitely (−1)-presented modules to be any module, then left 0-hereditary rings coincide with left hereditary. From now on, and unless otherwise noted, the term n-hereditary ring will mean left n-hereditary ring. A characterization of n-hereditary rings can be given in terms of the class of finitely n-presented modules, pd(M ), the projective dimension of an R-module M , wd(M ), the weak dimension of the R-module M . Lemma 8. Let n ≥ 2. The following statements are equivalent. (1) R is a n-hereditary ring. (2) pd(M ) ≤ 1 for all M ∈ F P n. (3) wd(M ) ≤ 1 for all M ∈ F P n. Furthermore if n = 1, then we have the following statement: R is 1-hereditary if and only if pd(M ) ≤ 1 for all M ∈ F P 1. Proof. (1) =⇒ (2): Suppose R is an n-herditary ring and let M ∈ F P n. Consider the following short exact sequence: 0 → ΩM → Rk → M → 0. Since ΩM ∈ F P n−1 and Rk is a finitely generated and projective module, then ΩM is also projective. Thus pd(M ) ≤ 1. (2) =⇒ (3): This is clear, since wd(M ) ≤ pd(M ) for any M ∈ R-Mod. TORSION PAIRS OVER n-HEREDITARY RINGS 5 (3) =⇒ (1): Suppose that N ∈ F P n−1 is a submodule of a finitely generated projective module P . Then P/N ∈ F P n, and so wd(P/N ) ≤ 1; hence N is flat. But n ≥ 2, so N is at least in F P 1. Thus, [EJ11, Proposition 3.2.12] gives that N is projective, showing that R is n-hereditary. Finally, for the last statement, observe that if I is a finitely generated ideal, then R/I ∈ F P 1 and so I is projective; thus R is semi-hereditary, or 1-hereditary. For the converse, note that the proof of (1) =⇒ (2) also applies. (cid:3) Using this last characterization, we give an example of a (commutative) 2- hereditary ring, that is not 1-hereditary. Example 9. Let R = Z ⊕ Li≥1 Z/2Z with addition defined component wise, and multiplication given by (m, a) · (n, b) = (mn, mb + na + ab) where m, n ∈ Z, a, b ∈ Li≥1 Z/2Z and m · a = (ma1, ma2, ma3, . . .). From [Vas76, Example 1.3(b)], we have that gl.wd(R) ≤ 1; here gl.wd(R) corre- sponds to the weak global dimension of the ring R. So in particular, wd(M ) ≤ 1, for any M ∈ F P n and any n ≥ 2. Thus by Lemma 8 we have that R is 2-hereditary. In Proposition (38) of the Appendix A section, we show that this ring is not a semi-hereditary ring, or 1-hereditary. Another immediate observation is that any ring such that gl.wd(R) ≤ 1 is 2- hereditary. The same proof as in the previous example works, and indeed we note that it is also n-hereditary for all n ≥ 2, but this is always the case as we observe next. As a consequence of Definition 7 and the chain (1.1), if R is n-hereditary, then R is also k-hereditary for all k ≥ n. Hence, if n-Her denotes the collection of all n-hereditary rings, then we get the following chain: 0-Her ⊂ 1-Her ⊂ 2-Her ⊂ · · · ⊂ n-Her ⊂ · · · ⊂ ∞-Her , (2.1) where ∞-Her is the corresponding definition using the class F P∞. We have given examples of rings that are 0-hereditary, 1-hereditary and 2- hereditary. Next we show how to get non-commutative n-hereditary rings from a given n-hereditary ring. Example 10. Let n be an integer such that n ≥ 1. Then R is an n-hereditary ring if and only if Mn(R) is an n-hereditary ring. Indeed, the key idea is to observe that there is an equivalence of between the module category of R-mod and Mn(R)-mod (see [Lam99, Theorem 17.20]) and that every such equivalence preserve homological properties such as projective modules and exact sequences, and thus finitely n- presented modules. We also give an example which shows that there are rings that are not ∞-Her. But to do that we need the following result. Corollary 11. Let n ≥ 0. If R is n-hereditary ring then every ideal I ∈ F P n−1 is projective. Proof. The cases n = 0 and n = 1 are known and can be found in [CE99] and [Meg70] respectively; recall that we are allowing F P −1 = R-Mod. 6 DANIEL BRAVO AND CARLOS E. PARRA Now let n > 1 and suppose R is n-hereditary. Then for any ideal I ∈ F P n−1, we have that R/I ∈ F P n. Applying Lemma 8 gives us that pd(R/I) ≤ 1, which implies that I is projective. (cid:3) Example 12. Let R = Z/4Z. Then the ideal I = 2Z/4Z of R is in F P∞ as the ·2−→ I −→ 0 shows. However, the ideal I is not resolution · · · → Z/4Z projective, and thus by Corollary 11, we see that R is not n-Her for any n ≥ 0. ·2−→ Z/4Z We still would like to find an explicit example of a ring that is ∞-Her , but not n-Her for any n ≥ 0. Corollary 11 can be thought as a first step in a characterization of n-hereditary rings in terms of its ideals. Since as noted in the begining of this section the reciprocal of this corollary works for n = 0, 1, thus it is natural to ask the following. Question 13. Is the reciprocal of Corollary 11 true? In other words, if every ideal I ∈ F P n−1 is projective, then is it true that R is n-hereditary? Also related to the idea of finitely n-presented modules, is the notion of n- coherent rings, which generalizes that of coherent rings. Recall that a ring R is said to be left coherent if every finitely generated left ideal of R also is finitely presented. Another equivalent definition for coherent ring is as follows: R is a left coherent ring if, and only if, every module in F P 1 is also in F P 2. Definition 14. A ring R is left n-coherent if F P n ⊆ F P n+1. Similarly, from now on, and unless otherwise noted, the term n-coherent ring will refer to left n-coherent ring. Thus coherent rings are just 1-coherent rings, and 0-coherent rings coincide with Noetherian rings. The rings in Example 1 and Example 9 are a 2-coherent ring, however a proof of the latter is given in Proposition 39. The following is an example of an n-coherent ring, for any n ≥ 1. Example 15. Let S = k[[∂1, ∂2, . . . , ∂n]] be the power series over a collection of n variables, and consider the S-module M = k[x1, x2, . . . , xn] with a linearly extended S-action given by ∂ixj = δij. Consider now the ring R = S ⋉ M given by the trivial extension of the ring S by the S-module M , defined over the set R = {(s, m) : s ∈ S and m ∈ M } and with product given by (s, m) · (s′, m′) = (ss′, sm′ + s′m). Then the ring R = k[[∂1, ∂2, . . . , ∂n]] ⋉ k[x1, x2, . . . , xn] is n-coherent. This is a concrete example of a more general result of J. Roos [Roo82, Theorem A]. Remark 16. Note that if R is n-coherent, then it is also k-coherent, for all k ≥ n. Thus, if n-Coh denotes the class of all n-coherent rings, and if by convention, we allow any ring to be ∞-coherent, then we obtain the following chain: 0-Coh ⊂ 1-Coh ⊂ 2-Coh ⊂ · · · ⊂ n-Coh ⊂ · · · ⊂ ∞-Coh. (2.2) Unlike the situation for n-hereditary rings, we observe that any ring can be thought as an ∞-coherent ring. Furthermore, there are rings that are never n- coherent for any n ≥ 0; see [BP17, Example 1.4]. The following theorem states equivalent conditions for the n-coherence of a ring in terms of finitely n-presented modules. Theorem 17 ([BP17, Theorem 2.4 and Corollary 2.6]). Let R be a ring and n ≥ 0. The following are equivalent. TORSION PAIRS OVER n-HEREDITARY RINGS 7 (1) R is n-coherent. (2) F P n is closed under kernels of epimorphisms. (3) F P n = F P∞. To end this section we establish a connection between n-hereditary rings and n-coherent rings. Corollary 18. Let n ≥ 1. If R is n-hereditary, then it is n-coherent. Proof. Suppose M ∈ F P n. Then from Lemma 8 we have that pd(M ) ≤ 1. Hence in the short exact sequence 0 → ΩM → F → M → 0, with F finitely generated and free, we get that the syzygy ΩM is also finitely generated and projective. Therefore ΩM ∈ F P∞, giving us that M ∈ F P ∞. (cid:3) The case n = 0, would say that any hereditary ring is Noetherian, but this is not the case. Consider, R = khx, yi, the polynomial ring over a field k in two nonconmuting variables; this is an (right and left) hereditary ring, but not Noetherian; see [Rot08, Example 4.12]. 3. Relative homological algebra over n-hereditary rings Having mentioned the class of finitely n-presented module, we discuss the relative homological algebra with respect to F P n and define the corresponding relative injective modules and relative flat modules. Definition 19. Let R be a ring, M ∈ R-Mod and n ≥ 0 (including the case n = ∞). (1) We say that M is FPn-injective if Ext1 R(F, M ) = 0 for all F ∈ F P n. We denote by F P n-Inj the class of all FPn-injective modules. (2) We say that M is FPn-flat if TorR 1 (F, M ) = 0 for all F ∈ F P n. We denote by F P n-Flat the class of all FPn-flat modules. With these definitions, M is injective if, and only if, M is FP0-injective, and M is FP-injective (as introduce by [Ste70]) if, and only if, M is F P1-injective. The usual flat modules coincide with the FP0-flat modules. Given that any module is the direct limit of finitely 1-presented modules, and that the functor Tor1(−, M ) commutes with direct limits, then we have that the FP1-flat modules also coincide with the usual flat modules. From the descending chain of inclusions (1.1), we get the following ascending chains of inclusions: F P 0-Inj ⊆ F P 1-Inj ⊆ · · · ⊆ F P n-Inj ⊆ · · · ⊆ F P∞-Inj (3.1) and F P 0-Flat = F P 1-Flat ⊆ · · · ⊆ F P n-Flat ⊆ · · · ⊆ F P∞-Flat. (3.2) For the rest of this article and motivated by these last chains of inclusion, we focus on the case when n > 1. The following two result appear in [BP17] and list several properties about F P n-Inj and F P n-Flat Proposition 20. Let n > 1. The classes F P n-Inj and F P n-Flat are closed under: 8 DANIEL BRAVO AND CARLOS E. PARRA (1) Direct summands and extensions. (2) Direct products and direct limits. (3) Pure submodules and pure quotients. Given a left R-module M , recall that the character module is defined as the right R-module M + = HomZ(M, Q/Z). Similarly, the character module of a right R-module M is defined in the same way, and it is a left R-module that will be also denoted by M +. The classes F P n-Inj and F P n-Flat relate well through the character modules. Proposition 21. Let n > 1. (1) M ∈ F P n-Flat if and only if M + ∈ F P n-Inj. (2) M + ∈ F P n-Flat if and only if M ∈ F P n-Inj. We provide a result regarding lifting properties of F P n-Inj. Proposition 22. Let n ≥ 0. M ∈ F P n-Inj if and only if for every diagram with P ′ ∈ F P n−1 and P finitely generated projective module, there is a homomorphism P h−→ M such that hg = f . P ′ g P f h M Proof. Suppose that M has this lifting property. Let F ∈ F P n, and consider the short exact sequence 0 → P ′ → P → F → 0, with P finitely generated free and P ′ ∈ F P n−1. Then applying HomR(−, M ) to this short exact sequence gives that Ext1(F, M ) = 0, since Hom(P, M ) → Hom(P ′, M ) is an epimorphism. Hence M ∈ F P n-Inj. Conversely the argument works similarlly, since for M ∈ F P n-Inj we have that Ext1(P/P ′, M ) = 0, given that P/P ′ ∈ F P n, and so Hom(P, M ) → Hom(P ′, M ) is an epimorphism. (cid:3) The next definition has recently been introduced by M. P´erez and T. Zhao in their study of syzygies and further generalizations to chain complexes; see [ZP17]. Nev- ertheless we recall these definitions here and investigate its relation to n-hereditary rings. Definition 23. Let M ∈ R-Mod and n ≥ 0. The FPn-injective dimension of M , which will be denote by FPn-id(M ), is given by the smallest integer k ≥ 0 such that Extk+1 R (F, M ) = 0 for every F ∈ F P n. Similarly, the FPn-flat dimension of M , which will be denote it by FPn-fd(n), is given by the smallest integer k ≥ 0 such that TorR k+1(F, M ) = 0, for every F ∈ F P n. With this definition at hand and motivated by the work of H. Cartan and S. Eilenberg [CE99] and C. Megibben [Meg70], we obtain a similar result regarding when F P n-Inj is closed under quotients and also investigate when F P n-Flat is closed under subobjects. Theorem 24. Let n ≥ 1. Then the following are equivalent. (1) R is an n-hereditary ring. (2) Quotients of FPn-injective modules are FPn-injective. TORSION PAIRS OVER n-HEREDITARY RINGS 9 (3) Quotients of injective modules are FPn-injective. (4) Submodules of FPn-flat modules are FPn-flat. (5) Submodules of flat modules are FPn-flat. (6) FPn-fd(M ) ≤ 1, for every M ∈ R-Mod. (7) FPn-id(M ) ≤ 1, for every M ∈ R-Mod. Furthermore, for n = 0 we have that (1) ⇐⇒ (2) ⇐⇒ (3) ⇐⇒ (7) and that (4) ⇐⇒ (5) ⇐⇒ (6). Proof. (1) =⇒ (2). Suppose M ∈ F P n-Inj and consider a short exact sequence with M in the middle: For each F ∈ F P n, we consider the following exact sequence 0 → K → M → M ′ → 0. · · · → Ext1 R(F, M ) → Ext1 R(F, M ′) → Ext2 R(F, K) → · · · . If R is n-hereditary, then pd(F ) ≤ 1, and therefore Ext2 M ∈ F P n-Inj, then Ext1 F ∈ F P n. R(F, M ) = 0. This gives us that Ext1 R(F, K) = 0. Also since R(F, M ′) = 0 for any (2) =⇒ (3). This implication is easy since every injective module is in F P n-Inj. (3) =⇒ (7). Let M ∈ R-Mod and denote its injective envelope by E(M ). Then, the short exact sequence 0 → M → E(M ) → E(M )/N → 0 gives us that Ext2 F P n-Inj. R(F, M ) = 0, for all F ∈ F P n, since by hypothesis E(M )/M ∈ (7) =⇒ (1). Suppose that N ∈ F P n−1 is a submodule of a finitely generated projective module P . Hence from the short exact sequence 0 → N → P → P/N → 0, we have that P/N ∈ F P n. Now, to this short exact sequence, apply the functor HomR(−, M ), for any M ∈ R-Mod, and obtain the following exact sequence in R-Mod: · · · → Ext1 R(P, M ) = 0 → Ext1 R(N, M ) → Ext2 R(P/N, M ) → · · · . By hypothesis, we have that FPn-id(M ) ≤ 1, hence Ext2 P is a projective module, we get that Ext1 projective module also. R(P/N, M ) = 0. Since R(N, M ) = 0 and therefore that N is a (4) =⇒ (5). Since every flat module is in F P n-Flat, we get this immediately. (5) =⇒ (6). Let M ∈ R-Mod, and consider a short exact sequence 0 → K → M h∈HomR(R,M) Rh → M → 0, where Rh = R. By hypothesis, we have that K ∈ F P n-Flat, and so from this sequence we obtain that TorR 2 (F, M ) = 0, for every F ∈ F P n. (6) =⇒ (4). Suppose that we have an exact sequence 0 → N → M → M/N → 0 with M ∈ F P n-Flat. By assumption, we know that TorR F ∈ F P n, and so from this short exact sequence we have TorR F ∈ F P n. 2 (F, M/N ) = 0, for every 1 (F, N ) = 0, for every For the rest of the proof, we suppose that n > 1. 10 DANIEL BRAVO AND CARLOS E. PARRA (4) =⇒ (2). Suppose that we have an exact sequence B → C → 0 with B ∈ F P n-Inj. Then we get an exact sequence 0 → C+ → B+, and Proposition 21 tells us that B+ ∈ F P n-Flat. From our hypothesis we get C+ ∈ F P n-Flat and therefore by Proposition 21 again, we obtain that C ∈ F P n-Inj. (2) =⇒ (4). Dual to the proof of (4) =⇒ (2). (cid:3) We would like to point out a new characterization of n-coherent rings in terms of the class F P n-Inj. Indeed, several such characterization already are given in [BP17, Theorem 5.5]. The following result also appears in that same article, but not precisely in the following format. Corollary 25 ([BP17, Lemma 5.2]). R is n-coherent if and only if Ext1 for all M ∈ F P n and all N ∈ F P n+1-Inj. R(M, N ) = 0 This characterization, given in terms of the class F P n-Inj, makes us wonder if perhaps a similar characterization of n-coherent rings can be given just in terms of injective modules. Indeed, this is what we do next, however first we introduce some notation. Let I be any injective cogenerator in R-Mod (for example I = HomZ(R, Q/Z), which can be thought as a left R-module, as explain in the paragraph after Propo- sition 21) and define the class I as follows: I =   Y h∈HomR(M,I) Ih : where Ih = I and M ∈ R-Mod  . Given a class of modules A, we denote by lim A the direct limit closure in R-Mod −→ of the class A. We denote by Ω−i(A) the class of all the i-th cozysygies of injective corresolutions from objects in A. Theorem 26. Let R be a ring, n > 0 and I as describe above. Then the following are equivalent. (1) R is n-coherent. (2) Extn (3) Ω−n+1(lim −→ R(M, X) = 0 for all M ∈ F P n and all X ∈ lim −→ I. Im(Ψ)) ⊆ F P n-Inj. A key result for the proof of this theorem is given in general in the Appendix B section and we refer to it in the proof given below. Proof. (1) ⇐⇒ (2). Let M ∈ F P n. From Corollary 43 we get that M ∈ F P n+1 if and only if lim R(M, −)). Since in this case Im (Ψ) = I, then −→ F P n ⊆ F P n+1 if and only if Extn R(M, X) = 0, for all M ∈ F P n and X ∈ lim −→ Im(Ψ) ⊆ Ker(Extn (2) ⇐⇒ (3). This follows from dimension shifting. I. (cid:3) A quick application of this theorem gives the following result. Corollary 27. If the injective dimension of the class lim −→ R is n + 1-coherent. Im(Ψ) is at most n, then TORSION PAIRS OVER n-HEREDITARY RINGS 11 4. Torsion pairs and n-hereditary rings As an application, we see that over n-hereditary ring the classes F P n-Inj and F P n-Flat define torsion classes and torsion-free classes respectively; this allows us to introduce new torsion pairs. Our approach to torsion pairs is that of B. Stenstrom [Ste75], and so is the general terminology used in the section. Definition 28. A torsion pair of a (co)complete and locally small abelian category A, is a pair (T , F ) of classes of A such that: (1) HomA(T, F ) = 0 for all T ∈ T and F ∈ F . (2) If HomA(C, F ) = 0 for all F ∈ F , then C ∈ T . (3) If HomA(T, C) = 0 for all T ∈ T , then C ∈ F . In this case T is called a torsion class and F is called a torsion-free class. The pair (T , F ) is called hereditary if T is closed under subobjects. Given a class C of object in A, we define C⊥ = {X ∈ A : HomA(C, X) = 0 for all C ∈ C} and similarly define ⊥C = {X ∈ A : HomA(X, C) = 0 for all C ∈ C}. This way, for any class C of A, the pair (⊥(C⊥), C⊥) is a torsion pair. Proposition 29. [Ste75, Theorem VI.2.1 and Proposition VI.2.2] Let T and F be classes of a (co)complete, locally small abelian category A. (1) T is a torsion class for some torsion pair if and only if T is closed under quotients, coproducts and extensions. (2) F is a torsion-free class for some torsion pair if and only if F is closed under subobjects, products and extensions. Let n > 1, and note that the classes F P n-Inj and F P n-Flat in R-Mod are closed under direct sums and extensions. Given that R is n-hereditary if and only if F P n-Inj is closed under quotients or F P n-Flat is closed under submodules (see Theorem 24), then in combination with Proposition 29 we get the following result. Theorem 30. Let n > 1. The following statements are equivalent: (1) R is an n-hereditary ring. (2) The pair (F P n-Inj, F P n-Inj⊥) is a torsion pair. (3) The pair (⊥F P n-Flat, F P n-Flat) is a torsion pair. Next we give several definitions available in the literature regarding tilting and cotilting modules (see [GT06], [AHHK07]) Let T ∈ R-Mod. We will say that T is a 1-tilting R-module if the following assertions hold: (1) T has projective dimension less or equal than 1. (2) Exti (3) There exists an exact sequence 0 → R → T0 → T1 → 0 such that Ti is R(T, T (I)) = 0, for each integer i ≥ 1 and all sets I. isomorphic to a direct summands of copies of T , for each i = 0, 1. If T is 1-tilting R-module, then the pair (Ker(cid:0)Ext1 R(T, −)(cid:1), Ker (HomR(T, −))) is a torsion pair in R-Mod which is called the 1-tilting torsion pair associated to T , and the class Ker(cid:0)Ext1 R(T, −)(cid:1) is called the 1-tilting class associated to T . Dually, C is a 1-cotilting R-module if it satisfies the following conditions: 12 DANIEL BRAVO AND CARLOS E. PARRA R(C I , C) = 0, for each integer i ≥ 1 and all sets I. (1) C has injective dimension less or equal than 1. (2) Exti (3) There exists an injective cogenerator Q of R-Mod and there exists an exact sequence 0 → C1 → C0 → Q → 0 such that Ci is isomorphic to a direct summands of a direct products of copies of C, for each i = 0, 1. Similarly, the pair (Ker (HomR(−, C)), Ker(cid:0)Ext1 R(−, C))(cid:1) is a torsion pair in R-Mod, called the 1-cotilting torsion pair associated to C whenever C is 1-cotilting R-module. The class Ker(cid:0)Ext1 R(−, C)(cid:1) is called the 1-cotilting class associated to C. We say that C is a 1-tilting (respectively 1-cotilting) class if there is some tilting (repectively cotilting) module M such that C = Ker(cid:0)Ext1 R(M, −)(cid:1) (respectively C = Ker(cid:0)Ext1 The next theorem follows as an application of [AHHT06, Theorem 2.2]. For completion we record a weak version of [AHHT06, Theorem 2.2] as the following proposition. R(−, M )(cid:1)). Proposition 31. Every resolving subclass C of finitely generated modules of pro- jective dimension at most 1 gives rise to a 1-tilting class of R-modules by assigning C to Ker(cid:0)Ext1 R(C, −)(cid:1). Here we say that a class of modules is resolving if it is closed under extensions, direct summands, kernels of epimorphisms in that class and contains the finitely generated projective modules. For example, the class of F P∞ is resolving. Theorem 32. Let n > 1. The following statements are equivalent: (1) R an n-hereditary ring. (2) F P n-Inj is a 1-tilting class. (3) F P n-Flat is a 1-cotilting class. Proof. (1) =⇒ (2). Since n > 1 and R is n-hereditary, then by Corollary 18 we have that R is n-coherent. Now, from Proposition 31 and Theorem 17 we get that Ker(cid:0)Ext1 R(F P ∞, −)(cid:1) = F P∞-Inj = F P n-Inj is a 1-tilting class. (2) =⇒ (3). We use that F P n-Inj is a 1-tiltitng class and apply [GT06, Theorem 8.1.2] to get that F P n-Flat is a 1-cotilting class. (3) =⇒ (1). If F P n-Flat is a 1-cotilting class, then it is a torsion-free class, and so it is closed under submodules. Hence by Theorem 24, the ring R is n- hereditary. (cid:3) We note that for the ring from Example 9 the torsion pairs from Theorem 30 are not trivial. This follows from the following result. Proposition 33. Let R = Z ⊕ Li≥1 Z/2Z as described in Example 9. Then we have that F P 2-Inj ( R-Mod and F P 2-Flat ( R-Mod. Proof. Consider an odd integer m 6= 1 and the ideal (m, a)R. Since (m2, a) = (m, a)(m, a), then we have the following short exact sequence: 0 → (m2, a)R → (m, a)R → C → 0, where C is the cokernel of the inclusion map i : (m2, a)R → (m, a)R. We note that this short exact sequence doesn't split. In fact, if there is a map q : (m, a)R → TORSION PAIRS OVER n-HEREDITARY RINGS 13 (m2, a)R such that qi = id, then we have that (m2, a) = q((m, a)(m, a)) = (m, a)q((m, a)) = (m3n, b), with n some integer; this can't be. From Lemma 41 in the Appendix A, we see that (m, a)R and (m2, a)R are finitely generated projective modules, and so they are in F P ∞, thus making C ∈ F P∞. Since R is 2-hereditary, it is also 2-coherent and so F P ∞ = F P 2. Thus we have that C ∈ F P 2 and that Ext1 R(C, (m2, a)R) 6= 0. Hence the R-module (m2, a)R 6∈ F P 2-Inj, giving us the first statement. The duality between F P n-Inj and F P n-Flat, gives the last statement. (cid:3) Furthermore if the ring is commutative, then the work of Hrbek [Hrb16] allows us to say a few more results about the torsion pair associated to F P n-Flat. Corollary 34. Let R be an n-hereditary and commutative ring with n > 1. Then we have that the torsion pair (⊥F P n-Flat, F P n-Flat) is an hereditary 1-cotilting torsion pair. Proof. This follows immediately from [Hrb16, Proposition 3.11]. (cid:3) Remark 35. As a consequence of this last result we have a far from obvious statement about the class F P n-Flat. Namely, that for n > 1 and over an n- hereditary ring the class F P n-Flat is closed under injective envelopes (see [Ste75, Proposition VI.3.2]) Corollary 34 also allows us to show that the torsion pair associated to F P n-Flat is a tCG torsion pair; for the definition of tCG-torsion pairs see [BP16]. Corollary 36. Let R be an n-hereditary and commutative ring with n > 1. Then we have that the torsion pair (⊥F P n-Flat, F P n-Flat) is a tCG torsion pair. Proof. From Corollary 34 we have that the torsion pair is hereditary. Since the class of F P n-Flat is closed under direct limits, then from [BP16, Corollary 3.8] we have the result. (cid:3) Appendix A. Modules over the ring Z ⊕ Li≥1 Z/2Z In this section we show some properties about the ring R = Z⊕Li≥1 Z/2Z, from Example 9. In particular we show that it is not a semi-hereditary (1-hereditary), 2-coherent ring, that is also B´ezout and therefore an arithmetical ring. We also include a property about the projectivity of some of its principal ideals which are of importance for the torsion pairs of the previous section. For notational purposes we let A = Li≥1 Z/2Z, thus R = Z ⊕ A. Recall that addition is defined component wise, and for m, n ∈ Z, and for a, b ∈ A multiplication is given by (m, a) · (n, b) = (mn, mb + na + ab), where ma = (ma1, ma2, ma3, . . .), and ab = (a1b1, a2, b2, a3, b3 . . .). Given (m, a) ∈ R, we define the support of a as supp(a) = {i ∈ Z : ai 6= 0}. We begin by observing that this ring is the standard unitification of A, when viewed as a ring without a unit (see [AF92, §1.1 Exercise 1]). Next, we have the following technical lemma. 14 DANIEL BRAVO AND CARLOS E. PARRA Lemma 37. For any a ∈ A, the ideal I = (2m, a)R ∈ F P 0 \ F P 1. Proof. Clearly I is finitely generated, hence I ∈ F P 0. Now consider the map R f −→ (2m, a)R given by f ((1, 0)) = (2m, a). A quick computation shows that: Ker (f ) = 0 ⊕ (Z/2Z)(N\supp(a)) , where (Z/2Z)(N\supp(a)) is the direct sum of Z/2Z in the positions given by N \ supp(a) and 0 otherwise. Note that (0, b)R = 0 ⊕ (Z/2Z)(supp(b)). Let us suppose that Ker (f ) is finitely generated, then we get that: Ker (f ) = (0, a1)R + · · · + (0, an)R ⊆ 0 ⊕ (Z/2Z)(∪ supp(ai)) . This is a contradiction, since Ker (f ) = 0 ⊕ (Z/2Z)(N\supp(a)). Hence Ker (f ) is not finitely generated. (cid:3) Proposition 38. R is not 1-hereditary. Proof. Lemma 37 shows a family of ideals I ∈ F P 0 \ F P 1; that is finitely generated ideals, that are not finitely presented. Hence R is not coherent, or 1-coherent, and the result follows from from Corollary 18. (cid:3) Regarding the observation about the non-coherency of this ring we say the fol- lowing. Proposition 39. R is a 2-coherent ring. Proof. This follows directly from Example 9 and Corollary 18. (cid:3) From [Vas76, Example 1.3(b)] we know that the localizations of R at its prime ideals are valuation domains. Thus from [Fai99, Section 6.4] we have that R is an arithmetical ring (that is, a ring such that the localization at all of its maximal ideal are valuation rings). However, we can say more. Indeed, we next show that R is a B´ezout ring, that is a ring such that all its finitely generated ideals are principal [Fai99, Section 5.4B]. Proposition 40. R is a B´ezout ring. Proof. Let us consider the following short exact sequence: 0 → M i≥1 Z/2Z → R π−→ Z → 0 where π(m, a) = m. Now let I = h(n1, a1), · · · , (nk, ak)i, be a finitely generated ideal of R. Since π(I) is an ideal of Z, we know is a principal ideal and indeed π(I) = (n1, . . . , nk) = (d). We now split the proof in two cases depending if either d is even or odd. Suppose that d is even. We consider the set S = Sn j=1 supp(aj) and define a = Pi∈S ei, where ei is the element of A with zero everywhere except in i-th position where it has a 1. Then we claim that I = h(d, a)i. To see this, let mi := ni/d and check that. (1) If mi is even, then (ni, ai) = (d, a)(mi, ai). (2) If mi is odd, then (ni, ai) = (d, a)(mi, a + ai). TORSION PAIRS OVER n-HEREDITARY RINGS 15 Hence I ⊂ h(d, a)i. For the other containment we just need a suitable combination of the generators giving (d, a), and this is not hard to obtain. This concludes the even case. Now, suppose that d is odd. We consider the set J = {a ∈ A : (d, a) ∈ I} and note that if a, b ∈ J, then ab ∈ J. Indeed, if (d, a), (d, b) ∈ I, then (d, ab) = (d, a) − (d, b) + (d, a)(1, b) ∈ I. Since supp(ab) ⊆ supp(a), then there is c ∈ J such that supp(c) ⊆ supp(a), for all a ∈ J. That is ca = c for all a ∈ J. We, now claim that I = h(d, c)i. From the definition we have that (d, c) ∈ I, which gives the first inclusion. To see the other inclusion, let mi := ni/d and check the following. (1) If mi is even, then (d, ai + c) = (ni, ai) − (d, c)(mi − 1, c) ∈ I. This means that ai + c ∈ J and thus cai = 0. Now, (ni, ai) = (ni, ai + cai) = (d, c)(mi, ai) (2) If mi is odd, then (d, ai) = (ni, ai) − (d, c)(mi − 1, c) ∈ I. This means that ai ∈ J and thus cai = c, that is, cai + c = 0. Now, (ni, ai) = (ni, ai + cai + c) = (d, c)(mi, ai) (cid:3) We add a result about the projectivity of certain principal ideals. Lemma 41. For any a ∈ A and m any odd integer, the ideal (m, a)R is projective. f −→ (m, a)R, which sends (1, 0) 7→ (m, a). Now Proof. Consider the epimorphism R g consider the homomorphism (m, a)R −→ R given by (m, a) 7→ (1, a). This is a well defined map since the equation (m, a)(n, b) = 0 implies that (1, a)(n, b) = 0, when m is odd. Hence any zero divisor of (m, a) is also a zero divisor of (1, a). Now, since (m, a)(1, a) = (m, a), we quickly check that the composition f g(x) = x for any x ∈ (m, a)R, giving us a splitting of R. (cid:3) Appendix B. A categorical result applied to finitely n-presented modules In this appendix we show a result in the general setting of Grothendieck cate- gories, which when applied to the setting of R-modules provides a characterization of finitely n-presented modules and therefore of n-coherent rings; namely Theorem 26. Although the proof can be done directly for R-modules, we prefer to show it in this generality. We recall some concepts of these type of categories; however, for a general treatment of Grothendieck categories, we refer the readers to [Ste75]. A Grothendieck category, G, is a cocomplete abelian category with a generator and the direct limits are exact. It is well-known that every Grothendieck category has a injective cogenerator, that is, an object I that is injective and such that the functor HomG(−, I) is faithful. Given that every Grothendieck category has products for every family of objects, we get that the faithful condition on the functor HomG(−, I) is equivalent to the condition that every object of G is isomorphic to a subobject of a product of I. 16 DANIEL BRAVO AND CARLOS E. PARRA Given any Grothendieck category G and an injective cogenerator I of G we consider the functor Ψ : G → G given by M 7→ I HomG(M,I) := Y h∈HomG (M,I) Ih, with Ih = I. Indeed, this assignment is functorial since if f : M → N is a morphism in G, then Ψ(f ) : I HomG(M,I) → I HomG (N,I) is given by the universal property of the product in G. That is, Ψ(f ) is the unique morphism in G such that πN h◦f , h : I HomG (N,I) → I is the h-projection morphism, where h ∈ HomG(N, G), where πN and similarly for πM h◦f . Also observe that the functor Ψ comes with a natural transformation ι : idG → Ψ which is monomorphic. h ◦ Ψ(f ) = πM We will denote by lim −→ of a direct system in Im (Ψ). Im (Ψ) the class the objects of G which are a direct limit Theorem 42. Let G be a Grothendieck category and let M be an object in G. Consider the functor Ψ described in the previous paragraph and let n > 1. Then, the functors Exti G(M, −) : G → Ab preserve direct limits, for each i = 0, . . . , n − 1, if and only if, the functors Exti G(M, −) : G → Ab preserve direct limits, for each i = 0, . . . , n − 2 and lim −→ Im (Ψ) ⊆ Ker(cid:0)Extn−1 (M, −)(cid:1). G G G Extn−1 (M, lim −→ (M, Nλ) → Extn−1 Proof. Let (Nλ)λ be a direct system in Im (Ψ) and suppose that the canonical Nλ) is an isomorphism. Since each morphism lim −→ Nλ is an injective object of G and that n−1 > 0, then we get that Extn−1 (M, Nλ) = 0. Therefore the previous isomorphism gives that Extn−1 Nλ) = 0; that is lim −→ Nλ ∈ Ker(cid:0)Extn−1 For the converse, we only need to check that the functor Extn−1 (M, −) : G → Ab preserves direct limits. To see this, we consider (Mλ)λ∈Λ, a direct system in G. Note that from the natural transformation ι of the mentioned before, we get a direct system of short exact sequences in G of the form: (M, −)(cid:1). (M, lim −→ G G G G 0 → Mλ ιM λ−−→ Ψ(Mλ) → Coker (ιMλ ) → 0. Since G is a Grothendieck category, we obtain the following exact sequence in G 0 → lim −→ Mλ λ lim−→ ιM −−−−−→ lim −→ Ψ(Mλ) → lim −→ Coker (ιMλ ) → 0. Now apply the functor HomG(M, −) : G → Ab to this last exact sequence and obtain the following commutative diagram with exact rows and where (X, Y )i G denotes Exti G(X, Y ). lim −→ (M, Ψ(Mλ))n−2 G lim −→ (M, Coker (cid:0)ιMλ )(cid:1)n−2 G lim −→ (M, Mλ)n−1 G lim −→ (M, Ψ(Mλ))n−1 G f1 f2 f3 (M, lim −→ Ψ(Mλ))n−2 G (M, lim −→ Coker (cid:0)ιMλ (cid:1)n−2 G (M, lim −→ Mλ)n−1 G (M, lim −→ Ψ(Mλ))n−1 G Note that both terms on the right of each row are 0, since Ψ(Mλ) is an injective object of G and n > 1, and also by hypothesis Extn−1 Ψ(Mλ)) = 0. Now, since f1 and f2 are isomorphisms, then we obtain from the Snake Lemma that f3 is also an isomorphism, thus completing the proof. (M, lim −→ G TORSION PAIRS OVER n-HEREDITARY RINGS 17 (cid:3) We apply of this last result to G = R-Mod. Corollary 43. Let M ∈ R-Mod. Then M ∈ F P n, if and only if, M ∈ F P n−1 and lim −→ Im (Ψ) ⊆ Ker(cid:0)Extn−1 G (M, −)(cid:1). Proof. This follows directly from the characterization of Theorem 4 and Theorem 42. (cid:3) References [AF92] Frank W. Anderson and Kent R. Fuller. Rings and categories of modules, volume 13 of Graduate Texts in Mathematics. Springer-Verlag, New York, second edition, 1992. [AHHK07] Lidia Angeleri Hugel, Dieter Happel, and Henning Krause, editors. Handbook of tilting theory, volume 332 of London Mathematical Society Lecture Note Series. Cambridge University Press, Cambridge, 2007. [AHHT06] Lidia Angeleri Hugel, Dolors Herbera, and Jan Trlifaj. Tilting modules and Gorenstein [AS05] rings. Forum Math., 18(2):211 -- 229, 2006. Ibrahim Assem and Manuel Saor´ın. Abelian exact subcategories closed under prede- cessors. Comm. Algebra, 33(4):1205 -- 1216, 2005. [BGH14] D. Bravo, J. Gillespie, and M. Hovey. The stable module category of a general ring. [BH09] [Bie76] [BP16] [BP17] [Bro82] [CE99] ArXiv e-prints, May 2014. Silvana Bazzoni and Dolors Herbera. Cotorsion pairs generated by modules of bounded projective dimension. Israel J. Math., 174:119 -- 160, 2009. Robert Bieri. Homological dimension of discrete groups. Mathematics Department, Queen Mary College, London, 1976. Queen Mary College Mathematics Notes. D. Bravo and C. E. Parra. tCG Torsion Pairs. ArXiv e-prints, October 2016. Daniel Bravo and Marco A. P´erez. Finiteness conditions and cotorsion pairs. J. Pure Appl. Algebra, 221(6):1249 -- 1267, 2017. K. S. Brown. Cohomology of Groups. Graduate Texts in Mathematics. Springer, 1982. Henri Cartan and Samuel Eilenberg. Homological algebra. Princeton Landmarks in Mathematics. Princeton University Press, Princeton, NJ, 1999. With an appendix by David A. Buchsbaum, Reprint of the 1956 original. [CGM07] Riccardo Colpi, Enrico Gregorio, and Francesca Mantese. On the heart of a faithful [Col99] [CT95] [Dic66] [EJ11] [Fai99] [GT06] torsion theory. J. Algebra, 307(2):841 -- 863, 2007. Riccardo Colpi. Tilting in Grothendieck categories. Forum Math., 11(6):735 -- 759, 1999. Riccardo Colpi and Jan Trlifaj. Tilting modules and tilting torsion theories. J. Algebra, 178(2):614 -- 634, 1995. Spencer E. Dickson. A torsion theory for Abelian categories. Trans. Amer. Math. Soc., 121:223 -- 235, 1966. Edgar E. Enochs and Overtoun M. G. Jenda. Relative homological algebra. Volume 1, volume 30 of De Gruyter Expositions in Mathematics. Walter de Gruyter GmbH & Co. KG, Berlin, extended edition, 2011. Carl Faith. Rings and things and a fine array of twentieth century associative algebra, volume 65 of Mathematical Surveys and Monographs. American Mathematical Society, Providence, RI, 1999. R Gobel and J. Trlifaj. Approximations and Endomorphism Algebras of Modules. De Gruyter Expositions in Mathematics. W. De Gruyter, 2006. [Hrb16] Michal Hrbek. One-tilting classes and modules over commutative rings. J. Algebra, 462:1 -- 22, 2016. [HRS96] Dieter Happel, Idun Reiten, and Sverre O. Smalø. Tilting in abelian categories and [Lam99] [Meg70] [PS15] quasitilted algebras. Mem. Amer. Math. Soc., 120(575):viii+ 88, 1996. T. Y. Lam. Lectures on Modules and Rings. Graduate Texts in Mathematics. Springer New York, 1999. Charles Megibben. Absolutely pure modules. Proc. Amer. Math. Soc., 26:561 -- 566, 1970. Carlos E. Parra and Manuel Saor´ın. Direct limits in the heart of a t-structure: the case of a torsion pair. J. Pure Appl. Algebra, 219(9):4117 -- 4143, 2015. 18 [Roo82] [Rot08] [Ste70] [Ste75] DANIEL BRAVO AND CARLOS E. PARRA Jan-Erik Roos. Finiteness conditions in commutative algebra and solution of a prob- lem of Vasconcelos. In Commutative algebra: Durham 1981 (Durham, 1981), vol- ume 72 of London Math. Soc. Lecture Note Ser., pages 179 -- 203. Cambridge Univ. Press, Cambridge-New York, 1982. J. J. Rotman. An Introduction to Homological Algebra. Universitext. Springer New York, 2008. B. Stenstrom. Coherent rings and fp-injective modules. J.London Math.Soc., 2(2):323 -- 329, 1970. Bo Stenstrom. Rings of quotients. Springer-Verlag, New York-Heidelberg, 1975. Die Grundlehren der Mathematischen Wissenschaften, Band 217, An introduction to methods of ring theory. [Vas76] Wolmer V. Vasconcelos. The rings of dimension two. Marcel Dekker, Inc., New York- [ZP17] Basel, 1976. Lecture Notes in Pure and Applied Mathematics, Vol. 22. T. Zhao and M. A. P´erez. Relative FP-injective and FP-flat complexes and their model structures. ArXiv e-prints, March 2017. Instituto de Ciencias F´ısicas y Matem´aticas, Facultad de Ciencias, Universidad Aus- tral de Chile, Valdivia, Chile E-mail address: [email protected] E-mail address: [email protected]
1201.5071	2	1201	2013-10-23T04:35:45	Leibniz Algebras and Lie Algebras	[ "math.RA" ]	This paper concerns the algebraic structure of finite-dimensional complex Leibniz algebras. In particular, we introduce left central and symmetric Leibniz algebras, and study the poset of Lie subalgebras using an associative bilinear pairing taking values in the Leibniz kernel.	math.RA	math	SIGMA 9 (2013), 063, 10 pages Symmetry, Integrability and Geometry: Methods and Applications Leibniz Algebras and Lie Algebras(cid:63) Geoffrey MASON † and Gaywalee YAMSKULNA ‡§ † Department of Mathematics, University of California, Santa Cruz, CA 95064, USA E-mail: [email protected] ‡ Department of Mathematical Sciences, Il linois State University, Normal, IL 61790, USA E-mail: [email protected] § Institute of Science, Walailak University, Nakon Si Thammarat, Thailand Received September 09, 2013, in f inal form October 19, 2013; Published online October 23, 2013 http://dx.doi.org/10.3842/SIGMA.2013.063 Abstract. This paper concerns the algebraic structure of f inite-dimensional complex Leib- niz algebras. In particular, we introduce left central and symmetric Leibniz algebras, and study the poset of Lie subalgebras using an associative bilinear pairing taking values in the Leibniz kernel. Key words: Leibniz algebras; Lie algebras 2010 Mathematics Subject Classification: 17A32 Introduction 1 Throughout the present paper, a left Leibniz algebra means a nonassociative C-algebra M with a product (or bracket ) [ ] satisfying the following identity for all a, b, c ∈ M : [a[bc]] = [[ab]c] + [b[ac]]. (1) (1) says that the left adjoint ada : b (cid:55)→ [ab] (a, b ∈ M ) is a derivation of M , so that ad : M → Der(M ), a (cid:55)→ ada , is a morphism of M to the algebra of derivations of M regarded as a Lie (or left Leibniz) algebra with the usual bracket. The Leibniz kernel is the subspace C (M ) spanned by [aa] (a ∈ M ); M is a Lie algebra if, and only if, C (M ) = 0. Leibniz algebras were f irst studied for their own sake by Loday [5] (see also [6, Section 10.6]). The rationale for the present work is partially motivated by the triangular decomposition V = (⊕n≤0Vn ) ⊕ V1 ⊕ (⊕n≥2Vn ) (2) of a vertex operator algebra (VOA) V . In the most widely studied case when V is of CFT-type, i.e. Vn = 0 for n < 0 and V0 = C1 is spanned by the vacuum vector, the summands in (2) are Lie algebras. In the general case (2) satisf ies only the weaker condition of being a decomposition into Z-graded left Leibniz algebras with respect to the 0th operation in V (cf. [4]). This decomposition plays a role in our work [7] where, among other things, we are interested in the Lie subalgebras of V1 and the vertex subalgebras of V that they generate. This leads directly to the main theme of the present paper, which is the study of the poset of Lie subalgebras of certain kinds of Leibniz algebras. Readers uninured to VOA theory need not be concerned – the present paper deals solely with Leibniz algebras, and no further mention of VOAs will be made. The def inition of Leibniz algebra is not left-right symmetric; a right Leibniz algebra, in which the map b (cid:55)→ [ba] for f ixed a is a derivation, is not necessarily a left Leibniz algebra. We call (cid:63)This paper is a contribution to the Special Issue on New Directions in Lie Theory. The full collection is available at http://www.emis.de/journals/SIGMA/LieTheory2014.html 2 G. Mason and G. Yamskulna an algebra that is both a left and right Leibniz algebra a symmetric Leibniz algebra. Notice from (1) that a left Leibniz algebra M satisf ies a, b ∈ M , [[aa]b] = 0, (3) and dually, a right Leibniz algebra satisf ies a, b ∈ M . [a[bb]] = 0, (4) We call M a left central Leibniz algebra if it is a left Leibniz algebra that also satisf ies (4). Equivalently, M is both the left and right centralizer of C (M ). There is a hierarchy of algebras {left Leibniz} ⊇ {left central Leibniz} ⊇ {symmetric Leibniz} ⊇ {Lie}, and in fact each containment is strict. We now describe the contents of the present paper, which deals solely with finite-dimensional, complex, left Leibniz algebras M . We tacitly assume this in everything that follows. In Section 2 we discuss some basic facts, in particular concerning Levi subalgebras and Levi decompositions of Leibniz algebras. Levi subalgebras (i.e., semisimple Lie subalgebras that complement the solvable radical) are readily seen to exist in M (see [3] and Section 2 below). Malcev’s theorem for Lie algebras does not extend to left Leibniz algebras, though it does for various special classes, including left central Leibniz algebras. Section 3 is devoted to these issues. The remainder of the paper revolves about the symmetric bilinear pairing ψ : M × M → C (M ), (a, b) (cid:55)→ [ab] + [ba]. This is a feature of all Leibniz algebras, but for left central Leibniz algebras it is particularly ef f icacious, because in this case it is associative. Then the radical R of ψ is a 2-sided ideal of M , and the poset L of Lie subalgebras of M coincides with the set of Leibniz subalgebras which are also isotropic subspaces. General properties of this set-up are developed in Section 4, including (Proposition 1) the fact that R coincides with the intersection of the maximal elements of L. We also give (Lemma 7) a general construction of a class of left central Leibniz algebras based solely on Lie-theoretic data. In Sections 5 and 6 we consider left central Leibniz algebras with dim C (M ) = 1 (so that ψ is a trace form in the usual sense). Such algebras arise, for example, as quotients M /U whenever U is a hyperplane of C (M ). We prove (Theorem 3) that there is a Lie subalgebra L ⊆ M that is also a maximal isotropic subspace of M (and therefore also a maximal element of L) and such that L/R is nilpotent. In Section 6 we assume that M is also symmetric, a property that holds precisely when M (cid:48) ⊆ R (Lemma 8). We prove (Theorem 4) that a symmetric Leibniz algebra with dim C (M ) = 1 has a 2-sided ideal of codimension at most 1 that arises from the construction of Lemma 7. In this way, we more-or-less obtain a characterization of such Leibniz algebras in terms of Lie-theoretic data. 2 Basic facts In this section, M is a left Leibniz algebra. For subsets A, B ⊆ M , we def ine [AB ] (or [A, B ]) to be the subspace spanned by all brackets [ab] (a ∈ A, b ∈ B ). If A = {a} is a singleton, we write [{a}B ] = [a, B ]. Introduce Z (M ) := {z ∈ M [M , z ] = [z , M ] = 0}, M (cid:48) := [M M ], C (M ) := span(cid:104)[aa] a ∈ M (cid:105) = span(cid:104)[ab] + [ba] a, b ∈ M (cid:105). Z (M ), M (cid:48) and C (M ) are 2-sided ideals of M called the center, derived subalgebra and Leibniz kernel of M respectively. (The second equality def ining C (M ) is equivalent to the f irst by Leibniz Algebras and Lie Algebras 3 polarization. That C (M ) is a 2-sided ideal then follows from (3) and (1) with b = c. See also [1, 2].) Obviously, M is a Lie algebra if, and only if, C (M ) = 0. Set M (0) := M and for n ≥ 0 inductively def ine M (n+1) := (M (n) )(cid:48) ; M is called solvable if M (n) = 0 for some n ≥ 0. M has a unique maximal solvable ideal, called the solvable radical of M and denoted by B (M ). M is called abelian if M (cid:48) = 0. By (3), M is the right centralizer of C (M ), i.e., [C (M )M ] = 0. (5) Consequently, C (M ) is a 2-sided abelian ideal of M and C (M ) ⊆ B (M ). Because C (M /C (M )) = 0, M /C (M ) is a Lie algebra and B (M )/C (M ) is its solvable radical. M is a left central Leibniz algebra if, and only if, C (M ) ⊆ Z (M ). It is not hard to see that the analog of the Levi decomposition for Lie algebras holds in M (cf. [3]). Indeed, set N = C (M ), B = B (M ), and let N ⊆ T ⊆ M with T /N a Levi subalgebra of the Lie algebra M /N . We have a Levi decomposition in M /N , M /N = B/N ⊕ T /N . Because T (cid:48) ⊆ T and N acts trivially on the left of M , T is a left T /N -module. Because T /N is semisimple, Weyl’s theorem of complete reducibility tells us that there is a left T -submodule that complement N in T , call it S . Therefore, S (cid:48) ⊆ [T S ] ⊆ S , so that S is a Leibniz subalgebra of M . Moreover, C (S ) ⊆ S ∩ N = 0, whence S is, in fact, a Lie subalgebra of M . It is clearly isomorphic to T /N , hence is semisimple. We have a direct sum decomposition M = B ⊕ S. (6) We call a Lie subalgebra S of M that complements B as in (6) a Levi factor, or Levi subalge- bra, of M ; (6) itself is called a Levi decomposition of M . The Levi subalgebras of M can be characterized as the Lie subalgebras of M of maximal dimension sub ject to being semisimple. 3 On Malcev’s theorem for Leibniz algebras Malcev’s theorem for f inite-dimensional complex Lie algebras includes the statements that all Levi subalgebras are conjugate by the exponential of an inner derivation, and every semisimple Lie subalgebra is contained in a Levi subalgebra. We shall see that both of these assertions are generally false for Leibniz algebras. On the other hand, versions of the Malcev theorem can be proved for certain classes of Leibniz algebras. This section is concerned with these questions. We begin with a general construction. Let S be a f inite-dimensional complex Lie algebra and N a f inite-dimensional left S -module with action S × N → N , (a, m) (cid:55)→ a.m (a ∈ S , m ∈ N ). Let M = S ⊕ N with bracket [(a, m)(b, n)] := ([ab], a.n). (7) Here, of course, [ab] denotes the bracket in S . (7) def ines the structure of a Leibniz algebra on M ; S is a Lie subalgebra of M , and C (M ) = [SN ]. Now suppose in addition that S is semisimple. Then S is a Levi subalgebra of M and N is the solvable radical. Suppose that S1 is any Levi subalgebra of M . Then M = S1 ⊕ N and S1 = {(x, c(x)) x ∈ S } for some linear map c : S → N . Moreover [(x, c(x))(y , c(y))] = ([xy ], x.c(y)) = ([xy ], c([xy ])), so that x.c(y) = c([xy ]). This says that c is a morphism of left S -modules, where S furnishes the left adjoint S -module. Conversely, given c ∈ HomS (S, N ), the set of pairs {(x, c(x)) x ∈ S } is a Levi subalgebra of M . 4 G. Mason and G. Yamskulna Suppose that S is simple and N a simple S -module that is not the adjoint module. Then HomS (S, N ) = 0, so that S is the unique Levi subalgebra of M . In the case of Lie algebras, if there is a unique Levi subalgebra then it is an ideal. However, as our construction shows, this is not necessarily true for Leibniz algebras. Suppose that T ⊆ S is a simple Lie subalgebra. Replacing S by T in the previous paragraphs, we see that if d ∈ HomT (T , N ) then T1 := {(x, d(x) x ∈ T } is a Lie subalgebra of M isomorphic to T . In order for T1 to be contained in a Levi subalgebra, it is necessary and suf f icient that d T (c) for some c ∈ HomS (S, N ). is the restriction d = ResS Example 1. S = sl3 , V the natural 3-dimensional module, N = S 2 (V ) the second symmetric square, T the sl2 -subalgebra of S corresponding to a simple root. Then HomS (S, N ) = 0 and HomT (T , N ) (cid:54)= 0. Thus there exists T1 not contained in the unique Levi subalgebra S . The theory of highest weight modules can be used to construct many similar examples. Example 2. Suppose that S is simple and there is an isomorphism of left S -modules S ∼= N . Then HomS (S, N ) (cid:54)= 0, and there are at least two distinct Levi subalgebras in M . On the other hand, because [N M ] = 0, every inner derivation of M leaves every Levi subalgebra invariant, so that conjugacy a la Malcev does not hold in M . (This is essentially Example 2 in [3].) Now we turn to some positive results in the direction of Malcev’s theorem. Theorem 1. Let M be a left Leibniz algebra with Leibniz kernel N and solvable radical B . Let S be a Levi subalgebra of M , and suppose that [SN ] = 0. If T ⊆ M is any semisimple Lie subalgebra, there is an automorphism α = exp(ad x) (x ∈ [M , B ]) such that α(T ) ⊆ S . Remark 1. If one Levi subalgebra S satisf ies [SN ] = 0, then al l Levi subalgebras have the same property. Proof of Theorem 1. Let T ⊆ M be any semisimple Lie subalgebra. By Malcev’s theorem for Lie algebras applied to M /N , we can f ind x + N ∈ [M /N , B/N ] so that α = exp(ad (x + N )) and α(T + N/N ) ⊆ S + N/N . Because of (5), we can identify α with exp(ad x), which we also denote by α. Then α(T ) ⊆ S + N . Because [M /N , B/N ] = ([M , B ] + N )/N , we may choose x ∈ [M , B ]. Because of the assumption that [SN ] = 0, S + N is a reductive Lie subalgebra, and S is the unique maximal semisimple subalgebra. Therefore α(T ) ⊆ S , and the theorem is (cid:4) proved. Theorem 2. Let M be a left Leibniz algebra with Leibniz kernel N and solvable radical B . Suppose that [BN ] = 0. Then given any pair of Levi subalgebras S1 and S2 of M , there is a derivation f of M such that exp(f )(S1 ) = S2 . Remark 2. The conditions of the theorem apply in the situation of Example 2, for example. Thus, while we may not have conjugacy of Levi subalgebras using only exponentials of inner derivations, conjugacy is restored in some cases if more general automorphisms are permitted. Proof of Theorem 2. First note that Si + N/N (i = 1, 2) are two Levi subalgebras of the Lie algebra L = M /N . By the Levi–Malcev theorem for M /N , there is a derivation δ ∈ Der(L) such that exp(δ) maps S1 + N/N onto S2 + N/N . Because [N M ] = 0, δ lifts to a derivation of M . So in proving the theorem, we may, and shall, assume that S1 + N = S2 + N . Arguing in the same way as the discussion preceding Example 1, we know that S2 = {(x, c(x)) x ∈ S1} for some c ∈ HomS1 (S1 , N ). We will show that the linear map f : M → M (cid:40) def ined by c(x), x ∈ S1 , x ∈ B , 0, f (x) = Leibniz Algebras and Lie Algebras 5 has the desired properties. f is well-def ined because M = B ⊕ S1 , and since f (M ) ⊆ N ⊆ B then f 2 = 0. To see that f is a derivation, note that for x, y ∈ S1 , a, b ∈ B , we have f ([(x + a)(y + b)]) = f ([xy ]) = c([xy ]), [(x + a)f (y + b)] + [(f (x + a))(y + b)] = [(x + a)c(y)] = c([xy ]), where we used [ac(y)] ∈ [BN ] = 0 to ensure the last equality. Since f 2 = 0, we have exp(f ) = I d+f , whence exp(f )(x) = x+c(x) for x ∈ S1 . Consequently, (cid:4) exp(f ) maps S1 onto S2 . This completes the proof of the theorem. 4 Left central Leibniz algebras In this section, M is a left central Leibniz algebra with Leibniz kernel N = C (M ). Recall that this means that N ⊆ Z (M ), i.e. [M N ] = 0. Note that M satisf ies the hypotheses of both Theorem 1 and Theorem 2, so that the Levi–Malcev theorem holds in M . This suggests that left central Leibniz algebras are more manageable than general Leibniz algebras. We introduce and study a very useful bilinear pairing attached to M . For a, b ∈ M , write ψM (a, b) = ψ(a, b) := [ab] + [ba]. Then ψ : M × M → N is a symmetric bilinear map. Even though ψ takes values in N , it still makes sense to use terminology associated with the more familiar situation of a C-valued bilinear form on M . Thus a ∈ M is isotropic if ψ(a, a) = 0 ([aa] = 0); otherwise, a is anisotropic. a and b are orthogonal if ψ(a, b) = 0. For a subspace U ⊆ M , def ine U ⊥ := {b ∈ M ψ(a, b) = 0, a ∈ U }, rad(U ) := U ∩ U ⊥ . U is total ly isotropic if U ⊆ U ⊥ (this is equivalent to every element of U being isotropic, and also to U = rad U ), and nondegenerate if rad U = 0. U, V ⊆ M are orthogonal if V ⊆ U ⊥ . The radical of ψ (or M ) is R := rad(M ) = M ⊥ = {x ∈ M ψ(x, a) = 0, a ∈ M }. These def initions apply to all Leibniz algebras. ψ is particularly useful in the study of central Leibniz algebras because of the next result. Lemma 1. Suppose that M is a left central Leibniz algebra. Then ψ is associative in the sense that for a, b, c ∈ M we have ψ([ab], c) = ψ(a, [bc]). (8) In particular, R is a 2-sided ideal of M . Proof . The identities ([ab] + [ba])c = a([bc] + [cb]) = 0 hold for all a, b, c ∈ M . Therefore, ψ([ab], c) = [[ab]c] + [c[ab]] = −([[ba]c] + [c[ba]]) = −([b[ac]] − [a[bc]] + [[cb]a] + [b[ca]]) = [a[bc]] + [[bc]a] = ψ(a, [bc]). (cid:4) This proves (8). That R is a 2-sided ideal is an immediate consequence. Because M is a left central Leibniz algebra, we have N ⊆ R. Thus, M /R is a Lie algebra equipped with the symmetric nondegenerate pairing induced by ψ . Introduce the poset (with respect to containment) L := {Lie subalgebras of M }. 6 G. Mason and G. Yamskulna Lemma 2. Let L ⊆ M be a Leibniz subalgebra. Then the fol lowing are equivalent: (a) L ∈ L; (b) L is a total ly isotropic subspace; (c) L + R ∈ L. Proof . The equivalence of (a) and (b) follows directly from the def initions. Next, if L + R is a Lie algebra then L is necessarily a Lie subalgebra, so that (c) ⇒ (a). Conversely, if (a) holds then L is a Lie subalgebra of M . In this case, L + R is itself a Leibniz subalgebra because R is a 2-sided ideal. Moreover, if a ∈ R, b ∈ L, then ψ(a + b, a + b) = ψ(a, a) + 2[bb] + 2ψ(a, b) = 0. Therefore, R + L is totally isotropic, and the equivalence of (a) and (b) applied to R + L shows (cid:4) that (c) holds. This completes the proof of the lemma. Lemma 3. Suppose that U ⊆ M is a total ly isotropic 2-sided ideal of M . Then U ∈ L and U (cid:48) ⊆ R. Proof . Because U is a totally isotropic Leibniz subalgebra of M , Lemma 2 tells us that U ∈ L. Now let a, b ∈ U , x ∈ M . Then [bx] ∈ U , and the associativity of ψ (Lemma 8) together with the fact that U is totally isotropic implies that ψ([ab], x) = ψ(a, [bx]) = 0. Therefore [ab] ∈ R for all a, b ∈ U , i.e. U (cid:48) ⊆ R. This completes the proof of the lemma. (cid:4) Lemma 4. Suppose that R ⊆ L ⊆ M such that L/R ⊆ M /R is a semisimple Lie subalgebra. Then L ∈ L. Proof . L is clearly a Leibniz subalgebra of M . Let S ⊆ L be a Levi subalgebra. Because L/R is semisimple we have L = R + S , and this is a Lie subalgebra thanks to the equivalence of (a) (cid:4) and (c) in Lemma 2. Lemma 5. M /R has no nonzero semisimple ideals. In particular, every minimal ideal of M /R is abelian. R = L. Proof . Suppose that U /R is an ideal of M /R that is also a semisimple Lie algebra. By Lemma 4, U ∈ L, and in particular it is totally isotropic. Then U (cid:48) ⊆ R by Lemma 3. Therefore U /R is both abelian and semisimple, whence it reduces to 0. This proves the f irst assertion of the lemma. Because every minimal ideal of the Lie algebra M /R is either abelian or semisimple, (cid:4) the second assertion follows. This completes the proof of the lemma. (cid:92) Proposition 1. Let L∗ be the set of maximal elements of L. Then L∈L∗ Proof . Let L ∈ L∗ . Then R + L is also a Lie subalgebra by Lemma 2, so that R ⊆ L because L is maximal. This shows that R is contained in every L ∈ L∗ , and hence also in their intersection. To prove the opposite containment, we use induction on dim N . There is nothing to prove if M is a Lie algebra, so we may, and shall, assume that this is not the case. Therefore, N (cid:54)= 0. Suppose that dim N ≥ 2. Then every hyperplane N0 ⊆ N is a 2-sided ideal in M , and is itself (cid:92) contained in every maximal Lie subalgebra L. Then by induction we obtain L∈L∗ L/N0 = R(M /N0 ), Leibniz Algebras and Lie Algebras 7 which says that if a ∈ ∩L then ψ(a, x) ∈ N0 for all x ∈ M . Then ψ(a, x) ∈ ∩N0 = 0 (the last intersection ranging over hyperplanes of N ), and thus a ∈ R, as required This reduces us to the case when dim N = 1, so that we can think of ψ as a C-valued bilinear form. We assume this for the remainder of the proof. Because M is not a Lie algebra, R (cid:54)= M . Suppose that R has codimension 1. Then every nonzero element of M /R is anisotropic, so that if a ∈ M \ R then a cannot be contained in a Lie subalgebra of M . Thus in this case, R is the unique element in L∗ , and the desired result is clear. Finally, suppose that M /R has dimension at least 2. Because ψ is a nondegenerate C-valued bilinear form on M /R, M /R is spanned by isotropic vectors, i.e., elements a + R with [aa] = 0. Such elements a are contained in some maximal Lie subalgebra; therefore, if b ∈ ∩L∈L∗ L, a and b generate a Lie subalgebra of M , so that [ab] + [ba] = 0. Because the isotropic vectors a + R span, we can conclude that [ab] + [ba] = 0 for all a ∈ M , whence b ∈ R. This completes the (cid:4) proof of the proposition. We will need the next lemma in Section 5. Lemma 6. Let R ⊆ B ⊆ M satisfy B/R := B (M /R). Then B⊥ ⊆ B . Proof . Write B⊥ = S0 ⊕ B0 , where S0 and B0 are a Levi factor and the solvable radical respectively for B⊥ . Because B is an ideal in M then so too is B0 . Therefore, B0 ⊆ B ∩ B⊥ . Thus B0 = rad(B ), and in particular B0 is totally isotropic. So we see that there is an orthogonal decomposition B⊥ = S0 ⊥ B0 , and since both summands are totally isotropic then B⊥ is a totally isotropic ideal of M . Now we can apply Lemma 3, with B⊥ playing the role of U , to conclude that S0 ⊆ R. Therefore, B⊥ ⊆ B , and the lemma is proved. (cid:4) We now describe the construction of a class of left central Leibniz algebras that depends only on Lie-theoretic data (L1 , L2 , R, α, π) satisfying (a)–(c) below. The set-up is as follows: (a) a pair of Lie algebras L1 , L2 with a common ideal R and L2/R abelian: 0 → R → L1 → L1/R → 0 (cid:107) 0 → R → L2 → L2/R → 0 Set Z := R ∩ Z (L1 ) ∩ Z (L2 ). (b) a morphism of Lie algebras α : L1 → Der(L2 ) with αR = adR , α(L1 )R = adL1 R . Setting α(x1 )(y1 ) = x1 .y1 , these assumptions amount to [x1x2 ].y1 = x1 .(x2 .y1 ) − x2 .(x1 .y1 ), x1 .[y1y2 ] = [(x1 .y1 )y2 ] + [y1 (x1 .y2 )], (x1 ∈ R or y1 ∈ R), x1 .y1 = [x1y1 ] for x1 , x2 ∈ L1 , y1 , y2 ∈ L2 , and where [ ] is the bracket in L1 or L2 . (c) an injective morphism of left L1 -modules π : L1/R → HomC (L2/R, Z ) such that im π annihilates no nonzero elements of L2/R. (L1 acts on L2/R via α, trivially on Z , and with the induced left action on HomC (L2 , Z )). Lifting π to a morphism of left L1 -modules π : L1 → HomC (L2 , Z ) and setting ψ (cid:48) : L1 × L2 → Z, (x1 , y1 ) (cid:55)→ π(x1 )(y1 ), these assumptions mean that ψ (cid:48) ([x1x2 ], y) = ψ (cid:48) (x1 , x2 .y), R = {x1 ∈ L1 ψ (cid:48) (x1 , L2 ) = 0} = {y1 ∈ L2 ψ (cid:48) (L1 , y1 ) = 0}. 8 G. Mason and G. Yamskulna Def ine M = (L1 ⊕ L2 , [ ]), where [(x1 , y1 )(x2 , y2 )] := ([x1x2 ], x1 .y2 − x2 .y1 + [y1y2 ] + ψ (cid:48) (x2 , y1 )). (9) Thus L1 , L2 are naturally Lie subalgebras of M . One calculates that M is a left Leibniz algebra, L2 is a 2-sided ideal, and Z ⊕ Z ⊆ Z (M ). Moreover, ψ((x1 , y1 ), (x2 , y2 )) = (0, ψ (cid:48) (x2 , y1 ) + ψ (cid:48) (x1 , y2 )) ∈ Z ⊕ Z . So M is a left central Leibniz algebra with radical R ⊕ R. Furthermore, D := {(a, −a) a ∈ R} is a 2-sided ideal. So M := M /D is itself a central Leibniz algebra, and the radical R ⊕ R/D of the induced bilinear form ψ on M is naturally identif ied with R. We state these conclusions as Lemma 7. Suppose that (L1 , L2 , R, α, π) satisfies assumptions (a)–(c). Then M = M (L1 , L2 , R, α, π) is a left central Leibniz algebra with radical R. 5 Left central Leibniz algebras of rank 1 For a Leibniz algebra M , we def ine the rank of M to be the dimension of N = C (M ), and denote it by rk(M ). This is a useful invariant for left central Leibniz algebras, because in this case any subspace N0 ⊆ N is a 2-sided ideal of M and M /N0 is a left central Leibniz algebra with rk(M /N0 ) = dim N − dim N0 . In this way, we can try to reduce questions about left central Leibniz algebras to those of rank 1. These are easier to deal with, because the form ψ may be considered to be a trace form in the usual sense, i.e. a C-valued associative bilinear form. This method was already used in the proof of Proposition 1. If M1 , M2 are two Leibniz algebras, their orthogonal sum M1 ⊥ M2 is the direct sum M1 ⊕M2 with product [(a1 , b1 )(a2 , b2 )] = ([a1a2 ], [b1 b2 ]). Then M1 ⊥ M2 is a Leibniz algebra, M1 and M2 are orthogonal ideals (with respect to ψM1⊥M2 ), and rk(M1 ⊥ M2 ) = rk(M1 ) + rk(M2 ). Thus rk is additive over orthogonal sums. If M has index r ≥ 1, we can f ind r hyperplanes N1 , . . . , Nr of N such that ∩r i=1Ni = 0. Then there is an injective morphism of Leibniz algebras M →⊥r a (cid:55)→ (a + N1 , . . . , a + Nr ). i=1 M /Ni , In this way, any left central Leibniz algebra of rank r ≥ 1 is isomorphic to a subalgebra of an orthogonal sum of r left central Leibniz algebras of rank 1. Theorem 3. Suppose that M is a left central Leibniz algebra of rank 1, and let B be as in Lemma 6. Then there is at least one maximal Lie algebra L ∈ L∗ with the fol lowing properties: (a) L is a maximal isotropic subspace of M ; (b) L/R is nilpotent; (c) L ⊆ B . Proof . If L ∈ L∗ then R ⊆ L (cf. Proposition 1). So part (b) makes sense. To prove the theorem, we use induction on dim M . If every nonzero element of M /R is anisotropic, then R is both the unique maximal Lie subalgebra of M and the unique maximal totally isotropic subspace, in which case all parts of the theorem are obvious. So we may, and shall, assume that M /R contains nonzero isotropic elements. As in the proof of Proposition 1, this implies that dim(M /R) ≥ 2. We assume f irst that there is a (nonzero) ideal U /R in M /R which is totally isotropic. If M /R is solvable, we choose any such ideal U /R; if M /R is nonsolvable we take U = B⊥ . Note that in the second case, B is a proper ideal of M , so that B⊥/R is nonzero thanks to the Leibniz Algebras and Lie Algebras 9 assumption that M has rank 1. Moreover, B⊥/R is totally isotropic in the second case because of Lemma 6. With U /R chosen in this way, U ⊥ is a proper ideal of M , U ⊥ ⊆ B , and U = rad(U ⊥ ). By induction there is a Lie subalgebra L ⊆ U ⊥ which is a maximal isotropic subspace of U ⊥ and such that L/U is nilpotent. If L1 is a maximal isotropic subspace of M that contains L, then L1 ⊆ U ⊥ , and this implies that L = L1 since L is maximal isotropic in U ⊥ . So L is maximal isotropic in M . It remains to show that L/R is nilpotent. Let a ∈ M , u ∈ U, x ∈ L. Then ψ(a, [ux]) = ψ([au], x) = 0, which shows that [LU ] ⊆ R. Another way to say this is U /R ⊆ Z (L/R). Since we already know that L/U is nilpotent, we can conclude that L/R is too. This completes the proof of the theorem in this case. Thus we may, and now shall, suppose that no nonzero ideal U /R of M /R is totally isotropic. Now take any minimal nonzero ideal U /R ⊆ M /R. Then U /R is nondegenerate, and be- cause M has rank 1 there is an orthogonal decomposition M /R = U /R ⊥ U ⊥/R. Continuing in this way, we obtain an orthogonal decomposition M /R = U1/R ⊥ · · · ⊥ Uk /R with each Uj /R a minimal, nondegenerate ideal of M /R. By Lemma 5, each Uj /R is abelian. Then M /R is itself an abelian Lie algebra, and every nonzero element of M /R generates a 1- dimensional ideal. In particular, since M /R contains nonzero isotropic elements, it also has (cid:4) a nonzero isotropic ideal. This contradiction completes the proof of the theorem. 6 Symmetric Leibniz algebras Lemma 8. Suppose that M is a left Leibniz algebra. Then M is a symmetric Leibniz algebra if, and only if, M (cid:48) ⊆ R. Proof . In any left Leibniz algebra we have [a[bc]] + [[ac]b] = [[ab]c] + [b[ac]] + [[ac]b] = [[ab]c] + ψ([ac], b). Therefore, M is a symmetric Leibniz algebra if, and only if, ψ([ab], c) = 0 for all a, b, c ∈ M . (cid:4) The assertion of the lemma is just a restatement of this. Consider the left central Leibniz algebra M = M (L1 , L2 , R, α, π) with product (9) and its quotient algebra M = M /D (cf. Lemma 7) introduced at the end of Section 4. Since D is contained in the radical of M , it follows from Lemma 8 that M is symmetric if, and only if, M is symmetric. From (9) and Lemma 8 once more, this holds if, and only if, L1/R is abelian (recall that L2/R is abelian by construction) and x1 .y2 − x2 .y1 ∈ R (x1 , x2 ∈ L1 , y1 , y2 ∈ L2 ). We assert that the second condition is a consequence of the f irst. Indeed, because π : L1/R → HomC (L2/R, Z ) is an injection of L1 -modules then L1 acts trivially on im π , i.e., im π annihilates each element x1 .y1 . Since im π has no nonzero invariants in its action on L2/R, it follows that x1 .y1 ∈ R, and since this holds for any x1 ∈ L1 , y1 ∈ L2 then the second condition indeed follows, as asserted. Thus we have proved Lemma 9. The left central Leibniz algebras M (L1 , L2 , R, α, π) and M = M /D are symmetric Leibniz algebras if, and only if, L1/R and L2/R are both abelian Lie algebras. Now suppose that M0 is a symmetric Leibniz algebra of rank 1. Thus N = C (M0 ) has dimension 1 and ψ def ines a trace form on M0 with radical R. Consequently, we can f ind a pair of maximal, totally isotropic subspaces L1 , L2 ⊆ M0 such that L1 ∩ L2 = R and L1 + L2 has 10 G. Mason and G. Yamskulna codimension at most 1 in M0 . (L1 + L2 = M0 if, and only if, dim(M0/R) is even.) Because M0/R is an abelian Lie algebra by Lemma 8, it follows that H := L1 + L2 is a 2-sided ideal with Lie subalgebras L1 , L2 ; indeed, L1 , L2 ∈ L∗ . We will show that H ∼= M (L1 , L2 , R, α, π) for suitably def ined maps α, π . Both L1 , L2 are ideals of M0 , in particular the left adjoint def ines a morphism of Lie algebras α : L1 → Der(L2 ) satisfying the assumptions of (b) at the end of Section 4. Similarly, R = rad(M ) and we def ine the morphism of L1 -modules π : L1/R → HomC (L2/R, N ) as π(x1 , y1 ) := ψ(x1 , y1 ) (x1 ∈ L1 , y1 ∈ L2 ). It is easily seen that the assumptions of (c) at the end of Section 4 also hold. (The main point is the associativity of ψ .) Thus the set-up discussed in Section 4 holds, and we can apply Lemma 7 to obtain the left central Leibniz algebras M (L1 , L2 , R, α, π) and M . We assert that H ∼= M . To see this, one checks that the map ϕ : M (L1 , L2 , R, α, π) → H, (x1 , y1 ) (cid:55)→ x1 + y1 , is a surjective morphism of Leibniz algebras with kernel {(x1 , y1 ) x1 + y1 = 0} = {(a, −a) a ∈ R}. Now our assertion follows from the very construction of M . We have proved Theorem 4. A symmetric Leibniz algebra of rank 1 has an ideal of codimension at most 1 isomorphic to M (L1 , L2 , R, α, π). Acknowledgements This work was supported by the NSF (G.M.) and the Simons Foundation (G.Y.). The authors thank the (anonymous) referees for helpful comments. References [1] Ayupov Sh.A., Omirov B.A., On Leibniz algebras, in Algebra and Operator Theory (Tashkent, 1997), Kluwer Acad. Publ., Dordrecht, 1998, 1–12. [2] Barnes D.W., On Engel’s theorem for Leibniz algebras, Comm. Algebra 40 (2012), 1388–1389, arXiv:1012.0608. [3] Barnes D.W., On Levi’s theorem for Leibniz algebras, Bul l. Aust. Math. Soc. 86 (2012), 184–185, arXiv:1109.1060. [4] Lepowsky J., Li H., Introduction to vertex operator algebras and their representations, Progress in Mathe- matics , Vol. 227, Birkhauser Boston Inc., Boston, MA, 2004. [5] Loday J.L., Une version non commutative des alg`ebres de Lie: les alg`ebres de Leibniz, Enseign. Math. 39 (1993), 269–293. [6] Loday J.L., Cyclic homology, Grund lehren der Mathematischen Wissenschaften, Vol. 301, 2nd ed., Springer- Verlag, Berlin, 1998. [7] Mason G., Yamskulna G., On the structure of N-graded vertex operator algebras, arXiv:1310.0545.
1109.2986	2	1109	2011-11-01T14:13:45	Automorphisms of path coalgebras and applications	[ "math.RA" ]	Our main purpose is to introduce the notion of trans-datum for quivers, and apply it to the study of automorphism groups of path coalgebras and algebras. We observe that any homomorphism of path coalgebras is uniquely determined by a trans-datum, which is the basis of our work. Under this correspondence, we show for any quiver $Q$ an isomorphism from $\aut(kQ^c)$ to $\Omega^*(Q)$, the group of invertible trans-data from $Q$ to itself. We point out that the coradical filtration gives to a tower of normal subgroups of $\aut(kQ^c)$ with all factor groups determined. Generalizing this fact, we establish a Galois-like theory for acyclic quivers, which gives a bijection between large subcoalgebras of the path coalgebra and their Galois groups, relating large subcoalgebras of a path coalgebra with certain subgroups of its automorphism group. The group $\aut(kQ^c)$ is discussed by studying its certain subgroups, and the corresponding trans-data are given explicitly. By the duality between reflexive coalgebras and algebras, we therefore obtain some structural results of $\aut(\hat{kQ^a})$ for a finite quiver $Q$, where $\hat{kQ^a}$ is the complete path algebra. Moreover, we also apply these results to finite dimensional elementary algebras and recover some classical results.	math.RA	math	Automorphisms of path coalgebras and applications Yu Ye Abstract Our main purpose is to introduce the notion of trans-datum for quivers, and apply it to the study of automorphism groups of path coalgebras and algebras. We observe that any homomorphism of path coalgebras is uniquely determined by a trans-datum, which is the basis of our work. Under this correspondence, we show for any quiver Q an isomorphism from Aut(kQc) to Ω∗(Q), the group of invertible trans-data from Q to itself. We point out that the coradical filtration gives to a tower of normal subgroups of Aut(kQc) with all factor groups determined. Generalizing this fact, we establish a Galois-like theory for acyclic quivers, which gives a bijection between large subcoalgebras of the path coalgebra and their Galois groups, relating large subcoalgebras of a path coalgebra with certain subgroups of its automorphism group. The group Aut(kQc) is discussed by studying its certain subgroups, and the corresponding trans-data are given explicitly. By the duality between reflexive coalgebras and algebras, we therefore obtain some structural results of Aut( dkQa) for a finite quiver Q, where dkQa is the complete path algebra. Moreover, we also apply these results to finite dimensional elementary algebras and recover some classical results. Contents 0 Introduction 1 Qivers, path algebras and coalgebras 1.1 Quivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Taft-Wilson Theorem and coalgebra filtration . . . . . . . . . . . . . . . . . . 1.3 Complete path algebra, the dual of path coalgebra . . . . . . . . . . . . . . . 1.4 Augmented quivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Trans-data and path coalgebra morphisms 2.1 Trans-data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Trans-data = Path coalgebra homomorphisms . . . . . . . . . . . . . . . . . . 2.3 Composition of trans-data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 The automorphism group of a path coalgebra 3.1 A tower of normal subgroups of Aut(kQc) . . . . . . . . . . . . . . . . . . . . 3.2 Factor groups and solvability of Aut(kQc) . . . . . . . . . . . . . . . . . . . . Keywords: path coaglebra, complete path algebra, trans-datum, automorphism MSC2010: 16W20, 16T15, 16G20 Supported by National Natural Science Foundation of China (No. 10971206) 2 5 5 6 7 8 9 9 12 14 17 17 19 1 4 A Galois theory for path coalgebras 4.1 The Galois group of a coalgebra extension . . . . . . . . . . . . . . . . . . . . 4.2 Galois extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Fundamental theorem of Galois extensions . . . . . . . . . . . . . . . . . . . . 5 Automorphisms of a complete path algebra 5.2 5.1 Aut(dkQa) ∼= Aut(kQc) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Aut0(dkQa) = Inn(dkQa) Aut•(dkQa) Inner automorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4 The finite dimensional case . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 Examples 6.1 Quivers of directed An type . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Generalized Kronecker quivers . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 n-subspace quivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4 Aut(k[x]) and Aut(k[[x]]) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 20 22 24 27 27 28 32 35 37 37 38 38 39 0 Introduction Automorphism group of a mathematical object is, in roughly speaking, the symmetry of the object. A good knowledge about automorphisms of an object is essential in understanding its structure. To determine an automorphism group is always fundamental in classification problems. The present work aims to investigate automorphism groups of an important class of coalgebras and algebras, say the ones obtained from quivers. Let A be an associative algebra over a field k. In case that A is finite dimensional, the automorphism group Aut(A) is known to be an affine algebraic group with Lie algebra Der(A), the algebra of derivations of A; see [OV, §1.2.3, ex.2], for example. Some other important groups related are the outer automorphism group Out(A) = Aut(A)/ Inn(A), the quotient group modulo the inner ones, and the Picard group Pic(A), the group of isoclasses of invertible A-A-bimodules. Frohlich studied the Picard group of associative rings systematically. The Picard group is shown to be invariant under Morita equivalence, and for any ring R there exists a map from Aut(R) to Pic(R) with kernel Inn(R); see [Fr]. Moreover, Bolla [Bo] proved that when R is basic semiperfect, then the map is also surjective, that is Pic(R) ∼= Out(R) in this case. How the assumption Aut(A) being reductive (respectively, semi-simple, toral, solvable, nilpotent) effects the algebra structure of A has been discussed by Pollack [Po]. It is worth mentioning that the solvability of the automorphism group is of great importance in the classification of isolated hypersurface singularities, see a series papers [Y1, Y2] by Yau. Usually to determine Aut(A) (respectively, Out(A), Pic(A)) for a finite dimensional alge- bra A is very difficult, even to find out all idempotent elements of A is not an easy task. Only a few scattered examples are known. A well studied example is the automorphism group of an incidence algebra beginning with Stanley [St] and continued by Scharlau [Sch], Baclawski [Ba], and Coelho [Co]. Other examples are exterior algebras by Djokovi´c [Dj], square-free algebras by Anderson and D'Ambrosia [AA], and monomial algebras by Guil-Asensio and Saor´ın [GS2]. In a series papers [GS1, GS2], Guil-Asensio and Saor´ın developed a strategy to compute the outer automorphism group and the Picard group for finite dimensional algebras. As 2 an application the Picard group of a split finite dimensional algebra is calculated in several special cases. The present work is intended to be from a different point of view. We begin with the study of automorphisms of coalgebras and then apply to algebras. Notice that algebra and coalgebra are dual categorically. In philosophy, algebra is easy to handle sometimes, while in some other cases, to deal with coalgebras is easier. The key observation is that automorphisms of a path coalgebra are much easier to characterize. This is our starting point. We make an attempt to study the automorphism group (respectively, outer automorphism group, Picard group) of a path algebra via its dual path coalgebra. Recall that in path algebra case, the outer automorphism group and the Picard group are the same. We remark that both incidence algebras and exterior algebras are elementary, which can be realized as quotient algebras of path algebras. It is possible to apply our general results on path algebras and coalgebras to these examples, although we will not go further in this direction in this paper. Throughout, k will be a fixed field and all unadorned ⊗ will mean ⊗k. It is worth men- tioning that most results hold true in general situation, and quivers considered are arbitrary quivers unless otherwise stated. More precisely, we need the "acyclic" condition in Section 4 for the establishment of the Galois theory, and in some scattered applications we need certain finiteness condition. The main results and the structure of the paper is outlined as follows. In Section 1, we recall some basic notions and give a brief introduction to quiver techniques needed. Section 2 deals with a characterization of path coalgebra homomorphisms. We introduce the notion of trans-datum for quivers, see Definition 2.1.1. Let Q and Q′ be quivers. We denote by Ω(Q, Q′) the set of trans-data from Q to Q′ and write for brevity Ω(Q) = Ω(Q, Q). The following fundamental result is given in Theorem 2.2.1. Theorem 1. Let Q, Q′ be quivers, and kQc and kQ′c the path coalgebras. Then we have mutually inverse maps Coalg(kQc, kQ′c) / Ω(Q, Q′) , here Coalg(kQc, kQ′c) denotes f the set of coalgebra homomorphisms from kQc to kQ′c. Under the above correspondence, a composition map of trans-data, which is compatible with the one of coalgebra homomorphisms, is given explicitly, making Ω(Q) a monoid which is isomorphic to Coalg(kQc, kQc); see Theorem 2.3.5. The following results is obtained in Theorem 2.2.6 and 5.4.8, which show us the advantage of using the notion of trans-datum, especially in the study of automorphisms of pointed coalgebras and elementary algebras. Theorem 2. (1) Let Q be any quiver and C a large subcoalgebra of kQc. Then any auto- morphism of C extends to an automorphism of kQc. (2) Let A be a finite dimensional elementary algebra and Q its extension quiver. Then any automorphism of A is induced from an automorphism of dkQc. The above theorem just says that the automorphism group of a finite dimensional elemen- tary algebra is somehow controlled by the automorphism group of a complete path algebra. We expect similar result holds true for any finite dimensional algebra which satisfies certain separable condition, although we do not have an argument at moment. 3 / o o We may compare the above theorem with Quebbemann's result [Qu], which connects automorphisms of a tensor algebra and the ones of its quotient algebras. Let A be a finite dimensional algebra with Jacobson radical J. Suppose that there is an algebra embedding A/J ֒→ A such that A = A/J ⊕ J as A/J-bimodules. Quebbemann showed that for any automorphism τ of A, there exists an automorphism τ ′ of the tensor algebra TA/J (J/J 2) and an algebra epimorphism φ : TA/J (J/J 2) −→ A, such that τ ◦ φ = φ ◦ τ ′. Notice that this correspondence does not give a group homomorphism, for φ varies for different τ . Section 3 is devoted to the study of Aut(kQc) for a quiver Q. The coradical filtration is shown to give a tower of normal subgroups of Aut(kQc) with all factor groups characterized explicitly, particularly we calculate dim(Aut(kQc ≤n is the n-truncated path coalgebra of Q. Consequently, finite dimensional truncated path coalgebras with solvable automorphism group are classified as follows; see Theorem 3.2.5 for detail. ≤n)) in Proposition 3.2.4, where kQc Theorem 3. Let Q be a finite quiver and n ≥ 2 an integer. Then Aut0(kQc if and only if Q is a Schurian quiver; Aut(kQc quiver and Aut(Q) is resolvable. ≤n) is solvable ≤n) is solvable if and only if Q is a Schurian Compare also with Proposition 5.4.2 which concerns truncated path algebras with solvable automorphism groups. We mention that the solvability of the identity component of the automorphism group of a monomial algebra has been discussed in [GS2]. Since truncated path algebras are obviously monomial and Aut0(kQc ≤n) (see Remark 5.3.6), thus the first part of Proposition 5.4.2 can also be deduced from [GS2, Corollary 2.22]. ≤n) is the identity component of Aut(kQc In Section 4 we develop a Galois-like theory for an acyclic quiver Q, connecting admissible subgroups of Aut(kQc) and large subcoalgebras of kQc. The following result, which we call the fundamental theorem for Galois extensions, is given in Theorem 4.3.2. Theorem 4. Let Q be an acyclic quiver. Let D be a large subcoalgebra of kQc and E/D a Ga- lois extension. Then there is a bijection between C (D, E), the set of intermediate coalgebras, and G (D, E), the set of admissible subgroups of Gal(E/D); Moreover, for any intermediate coalgebra D ⊆ M ⊆ E, M/D is a Galois extension if and only if Gal(E/M ) ⊳ Gal(E/D), and in this case, Gal(E/D)/ Gal(E/M ) ∼= Gal(M/D). We refer to Section 4 for an explanation of notations. Some applications of the funda- mental theorem are also given there. Section 5 concerns with the study of automorphisms of path coalgebra and complete path algebras. The idea is to consider for a quiver Q the following natural subgroups of Ω∗(Q), say IΩ∗(Q), IΩ∗ ◦(Q) ⊳ Ω∗(Q) and IΩ∗(Q) = IΩ∗ •(Q). We show that IΩ∗(Q) ⊳ Ω∗(Q), IΩ∗ 0(Q), see Proposition 5.2.4 for detail. ◦(Q) ⋊ IΩ∗ 0(Q), IΩ∗ ◦(Q) and Ω∗ We set Inn(kQc) = f(IΩ∗(Q)), Inn0(kQc) = f(IΩ∗ ◦(Q)) and •(Q)). We mention that for µ ∈ IΩ∗(Q), a useful formula for fµ is ob- Aut•(kQc) = f(Ω∗ tained in Proposition 5.2.1. Consequently each σ ∈ Inn(kQc) acts invariantly on any large subcoalgebra of kQc; see Corollary 5.2.2. 0(Q)), Inn◦(kQc) = f(IΩ∗ Let D be a finitary pointed coalgebra, here finitary means that D(1) is finite dimensional. D is realized as a large subcoalgebra of kQc for some finite quiver Q. Due to Taft we have a group isomorphism Aut(D) ∼= Aut(D∗), particularly Aut(dkQa) ∼= Aut(kQc) and hence Aut(dkQa) ∼= Ω∗(Q) for any finite quiver Q; see Section 5.1.1. We rely heavily on this basic fact, which enables to apply the technique developed for path coalgebras. 4 Denote by Inn(dkQa) the inner automorphism group of dkQa. Consider also Inn◦(dkQa), the normal subgroup of Inn(dkQa) induced by elements in 1 + Rad(dkQa), and the subgroup Inn0(dkQa) of inner automorphisms induced by elements in (kQ0)×. Another subgroup taken into account is Aut•(dkQa), the subgroup of automorphisms fixing kQ0. Similar subgroups are defined for quotient algebras of dkQa. Notice that the subgroups Inn◦(kQa) and Aut•(kQa) By Proposition 5.2.6 and 5.3.1, Inn(dkQa), Inn◦(dkQa), Inn0(dkQa) and Aut•(dkQa) corre- spond to Inn(kQc), Inn◦(kQc), Inn0(kQc) and Aut•(kQc) respectively. In this sense, elements in Inn(kQc) are also called inner automorphisms of kQc. Now we have the following charac- terization, which is given in Theorem 5.3.8. have been studied by many authors, see for example [Po, Section 1] and [GS1]. Theorem 5. Let Q be a finite quiver, A = dkQa and B = A/I, where I ⊆ dkQa ideal of dkQa. Then Aut0(B) = Inn◦(B) Aut•(B) and Inn(B) ∩ Aut•(B) = χ(Z × 0 , then Aut0(B) = Inn◦(B) ⋊ Aut•(B). ≥2 is an B(B0)). If moreover, Z × B(B0) = B× Apply to finite dimensional case, we reobtain some classical results. For instance, we cal- culate the dimension of Out(kQa) for acyclic quivers, and consequently show that Out(kQa) is finite if and only if the quiver is a tree; see Corollary 5.4.4 and 5.4.7 for more detail. Examples are provided in Section 6. We discuss the quivers of directed An-type, n- Kronecker quivers and n-subspace quivers there. It is known that for a non-acyclic quiver, the automorphism group of the path algebra and the one of the path coalgebra are not isomorphic under the natural correspondence. We also illustrate this fact with a typical example. 1 Qivers, path algebras and coalgebras We will give a brief introduction to quiver techniques in this section. For more details we refer to [ARS] and [Ri]. 1.1 Quivers Recall that a quiver Q = (Q0, Q1, s, t : Q1 −→ Q0) is a directed graph, where Q0 is the set of vertices, Q1 the set of edges (usually called arrows), and s, t : Q1 −→ Q0 are two maps assigning for each arrow its starting and terminating vertex respectively. A quiver Q is said to be finite if Q0 and Q1 are both finite, and acyclic if Q contains no oriented cycles as subquiver. By a nontrivial path in Q we mean a sequence of arrows p = α1α2 · · · αn with s(αi) = t(αi−1) for all 2 ≤ i ≤ n, n is called the length of p, denoted by l(p) = n, pictorially α1−→ • α2−→ • α3−→ · · · αn−−→ •. • We use s(p) = s(α1) and t(p) = t(αn) to denote the starting and terminating vertex of p respectively and say that p is a path from s(p) to t(p). For each vertex i ∈ Q0, we use ei to denote the trivial path, i.e., a path of length 0, from vertex i to itself. By abuse of notations, we also use Q to denote the set of all paths in Q and usually P the set of all nontrivial paths in Q. For i, j ∈ Q0, Qi,j (respectively, Pi,j) denotes the set of paths (respectively, nontrivial paths) from i to j. 5 As usual kQ denotes the k-space spanned by all paths in Q. Clearly, kQ =Ln≥0 kQn is a positively graded space, here Qn is the set of paths in Q which are of length n. Similarly set (Qn)i,j ⊆ Qn to be the subset consisting of those paths from i to j for any i, j ⊆ Q0. Clearly kQ is finite dimensional if and only if Q is a finite acyclic quiver. The path algebra of the quiver Q has the underlying vector space kQ and multiplication given by concatenation of paths in an obvious way, that is, the product of two paths x and y is the path xy if t(x) = s(y) and 0 otherwise. We denote the path algebra by kQa. By definition ei is an idempotent for each i ∈ Q0, and kQa has an identity element if and only if ei. The path algebra is a positively graded Q0 is a finite set, and in this case, 1kQa =Pi∈Q0 algebra with respect to the length grading. The path coalgebra of Q, denoted by kQc, is in some sense the dual of the path algebra. As a vector space kQc = kQ. The coproduct ∆ is given by splitting the path at all possible position, to be precise, ∆(ei) = ei ⊗ ei for each i ∈ Q0 and ∆(p) = s(p) ⊗ p + X1≤i≤n−1 α1α2 · · · αi ⊗ αi+1αi+2 · · · αn + p ⊗ t(p) for each nontrivial path p = α1α2 · · · αn. The counit is given by ε(ei) = 1 for each vertex i and ε(p) = 0 for each nontrivial path p. Again the path coalgebra is a positively graded coalgebra with respect to the length grading. Remark 1.1.1. Recall that there is a more general construction by using tensor product and cotensor product. Let C be a coalgebra and M C and C N be C-comodules with the coaction given by φ : M → M ⊗ C and ψ : N → C ⊗ N . The cotensor product M (cid:3)CN is defined to be the kernel of φ ⊗ IdN − IdM ⊗ψ : M ⊗ N −→ M ⊗ C ⊗ N, we refer to [EM] for more detail about cotensor products. Note that for a quiver Q, the path algebra kQa is isomorphic to TkQ0 (kQ1), the tensor algebra of the kQ0-bimodule kQ1 over kQ0; and the path coalgebra kQc is isomorphic to the cotensor coalgebra CotkQ0 (kQ1) of the kQ0-bicomodule kQ1 over kQ0. We therefore identify the k-space kQn with the tensor space Tn kQ0 (kQ1). kQ0 (kQ1) as well as the cotensor space Cotn 1.2 Taft-Wilson Theorem and coalgebra filtration Let C be a coalgebra. A nonzero element g in C is called a group-like element if ∆(g) = g ⊗ g. The set of group-like elements in C is denoted by G(C). Let g, h be two group-likes, a (g, h)-primitive element is by definition an element α such that ∆(α) = g ⊗ α + α ⊗ h. We denote the set of (g, h)-primitive elements by Pg,h, which is easy shown to be a k-linear space. For consistency of notations, we set Pg,h = 0 if either g = 0 or h = 0. In case C = kQc is the path coalgebra of some quiver Q, then we simply write Ps,t = Pes,et for any s, t ∈ Q0. We have the following easy characterization for path coalgebras, which is known as the Taft-Wilson Theorem, see also [DNR, Theorem 5.4.1] for a proof. Lemma 1.2.1. ([TW]) Let Q be an arbitrary quiver and kQc the path coalgebra. Then (1) G(kQc) = {eii ∈ Q0}; (2) Ps,t = k(es − et)L k(Q1)s,t for each pair of vertices s, t ∈ Q0; in particular, Ps,s = k(Q1)s,s for each vertex s ∈ Q0. 6 Let C be an arbitrary coalgebra. There exists a unique filtration of subcoalgebras of C, say the coradical filtration C(0) ⊆ C(1) ⊆ C(2) ⊆ C(3) ⊆ · · · , such that C(0) is the coradical of C, that is the sum of all simple subcoalgebras of C, and (n) = C(i) ⊗ C(n−i). If C = kQc, the path coalgebra of a quiver Q, then kQc kQi. The following useful lemma is due to Heyneman and Radford ([HR]); ∆(C(n)) ⊆ P0≤i≤n kQ≤n = P0≤i≤n see also [DNR, p.65] for a proof. Lemma 1.2.2. Let C and D be coalgebras and f : C −→ D a coalgebra morphism. Then f (C(n)) ⊆ D(n), and f is injective if and only if f C(1) is injective. In particular, f (G(C)) ⊆ G(D) and f (Pg,h) ⊆ Pf (g),f (h). 1.3 Complete path algebra, the dual of path coalgebra In this subsection Q is assumed to be a finite quiver. Note that the path algebra kQa and path coalgebra kQc have the same underlying space kQ, thus they share the same basis Q consisting of all paths. To avoid confusion, we use Q = {pp ∈ Q} to denote the corresponding basis of kQa. We may introduce a non-degenerate pairing (−, −) : kQa×kQc −→ k by setting (p, q) = δp,q, for any p, q ∈ Q. Thus for each n ≥ 0, the homogeneous components kQa n and n are dual to each other as vector spaces via (−, −). Moreover, if Q is acyclic, then kQa kQc and kQc are dual to each other. Weyman and Zelevinsky to study the mutation of quivers with potentials; see [DWZ, Defi- The notion of complete path algebra dkQa for a quiver Q is introduced by Derksen, nition 2.2]. As k-spaces, dkQa =Qp∈Q k = {(ap)p∈Q, ap ∈ k}, and the multiplication in dkQa is given by (ap)p∈Q · (bp)p∈Q = (cp)p∈Q, cp = Xp=p1p2 p1,p2∈Q, ap1bp2 . Set J = kQa a finite acyclic quiver. Here by p = p1p2 we mean a split of the path p such that the concatenation of p1 and p2 is p. ≥1 be the Graded Jacobson ideal of kQa. Then we have a filtration of ideals An element (ap)p∈Q ∈ dkQa can also be written as Pp∈Q app, an infinite linear combinations of paths in Q. In this sense, kQa is a subalgebra of dkQa and kQa = dkQa if and only if Q is ≥n. One shows that dkQa is nothing There is a uniquely determined non-degenerate pairing (−, −) : dkQa × kQc −→ k, extend- ing the one between kQa and kQc introduced above. Under this pairing, dkQa is isomorphic kQa ⊇ J ⊇ J 2 ⊇ J 3 ⊇ · · · of kQa, where each J n = kQa but the completion of kQa at the ideal J, i.e., the inverse limit lim ←− to (kQc)∗, the dual vector space of kQc. The following basic facts can be found in [DNR, Chapter5], and we list them without a proof for later use. kQa/J n. Lemma 1.3.1. Let (C, ∆, ε) be a coalgebra and C ∗ its dual vector space. (1) C ∗ is an associative algebra with the identity element ε and the multiplication given by the convolution ∗, that is, (f ∗ g)(c) :=P f (c(1))g(c(2)) for any f, g ∈ C ∗ and c ∈ C, here we use the Sweedler's notation ∆(c) =P c(1) ⊗ c(2). 7 (2) Let D be a subcoalgebra of C. Then D∗ ∼= C ∗/D⊥ as an associative algebra, where D⊥ = {f ∈ C ∗ f (d) = 0, ∀d ∈ D} is an ideal of C ∗. (3) Let C(0) ⊆ C(1) ⊆ C(2) ⊆ · · · be the coradical filtration of C. Then Radn(C ∗) = (n−1) for each n ≥ 1, here Rad(C ∗) is the Jacobson radical of C ∗ and Radn(C ∗) = C ⊥ Rad(C ∗) Radn−1(C ∗) for each n ≥ 2. Turning back to quiver case, one has the following results. Lemma 1.3.2. Let Q be a finite quiver. Then (1) dkQa ∼= (kQc)∗ as associative algebras, and kQa ∼= (kQc)∗ if Q is acyclic; (2) Radn(dkQa) = {Pp∈Q,l(p)≥n app ap ∈ k}; (3) A subcoalgebra D ⊆ kQc is large if and only if D⊥ ⊆ Rad(dkQa). Recall that a subcoalgebra D of a path coalgebra kQc is said to be large if D ⊇ kQ≤1. A classical result by Chin and Montgomery says that any pointed coalgebra D can be realized as a large subalgebra of the path coalgebra kQc of some quiver Q, and such Q is uniquely determined by D and usually called the extension quiver of D; see [CM, Theorem 4.3]. An ideal I of kQa is called an admissible ideal if kQa ≥2 for some n ≥ 2. Note that kQa/I is a finite dimensional elementary algebra if Q is a finite quiver and I is admissible, here elementary means that each simple module is one dimensional. In this case, ≥n ⊆ I ⊆ kQa dkQa/bI ∼= kQa/I, here bI = IdkQa is an ideal of dkQa satisfying Radn(dkQa) ⊆ bI ⊆ Rad2(dkQa). We also call an ideal of dkQa with this property an admissible ideal. Conversely, a famous result by Gabirel says that any finite dimensional elementary algebra A has the form kQa/I, where Q is a finite quiver uniquely determined by A and I an admissible ideal of kQa; see [DK, Theorem 3.6.6]. Q is usually called the extension quiver of A. Clearly, the dual coalgebra C = A∗ is a large subcoalgebra of kQc. 1.4 Augmented quivers We may associate each quiver an auxiliary quiver, which plays an important role in our characterization of path coalgebra homomorphisms. Definition 1.4.1. Let Q be a given quiver. The augmented quiver of Q, denoted by Qτ , is a quiver obtained from Q by adding some new arrows, precisely Qτ 0 = Q0, and Qτ 1 = Q1 ∪ {es,t s, t ∈ Q0, s 6= t} with s(es,t) = s and t(es,t) = t for each es,t. Remark 1.4.2. For a given quiver Q, we can identify Ps,t with k(Qτ here (Qτ 1 )s,t denotes the set of arrows in the augmented quiver Qτ from i to j. 1)s,t for any s, t ∈ Q0, Example 1.4.3. Let Q be the quiver •1 is given by α / •2 β / •3 , then the augmented quiver Qτ e1,3 _ Y V S P •1 q t g e ZW J M n P k h e e1,2 _ α _ e2,1 YVS Z W ge •2 g e ZW e2,3 _ β _ e3,2 M J Z W •3 ge t q , nkhe _ e3,1 here we use dashed arrows to denote the new added ones. 8 / / / / + + $ $ k k / / + + k k d d As before let Q denote the set of all paths in Q, P ⊂ Q the set of all nontrivial paths in Q and Qτ the set of all paths in Qτ . Q and P are identified as subsets of Qτ in an obvious way. We also use E to denote the set of all nontrivial paths in Qτ which are compositions of new arrows. We define a linear map F : k(Qτ )c −→ kQc by setting F (p) = p, F (e) = ei1 − ei2, F (pe) = p, ∀ p ∈ Q; ∀ e = ei1,i2 ei2,i3 · · · ein,in+1 ∈ E; ∀ p ∈ P, e ∈ E; ∀ p ∈ P, s 6= s(p) ∈ Q0; F (es,s(p)p) = −p, F (es,s(p)pe) = −p, ∀ p ∈ P, s 6= s(p) ∈ Q0, e ∈ E; F (p) = 0, otherwise. To characterize homomorphisms of path coalgebras we need the following key lemma. The proof is given by routine check and we omit it here. Lemma 1.4.4. Let Q and F be as above. Then F is a homomorphism of coalgebras such that F kQc = Id and F (es,t) = es − et for any s, t ∈ Q0. Remark 1.4.5. We emphasize that a coalgebra homomorphism from k(Qτ )c to kQc satisfying the conditions in Lemma 1.4.4 is not unique in general. 2 Trans-data and path coalgebra morphisms We introduce the notion of trans-datum for quivers in this section. As we will show, homomor- phisms of path coalgebras correspond to trans-data bijectively. Under this correspondence, trans-data can also be composed and the composition map is given explicitly. 2.1 Trans-data Definition 2.1.1. Let Q, Q′ be quivers. A trans-datum µ from Q to Q′ is by definition a pair (µ0, µ+), where µ0 : kQc 0 is a coalgebra homomorphism and µ+ = (µp)p∈P ∈ Qp∈P Pµ0(es(p)),µ0(et(p)) a family of primitive elements in kQ′c indexed by P . Trans-data are usually denoted by λ, µ, ν, · · · . The set of all trans-data from Q to Q′ is denoted by Ω(Q, Q′) and we write Ω(Q) = Ω(Q, Q) for brevity. 0 −→ Q′c Let µ = (µ0, µ+) ∈ Ω(Q, Q′). Write µ+ = µ0 + µ1 in the sense that µ0 and µ1 are both families of primitives indexed by P with µ0 p for any p ∈ P . Clearly µ0 and µ1 are uniquely determined by µ+ and hence µ can also be written as a triple (µ0, µ0, µ1). Moreover, since µ0 p is a multiple of µ0(es(p)) − µ0(et(p)), we denote by µp the coefficient in the expression µ0 p = µp(µ0(es(p)) − µ0(et(p))). Clearly we may set µ = 0 whenever s(p) = t(p). We also view µ0 and µ1 as linear maps from kQ≥1 to kQ′ 0 and kQ′ 1 in an obvious way. Now we have the following k-linear maps associated to a trans-datum µ: 1 and µp = µ0 p ∈ kQ′ p + µ1 0, µ1 p ∈ kQ′ µ0 : kQ −→ kQ′ 0, µ0(p) = 0 ∀p ∈ P ; µ0 : kQ −→ kQ′ 0, µ0(ei) = 0 ∀i ∈ Q0, µ0(p) = µ0 p ∀p ∈ P ; 9 µ : kQ −→ k, µ(ei) = 0 ∀i ∈ Q0, µ(p) = µp ∀p ∈ P ; µ1 : kQ −→ kQ′ 1, µ1(ei) = 0 ∀i ∈ Q0, µ1(p) = µ1 1, µ+ = µ0 + µ1. 0 ⊕ kQ′ µ+ : kQ −→ kQ′ p ∀p ∈ P ; Proposition 2.1.2. Let Q, Q′ be any quivers and Qτ , Q′τ the augmented quivers. Denote by P the set of all nontrivial paths in Q. Let µ = (µ0, µ+) ∈ Ω(Q, Q′). We extend µ0 to a linear map f : kQc −→ kQ′c by setting f kQ0 = µ0 and f (p) = Xp=p1p2···pr pi∈P,i=1,··· ,r F (µp1 (cid:3)µp2 · · · (cid:3)µpr ) (2.1) for any p ∈ P , here we view the path coalgebra as a cotensor coalgebra and (cid:3) = (cid:3)kQ′ is the cotensor product. Then f is a homomorphism of coalgebras. Moreover, f is injective if and only µ0 and µ1kQ1 are both injective. 0 Proof. It suffices to prove that f is a coalgebra homomorphism, and the rest part follows from Lemma 1.2.2 directly. We will show that for any p ∈ P , (f ⊗ f )(∆(p)) = ∆(f (p)). By direct calculation, (cid:3) · · · (cid:3)µpr )) (F ⊗ F )(∆(µp1 (cid:3) · · · (cid:3)µpr )) (cid:0)(f (es(p)) ⊗ F (µp1 (cid:3) · · · (cid:3)µpr ) F (µp1 (cid:3) · · · (cid:3)µpi) ⊗ F (µpi+1 (cid:3) · · · (cid:3)µpr ) (cid:3) · · · (cid:3)µpr ) ⊗ f (et(p))(cid:1) F (µp1 (cid:3) · · · (cid:3)µpr ) pi∈P pi∈P pi∈P ∆(F (µp1 RHS = Xp=p1p2···pr = Xp=p1p2···pr = Xp=p1p2···pr + X1≤i≤r−1 =f (es(p)) ⊗ Xp=p1p2···pr X1≤i≤r−1 + Xp=p1p2···pr + Xp=p1p2···pr + (µp1 (µp1 pi∈P pi∈P pi∈P F (µp1 (cid:3) · · · (cid:3)µpi ) ⊗ F (µpi+1 (cid:3) · · · (cid:3)µpr ) (cid:3) · · · (cid:3)µpr ) ⊗ f (et(p)) =f (es(p)) ⊗ f (p) + Xp=p1p2···pr pi∈P X1≤i≤r−1 F (µp1 (cid:3) · · · (cid:3)µpi ) ⊗ F (µpi+1 (cid:3) · · · (cid:3)µpr ) + f (p) ⊗ f (et(p)). 10 and LHS =(f ⊗ f )(es(p) ⊗ p + Xp=p1p2 =f (es(p)) ⊗ f (p) + Xp=p1p2 pi∈P pi∈P p1 ⊗ p2 + p ⊗ et(p)) f (p1) ⊗ f (p2) + f (p) ⊗ f (et(p)), Now by comparing different ways to write a path as a composition of subpaths, we have f (p1) ⊗ f (p2) pi∈P Xp=p1p2 = Xp=p1p2 = Xp=p1p2···pr pi∈P pi∈P p2=y1y2 ···yv Xp1=x1x2···xu X1≤i≤r−1 xi ,yj ∈P F (µx1 (cid:3) · · · (cid:3)µxu) ⊗ F (µy1 (cid:3) · · · (cid:3)µyv ) F (µp1 (cid:3) · · · (cid:3)µpi) ⊗ F (µpi+1 (cid:3) · · · (cid:3)µpr ), and it follows that LHS=RHS, which completes the proof. The formula 2.1 can also be rewritten as follows, which will be more practical in some special cases. Proposition 2.1.3. Let Q, Q′ be quivers and µ ∈ Ω(Q, Q′). Then the coalgebra morphism f obtained from µ is given by fµ(x) = µ0(x) +Xr≥1 F (µ+(x(1))(cid:3) · · · (cid:3)µ+(x(r))) for any x ∈ kQc ≥1, again we use the Sweedler's notation ∆n(x) =P x(1) ⊗ · · · ⊗ x(n+1). Proof. We need only to prove the case x = p for some nontrivial path p. By definition ∆r−1(p) = Xp=q1q2 ···qr qi ∈Q q1 ⊗ q2 ⊗ · · · ⊗ qr, where qi's are allowed be trivia paths. Note that µ+(qi) = 0 if qi is a trivial path. Therefore F (µ+(q1)(cid:3)µ+(q2)(cid:3) · · · (cid:3)µ+(qr)) = 0 whenever one of qi's is trivial. It follows that F (µ+(q1)(cid:3)µ+(q2) · · · (cid:3)µ+(qr)) qi ∈Q µ0(p) + Xp=q1q2 ···qr = Xp=p1p2···pr = Xp=p1p2···pr F (µp1 pi∈P pi∈P F (µ+(p1)(cid:3)µ+(p2) · · · (cid:3)µ+(pr)) (cid:3)µp2 · · · (cid:3)µpr ) = fµ(p). The assertion follows from the linearity of both sides in the equality. 11 2.2 Trans-data = Path coalgebra homomorphisms Next we show that any coalgebra homomorphism of path coalgebras is obtained from a trans- datum in a way as given in Proposition 2.1.2. Recall that a subcoalgebra D of kQc is said to be monomial if D has a basis consisting of paths in Q. Note that if p = α1α2 · · · αr ∈ D, then αiαi+1 · · · αj ∈ D for any i ≤ j, especially es(p), et(p) ∈ D. Theorem 2.2.1. Let Q, Q′ be given quivers, D ⊆ kQc a monomial subcoalgebra with a basis X consisting of paths in Q and f : D −→ kQ′c a coalgebra homomorphism. Then we can assign each nontrivial path p ∈ X a primitive element µp ∈ Pf (es(p)),f (et(p)) in a unique way, such that (2.1) holds for any p ∈ P ∩ X. In particular, we have mutually inverse bijections Coalg(kQc, kQ′c) / Ω(Q, Q′) , f here Coalg(kQc, kQ′c) is the set of coalgebra homomorphisms, is given as above and f given as in Proposition 2.1.2. Proof. Set Xn = {p ∈ X l(p) = n} for each n ≥ 1, here l(p) denotes the length of p. We use induction on l(p) to prove the existence and uniqueness of µp. For p ∈ X1, we set µp = f (p), and p ∈ Ps(p),t(p) implies that f (p) ∈ Pf (es(p)),f (et(p)). In this case, the formula (2.1) holds automatically and it also determines µp uniquely. Now assume that for each p ∈ X with 1 ≤ l(p) ≤ n − 1, we have assigned a primitive element µp ∈ Pf (es(p)),f (et(p)) such that (2.1) holds for all p ∈ X with 1 ≤ l(p) ≤ n − 1. Consider p ∈ Xn. We set µp = f (p) − Xp=p1p2···pr ,r≥2 pi∈P,i=1,··· ,r F (µp1 (cid:3)µp2 (cid:3) · · · (cid:3)µpr ). Again (2.1) holds automatically for p and µp is uniquely determined. The only left is to show that µp ∈ Pf (es(p)),f (et(p)). It follows from the proof of Propostion 2.1.2 easily that ∆(µp) = f (es(p)) ⊗ µp + µp ⊗ f (et(p)), this just says µp is a primitive element and the proof is completed. Example 2.2.2. Let Q be a subquiver of Q′ and ι : kQc −→ kQ′c the obvious embedding. Then the trans-datum ι ∈ Ω(Q, Q′) is given by (ι)0(ei) = ei for any i ∈ Q0, (ι)α = α for any α ∈ Q1 and (ι)p = 0 for any p ∈ Q with l(p) ≥ 2. In particularly, the trans-datum IdkQc ∈ Ω(Q) corresponding to the identity map of kQc is denote by 1Q, or simply 1 when there is no confusion. Remark 2.2.3. The correspondence is given by using induction. It comes to a naive but interesting question: whether there exists an explicit reciprocity formula for (2.1). Such a formula is helpful in finding the inverse element of an automorphism of a path coalgebra, which could be useful in the study of Jacobian conjecture. Remark 2.2.4. One defines n-truncated trans-datum for quivers Q and Q′ and any integer n ≥ 1, which characterizes the coalgebra morphisms Coalg(kQc ≤n). By definition an n-truncated trans-datum from Q to Q′ is a pair µ = (µ0, µ+), where µ0 = Coalg(kQ0, kQ′ 0) is a coalgebra morphism and µ+ = {µp ∈ Pµ0(es(p)),µ0(et(p))}p∈P≤n , a family of primitive elements in kQ′c indexed by P≤n, here P≤n denotes the nontrivial paths in Q with length less than or equal to n. ≤n, kQ′c 12 / o o Denote by Ω(Q, Q′; n) the set of n-truncated trans-datum from Q to Q′. Similar to Theorem 2.2.1, we have a bijection from Ω(Q, Q′; n) to Coalg(kQc ≤n)). Obviously each µ ∈ Ω(Q, Q′; n) gives rise to an element µ in Ω(Q, Q′) with µp = 0 for all path p in Q with length strictly greater than n. In other words, any coalgebra homomorphism from ≤n extends to a coalgebra homomorphism from kQc to kQ′c. This interesting kQc result has a more generalized version. ≤n to kQ′c ≤n, kQ′c Proposition 2.2.5. Let Q, Q′ be quivers and D ⊆ kQc a subcoalgebra. Then any f ∈ Coalg(D, kQ′c) extends to a coalgebra homomorphism from kQc to kQ′c. Proof. First of all, we can extend f D0 to a coalgebra homomorphism µ0 : kQ0 −→ kQ′ 0 easily by setting µ0(ei) = f (ei) for ei ∈ D and µ0(ei) = 0 for ei /∈ D. Thus f extends to a coalgebra homomorphism f0 : D + kQ0 −→ kQ′c with f kQ0 = µ0. Let S = {(E, g)D + kQ0 ⊆ E ⊆ kQ′c, g : E −→ kQc, f D = gD} be the set of pairs (E, g), where E is subcoalgebra of kQc containing D + kQ0 and g : E −→ kQ′c a colagebra homomorphism extending f . The above argument says that S is nonempty. We may define a partial ordering (cid:22) on S, by definition (E, g) (cid:22) (E′, g′) if and only if E ⊆ E′ and g = g′E. Let (E1, g1) (cid:22) (E2, g2) (cid:22) · · · (cid:22) (En, gn) (cid:22) · · · be an arbitrary ascending chain in S with respect to the ordering (cid:22). We set E =Sn≥1 En. It is easy to show that E is a subcoalgebra of kQc which contains all En's. We can also define a coalgebra homomorphism g : E −→ kQ′c as follows. For any c ∈ E, there exists some n with c ∈ En, then we set g(c) = gn(c). Note that g(c) does not depend on the choice of n. Thus we have obtained a (E, g) ∈ S with (En, gn) (cid:22) (E, g) for all n ≥ 1. By Zorn's Lemma, there exists some maximal element (E, g) in S. We claim that E = kQc. For this it suffices to show that kQ≤n ⊆ E for all n ≥ 0. We use induction on n. By definition kQ0 ⊆ E. Now suppose that kQ≤n−1 ⊆ E. By Theorem 2.2.1, we may assign each p ∈ Q≤n−1 a primitive element µp ∈ Pf (es(p)),f (et(p)) in a unique way, such that (2.1) holds for any p ∈ Q≤n−1. Assume that we have some p ∈ Qn such that p /∈ E. We set µp = 0 and E = E ⊕ Kp. We can extend g to a linear map g : E −→ kQ′c by setting g(p) = Xp=p1p2···pr pi∈Q F (µp1 (cid:3)µp2 (cid:3) · · · (cid:3)µpr ). g is a coalgebra morphism since it is when restricted to the subcoalgebra E and to the subcoalgebra generated by p. Hence (E, g) (cid:22) ( E, g), contrary to the maximality of (E, g). Therefore kQn ⊆ D, and the proof is completed by induction. Combined with Lemma 1.2.2, we draw the following consequence, a special case of which plays a very important role in the classification of not necessarily coradically graded pointed Hopf algebras; see for instance [HYZ]. In fact, this is also one of our main motivations to start with the present work. Theorem 2.2.6. Let Q be a finite quiver and D ⊆ kQc a large subcoalgebra. Then any au- tomorphism of D extends to an automorphism of kQc. Particularly, Aut(D) is a subquotient group of Aut(kQc), here Aut(kQc) and Aut(D) are the automorphism group of kQc and D respectively. 13 Remark 2.2.7. Note that the theorem is not true in general if D is not a large subcoalgebra. For example, let Q be the quiver 1• α β γ / •2 •3 , and kQc the path coalgebra. Let D ⊆ kQc be the subcoalgebra spanned by {e1, e2, e3, α, γ}. Let σ ∈ Aut(D) given by σ(e1) = e3, σ(e2) = e2, σ(e3) = e1, σ(α) = γ and σ(γ) = α. Clearly σ does not extend to an automorphism of kQc. 2.3 Composition of trans-data Since trans-data correspond to path coalgebra homomorphisms bijectively, the composition of coalgebra homomorphisms will induced a composition of trans-data. In this subsection we will give the composition map explicitly. Let Q, Q′ be quivers and Qτ , Q′τ the augmented quivers. As before P and P τ denote the set of nontrivial paths in Q and Qτ respectively. Let µ = (µ0, µ+) ∈ Ω(Q, Q′) be a trans- datum. µ gives rise to an element ¯µ ∈ Ω(Q, Q′τ ) in an obvious way, say ¯µ0 = µ0, ¯µ0 = 0 and ¯µ1 p = µp for all p ∈ P , here each µp is viewed as an element in kQ′τ 1 . The following result is easily deduced from the definition and we omit the proof. Lemma 2.3.1. Let Q, Q′ be quivers and µ ∈ Ω(Q, Q′). Then we have fµ = F ◦ f¯µ. fµ kQc k(Q′τ )c F f ¯µ / kQ′c We may also lift µ to µ ∈ Ω(Qτ , Q′) in the following way: µ0 = µ0, µp = µp for p ∈ P , µes,t = µ0(es − et) for any s, t ∈ Q0, µes,s(p)p = −µpµ0(es − et(p)) for p ∈ P and µq = 0 for any other q ∈ P τ . Recall that µp denotes the coefficient in the expression µ0 p = µpµ0(es(p) −et(p)). Lemma 2.3.2. Keep the above notation. Then fµ = fµ ◦ F . k(Qτ )c F f µ kQc / kQ′c fµ Proof. We need only to prove fµ(q) = fµ(F (q)) for any q ∈ P τ and this can be checked case by case directly. Case 1. fµ(q) = fµ(F (q)). If q ∈ P or q has the form ei1,i2 ei2,i3 · · · ein−1,in , then it is easy to check that 14 / / / o o / / ) ) / / / ) ) / Case 2. Let q = pei1,i2 ei2,i3 · · · ein−1,in with p ∈ P . Then by definition qi ∈P τ fµ(q) = Xq=q1 q2 ···qm = Xp=p1p2···pm = Xp=p1p2···pm = Xp=p1p2···pm pi∈P pi∈P pi∈P F (µq1 (cid:3) · · · (cid:3)µqm ) F (µp1 (cid:3) · · · (cid:3)µpm (cid:3)µei1 ,i2 (cid:3) · · · (cid:3)µein−1,in ) F (µp1 (cid:3) · · · (cid:3)µpm) F (µp1 (cid:3) · · · (cid:3)µpm) = fµ(p) = fµ(F (q)). Case 3. Let q = es,s(p)p with p ∈ P . F (µq1 (cid:3)µq2 (cid:3) · · · (cid:3)µqm ) (F (µes,s(p)p1 (cid:3)µp2 (cid:3) · · · (cid:3)µpm) + F (µes,s(p) (cid:3)µp1 (cid:3)µp2 (cid:3) · · · (cid:3)µpm )) pi∈P qi∈P τ fµ(q) = Xq=q1 q2 ···qm = Xp=p1p2···pm = − Xp=p1p2···pm + Xp=p1p2···pm + Xp=p1p2···pm pi∈P pi∈P pi∈P µp1 F (µes,t(p1 ) (cid:3)µp2 (cid:3)µp3 (cid:3) · · · (cid:3)µpm) F (µes,s(p) (cid:3)µ0 p1 (cid:3)µp2 (cid:3) · · · (cid:3)µpm) F (µes,s(p) (cid:3)µ1 p1 (cid:3)µp2 (cid:3) · · · (cid:3)µpm). Now by definition of F we have F (µes,s(p) (cid:3)µ1 p1 (cid:3)µp2 (cid:3) · · · (cid:3)µpm) = −F (µ1 p1 (cid:3)µp2 (cid:3) · · · (cid:3)µpm) F (µes,s(p) (cid:3)µ0 p1 (cid:3)µp2 (cid:3) · · · (cid:3)µpm) = µp1 µp2 · · · µpm µes,s(p) . and Hence we have (cid:3)µp3 (cid:3)µ0 p3 (cid:3)µp2 (cid:3)µ0 p2 (cid:3) · · · (cid:3)µpm) + F (µes(p1 ),t(p1 ) − F (µes,t(p1 ) (cid:3) · · · (cid:3)µ0 pm) + F (µes(p1 ),t(p1 ) = − F (µes,t(p1 ) = − µp2 µp3 · · · µpm µes,t(p1 ) + µp2 µp3 · · · µpm µes(p1 ),t(p1 ) = − µp2 µp3 · · · µpm µes,s(p1 ) . (cid:3)µp2 (cid:3)µ0 p2 (cid:3) · · · (cid:3)µpm) (cid:3) · · · (cid:3)µ0 pm) Therefore we obtain that fµ(q) = Xp=p1p2···pm = Xp=p1p2···pm pi∈P pi∈P (−F (µ0 p1 (cid:3)µp2 (cid:3) · · · (cid:3)µpm) − F (µ1 p1 (cid:3)µp2 (cid:3) · · · (cid:3)µpm)) −F (µp1 (cid:3)µp2 (cid:3) · · · (cid:3)µpm) = − fµ(p) = fµ(F (q)) 15 The same argument also works in the case q = es,s(p)pet(p),i1 ei1,i2 · · · ein−1,in . Case 4. Let q = q′et(q′),s(p)p with q′ ∈ P τ and p ∈ P . fµ(q) = Xq=q1 q2 ···qm = Xq′ =q′ qi∈P τ 1q′ 2 ···q′ l p=p1p2···pn q′ ∈P τ ,pi∈P j F (µq1 (cid:3)µq2 (cid:3) · · · (cid:3)µqm ) nF (µq′ 1 (cid:3) · · · (cid:3)µq′ l (cid:3)µet(q′ ),s(p) (cid:3)µp1 (cid:3) · · · (cid:3)µpm ) −F (µq′ 1 (cid:3) · · · (cid:3)µq′ l (cid:3)µet(q′ ),s(p)p1 (cid:3)µp2 Note that (cid:3) · · · (cid:3)µpm )o . F (µq′ 1 (cid:3) · · · (cid:3)µq′ l =F (µq′ 1 (cid:3) · · · (cid:3)µq′ l =F (µq′ 1 (cid:3) · · · (cid:3)µq′ l (cid:3)µet(q′ ),s(p) (cid:3)µet(q′ ),s(p) (cid:3)µ0 (cid:3) · · · (cid:3)µpc) (cid:3)µp1 (cid:3) · · · (cid:3)µ0 (cid:3)µ0 p1 (cid:3)µ0 p1 µet(q′ ),t(p1 ) p2 pc) (cid:3) · · · (cid:3)µ0 pc), it follows that fµ(q) = 0 = fµ(F (q)) in this case. The same argument works for the case q = q′et(q′),s(p)pet(p),s(q′′)q′′ with q′, q′′ ∈ P τ and p ∈ P . Now we have exhausted all cases and the proof is completed. Definition 2.3.3. Let Q, Q′, Q′′ be quivers, µ ∈ Ω(Q, Q′) and ν ∈ Ω(Q′, Q′′). The compo- sition of µ and ν, denoted by ν ◦ µ, is defined to be the trans-datum in Ω(Q, Q′′) given by (ν ◦ µ)0 = ν0 ◦ µ0, and for each nontrivial path p, (ν ◦ µ)p = Xp=p1p2···pn pi∈P ν(µp1 (cid:3) · · · (cid:3)µpn ), (2.2) here each µp1 is viewed as an element in kQ′τ above lemma. 1 and ν ∈ Ω(Q′τ , Q′′) is the lift of ν as in the Before stating our main theorem in this section, we give a result in special case. Lemma 2.3.4. Let Q, Q′, Q′′ be quivers, µ ∈ Ω(Q, Q′) and ν ∈ Ω(Q′, Q′′). Assume that µ0 = 0. Then fν ◦ fµ = fν◦µ. Proof. Since µp1 have F (µp1 the lemma follows easily by direct calculation. In fact, we have (cid:3) · · · (cid:3)µpm is a linear combination of paths in kQ′, by definition we (cid:3) · · · (cid:3)µpm). Now (cid:3) · · · (cid:3)µpm and ν(µp1 (cid:3) · · · (cid:3)µpm) = ν(µp1 (cid:3) · · · (cid:3)µpm) = µp1 (cid:3)µp2 (fν ◦ fµ)(p) = Xp=p1p2···pm,pi∈P = Xp=p1p2···pm X1≤i1<···<ir<m pi∈P fν (µp1 (cid:3)µp2 (cid:3) · · · (cid:3)µpm) F (ν(µp1 (cid:3) · · · (cid:3)µpi1 )(cid:3) · · · (cid:3)ν(µpir +1 (cid:3) · · · (cid:3)µpm)) 16 F ((ν ◦ µ)p1 (cid:3)(ν ◦ µ)p2 (cid:3) · · · (cid:3)(ν ◦ µ)pm ) and fν◦µ(p) pi∈P = Xp=p1p2 ···pm = Xp=p1p2 ···pm = Xp=p1p2 ···pm pi∈P pi=pi1pi2···piri Xi=1,2,··· ,m Xi=1,2,··· ,m pi∈P pi=pi1pi2···piri F (ν(µp11 (cid:3) · · · (cid:3)µp1r1 )(cid:3) · · · (cid:3)ν(µpm1 (cid:3) · · · (cid:3)µpmrm )) F (ν(µp11 (cid:3) · · · (cid:3)µp1r1 )(cid:3) · · · (cid:3)ν(µpm1 (cid:3) · · · (cid:3)µpmrm )) for each p ∈ P , and therefore fν◦µ(p) = (fν ◦ fµ)(p) by comparing the summations. Combining the above lemmas, we have the following characterization. Theorem 2.3.5. Let Q, Q′, Q′′ be arbitrary quivers and µ ∈ Ω(Q, Q′) and ν ∈ Ω(Q′, Q′′) trans-data. Then fν ◦ fµ = fν◦µ. In particular, (Ω(Q), ◦) is a monoid with identity 1 and f : Ω(Q) ∼=−−→ Coalg(kQc, kQc) is an isomorphism of monoids. Proof. The equality follows easily from the commutative diagram f e d b a ` _ ^ ] \ [ Y X k(Q′τ )c F kQc fµ f ¯µ X Y Z \ ] ^ _ ` a b d e f kQ′c fν / kQ′′c , note that here we use the fact ν ◦ µ = ν ◦ ¯µ. The rest part is easy. fν Remark 2.3.6. Since the composition of coalgebra morphisms obeys the associative law, so is the composition of trans-data. Explicitly, let Q, Q′, Q′′, Q′′′ be any given quivers and µ ∈ Ω(Q, Q′), ν ∈ Ω(Q′, Q′′) and λ ∈ Ω(Q′′, Q′′′) be trans-data. Then λ ◦ (ν ◦ µ) = (λ ◦ ν) ◦ µ. This can also be checked directly from the definition. 3 The automorphism group of a path coalgebra Homomorphisms of path coalgebras are characterized in terms of trans-data in last section. In the rest part of this paper, we will restrict our interest to the automorphism group. We begin with the study of certain subgroups, which we suppose to be helpful in understanding the structure of the whole group. 3.1 A tower of normal subgroups of Aut(kQc) Recall that for any coalgebra C, we denote by Aut(C) the automorphism group of C and set Aut0(C) = {σ ∈ Aut(C) σC(0) = IdC(0) }, where C(0) is the coradical of C. Let Q be an arbitrary quiver. Set Ω∗(Q) to be the group of multiplicative invertible elements in Ω(Q), and hence Ω∗(Q) ∼= Aut(kQc) under f and . By Proposition 2.1.2, Ω∗(Q) = {(µ0, µ+) µ0 ∈ Aut(kQ0), µ1kQ1 ∈ GLk(kQ1)}. 17 / / + + / / 3 3 / Notice that each f ∈ Aut(kQc) induces a permutation of Q0. We may consider the subgroup Aut0(kQc) = {σ ∈ Aut(kQc) σ(ei) = ei ∀i ∈ Q0}. Denote by Ω∗ 0(Q) the subgroup of Ω∗(Q) corresponding to Aut0(C). Obviously Ω∗ 0(Q) = {(µ0, µ+) ∈ Ω∗(Q) µ0 = IdkQ0 }. By an automorphism of a quiver Q, we mean a permutation σ of Q0 such that (Q1)s,t and (Q1)σs,σt have the same cardinality for any s, t ∈ Q0. Denote by Aut(Q) the group of automorphisms of Q. Recall a classical result which says that the automorphism group of a finite dimensional associative algebra is a finite dimensional linear algebraic group. Thus if Q is finite acyclic, then kQ is finite dimensional and hence Aut(kQc) is also a finite dimensional linear algebraic group, comparing with Lemma 5.1.1 below. The following result is easy. Proposition 3.1.1. Let Q be any quiver and C = kQc. Then Aut0(C) is a normal subgroup of Aut(C) and the quotient group Aut(C)/ Aut0(C) ∼= Aut(Q). If in addition, Q is finite acyclic, then Aut0(C) is the identity component of Aut(C). Proof. Consider the restriction map Res0 : Aut(C) −→ Aut(C0). By definition, any auto- morphism of Q extends to an automorphism of C and hence Res0 is surjective. Clearly Ker(Res0) = Aut0(C) and it follows that Aut(C)/ Aut0(C) ∼= Aut(Q). Now assume that Q is finite acyclic. To show the connectedness of Aut0(C) one uses the fact that GL(V ) is connected for any finite dimensional vector space V . Let σ ∈ Aut0(C) and µ = σ the corresponding trans-datum. Now we define a family of trans-data µt = (Id, (µt)+) parameterized by t ∈ [0, 1], where µt is given by setting (µt)α = tµ0 α for each α ∈ Q1, and (µt)p = tµp for any p ∈ Q≥2. Thus µ transforms continuously to a datum (Id, ν+) such that να ∈ kQ1 for α ∈ Q1 and νp = 0 for p ∈ Q≥2. Such data form a subgroup of Aut(C) α + µ1 which is isomorphic to Qs,t∈Q0 GL((kQ1)s,t). The connectedness of Aut0(C) follows. If Q is finite acyclic, then Aut(C) is a finite dimensional linear group. We know that Aut0(C) is a closed subgroup, for "fixing points" is a closed condition. Thus Aut0(C) is a connected closed normal subgroup, and hence the identity component of Aut(C), which completes the proof. For any coalgebra D and each integer n ≥ 1, we denote by Autn(D) the subgroup of Aut(D) consisting of automorphisms which act on D(n) trivially. Clearly, for a quiver Q, Autn(kQc) corresponds to Ω∗ n(Q) = {(IdQ0 , µ+) ∈ Ω∗(Q) µα = α, ∀α ∈ Q1, µp = 0, ∀p ∈ Q, 2 ≤ l(p) ≤ n}, here l(p) denotes the length of p. For consistence of notations, we also set Ω∗ 1/2(Q) = {(IdQ0, µ+) ∈ Ω∗ 0(Q) µ1 α = α, ∀α ∈ Q1}. Now we have a more generalized result of Propostion 3.1.1. Proposition 3.1.2. Let Q be a quiver and set C = kQc. Then we have a tower of normal subgroups of Ω∗(Q) 0(Q) ⊲ Ω∗ 1/2(Q) ⊲ Ω∗ Ω∗(Q) ⊲ Ω∗ 1(Q) ⊲ Ω∗ n(Q) ∼= Aut(C(n)) for each n ≥ 1 and Ω∗(Q)/Ω∗ 2(Q) ⊲ · · · , 1/2(Q) ∼= GrAut(C(1)). More- and Ω∗(Q)/Ω∗ 1/2(Q) ∼= GrAut0(C(1)), here the over, we have Ω∗ group GrAut(C(1)) consists of graded automorphisms of C(1) and GrAut0(C(1)) the subgroup of graded automorphisms fixing C0. n(Q) ∼= Aut0(C(n)) and Ω∗ 0(Q)/Ω∗ 0(Q)/Ω∗ 18 Proof. By Lemma 1.2.2, each automorphism of C remains to be an automorphism when restricted to each C(n). Thus the restriction map gives a natural group homomorphism Resn : Ω∗(Q) −→ Ω∗ n(Q) for each n ≥ 1. Again Theorem 2.2.6 tells us that the map Resn is surjective, and clearly Ker(Resn) = Ω∗ n(Q), thus Ω∗(Q)/Ω∗ n(Q) ∼= Aut(C(n)). We can associate each trans-datum µ ∈ Ω∗(Q) a linear endomorphism Φµ : C(1) −→ C(1), ei 7→ µ0(ei), α 7→ µ1 α. In fact, Φµ is It is direct to show that Φµ is a graded coalgebra automorphism of C(1). given by the restriction of the automorphism of C which corresponds to the trans-datum (µ0, µ1). Thus we get a map Φ : Ω∗(Q) −→ GrAut(C(1)), which is easily shown to be a group homomorphism. Again Φ is an epimorphism by Theorem 2.2.6 and Ker(Φ) = Ω∗ 1/2(Q), thus Ω∗(Q)/Ω∗ 1/2(Q) ∼= GrAut(C(1)). For the rest part we use the same argument as in the proof of Proposition 3.1.1. 3.2 Factor groups and solvability of Aut(kQc) In the sequel, the factor groups of the above filtration will be discussed in more detail. The quiver Q is assumed to be finite, so the factor groups can be characterized by the dimension. The following easy lemmas are useful in studying the solvability of the automorphism group of path coalgebras. Recall that an algebraic group is solvable if and only if it is solvable as an abstract group. Lemma 3.2.1. Assume Q is finite. Then Ω∗ for any n ≥ 2, where d1 = Ps,t∈Q0 s6=t (Q1)s,t and dn = Ps,t∈Q0 denotes the additive group of a di-dimensional k-vector space. 1/2(Q)/Ω∗ 1(Q) ∼= kd1 and Ω∗ (Qn)s,t · (Qτ n−1(Q)/Ω∗ n(Q) ∼= kdn 1 )s,t, and each kdi Proof. The proof is given by direct calculation. Note that Formula (2.2) will be quite sim- plified in this case. Let n ≥ 2 and µ, ν be in Ω∗ n−1(Q). Then µ ◦ ν ∈ Ω∗ n−1(Q) and (µ ◦ ν)p = Xp=p1p2···pm,pi∈P µ(νp1 (cid:3)νp2 (cid:3) · · · (cid:3)νpm) = µp + νp n−1(Q)/Ω∗ for all p ∈ P with length n. Thus the quotient group Ω∗ group of some k-vector space. The dimension dn is computed by n(Q) is given by the additive dn = Xp∈Q,l(p)=n dimk Ps(p),t(p) = Xs,t∈Q0 (Qn)s,t · (Qτ 1)s,t. The case Ω∗ 1/2(Q)/Ω∗ 1(Q) is proved similarly. Remark 3.2.2. In fact, for any quiver Q we always have (abstract) group isomorphisms Ω∗ 1/2(Q)/Ω∗ Homk(k(Q1)s,t, Ps,t), 1(Q) ∼= Ys,t∈Q0 n(Q) ∼= Ys,t∈Q0 and Ω∗ n−1(Q)/Ω∗ Homk(k(Qn)s,t, Ps,t) for all n ≥ 2, where the right hand sides are both viewed as an additive group. 19 Lemma 3.2.3. Suppose Q is finite. Then GrAut0(C(1)) ∼=Qs,t∈Q0 GL(k, (Q1)s,t). Combining Proposition 3.1.2, Lemma 3.2.1 and Lemma 3.2.3, we can calculate the di- mension of the automorphism group of path coalgebras. Proposition 3.2.4. Assume Q is finite. Then Aut(kQc ≤n) is a linear algebraic group with dim(Aut(kQc ≤n)) = Xs,t∈Q0 (Qτ 1 )s,t · (P≤n)s,t for any n ≥ 1, here (P≤n)s,t = {p ∈ Q s(p) = s, t(p) = p, 1 ≤ l(p) ≤ n}. If moreover, Q is finite acyclic, then dim(Aut(kQc)) = Xs,t∈Q0 ((Q1)s,t + 1) · Ps,t. Recall that by a Schurian quiver we mean a quiver Q such that (Q1)s,t ≤ 1 for any s, t ∈ Q0. We end this section with the following characterization. Theorem 3.2.5. Let Q be a finite quiver. Then Aut0(kQc if and only if Q is a Schurian quiver, if and only if Aut0(kQc Aut(kQc resolvable, if and only if Aut(kQc ≤n) is solvable for some n ≥ 2 ≤n) is solvable for all n ≥ 2; ≤n) is solvable for some n ≥ 2 if and only if Q is a Schurian quiver and Aut(Q) is ≤n) is solvable for all n ≥ 2. Proof. Given n ≥ 2, by Proposition 3.1.2 and Lemma 3.2.1 we have Aut0(kQc ≤n) is solvable if and only if GrAut(C(1)) is solvable. Note that the general linear group GL(k, d) is solvable if and only if d = 1. Thus by Lemma 3.2.3, GrAut(C(1)) is solvable if and only if for each pair (s, t) in Q0, (Q1)s,t ≤ 1, if and only if Q is Schurian. Now by Proposition 3.1.1, Aut(Q) is solvable if and only Aut0(kQc) and Aut(Q) are both solvable, if and only if Q is Schurian with Aut(Q) solvable. This completes the proof. Consequently we may obtain a dual version of the result, say a classification of truncated path algebras with solvable automorphism group; see Proposition 5.4.2 below. 4 A Galois theory for path coalgebras In this section, we will develop a Galois-like theory for path coalgebras extensions, which aims to give a connection between the automorphism groups of subcoalgebras of a path coalgebra and certain subgroups of its automorphism group. Recall that in Propostion 3.1.2, we have already shown for any path coalgebra a tower of normal subgroups of its automorphism group induced by the coradical filtration. Our Galois-like theory will generalize this fact. Note that Theorem 4.3.2 and some useful lemmas in this section hold true only for acyclic quivers. 4.1 The Galois group of a coalgebra extension We begin with a more general setup, although we mainly deal with Galois groups of pointed coalgebra extensions. 20 Definition 4.1.1. Let E be a coalgebra and D ⊆ E a subspace, the set of automorphisms {σ ∈ Aut(E) σ(d) = d, ∀d ∈ D} forms a subgroup of Aut(E), which we call the Galois group of E over D and denote by Gal(E/D). Conversely, for any H ≤ Aut(E), we set Inv(H) = {c ∈ E σ(c) = c, ∀σ ∈ H}, the subspace of fixed points of H. The following lemma is classical. The proof is standard and we omit it here. Lemma 4.1.2. Let E be a coalgebra. Let H, H ′ be subgroups of Aut(E) and D, D′ subspaces of E. Then (1) H ≤ H ′ =⇒ Inv(H ′) ⊆ Inv(H) and D ⊆ D′ =⇒ Gal(E/D′) ⊆ Gal(E/D); (2) Inv(Gal(E/D)) ⊇ D and Gal(E/ Inv(H)) ≥ H; (3) Gal(E/) ◦ Inv ◦ Gal(E/) = Gal(E/) and Inv ◦ Gal(E/) ◦ Inv = Inv. Remark 4.1.3. Note that in general, Inv(H) is not a subcoalgebra of E. For example, let Q be the quiver 1• α−−→ 2• −−→ 3• and E = kQc the pathcoalgebra. Assume that there exists some a ∈ k, a /∈ {0, 1}. Consider the automorphism σ ∈ Aut(E) given by σ(ei) = ei, i = 1, 2, 3, σ(α) = aα, σ(β) = a−1β and σ(αβ) = αβ. Let H = hσi the subgroup generated by σ. Then it is direct to check that Inv(H) = ke1 ⊕ ke2 ⊕ ke3 ⊕ kαβ, which is not a subcoalgebra. β Now we turn to the path coalgebra case. Let Q be a quiver and C = kQc. We are only interested in large subcoalgebras of C and their Galois groups, since each pointed coalgebra can be realized as a large subcoalgebra. Set C (C) = {D C(1) ⊆ D ⊆ C}, the set of large subcoalgebras of C. A subgroup H ≤ Aut(C) is called an admissible subgroup if H = Gal(C/D) for some D ∈ C (C), and the set of all admissible subgroups of Aut(C) is denoted by G (C). Set Autn(C) = Gal(C/C(n)) for each n ≥ 0, comparing with the notation Ω∗ n(Q) in Section 3. Clearly any H ∈ G (C) is a subgroup of Gal(C/C(1)). For D ∈ C (C), we have D = kQ0 ⊕ kQ1 ⊕ D≥2, where D≥2 = D ∩ kQ≥2. Recall that each µ ∈ Ω(Q) gives k-linear maps µ0 : kQ → kQ0 and µ+ : kQ −→ kQ0 ⊕ kQ1, see Section 2.1. Then we have the following characterization for Galois groups. Proposition 4.1.4. Let Q be an arbitrary quiver, C = kQc the path coalgebra and D ∈ C (C) a large subcoalgebra. Then (Gal(C/D)) = {µ = (IdkQ0 , µ+) µα = α ∀α ∈ Q1, µ+(x) = 0 ∀x ∈ D≥2}. Proof. Suppose that µα = α for all α ∈ Q1 and µ+(x) = 0 for all x ∈ D≥2. Then by Proposition 2.1.3, fµ(x) =Xr≥1 F (µ+(x(1))(cid:3) · · · (cid:3)µ+(x(r))) = f 1(x), for in this case µ+(x) = 1+(x) for any x ∈ kQ. Recall that 1 is the identity trans-datum which corresponds to IdkQc . Conversely, suppose that σ ∈ Gal(C/D) and set µ = σ. For x ∈ D, we denote by l(x) the minimal integer such that x ∈ D(l(n)), where D(l(n)) is the l(n)-th coradical of D. Now we prove that µ(x) = 0 for all x ∈ D≥2 by induction on l(x). Clearly µ0 = IdQ0 and µα = α for all α ∈ Q1. Assume that µ+(x) = 0 = 1+(x) for all x ∈ D≥2 with l(x) < n. Take any x ∈ D≥2 with l(x) = n, Proposition 2.1.3 says that σ(x) =Xr≥1 F (µ+(x(1))(cid:3) · · · (cid:3)µ+(x(r))) = µ+(x) +Xr≥2 F (µ+(x(1))(cid:3) · · · (cid:3)µ+(x(r))), 21 and 1(x) =Xr≥1 F (1+(x(1))(cid:3) · · · (cid:3)1+(x(r))) = 1+(x) +Xr≥2 f F (1+(x(1))(cid:3) · · · (cid:3)1+(x(r))). Note that for r ≥ 2, we always have X F (µ+(x(1))(cid:3) · · · (cid:3)µ+(x(r))) =X F (1+(x(1))(cid:3) · · · (cid:3)1+(x(r))), the reason is that for each term in the sum, either l(x(i)) < n for all i, or there exists some xi with l(xi) = 0. Since σ ∈ Gal(C/D), σ(x) = x = f 1(x), and hence µ+(x) = 1+(x) = 0. To give the following key lemma, we need some notation. We define a partial ordering on the set of all paths in Q. For each pair of paths p, q ∈ Q, by p ≤ q we mean that p is a subpath of q, and by p < q we mean that p is a proper subpath of q. Lemma 4.1.5. Let Q be an acyclic quiver, C = kQc and D a large subcoalgebra of C. Then Inv(Gal(C/D)) = D. Consequently, we have a well-defined map Inv : G (C) −→ C (C). Proof. Clearly D ⊆ Inv(Gal(C/D)). We need only to show that Inv(Gal(C/D)) ⊆ D. Assume that x ∈ Inv(Gal(C/D)) \ D. We may write x =Pj λj pj, a linear combination of paths, here pj ∈ Q and 0 6= λj ∈ k for each j. Since x /∈ D, there exists some i such that pi /∈ D. We choose such a pi with maximal length. Clearly pi (cid:2) pj for any other pj, otherwise we have either pj /∈ D with length strictly greater than pi, or pj ∈ D and hence pi ∈ D, which leads to a contradiction in either case. Moreover, by assumption Q is acyclic and D is a large subcoalgebra, we know that s(pi) 6= t(pi). Let D′ denote the subcoalgebra generated by D and all pj's. We define σ ∈ Aut(D′) as follows: σ(pi) = pi + es(pi) − et(pi); σ(x) = x for any x < pi; σ(x) = x, ∀j 6= i, ∀x ≤ pj; It is direct to check that σ is an automorphism of D′, and by and σ(d) = d, ∀d ∈ D. Theorem 2.2.6, σ extends to σ′ ∈ Aut0(C). By construction σ′ ∈ Gal(C/D) and σ′(x) = x + es(pi) − et(pi) 6= x and hence x /∈ Inv(Gal(C/D)). Remark 4.1.6. The lemma does not hold true in general when the quiver Q contains oriented cycles. For example, consider the quiver 1• α β / •2 . Let C be the path coalgebra. Then by direct calculation one shows that C(2) ⊆ Inv(Gal(C/C(1))). 4.2 Galois extensions Let D, E be coalgebras over a field k. If D is a subcoalgebra of E with E(1) ⊆ D, then we say that E is an admissible extension over D, and write as E/D. Let Q be a quiver, C = kQc the path coalgebra and D ≤ C a large subcoalgebra. Assume E/D is an admissible extension. By the universal property of C, there exists an embedding E ⊆ C which is compatible with the embedding D ⊆ C. Roughly speaking, C plays a similar role as an algebraic closure of a field in field theory. Based on this fact, by an admissible extension E/D, we always mean a intermediate subcoalgebra D ⊆ E ⊆ C. The following definition makes sense now. Definition 4.2.1. Let D be a large subcoalgebra of C. An admissible extension E/D is called a Galois extension, if for any σ ∈ Gal(C/D), σ(E) = E. 22 / o o Note that C/D is a Galois extension for any large subcoalgebra D. Applying Lemma 4.1.5, we have the following Galois-like correspondence. Proposition 4.2.2. Let Q be an acyclic quiver and C = kQc the path coalgebra. Then (1) The maps Inv : G (C) −→ C (C) and Gal(C/) : C (C) −→ G (C) are well defined and inverse to each other. (2) Assume that D, E ∈ C (C) and set H = Gal(C/D) and G = Gal(C/E). Then Inv(H ∪ G) = D ∩ E and Inv(H ∩ G) = D + E, here H ∪ G denotes the subgroup of Aut(C) generated by H and G. (3) Let H be in G (C). Then H ⊳ Gal(C/C(1)) if and only if Inv(H) is invariant under Inv(H) is a Galois extension over C(1); and in this case, the action of Gal(C/C(1)), i.e. Gal(C/C(1))/H ∼= Gal(Inv(H)/C(1)). Proof. (1) Lemma 4.1.5 shows that Inv(Gal(C/D)) = D and Inv is well defined. We need only to show that H = Gal(C/ Inv(H)) for any H ∈ G (Q). This is easy, for each H ∈ G (C) is of the form Gal(C/D) for some D ∈ C (C). Thus by Lemma 4.1.2, H = Gal(C/D) = Gal(C/ Inv(Gal(C/D))) = Gal(C/ Inv(H)). (2) First we show that Inv(H ∪ G) = D ∩ E. By Lemma 4.1.2, Inv(H ∪ G) ⊆ Inv(H) and Inv(H ∪G) ⊆ Inv(H) and hence Inv(H ∪G) ⊆ D ∩E. Conversely, for any c ∈ D ∩E, c is fixed by any elements in H and G and hence fixed by H ∪ G, which implies that c ∈ Inv(H ∪ G). Thus we have proved Inv(H ∪ G) = D ∩ E. Next we show H ∩ G = Gal(C/(D + E)). Again by Lemma 4.1.2, Gal(C/(D + E)) ⊆ H and Gal(C/(D + E)) ⊆ G and hence Gal(C/(D + E)) ⊆ H ∩ G. Now let σ ∈ H ∩ G, σ fixes D and E and hence fixes D + E. It follows that H ∩ G = Gal(C/(D + E)) and hence Inv(H ∩ G) = D + E. (3) If Inv(H) is an invariant subspace under the action of Gal(C/C(1)), then for any σ ∈ Gal(C/C(1)) and any h ∈ H, (σ−1hσ)(c) = (σ−1(σ(c)) = c for any c ∈ Inv(H), which implies that σ−1hσ ∈ Gal(C/ Inv(H)) = H and hence H ⊳ Aut1(C). Conversely, suppose that H ⊳ Gal(C/C(1)). For any σ ∈ Gal(C/C(1)), we claim that σ(Inv(H)) ⊆ Inv(H). Otherwise there exists some c ∈ Inv(H) with σ(c) /∈ Inv(H). Now there exists some h ∈ H such that h(σ(c)) 6= σ(c). Thus σ−1(h(σ(c))) 6= σ−1(σ(c)), and hence σ−1hσ /∈ H, which leads to a contradiction. Similarly σ−1(Inv(H)) ⊆ Inv(H) and hence σ(Inv(H)) = Inv(H). We are only left to show the isomorphism of Galois groups. Set D = Inv(H). We have a group homomorphism f : Gal(C/C(1)) −→ Gal(D/C(1)) induced by restriction, which is well-defined since D is invariant under the action of Gal(C/C(1)). Theorem 2.2.6 says that any automorphism of D extends to an automorphism of C, which implies that f is surjective. Now Gal(C/C(1))/H ∼= Gal(D/C(1)) follows easily from the fact Ker(f ) = Gal(C/D). γ Example 4.2.3. Let Q be the quiver 1• α−→ 2• −→ 4• and C = kQc the path coalgebra. There are exactly 5 subcoalgebras of C containing C(1). Say C0 = C(1), C1 = C(1) + kαβ, C2 = C(1) + kβγ, C1,2 = C(2), and C3 = C. β −→ 3• The automorphism group Aut1(C) ∼= k3 = {(x, y, z) x, y, z ∈ k}, the additive group of 3-dimensional k-space, where (x, y, z) corresponds to the automorphism given by αβ 7→ αβ + x(e1 − e3), βγ 7→ βγ + y(e2 − e4) and αβγ 7→ αβγ − xγ + yα + z(e1 − e4). Compare with Proposition 5.2.6 and Example 6.1.1. 23 Set H 0 = {(x, y, z)}, H 1 = {(0, y, z)}, H 2 = {(x, 0, z)}, H 1,2 = {(0, 0, z)} and H 3 = {(0, 0, 0)}. By direct calculation, Gal(C/C ∗) = H ∗ for any ∗ ∈ {0, 1, 2, (1, 2), 3}. Pictorially we have the following correspondence. C (C) C G (C) {(0, 0, 0)} C(1) + k{αβ, βγ} C(1) + kαβ }}}}}}}} AAAAAAAA C(1) AAAAAAA C(1) + kβγ }}}}}}}} {(0, 0, z)} {(x, 0, z)} {(0, y, z)} <<<<<<< <<<<<<< {(x, y, z)} Remark 4.2.4. Recall that Proposition 4.1.4 has determined the trans-data given by Gal(C/D) for a large subcoalgebra D. It is natural to ask for an intrinsic characterization of admissibil- ity of subgroups of Aut(C), which could be helpful in our Galois-like theory. For instance, an admissible subgroup H ∈ G (C) must be a closed subgroup in the Zariski sense, for the con- dition fixing a point is a "closed" condition and therefore a stability group is always closed. Up to now, no sufficient and necessary condition is known to us. 4.3 Fundamental theorem of Galois extensions We move to a more general case now. Again we set C = kQc, the path coalgebra of a given quiver Q. Let D be a large subcoalgebra of D and E/D an admissible extension. Set C (D, E) = {M D ⊆ M ⊆ E}, the set of intermediate subcoalgebras of D and E, and G (D, E) = {Gal(E/M ) M ∈ C (D, E)}, the set of admissible subgroups of Aut(E). Lemma 4.3.1. Let Q be an arbitrary quiver, C = kQc and D ∈ C (C). Assume that E/D is a Galois extension and M ∈ C (D, E). Then E/M is a Galois extension, and Gal(E/M ) ∼= Gal(C/M )/ Gal(C/E). Proof. Firstly we show that E/M is a Galois extension. By definition, it suffices to show that any for σ ∈ Gal(C/E), σ(E) ⊆ E. This follows easily from the facts Gal(C/E) ≤ Gal(C/D) and E/D is a Galois extension. To show Gal(E/M ) ∼= Gal(C/M )/ Gal(C/E), we use the same argument as in the proof of Proposition 4.2.2. Consider the group homomorphism Res : Gal(C/M ) −→ Gal(E/M ) induced by restriction, which is well defined since E/M is a Galosis extension and E is invariant under the action of Gal(C/M ). Applying Theorem 2.2.6 again we show that Res is surjective. Moreover, it is easy to show that Ker(Res) = Gal(C/E) and hence Gal(E/M ) ∼= Gal(C/M )/ Gal(C/E) by the isomorphism theorem of groups. Now we have a more general version of Proposition 4.2.2, which we call the fundamental theorem of Galois extensions. 24 Theorem 4.3.2. Let Q be an acyclic quiver and C = kQc the path coalgebra. Let D be in C (C) and E/D a Galois extension. Then (1) The map Inv : G (D, E) → C (D, E) is well defined and gives the inverse of the map Gal(E/) : C (D, E) → G (D, E). (2) Assume that M, N ∈ C (D, E) and set H = Gal(E/M ) and G = Gal(E/N ). Then Inv(H ∪ G) = M ∩ N and Inv(H ∩ G) = M + N , here H ∪ G denotes the subgroup of Aut(E) generated by H and G. (3) Let H be in G (D, E). Then H ⊳ Gal(E/D) if and only if Inv(H) is a Galois extension over D; and in this case, Gal(E/D)/H ∼= Gal(Inv(H)/D). Proof. (1) and (2) follow directly from Proposition 4.2.2 and the last lemma. We need only to check (3). The argument is similar to the one in the proof of Proposition 4.2.2(3). Set M = Inv(H). First assume that H ⊳ Gal(E/D). Let σ ∈ Gal(C/D). We claim that σ(M ) ⊆ M . Otherwise, there exists some m ∈ M such that σ(m) /∈ M , and hence some h ∈ Gal(C/M ) such that h(σ(m)) 6= σ(m). Thus σ−1hσ(m) 6= σ−1σ(m) = m, and hence σ−1hσ /∈ Gal(C/M ). We denote by ¯σ and ¯h the restriction of σ, h to E respectively. Thus ¯σ and ¯h can be viewed as elements in Gal(E/D), since E/D is a Galois extension. Moreover, ¯h ∈ H by the choice of h. Now ¯σ−1¯h¯σ /∈ H, which is contradict to the assumption H ⊳ Gal(E/D). Thus M is Gal(C/D)-invariant, hence M/D is a Galois extension. Conversely, if M is Gal(C/D)-invariant, then clearly M is Gal(E/D)-invariant. Thus (σ−1hσ)(m) = σ−1(σ(m)) = m for any σ ∈ Gal(E/D), h ∈ H and m ∈ M . It follows that σ−1hσ ∈ H and hence H ⊳ Gal(E/D). To show the group isomorphism we need the map Res: Gal(E/D) → Gal(M/D), the group homomorphism induced by restriction. Res is well defined since M/D is a Galois extension. To show Res is surjective, we use the fact that any σ ∈ Gal(M/D) extends to some σ′ ∈ Gal(C/D). By assumption E/D is a Galois extension, the restriction of σ′ to E can be viewed as an element in Gal(E/D), which maps to σ under Res. It is easy to show that Ker(Res) = Gal(E/M ) and Gal(E/D)/ Gal(E/M ) ∼= Gal(M/D) follows. Remark 4.3.3. The requirement that Q is acyclic is essential in giving one to one correspon- dence between G (C) and C (C); see Remark 4.1.6. For general quivers, one also expects a similar correspondence between certain class of subcoalgebras of C and certain class of subgroups of Aut(C), but what subcoalgebras and what subgroups should be involved is still unclear to us. Remark 4.3.4. We mention that even for a Galois extension E/D, an automorphism of D may not extend to an automorphism of E. Define ResE Ker(ResE D) = Gal(E/D). The above remark just says that ResE Let E/D be an admissible extension and Aut(E; D) = {σ ∈ Aut(E) σ(D) = D}. D : Aut(E; D) −→ Aut(D) to be the map induced by restriction. Obviously, D is not surjective in general. Before going further we introduce some notation. Let E/D be an admissible extension and M ⊆ E a large subcoalgebra. Set Gal(E/M ; D) = Gal(E/M ) ∩ Aut(E; D). Now we have the following result on automorphism groups of pointed coalgebras, by applying the correspondence given by taking Galois group and fixed points. Proposition 4.3.5. Let Q be an arbitrary quiver and C = kQc. Let D ∈ C (C). Then 25 (1) Assume that E/D is an admissible extension such that E is Aut(C; D)-invariant. If in addition, E/D is a Galois extension, then Then Aut(D) ∼= Aut(E; D)/ Gal(E/D). ResE D is surjective if and only if E is Aut(C; D)-invariant. (2) Aut(D) ∼= Aut(C; D)/ Gal(C/D). Moreover, if D ⊆ C(n) for some n ≥ 1, then Aut(D) ∼= Aut(C(n); D)/ Gal(C(n)/D). (3) There is a tower of normal subgroups Aut(D) ⊲ Gal(D/D(0)) ⊲ Gal(D/D(1)) ⊲ · · · , and Aut(D)/ Gal(D/D(n)) ∼= Im(ResD Gal(C/C(n);D) D(n) ) for each n ≥ 0. (4) Gal(D/D(n)) Gal(D/D(m)) ∼= quotient group Gal(D/D(n)) Gal(C/(C(n)+D(m));D) , ∀m ≥ n ≥ 0. Particularly, for each n ≥ 1, the Gal(D/D(n+1)) is a subquotient group of Gal(C(n+1)/C(n)) and hence abelian. Proof. (1) Suppose that E is Aut(C; D)-invariant. Then ResE D is surjective, since any σ ∈ Aut(D) extends to some σ′ ∈ Aut(C; D), and E is σ′-invariant means that σ′ is an automorphism of E when restricted to E. Clearly Ker(ResE D) = Gal(E/D) and the isomor- phism follows. Conversely, assume that E/D is a Galois extension and ResE D is surjective. We need to show that for any σ ∈ Aut(C; D), σ(E) ⊆ E. Since ResE D is surjective, then there exists some h ∈ Aut(E, D) such that ResE E(h′) for some h′ ∈ Aut(C; E). Clearly h′ ∈ Aut(C; D). Now we have σ−1h′ ∈ Gal(C/D). By definition, E/D is a Galois extension implies that E is σ−1h′-invariant, together with the fact E is h′-invariant we obtain that E is σ-invariant. D(σ). Again by Theorem 2.2.6, h = ResC D(h) = ResC (2) This is a special case of (1). Since C and C(n), n ≥ 0 are all Aut(C)-invariant. (3) Observe that Aut(D; D(n)) = Aut(D) for any n ≥ 0. Applying the isomorphism theorem of groups to ResD D(n) (4) Note that we have ResC , we get the required isomorphism. D induces an epimorphism Gal(C/C(n); D) ։ Gal(D/D(n)), the reason is as follows. For any σ ∈ Gal(D/D(n)), we extend σ to σ′ ∈ Aut(D + C(n)) by setting σ′D = σ and σ′C(n) = IdC(n), and extend σ′ to σ′′ ∈ Aut(C). Clearly σ′′ ∈ Gal(C/C(n); D) and ResC D(σ′′) = σ. The isomorphism follows easily. It is direct to show that Gal(C/C(n);D) to be isomorphic to Gal(C(m)/C(n)). Now Gal(C/(C(m));D) is a subgroup of Gal(C/C(n)) Gal(C/(C(m))) , which is easy shown Gal(D/D(n)) Gal(D/D(m)) ∼= Gal(C/C(n); D) Gal(C/(C(n) + D(m)); D) ∼= Gal(C/C(n);D) Gal(C/C(m);D) Gal(C/(C(n)+D(m));D) Gal(C/C(m);D) , Gal(D/D(m)) is a subquotient group of Gal(C(m)/C(n)). Particularly, Gal(D/D(n)) and hence Gal(D/D(n)) Gal(D/D(m)) is abelian, for Gal(C(n+1)/C(n)) is isomorphic to the additive group of some k-vector space by the same argument used in the proof of Lemma 3.2.1. Note that Gal(C/C(n)) is just the same as Ω∗ n(Q) defined in last section, and therefore Gal(C(n+1)/C(n)) ∼= Ω∗ n(Q)/Ω∗ n+1(Q). Given a large subcoalgebra D of kQc, we set D+ = kQ≥1 ∩D and (D+)s,t = (kQ≥1)s,t ∩D (D+)s,t. By applying Proposition 4.1.4, we obtain for any s, t ∈ Q0. Clearly D = D0L Ls,t∈Q0 a generalization of Proposition 3.2.4. 26 Corollary 4.3.6. Let Q be a finite quiver, C = kQc and D ⊆ C a finite dimensional large subcoalgebra. Assume Aut(C; D) = Aut(C). Then dimk(Aut(D)) = Xs,t∈Q0 (Qτ 1)s,t(D+)s,t. 5 Automorphisms of a complete path algebra In this section we study the automorphism group of the complete path algebras of a finite quiver. The key point is that for a finite quiver, the automorphism group of the path coalgebra and the one of the complete path algebra are isomorphic, hence the tool we developed for automorphisms of path coalgebras applies in this case. Notice that most results on path coalgebras in this section hold true for any quiver, while when dealing with algebras, we always require the quiver to be finite. 5.1 Aut(dkQa) ∼= Aut(kQc) Let Coalg and Alg denote the category of coalgebras and the one of algebras respectively. Sweedler showed that there is a contravariant functor ()∗ : Coalg −→ Alg. For any coalgebra C, the dual algebra C ∗ has underlying vector space homk(C, k) and multiplication given by the convolution map, see [Sw, Section 1]; and for each coalgebra map f , f ∗ is the usual transpose map of f . Let A be any algebra and set A◦ = {f ∈ A∗ ker(f ) contains a cofinite ideal}. Sweedler showed that for any algebra map f : A −→ B, the transpose map f ∗ maps B◦ into A◦. By setting f ◦ : B◦ −→ A◦ to be the restriction of f ∗, we get a contravariant functor ()◦ : Alg −→ Coalg which is adjoint to ()∗, see [Sw, Theorem 6.0.5]. The following crucial result is due to Taft [Ta, Proposition 7.1]. Recall that a coalgebra C is said to be reflexive if (C ∗)◦ = C. Lemma 5.1.1. Let C be any coalgebra and A = C ∗ the dual algebra. Then ()∗ defined above induces a group monomorphism ()∗ : Aut(C) −→ (Aut(A))op, here Aut(A) is the automor- phism group of A. If moreover, C is reflexive, then Aut(A) ∼= (Aut(C))op under the map ()∗ and hence Aut(A) ∼= Aut(C). Note that the last isomorphism follows from the fact that G ∼= Gop under the map g 7→ g−1 for any group G. Heyneman and Radford showed that if C is a finitary coalgebra, that is C(1) is finite dimensional, then C is reflexive, see [HR, Theorem 4.1.1]. Thus we have the following consequence. Corollary 5.1.2. Let Q be a finite quiver. Set C = kQc and A = C ∗. Let D be a large subcoalgebra of C and B = D∗ = A/D⊥ the dual algebra. Then ()∗ : Aut(D) −→ Aut(B) is Remark 5.1.3. An easy consequence is that Aut(kQc) ∼= Aut(kQa) for any finite acyclic isomorphism in general case when Q is not acyclic, ()∗ even does not give a well-defined map an anti-isomorphism of groups. Particularly, Aut(dkQa) ∼= Aut(kQc) ∼= Ω∗(Q). quiver Q, for in this case kQa = dkQa. It is worth emphasizing that we do not have such an Aut(kQc) −→ Aut(kQa). The reason is that for any σ ∈ Aut(kQc), σ∗ ∈ Aut(dkQa) induces an automorphism of the subalgebra kQa if and only if σ∗(kQa) = kQa, while this does not hold true in general; see Example 6.4.1 below. 27 Recall that for any finite quiver Q, an element (ap)p∈Q in dkQa can be written as an infinite sum Pp∈Q ap ¯p. For any σ ∈ Aut(C), σ∗ is determined by σ∗(x) = Pp∈Q(x, σ(p))p one obtains σ(p) =Px∈Q(σ∗(x), p)x. Consequently, the automorphism of the complete path for any path x ∈ Q, here (−, −) is the paring given in Section 1.3. Applying the pairing again algebra corresponding to a trans-datum µ is given by f ∗ µ(x) =Pp∈Q(x, fµ(p))p. 5.2 Inner automorphisms Let Q be an arbitrary quiver and C = kQc. Consider the following subgroups of Ω∗ 0(Q), say IΩ∗(Q) = {(IdQ0 , µ+) ∃ki ∈ k×, i ∈ Q0, µ1 α = ks(α) kt(α) α, ∀α ∈ Q1, µ1 p = 0, ∀p ∈ Q≥2}, IΩ∗ ◦(Q) = {(IdQ0 , µ+) µ1 α = α, ∀α ∈ Q1, µ1 p = 0, ∀p ∈ Q≥2}, IΩ∗ 0(Q) = {(IdQ0 , µ+) ∃ki ∈ k×, ∀i ∈ Q0, µα = ks(α) kt(α) α, ∀α ∈ Q1, µp = 0, ∀p ∈ Q≥2}, which are of special interest to us. The reason is that calculations within such groups are much simplified, and more importantly, they are quite helpful in understanding the whole automorphism group. In case Q is finite, IΩ∗(Q) corresponds to the inner automorphism group of dkQa. For this reason, we call elements in IΩ∗(Q) inner trans-data and the corresponding automorphisms of kQc inner automorphisms. The following result is practical. Recall that in Section 2.1, we have associated k-linear maps µ0, µ0, µ1, µ+ and µ to each trans-datum µ . Proposition 5.2.1. Let Q be an arbitrary quiver and C = kQc. (1) Given µ ∈ IΩ∗ ◦(Q), then for any x ∈ C, fµ(x) = µ0(x) + µ+(x) + µ(x(2))x(1) − µ(x(2))x(1) +Xr≥3(cid:2)µ(x(3)) · · · µ(x(r))(cid:3) ·(cid:2)µ(x(2))µ0(x(1)) + µ(x(2))x(1) − µ(x(1))x(2)(cid:3) . (2) Given µ ∈ IΩ∗ 0(Q). Suppose there exists some {ki ∈ k}i∈Q0 such that µα = ks(α) kt(α) α for all α ∈ Q1. Then for any s, t ∈ Q0 and x ∈ kQs,t, f (x) = ks kt x. Again we use the Sweedler's notation ∆n(x) =P x(1) ⊗ · · · ⊗ x(n+1) here. For a proof we need only to check the case that x is a path, which can be done by using a similar argument as in Proposition 2.1.3 and we omit it here. The subgroup f(IΩ∗(Q)) is called the inner automorphism group of C and denoted by 0(Q)). Now a basic property ◦(Q)) and Inn0(C) = f(IΩ∗ Inn(C). We also set Inn◦(C) = f(IΩ∗ of Inn(C) follows easily. Corollary 5.2.2. Let Q be an arbitrary quiver and C = kQc. Then Inn(C) ⊆ Aut(C; D) for any subcoalgebra D ⊆ C. Remark 5.2.3. We remark that the converse of the corollary may not hold, say an automor- Aut(C; D) needs not to be an inner automorphism. For example, consider the phism σ ∈ TD⊆C quiver Q given by exactly one vertex with a loop attached. In this case, all subcoalgebras of kQc are given by kQc (n), n ≥ 0. Obviously each automorphism of kQc gives an automorphism of any kQc (n), while kQc has no nontrivial inner automorphism. 28 Proposition 5.2.4. Let Q be an arbitrary quiver. Then IΩ∗(Q) ⊳ Ω∗(Q), IΩ∗ and IΩ∗(Q) = IΩ∗ ◦(Q) ⋊ IΩ∗ 0(Q). ◦(Q) ⊳ Ω∗(Q) Proof. We first show that IΩ∗ ◦(Q) ⊳ Ω∗(Q). For any µ ∈ IΩ∗ ◦(Q) and ν ∈ Ω∗(Q), (µ ◦ ν)p = Xp=p1···pr ,pi∈P µ(νp1 (cid:3) · · · (cid:3)νpr ), and the degree 1 component of (µ ◦ ν)p appears only in the term µ(νp), and therefore (µ ◦ ν)p − νp ∈ k(es(p) − et(p)). Now it is direct to show that for any nontrivial path p, (ν−1 ◦ (µ ◦ ν))p − (ν−1 ◦ ν)p ∈ kQ0, which means that ν−1 ◦ µ ◦ ν ∈ IΩ∗ that IΩ∗(Q) ⊳ Ω∗(Q). ◦(Q) and hence IΩ∗ ◦(Q) ⊳ Ω∗(Q). Similarly, one proves It is easy to show that IΩ∗(Q) = IΩ∗ ◦ is normal and IΩ∗ ◦(Q) ∩ IΩ∗ 0(Q) = {1}. IΩ∗ ◦(Q) IΩ∗ 0(Q), and the last assertion is obvious since Let B be an algebra. Denote by B× the group of multiplicative invertible elements in B. For any b ∈ B×, consider the map χb : B −→ B, χb(x) = bxb−1 for any x ∈ B. It is direct to show that χb is an automorphism of B, and we call it the inner automorphism induced by b. We denote by Inn(B) = {χb b ∈ B×} the inner automorphism group. Inn(B) is known to be a normal subgroup of Aut(B), and the quotient group Out(B) = Aut(B)/ Inn(B) is called the outer automorphism group of B. The map χ : B× −→ Inn(B) is usually called the characteristic map. Clearly, Inn(B) ∼= B×/Z(B)×, here Z(B) denotes the center of B. Suppose B is the dual algebra of some coalgebra D. The map ()∗ : Aut(D) −→ Aut(B) considered in Lemma 5.1.1 may not be surjective in general, while any inner automorphism of B is indeed obtained from an automorphism of D, just as shown in the following lemma. Lemma 5.2.5. Let D be a coalgebra and B = D∗ the dual algebra. Then there exists a group homomorphism Φ : Inn(B) −→ (Aut(D))op, such that (Φ(σ))∗ = σ holds for any σ ∈ Inn(B). In particular, if D is reflexive, then Φ = ()◦. Proof. Define Φ : Inn(B) −→ Aut(D) as follows. For any g ∈ B×, Φ(χg) : D −→ D is given independent of the choice of g, and hence Φ gives a well-defined map, which is easily shown to be a group homomorphism. Moreover, by setting Φ(χg)(c) =P g(d(1))d(2)g−1(d(3)) for any d ∈ D. It is easy to check that Φ(χg) is (Φ(χg))∗(f )(d) = f ((Φ(χg))(d)) =X g(d(1))f (d(2))g−1(d(3)) = (gf g−1)(d) = (χg(f ))(d) for any f ∈ B and d ∈ D, which implies that (Φ(χg))∗ = χg. In the rest part of this subsection, Q is assumed to be a finite quiver. Set C = kQc and figure out the subgroups (Inn(C))∗, (Inn◦(C))∗ and (Inn0(C))∗ of Aut(A). A = (kQc)∗ = dkQa. Recall the isomorphism Aut(C) ∼= Aut(A). A natural question is to An to denote the subspace kQn, and A≥n the subspace {Pl(p)≥n app} for each n ≥ 0. Thus To give an answer, we need some notations. Since kQa is a subalgebra of A, we also use Rad(A) = A≥1, and A0 is a subalgebra with A× the subgroup A× Inn◦(A) = χ(A× ◦ = 1A + Rad(A). Clearly A× = A× ◦ ) and Inn0(A) = χ(A× 0 ). 0 = {Pi∈Q0 aiei ai ∈ k×, ∀i}. We need also 0 . Set 0 + Rad(A) and A× = A× ◦ ⋊ A× 29 Proposition 5.2.6. Let Q be a finite quiver. Set C = kQc and A = (kQc)∗. Then we have the following one to one correspondences: (1) Inn◦(A) (2) Inn0(A) (3) Inn(A) ()◦ ()∗ ()◦ ()∗ ()◦ ()∗ / Inn◦(C) / Inn0(C) f f IΩ∗ ◦(Q) ; IΩ∗ 0(Q) ; / Inn(C) f IΩ∗(Q) . Proof. (1) First consider Inn◦(A) and IΩ∗ be the inner automorphism of A induced by a and σa = χ◦ of C. We claim that σa = µ = (IdQ0, µ+) ∈ IΩ∗ each p ∈ P . To show this we need the fact ◦(Q). Let a = 1 −Pp∈P λpp ∈ 1 + Rad(A). Let χa a the corresponding automorphism p = λpes(p),t(p) for ◦(Q) with µ+ given by µ0 a−1 = 1 + (1 − a) + (1 − a)2 + · · · = 1 +Xp∈P Xp=p1p2···pm pi∈P λp1 λp2 · · · λpm p. By direct calculation we have for each x ∈ P , χa(x) = (1 − a)x(1 + (1 − a) + (1 − a)2 + · · · ) = x − Xq∈P − Xq∈P t(q)=s(x) t(q)=s(x) p=p1p2···pn Xp∈P,s(p)=t(x) λqqx + Xp∈P pi∈P s(p)=t(x) λqλp1 λp2 · · · λpm qxp Xp=p1p2···pm pi∈P λp1 λp2 · · · λpm xp ∗ µ be the automorphism of C and A corresponding to µ respectively. We will Let fµ and f ∗ prove χa = f µ. First assume that x = x1x2 · · · xr ∈ P , here each xi ∈ Q1. Then f ∗ µ(x) = Xi∈Q0 = Xi∈Q0 = Xi∈Q0 (x, fµ(p))p (x, ei)ei +Xp∈P (x, ei)ei +Xp∈P Xp=p1p2···pm (x, ei)ei + Xp1p2 ···pm pi∈P pi∈P (x, F (µp1 (cid:3)µp2 (cid:3) · · · (cid:3)µpm))p (x1x2 · · · xr, F (µp1 (cid:3)µp2 (cid:3) · · · (cid:3)µpm))p1p2 · · · pm. By definition of F , (x1x2 · · · xr, F (µp1 (cid:3)µp2 (cid:3) · · · (cid:3)µpm)) = 0 unless p1p2 · · · pr = x or 30 / / / o o o o / / / o o o o / / / o o o o p2p3 · · · pr+1 = x, thus ∗ µ(x) = (x1x2 · · · xr, F (µx1 f (cid:3) · · · (cid:3)µxr))x (x1x2 · · · xr, F (µq (cid:3)µx1 (cid:3) · · · (cid:3)µxr )qx (x1x2 · · · xr, F (µx1 (cid:3) · · · (cid:3)µxr (cid:3)µp1 (cid:3) · · · (cid:3)µpm))xp1p2 · · · pm (x1x2 · · · xr, F (µq (cid:3)µx1 (cid:3) · · · (cid:3)µxr (cid:3)µp1 (cid:3) · · · (cid:3)µpm))qxp1p2 · · · pm pi∈P +Xq∈P + Xxp1p2···pm + Xqxp1 p2···pm = x − Xq∈P − Xqxp1 p2···pm t(q)=s(x) q,pi ∈P q,pi ∈P λqqx + Xxp1p2···pm pi∈P λp1 λp2 · · · λpm xp1p2 · · · pm λqλp1 λp2 · · · λpm qxp1p2 · · · pm. Note that when we write qxp1p2 · · · pm, we require that t(q) = s(x), t(x) = s(p1) and t(pi) = s(pi+1) for all 1 ≤ i ≤ m − 1. It follows that χa(x) = f ∗ µ(x) in case x ∈ P . Similarly, one shows easily that χa(ei) = f ∗ µ(ei) for any i ∈ Q0. Thus χa = f ∗ µ and hence (2) For any x =Pi∈Q0 σa = (IdQ0 , µ+). It follows that IΩ∗ kiei ∈ A× ◦(Q) corresponds to Inn◦(A). 0 , denote by χx and σx the corresponding automorphism of A and C respectively. It is direct to check that τ = σx = (IdQ0 , τ+), where τ+ is given by τα = ks(α) kt(α) α, ∀α ∈ Q1, and τp = 0, ∀p ∈ Q≥2. In fact, for any p = α1α2 · · · αn, fτ (p) = ks(α1) kt(α1) ks(α2) kt(α2) · · · ks(αn) kt(αn) α1α2 · · · αn = ks(p) kt(p) p = xpx−1 = χx(p). (3) Now we consider a general case. Let a = Pi∈Q0 a ∈ A× if and only if ki 6= 0 for all i ∈ Q0. We may write ki ¯ei −Pp∈P λp ¯p ∈ A. Note that kiei) · (1 −Xp∈P 1 ks(p) λpp) = xy, a = (Xi∈Q0 kiei), y = Pp∈P here we set x = (Pi∈Q0 λpp. Let χa be the inner automorphism of A induced by a and σa the corresponding automorphism of C. By (1) and (2) we know that τ = σx and ω = σy are given by ks(p) 1 σx = (IdQ0, τ+), τα = ks(α) kt(α) α ∀α ∈ Q1, τp = 0 ∀p ∈ Q≥2, σy = (IdQ0, ω+), ωα = α + λα ks(α) es(α),t(α) ∀α ∈ Q1, ωp = λp ks(p) es(p),t(p) ∀p ∈ Q≥2. Clearly σa = (IdQ0 , µ+) for some µ+ and a = xy implies that σa = σy ◦ σx . By (2.2), µ is given by setting µα = ks(α) es(p),t(p) kt(α) for each p ∈ Q≥2. The correspondence between Inn(A) and IΩ∗(Q) follows and the proof is completed. es(α),t(α) for each α ∈ Q1, and µp = λp kt(p) α + λα kt(α) 31 Remark 5.2.7. Recall that in Proposition 5.2.4, we have showed the normality of Inn(C) and Inn◦(C) and the semiproduct Inn(C) = Inn◦(C)⋊Inn0(C). By the above correspondence, the normality of Inn(C) and Inn◦(C) follows easily from the normality of Inn(A) and Inn◦(A). While for a direct proof of the semiproduct Inn(A) = Inn◦(A) ⋊ Inn0(A), it is not quite obvious and takes some works. 5.3 Aut0(dkQa) = Inn(dkQa) Aut•(dkQa) The automorphisms of C which act invariantly on the coideal C≥1 form an important sub- group Aut•(C) of Aut(C), say Aut•(C) = {σ ∈ Aut0(C) σ(C≥1) ⊆ C≥1}. Consider the following subset of Ω∗(Q): Ω∗ •(Q) = {(IdQ0, µ) ∈ Ω∗ 0(Q) µ+ = µ1}. Then we have the following correspondence. Proposition 5.3.1. Let Q be an arbitrary quiver and C = kQc. Then (Aut•(C)) = Ω∗ and f(Ω∗ •(Q)) = Aut•(C). •(Q) Proof. We need only to show that (Aut•(C)) ⊆ Ω∗ follows easily from the construction of f and . The argument is as follows. •(Q) and f(Ω∗ •(Q)) ⊆ Aut•(C), and this Let µ = (IdQ0 , µ+) ∈ Ω∗ •(Q). Then by definition, for any p ∈ P we have fµ(p) = Xp=p1p2 ···pm pi∈P F (µp1 (cid:3)µp2 (cid:3) · · · (cid:3)µpm), now µ+ = µ1 implies each F (µp1 (cid:3)µp2 Conversely, if µ = (IdQ0 , µ+) /∈ Ω∗ (cid:3) · · · (cid:3)µpm) ∈ C≥1 and hence fµ ∈ Aut•(C). •(Q), then there exists some p ∈ P , such that µ0 p 6= 0. We take such a p with minimal length. Then fµ(p) = Xp=p1p2···pm m≥2,pi∈P F (µp1 (cid:3)µp2 (cid:3) · · · (cid:3)µpm) + µp, since µq = µ1 hence fµ /∈ Aut•(C). q for any q with length strictly less than p, we know that fµ(p) ∈ µ0 p + C≥1, and Remark 5.3.2. Note that when we use the notation C≥1, we have already fixed a natural projection C ։ C(0) of coalgebras, or equivalently, we fixed a decomposition C = C(0) ⊕ C≥1. Note that the subgroup Aut•(C) depends on the choice of the projection C ։ C0. The following result is given by direct calculation. Proposition 5.3.3. Let Q be an arbitrary quiver. Then Ω∗ quently, Aut0(kQc) = Aut•(kQc) Inn◦(kQc). 0(Q) = Ω∗ •(Q) IΩ∗ ◦(Q). Conse- Proof. Take µ = (Id, µ+) ∈ Ω∗ that µ ◦ ν ∈ Ω∗ conclusion follows. ◦(Q) such ◦(Q), and the •(Q). If this has been done, then µ = (µ ◦ ν) ◦ ν−1 ∈ Ω∗ •(Q) IΩ∗ 0(Q). We will construct inductively some ν ∈ IΩ∗ 32 In fact, set ν0 = Id and να = α for any α ∈ Q1. For each p with l(p) = 2, we set νp = −µ0 p. Then clearly, Xp=p1p2···pr pi∈P holds whenever l(p) = 2. µ(νp1 (cid:3) · · · (cid:3)νpr ) = µνp + µp = µ1 p ∈ kQ1 Assume that for each p with 2 ≤ l(p) < n, we have fixed a νp ∈ k(es(p) − et(p)), such (cid:3) · · · (cid:3)νpr ) ∈ kQ1 holds for any p with l(p) < n. Now for a path p with µ(νp1 that Pp=p1p2···pr pi∈P l(p) = n, we set νp = −µ0 p − Xp=p1p2···pr r≥2,pi ∈P µ(νp1 (cid:3) · · · (cid:3)νpr ). By definition of µ, it is direct to show that νp ∈ k(es(p) − et(p)), and Xp=p1p2···pr pi∈P µ(νp1 (cid:3) · · · (cid:3)νpr ) = µ1 p ∈ kQ1. By induction, we obtain some ν ∈ IΩ∗ p, that is µ ◦ ν ∈ Ω∗ •(Q). The proof is completed. ◦(Q), such that (µ ◦ ν)p ∈ kQ1 for any nontrivial path Now let D ⊆ C be a large subcoalgebra of C. By Corollary 5.2.2, we have Inn(C) ≤ D(Inn(C)), and call it the inner automorphism group D(Inn◦(C)) and Inn0(D) = ResC D(Inn0(C)). We have a Aut(C; D). We set Inn(D) = ResC of D. We also set Inn◦(D) = ResC generalized version of Proposition 5.2.4. Proposition 5.3.4. Let Q be a quiver, C = kQc and D ⊆ C a large subcoalgebra. Then (1) Inn(C) ≤ Aut(C; D), and the restriction map induces isomorphisms of groups Inn(C)/(Inn(C) ∩ Gal(C/D)) ∼= Inn(D), Inn◦(C)/(Inn◦(C) ∩ Gal(C/D)) ∼= Inn◦(D) and Inn0(C)/(Inn0(C) ∩ Gal(C/D)) ∼= Inn0(D); (2) (3) Inn(D)∗ = Inn(D∗), Inn0(D)∗ = Inn0(D∗) and Inn◦(D)∗ = Inn◦(D∗); Inn(D) ⊳ Aut(D), Inn◦(D) ⊳ Aut(D) and Inn(D) = Inn◦(D) ⋊ Inn0(D). Proof. We need only to prove (3). Recall that we have already shown that IΩ∗ ◦(Q) ⊳ Ω∗(Q) and IΩ∗(Q) ⊳ Ω∗(Q) in Proposition 5.2.4, and hence Inn(C) ⊳ Aut(C) and Inn◦(C) ⊳ Aut(C). Now Inn(D) ⊳ Aut(D) and Inn◦(D) ⊳ Aut(D) follows from (1). For the last assertion, it suffices to prove that Inn◦(D) ∩ Inn0(D) = {IdD}, which is equivalent to show that for any σ ∈ Inn0(C) and τ ∈ Inn◦(C), σ ◦ τ ∈ Gal(C/D) if and only if σ and τ are both in Gal(C/D). This is obvious since D is large. Consider the decomposition D = D(0) ⊕ D+, where D+ = D ∩ C≥1 of D is a coideal of D. Set Aut0(D) = Gal(D/D(0)) and Aut•(D) = {σ ∈ Aut0(D) σ(D+) ⊆ D+}. By a similar argument as in the proof of Proposition 2.2.5, any σ ∈ Aut•(D) extends to an element in Aut•(C). Thus if we set Aut•(C; D) = Aut•(C) ∩ Aut(C; D), then Aut•(D) = ResC D(Aut•(C; D)). Applying Proposition 5.3.4, we have the following characterization of Aut0(D), generalizing Proposition 5.3.3. Proposition 5.3.5. Let Q be an arbitrary quiver and D ⊆ kQc a large subcoalgebra. Then Aut0(D) = Aut•(D) Inn◦(D). 33 Remark 5.3.6. Recall that in Proposition 3.1.1, Aut0(D) is shown to be the identity com- ponent of Aut(D) if D = kQc for some finite acyclic quiver Q. A naive question is to ask whether this property holds for any finite dimensional pointed coalgebra D. We mention that the answer is yes, if D is given by a truncated subcoalgebra of a finitary path coalgebra, for this one uses a similar argument as in the proof of Proposition 3.1.1. When concerning with an arbitrary finitary pointed coalgebra D, the answer is no, the reason is that Aut0(D) is not connected in general. For instance, consider the quiver 1• α1 α2 / •2 β1 β2 / •3 , and the large subcoalgebra D = k{e1, e2, e3, α1, α2, β1, β2, α1β1, α2β2}. Direct calculation shows that Aut0(D) is not connected. In the rest part of this subsection, we will focus on algebras and the quiver Q is assumed to be finte. Let D be a large subcoalgebra of C = kQc, and B = A/D⊥ the dual algebra of D, here D⊥ = {f ∈ A f (d) = 0, ∀d ∈ D}. Note that D is large if and only if D⊥ ⊆ Rad2(A). Note that A0 is embedded to B via the map A0 ֒→ A ։ B, and the image is denoted by B0. Then we have a decomposition B = B0 ⊕ Rad(B), which just corresponds to the decomposition D = D(0) ⊕ D+. Set Inn0(B) = χ(B× 0 ) and Inn◦(B) = χ(1 + Rad(B)). We claim that any inner automorphism of B is induced from an inner automorphism of A. The reason is that for any a ∈ A, a + D⊥ is invertible in B if and only if a is invertible in A. This can be summarized as follows. Lemma 5.3.7. Let Q be a finite quiver, C = kQc the path coalgebra and A = (kQc)∗. Let D be a large subcoalgebra of C and B = D∗ its dual algebra. Then Inn(B) = (Inn(D))∗, Inn◦(B) = (Inn◦(D))∗ and Inn0(B) = (Inn0(D))∗. Similarly, set Aut0(B) = {σ ∈ Aut(B) σ(ei) ∈ ei + Rad(B), ∀i ∈ Q0}, and Aut•(B) = B(B0) = {a ∈ B× ab = ba, ∀b ∈ B0} the {σ ∈ Aut(B) σB0 = IdB0 }. Denote by Z × centralizer of B0 in B×. Then we have the following structural result for Aut(B). Theorem 5.3.8. Let Q be a finite quiver, C = kQc and A = C ∗. Let D ⊆ C a large subcoalgebra and B = D∗ = A/D⊥ the dual algebra. Then (1) (Aut0(D))∗ = Aut0(B). (2) (Aut•(D))∗ = Aut•(B). (3) Inn(B) ⊳ Aut(B), Inn◦(B) ⊳ Aut(B) and Inn(B) = Inn◦(B) ⋊ Inn0(B). (4) Aut0(B) = Inn◦(B) Aut•(B) and Inn(B) ∩ Aut•(B) = χ(Z × B(B0)). If in addition, Z × B(B0) = B× 0 , then Aut0(B) = Inn◦(B) ⋊ Aut•(B). (5) There is an exact sequence of groups 1 −→ Aut•(B)/χ(Z × B(B0)) −→ Aut(B)/ Inn(B) −→ Aut(B)/ Aut0(B) −→ 1. Proof. (1) Let σ ∈ Aut(D). By definition, for each i ∈ Q0 we have σ∗(ei) = Xj∈Q0 (ei, σ(ej ))ej + Xs∈S,l(s)≥1 (ei, σ(s))s ∈ σ(ei) + Rad(B), it follows that σ∗ ∈ Aut0(B) if and only if σ ∈ Aut0(D). Hence (Aut0(D))∗ = Aut0(B). 34 / / / / / / (2) We prove that for any σ ∈ Aut(D), σ∗ ∈ Aut•(B) if and only if σ ∈ Aut•(D). Assume that σ ∈ Aut•(D). By definition σ(D+) ⊆ D+ and hence (ei, σ(s)) = 0 for any s ∈ S with l(s) ≥ 1 and i ∈ Q0. Thus σ∗(ei) = Xj∈Q0 (ei, σ(ej))ej + Xs∈S,l(s)≥1 (ei, σ(s))s = ei for any i ∈ Q0, which implies that σ∗ ∈ Aut•(B). If σ /∈ Aut•(D), then there exists some x ∈ D+ such that σ(x) /∈ D+. Then there exists some i ∈ Q0 such that (ei, x) 6= 0. Hence (σ∗(ei), x) = (ei, x) 6= 0, which implies that σ∗(ei) /∈ B0 and σ∗ /∈ Aut•(B). Thus we have (Aut•(D))∗ = Aut•(B). (3) follows directly from Proposition 5.3.4 and Lemma 5.3.7. (4) The first assertion follows from (2) and Proposition 5.3.5. We next show that Inn(B)∩ Aut•(B) = χ(ZB×(B0)). For any b ∈ B×, χb ∈ Aut•(B) if and only if χb(ei) = beib−1 = ei for any i ∈ Q0, if and only if b ∈ ZB(B0), it follows that Inn(B) ∩ Aut•(B) = χ(Z × B(B0)). Since Inn0(B) ⊆ Inn(B) ∩ Aut•(B), we obtain that Aut0(B) = Inn◦(B) Aut•(B). If Z × B(B0) = B× 0 , then Inn◦(B) ∩ Aut•(B) = {Id}, and the semi-product follows. (5) follows easily from (4) and the second isomorphism theorem of groups, which says that Inn(B) Aut•(B)/ Inn(B) ∼= Aut•(B)/ Inn(B) ∩ Aut•(B). Remark 5.3.9. The parts (3), (4) and (5) of the theorem should be known to experts. For a direct proof of (4), one needs the following classical result, which can be found in any standard textbook on ring theory, see for instance, [La, Section 21], or [DK, Theorem 3.4.1] for finite dimensional algebra case: Let R be a ring with identity and 1 = x1 + x2 + · · · + xm = y1 + y2 + · · · + yn be two primitive orthogonal decompositions of identity, then m = n and there exists r ∈ R× and a permutation τ of {1, 2, · · · , n} such that xi = ryτ (i)r−1 for all 1 ≤ i ≤ n. Let σ ∈ Aut0(B). Then we have primitive orthogonal decompositions of identity 1 = σ(ei). By the above result, there exists some b ∈ B× and a permutation τ of Q0, such that χb(ei) = b · ei · b−1 = σ(eτ (i)) for any i ∈ Q0. Now Pi∈Q0 ei =Pi∈Q0 χb(ei) = σ(eτ (i)) ∈ eτ (i) + Rad(B) implies that τ = IdQ0 and hence χb(ei) = σ(ei). Aut0(B) = Inn(B) Aut•(B). It follows that χ−1 b σ ∈ Aut•(B) and Note that the above argument is only an existence proof, while our proof of Proposition 5.3.3 is in some extend a constructive one. 5.4 The finite dimensional case We will apply to finite dimensional case and recover some classical results. Note that when the quiver Q is finite acyclic, the path algebra is finite dimensional and coincides with the complete path algebra, and hence Aut(kQc) ∼= Aut(kQa). Proposition 5.4.1. Let Q be a finite acyclic quiver and A = kQa the path algebra. Then (1) Aut0(A) is the identity component of Aut(A) and Aut(A)/ Aut0(A) ∼= Aut(Q); (2) Inn(A) ∩ Aut•(A) = Inn0(A) and hence Aut0(A)/ Inn(A) ∼= Aut•(A)/ Inn0(A); (3) Aut0(A) = Inn◦(A) ⋊ Aut•(A); 35 Proof. Let C = kQc. It is direct to show that Aut0(A) = {σ∗ σ ∈ Aut0(C)}. Thus the assertion (1) follows from Proposition 3.1.1 together with the fact Aut(C) and Aut(A) are anti-isomorphic. (2) and (3) follow from Theorem 5.3.8. Note that ZA(A0) = A0 in case Q is acyclic. Moreover, by Theorem 3.2.5 we have the following characterization for solvability of the automorphism group of a truncated path algebra. Proposition 5.4.2. Let Q be a finite quiver, A = kQa and J = kQ≥1 the graded Jacobson ideal. Then Aut0(A/J n) is solvable for some n ≥ 2 if and only if Q is a Schurian quiver, if and only if Aut0(A/J n) is solvable for all n ≥ 2; Aut(A/J n) is solvable for some n ≥ 2 if and only if Q is a Schurian quiver with Aut(Q) being solvable, if and only if Aut(A/J n) is solvable for all n ≥ 2. Remark 5.4.3. The results should be known to experts. For instance, the solvability of the identity component of the automorphism group of a monomial algebra has been discussed in [GS2, Corollary 2.23], from which the first part of the above proposition can also be deduced. Combined with Proposition 5.2.6 and 5.3.1, we reobtain the following known dimension formula, which is essentially given in [GS2, Remark 4.10]. Corollary 5.4.4. Let Q be a finite acyclic quiver and A = kQa. Then dim(Out(A)) = Xs,t∈Q0 (Q1)s,t · (Q≥1)s,t − Q0 + r, where r denotes the number of connected components of the quiver Q. Proof. For a proof we use the correspondence given in Proposition 5.3.1. One shows that (Q1)s,t · (Q≥1)s,t and dim(Inn0(A)) = Q0 − r. By Proposition dim(Aut•(A)) =Ps,t∈Q0 5.4.1(3) we have dim(Aut0(A)/ Inn(A)) = Xs,t∈Q0 (Q1)s,t · (Q≥1)s,t − Q0 + r. Now Aut(A)/ Aut0(A) is a discrete group implies that dim(Out(A)) = dim(Aut(A)/ Inn(A)) = dim(Aut0(A)/ Inn(A)) and the equality follows. Remark 5.4.5. Note that Out(A) is not invariant under Morita equivalence in general, while the group Aut0(A)/ Inn(A) is always a Morita invariance for finite dimensional algebras [Po]. Remark 5.4.6. Compare with the dimension formula of the first Hochschild cohomology of a path algebra [Ha, Proposition 1.6], which computes the dim(Der(A) / InnDer(A)), here Der(A) and InnDer(A) denote the space of derivations and inner derivations respectively. Recall that Der(A) and InnDer(A) are the Lie algebras of Aut(A) and Inn(A) respectively. By [Ha, Corollary 1.6], dim(Der(A) / InnDer(A)) = 0 if and only if Q is a tree, that is, the underlying graph of Q contains no cycles. Moreover, Aut0(A)/ Inn(A) is connected since Aut0(A) is. Combining with Proposition 3.1.1, we have Corollary 5.4.7. Let Q be a finite acyclic quiver. Then Aut0(kQa) = Inn(kQa) if and only if Q is a tree; if and only if Out(A) is a finite group. 36 Now we turn to a more general situation. Let B be an arbitrary finite dimensional elementary algebra. Elementary means that each simple B-module is of one dimension. As we mentioned in Section 1.3, B ∼= kQa/I for some finite quiver Q and an admissible ideal I of kQa. Let D = B∗ be the dual coalgebra of A, then D is a subcoalgebra of kQc. By Theorem 2.2.6, we have the following result. Theorem 5.4.8. Let B = kQa/I, where Q is a finite quiver and I an admissible ideal of kQa. Then Aut(B) is a subquotient group of Aut(dkQa). In particular, if Q is acyclic, then Aut(B) is a subquotient group of Aut(kQa). Remark 5.4.9. When the quiver Q is not acyclic, an automorphism of B needs not to be induced from an automorphism of kQa in general. An easy example is as follows. Let Q be the quiver with exactly one vertex and one arrow. Then kQa ∼= k[x]. Let I = hx3i and B = k[x]/I. Consider the automorphism τ of B given by τ (¯x) = ¯x + ¯x2. Then τ is not obtained from any automorphism of k[x]. Compare with Example 6.4.1. 6 Examples In this section, we consider some typical quivers and calculate the automorphism group and outer automorphism group of their path algebras. 6.1 Quivers of directed An type Example 6.1.1. Let Q be the quiver of directed An type, pictorially Q is of the shape and A = kQa the path algebra. We claim that Aut(A) ∼= Tn/k×, where 1 • −→ 2 • −→ 3 • −→ · · · −→ n •, Tn = ⊆ GLn,k k1 ∗ k2   ∗ ∗ . . . ∗ ∗ ∗ kn   , ki ∈ k× is the group of invertible n×n upper triangular matrices with entries in k, and k× is identified with the subgroup of Tn consisting of multiples of identity. Clearly Tn/k× has a section eTn 1  ∗ k2 ∗ ∗ . . . ∗ ∗ ∗ kn   , ki ∈ k×, i ≥ 2 in Tn, and this froms a subgroup which is isomorphic to Tn/k×. The claim is easy to prove. In fact, A is isomorphic to the algebra of upper triangular n × n matrices with entries in k and A× = Tn. Moreover, Q is a tree with Aut(Q) = {1} implies that Aut(A) = Inn(A) = A×/Z(A×) and the conclusion follows. 37          Next we consider large subcoalgebras of C = kQc and their Galois groups. Recall that for each 1 ≤ i ≤ j ≤ n, there exists an unique path starting at ei and terminating at ej, which we denote by Ei,j. Let D be a large subcoalgebra of C. For each 1 ≤ i ≤ n, let ri denote the maximal integer such that Ei,ri ∈ D. Clearly rn−1 = rn = n, i + 1 ≤ ri ≤ n for each 1 ≤ i ≤ n − 1 and ri ≥ ri−1 for each 2 ≤ i ≤ n. One shows easily that D is uniquely determined by the vector (r1, r2, · · · , rn) ∈ Nn. By direct calculation we show that Gal(C/D) = {(Xi,j) ∈ eTn Xi,i = 1, Xi,l = 0, 1 ≤ i ≤ n, i + 1 ≤ l ≤ ri}. Moreover, one shows easily Aut(C; D) = Aut(C), i.e., σ(D) = D for each σ ∈ Aut(C), and hence Aut(D) ∼= Aut(C)/ Gal(C/D) follows. 6.2 Generalized Kronecker quivers Example 6.2.1. Let Q = Kn be the generalized Kronecker quiver 1• α1 ...αn •2 with n arrows and A the path algebra. Thus Aut(A) ∼= GL(V )⋉V , where V = ⊕1≤j≤nkαj = kQ1 and GL(V ) acts on V canonically. In fact, in this case Aut(Q) = {1} again and hence Aut(A) = Aut0(A). For any v ∈ V , we denote by tv ∈ Aut(A) the automorphism with tv(e1) = e1+v, tv(e2) = e2−v and tv(αj ) = αj for 1 ≤ j ≤ n. For each X ∈ GL(V ), we denote by σX ∈ Aut(A) the automorphism with σX (ei) = ei, i = 1, 2 and σX (αi) = X · αj for any 1 ≤ j ≤ n. Now we have Aut•(A) = {σX X ∈ GL(V )} ∼= GL(V ) and Inn◦(A) = {tv v ∈ V } ∼= V , and the action of Aut•(A) on Inn◦(A) is given by the canonical action of GL(V ) on V . More explicitly, Aut(A) is given by the matrix group (cid:26)(cid:18)GLn(k) ∗ 1(cid:19)(cid:27) ⊆ GLn+1(k). 0 It is direct to show that Inn0(A) = {σa·IdV a ∈ k×}. Thus by Theorem 5.3.8(5), we have Pic(A) ∼= Out(A) ∼= P GL(V ), the projective general linear group of V . 6.3 n-subspace quivers Example 6.3.1. Let Q be the n-subspace quiver, that is a star-like quiver of the shape 1• !BBBBBBBB α1 · · · •0 , •n }{{{{{{{{ αn and A its path algebra, here n ≥ 1 is a positive integer. Then Aut(Q) ∼= Sn, the symmetric group of {1, 2, · · · , n} and Aut0(A) = k1 . . .     λ1 ... kn λn 1   38 , λi ∈ k, ki ∈ k×   / / / / / / ! } a subgroup of GLn+1,k. Now Aut(A) = Sn ⋉ Aut0(A) with the action of σ ∈ Sn on Aut0(A) given by . . . λ1 ... kn λn 1   =   σ · k1   kσ−1 1 . . . kσ−1 n 1 λσ−1 ... λσ−1 n 1   . By comparing Ω•(Q) and Inn0(Q), one shows easily that Aut•(A) = Inn0(A). Again by Theorem 5.3.8(5), we have Pic(A) ∼= Out(A) ∼= Sn. 6.4 Aut(k[x]) and Aut(k[[x]]) As we mentioned before, if the quiver contains oriented cycles, then the natual correspondence between the automorphism group of the path algebra and the one of the path coalgebra given as in Lemma 5.1.1 may not be one to one. We will discuss this in more detail with the following example. Example 6.4.1. Let Q be the quiver given by one vertex with a loop attached. Then the path algebra kQa is isomorphic to the polynomial algebra k[x] in one variable. The complete path algebra dkQa is given by the algebra of power series k[[x]] in one variable. It is easy to show that Aut(k[x]) ∼=(cid:26)(cid:18)1 µ 0 λ(cid:19) , µ ∈ k, λ ∈ k×(cid:27), here each (cid:18)1 µ 0 λ(cid:19) corre- sponds to the automorphism of k[x] mapping x to λx + µ. On the other hind, Ω∗(Q) = {(λn)n≥1, λ1 ∈ k×, λi ∈ k, i = 2, 3, · · · }, here each (λn)n≥1 corresponds to the trans-datum λ given by λxn = λnx. Let λ = (λn)n≥1 ∈ Ω∗. Let fλ ∈ Aut(kQc) and f ∗ λ ∈ Aut(k[[x]]) be the corresponding ∗ λ(k[x]) ⊆ k[x] if and only if λn = 0 for sufficiently large λ(k[x]) = k[x] if and only if λn = 0 for any n ≥ 2, and such automorphisms of kQc automorphisms. One shows that f n; f ∗ correspond to the subgroup (cid:26)(cid:18)1 0 λ(cid:19) , λ ∈ k×(cid:27) of Aut(k[x]). 0 Remark 6.4.2. The above example shows that for a general quiver, the construction as in Lemma 5.1.1 gives only a correspondence between certain subgroup of Aut(kQa) and certain subgroup of Aut(kQc). To determine this subgroup is still open for us. Acknowledgements: The work is partly done during the author's visit at the University of Paderborn with a support by Alexander von Humboldt Stiftung. He would like to thank Professor Henning Krause and the faculty of Institut fur Mathematik for their hospitality. Thanks should also go to Professor Manuel Saor´ın for helpful communications, especially for pointing out the reference [GS2]. References [AA] F.W. Anderson and B.K. DAmbrosia, Square-free algebras and their automorphism groups, Comm. Alg., 24 (10) (1996), 3163-3191. [ARS] M. Auslander, I. Reiten, and S.O. Smalφ, Representation Theory of Artin Algebras, Cambridge Studies in Adv. Math., Vol. 36, Cambridge Univ. Press, 1995. 39 [Ba] K. Baclawski, Automorphisms and derivations of incidence algebras, Proc. Amer. Math. Soc., 36 (1972), 351-356. [Bo] M.L. Bolla, Isomorphisms between endomorphisms rings of progenerators, J. Algebra 87 (1984), 261C281. [CM] W. Chin and S. Montgomery, Basic coalgebras, Modular interfaces (Reverside, CA, 1995), 41-47, AMS/IP Stud. Adv. Math. 4, Providence, RI: Amer. Math. Soc., 1997. [Co] [Dj] S.P. Coelho, The automorphism group of a structural matrix algebra, Linear Algebra Appl., 195 (1993), 35-58. D.Z. Djokovi´c, Derivations and automorphisms of exterior algebras, Can. J. Math., 30 (1978), 1336-1344. [DK] Yu. A. Drozd and V.V. Kirichenko, Finite dimensional algebras (english edition), Springer Verlag (1994). [DNR] S. Dascalescu, C. Nastasescu and S¸. Raianu, Hopf algebras: An introduction. Monogr. Textbooks Pure and Appl. Math. 235, Dekker, 2001. [DWZ] H. Derksen, J.Weyman and A. Zelevinsky, Quivers with potentials and their repre- sentations I: Mutations, Selecta Mathematica 14(1) (2008), 59-119. [EM] S. Eilenberg and J.C. Moore, Homology and fibrations I: Coalgebras, cotensor product and its derived functors, Comment. Math. Helv. 40 (1966), 199–236. [Fr] A. Frolich, The Picard group of nonconmutative rings in particular of orders, Trans. Amer. Math. Soc. 180 (1973), 1-45. [Ha] D. Happel, Hochschild cohomology of finite-dimensional algebras, S´eminaire d'alg`ebre Paul Dubreil et Marie-Paul Malliavin, Lecture Notes in Mathematics 1404 (Springer, Berlin, 1989) 108-126. [HR] R.G. Heyneman and D.E. Radford, Reflexivity and coalgebras of finite type, J. Alge- bra 28 (1974), 215-246. [HYZ] H.-L. Huang, Y. Ye and Q. Zhao, Hopf Structures on the Hopf Quiver Q(hgi, g), Pac. J. Math., 248 (2) (2010), 317-334. [GS1] F. Guil-Asensio and M. Saor´ın, The group of outer automorphisms and the Picard group of an algebra, Algebr. Represent. Theory 2 (4) (1999), 313-330. [GS2] F. Guil-Asensio and M. Saor´ın, The automorphism group and the Picard group of a monomial algebra, Comm. Algebra 27 (1999), 857-887 . [La] T.Y. Lam, A first course in noncommutative rings. Graduate Texts in Mathematics, 131. Springer-Verlag, New York, 1991. [OV] A.L. Onishchik and E.B. Vinberg, Lie Groups and Algebraic Groups, Springer-Verlag, Berlin, 1990. [Po] R.D. Pollack, Algebras and their automorphism groups, Comm. Alg. 17(8) (1989), 1843-1866. 40 [Qu] H.G. Quebbeman, Automorphismen und antiautomorphismen von tensoralgcbren, Math. Z. 158 (1978), 195-198. [Ri] C.M. Ringel, Tame Algebras and Integral Quadratic Forms, Lecture Notes in Math., Vol. 1099, Springer-Verlag, Berlin, Heidelberg, New York, Tokyo, 1984. [Sch] W. Scharlau, Automorphisms and involutions of incidence algebras, in Representa- tions of Algebras, Ottawa, 1974, Lecture Notes in Math. No. 488, Springer-Verlag, New York-Berlin-Heidelberg, 1975. [St] R.P. Stanley, Structure of incidence algebras and their automorphism groups, Bull. Amer. Math. Soc. 76 (1970), 1236-1239. [Sw] M.E. Sweedler, Hopf Algebras, W. A. Benjamin, New York, 1969. [Ta] E.J. Taft, Reflexivity of algebras and coalgebras, Amer. J. Math. 94 (1972), 1111- 1130. [TW] E.J. Taft and R.L. Wilson, On antipodes in pointed Hopf algebras, J. Algebra 29 (1974), 27-32. [Y1] [Y2] S.S.-T. Yau, Continuous family of finite dimensional representations of a solvable Lie algebra arising from singularities, Proc. Nat. Acad. Sci. USA 80 (1983), 7694-7696. S.S.-T. Yau, Solvable Lie algebras and generalized Cartan matrices arising from iso- lated singularities, Math. Z. 191 (1986), 489-506. Department of Mathematics, University of Science and Technology of China, China Wu Wen-Tsun Key Laboratory of Mathematics, USTC, Chinese Academy of Sciences, China Email: [email protected] 41
1406.0299	2	1406	2016-03-01T09:56:00	The Larson-Sweedler theorem for weak multiplier Hopf algebras	[ "math.RA" ]	The Larson-Sweedler theorem says that a finite-dimensional bialgebra with a faithful integral is a Hopf algebra. The result has been generalized to finite-dimensional weak Hopf algebras by Vecserny\'es. In this paper, we show that the result is still true for weak multiplier Hopf algebras. The notion of a weak multiplier bialgebra was introduced by B\"ohm, G\'omez-Torrecillas and L\'opez-Centella. In this note it is shown that a weak multiplier bialgebra with a regular and full coproduct is a regular weak multiplier Hopf algebra if there is a faithful set of integrals. This result is important for the development of the theory of locally compact quantum groupoids in the operator algebra setting. Our treatment of this material is motivated by the prospect of such a theory.	math.RA	math	The Larson-Sweedler theorem for weak multiplier Hopf algebras Byung-Jay Kahng (1) and Alfons Van Daele (2) Abstract The Larson-Sweedler theorem says that a finite-dimensional bialgebra with a faithful in- tegral is a Hopf algebra [L-S]. The result has been generalized to finite-dimensional weak Hopf algebras by Vecserny´es [Ve]. In this paper, we show that the result is still true for weak multiplier Hopf algebras. The notion of a weak multiplier bialgebra was introduced by Bohm, G´omez-Torrecillas and L´opez-Centella in [B-G-L]. In this note it is shown that a weak multiplier bialgebra with a regular and full coproduct is a regular weak multiplier Hopf algebra if there is a faithful set of integrals. Weak multiplier Hopf algebras are introduced and studied in [VD-W3]. Integrals on (regular) weak multiplier Hopf algebras are treated in [VD-W5]. This result is important for the development of the theory of locally compact quantum groupoids in the operator algebra setting, see [K-VD2] and [K-VD3]. Our treatment of this material is motivated by the prospect of such a theory. Date: 1 March 2016 (Version 2.0) (1) Department of Mathematics and Statistics, Canisius College Buffalo, NY 14208, USA. E-mail: [email protected] (2) Department of Mathematics, University of Leuven, Celestijnenlaan 200B, B-3001 Hev- erlee, Belgium. E-mail: [email protected] 1 0. Introduction In this paper we will prove the Larson-Sweedler theorem for weak multiplier Hopf algebras. In order to explain our result, in this introduction we first recall some basic notions and related results. i) To begin with, there is the well-known notion of a bialgebra (A, ∆). In this case A is an (associative) algebra with an identity and ∆ is a coproduct on A. A coproduct (or comultiplication) is a unital homomorphism ∆ : A → A ⊗ A that satisfies coassociativity, i.e. so that (0.1) (∆ ⊗ ι)∆(a) = (ι ⊗ ∆)∆(a) for all a ∈ A, where we use ι for the identity map on A. It is also assumed that there is a counit. This is a homomorphism ε from the algebra A to the scalar field such that (0.2) (ε ⊗ ι)∆(a) = a and (ι ⊗ ε)∆(a) = a for all a. Because the unit in an algebra is unique if it exists and because the same is true for a counit, there is no need to include these objects in the notation for a bialgebra. For the notion of a bialgebra, there are several standard references. We will collect some of them later in the introduction in the separate item 'Various references'. An easy but important result says that the linear dual of a finite-dimensional bialgebra is again, in a natural way, a bialgebra. The result is no longer true if the algebra is not assumed to be finite-dimensional. Again see the standard references about bialgebras. ii) Next we recall the notion of a weak bialgebra. In this case, it is assumed that A is a unital algebra and that ∆ is a coassociative homomorphism from A to A ⊗ A but it is no longer required that the coproduct ∆ is a unital homomorphism. On the other hand, one still assumes that there is a counit ε on A satisfying (0.2), but it is no longer required to be a homomorphism. Instead some weaker conditions are assumed. We refer to the standard references about weak bialgebras, again see further in this introduction. Also here, the dual of a finite-dimensional weak bialgebra is again a weak bialgebra. iii) We can generalize the case of a bialgebra, as considered in item i), into another direction. Now it is no longer assumed that the algebra A has an identity. This condition is replaced by the weaker requirement that the product is non-degenerate and that the algebra is idempotent. For the coproduct one now takes a homomorphism ∆ from A to the multiplier algebra M (A⊗A). The homomorphism ∆ is assumed to be non-degenerate and this implies that there is a unique extension to a unital homomorphism from M (A) → M (A ⊗ A). Also the homomorphisms ∆ ⊗ ι and ι ⊗ ∆ have unique extensions and therefore it makes sense to require again coassociativity as in (0.1). Still the counit is a homomorphism ε from A to the field satisfying the conditions (0.2). In order to make these conditions precise, also the maps ε ⊗ ι and ι ⊗ ε have to be extended to the multiplier algebra M (A ⊗ A). This results in the notion of a multiplier bialgebra. This case was first considered in [J-V]. However more recently, the notion has been defined also in [B-G-L] (see Theorem 2.11 in 2 that paper and a remark made after the proof). The approach is different from [J-V] and it is considered as a special case of a weak multiplier bialgebra (see further in item iv)). The basic notions (like that of a multiplier algebra) that are needed here are found in e.g. [VD1]. See also the other references further in this introduction. There is no relevant duality result here in the finite-dimensional case, because if (A, ∆) is a multiplier bialgebra and if A is finite-dimensional, then A has a unit and we are back in the case of a bialgebra as considered in item i). iv) The last case is that of a weak multiplier bialgebra (A, ∆). The idea is clear. The axioms are weakened in both directions. So, roughly speaking, the algebra is not assumed to be unital and the coproduct is no longer assumed to be non-degenerate. This case is considered in [B-G-L]. Because this is the basic concept for the current paper, we will recall this notion more precisely later, in Section 1. In all these four cases, one can require the existence of an antipode. In the first case, for a bialgebra, it is an anti-homomorphism S from A to itself satisfying the conditions (0.3) m(S ⊗ ι)∆(a) = ε(a)1 and m(ι ⊗ S)∆(a) = ε(a)1 where m is the multiplication map, considered from A ⊗ A to A. If a bialgebra has an antipode, it is called a Hopf algebra. Also in the three other more general cases, one can assume the existence of an antipode, but the notion has to be adapted for each of these cases. A weak bialgebra with an antipode is a weak Hopf algebra, a multiplier bialgebra with an antipode is a multiplier Hopf algebra and a weak multiplier bialgebra with an antipode is a weak multiplier Hopf algebra. References for these cases are again given further in this introduction. In general, the antipode need not be a bijective map from A to itself. If that is the case, we call these objects regular. In the finite-dimensional case, if the bialgebra is a Hopf algebra, the dual bialgebra is again a Hopf algebra. Similarly for a finite-dimensional weak Hopf algebra. Again, this is no longer true in the infinite-dimensional case. Next there is the notion of integrals. If (A, ∆) is a bialgebra as in item i), a non-zero linear functional ϕ on A is called a left integral and a non-zero linear functional ψ on A is called a right integral if (ι ⊗ ϕ)∆(a) = ϕ(a)1 and (ψ ⊗ ι)∆(a) = ψ(a)1 for all a ∈ A. Integrals are unique (up to a scalar) if they exist. For a finite-dimensional Hopf algebra, we always have the existence of integrals. They are automatically faithful. The notion of integrals also exists for the other cases but again, it has to be modified so as to fit with the other axioms. In the case of multiplier bialgebras, integrals are also unique. However, in the case of a weak bialgebra integrals in general are not unique and the same is true for weak multiplier bialgebras. 3 Again, in the case of a finite-dimensional weak Hopf algebra, integrals always exist. A single integral is not necessarily faithful here, but there is a set of faithful integrals (cf. Definition 2.10 in Section 2 for a precise formulation of this property). A necessary and sufficient condition for the existence of a single faithful integral is that the underlying algebra has a faithful functional, i.e. that it is a Frobenius algebra. See Section 3 in [B-N-S]. With integrals, duality can be extended to the non finite-dimensional case in a certain sense. If (A, ∆) is a regular multiplier Hopf algebra with integrals, there exists a suitable subspace of the linear dual of A that can be made in a natural way into a regular multiplier Hopf algebra with integrals. Biduality holds in the sense that the dual of the dual is canonically isomorphic with the original. A similar result holds for regular weak multiplier Hopf algebras. For references about this duality, see later under the item 'Various references'. Finally, we are ready to describe the content of the Larson-Sweedler theorem as it can be proven in all these cases. Roughly speaking, the result says that, if a bialgebra has (enough) integrals, then there is an antipode. More precisely, for a bialgebra we have that it is a Hopf algebra if it has integrals. For a multiplier bialgebra we get a multiplier Hopf algebra if there is an integral. For weak bialgebras one needs enough integrals in order to get a weak Hopf algebra and the same is true in the case of multiplier weak bialgebras. We will again give references for the first three cases later in the introduction. The last result is precisely the one that will be proven in this paper. Content of the paper In Section 1 we recall the notion of a weak multiplier bialgebra (as it has been introduced recently in [B-G-L]) and see how this generalizes multiplier bialgebras, weak bialgebras and of course also bialgebras. We will formulate the basic results that are needed for the proof of the main theorem in this paper. We will not give proofs, we refer to [B-G-L]. In Section 2 we will use these results about weak multiplier bialgebras to prove the Larson- Sweedler theorem in this setting. So, we will show that for any weak multiplier bialgebra, if there are enough integrals, there is an antipode and so it is a weak multiplier Hopf algebra. On the other hand, we will make the arguments more or less independent from the results obtained in the first section. In fact, we will start with a weaker set of assumptions than what is obtained for weak multiplier bialgebras in the first section. We will e.g. not assume the existence of a counit here. The motivation for this lies in the hope to use this approach for the further development of a theory of locally compact quantum groupoids (where counits in general are not considered). See the discussion in Section 3. Indeed, in Section 3 we will discuss the results obtained in Section 1 and Section 2 fur- ther and make some remarks about the importance of the results for further research, in particular for the link of the algebraic theory with the operator algebraic approach. As mentioned already, the treatment in Section 2 is such that it is already preparing for this generalization into the direction of the operator algebra setting. In Appendix A we discuss the differences in notation and terminology in this paper with the existing literature. We provide some kind of dictionary. Recall that this paper is, in 4 some sense, preparing for a generalization to the operator algebraic context and that the algebraists and the operator algebraists often use different conventions and notations. In Appendix B we look at the notions of separability and the Frobenius property as they appear in this note. Here again we also discuss the relation of our conventions with the commonly used ones in the algebraic literature. Conventions and notations In this note, algebras are associative algebras over the field of complex numbers. They are in general not required to be unital, but the product is always assumed to be non- degenerate (as a bilinear form). We will also only work with idempotent algebras A, i.e. where A = A2. The two conditions are fulfilled if the algebras have local units, but we will not need to assume that. It will follow from the other requirements. We use Aop for the algebra obtained from A by reversing the product. For any non-degenerate algebra A we can define its multiplier algebra M (A). This can be characterized as the largest algebra containing A as an essential ideal. If A has an identity, then A = M (A). We will also consider A ⊗ A and its multiplier algebra M (A ⊗ A). We have natural embeddings A ⊗ A ⊆ M (A) ⊗ M (A) ⊆ M (A ⊗ A) and in general, these inclusions are strict. We will use 1 for the identity in M (A). On the other hand, we use ι for the identity map from A to itself. A coproduct ∆ on A is a homomorphism from A to the multiplier algebra M (A ⊗ A) satisfying certain properties. We will use ∆cop for the coproduct on A that is obtained by composing ∆ with the flip map (extended to the multiplier algebra). For a coproduct, we will sometimes use the Sweedler notation and write ∆(a) = P(a) a(1) ⊗ a(2) etc. This has to be done with care and it will be explained when we use this. There is also a new note devoted to the use of the Sweedler notation for multiplier Hopf algebras and weak multiplier Hopf algebras. See reference [VD6]. We will use the leg-numbering notation. If e.g. E is an element in M (A ⊗ A) we will use E12 and E23 for the elements E ⊗ 1 and 1 ⊗ E in M (A ⊗ A ⊗ A). Also we will use E13 for E as 'sitting' in this multiplier M (A ⊗ A ⊗ A) with its 'first leg' in the first factor and with its 'second leg' in the third factor. More precisely E13 = ζ12E23 where ζ12 flips the first two factors in A ⊗ A ⊗ A and is extended to the multiplier algebra. Given a coproduct ∆ on an algebra A we can define four canonical maps T1, T2, T3 and T4 from A ⊗ A to M (A ⊗ A). The first one e.g. is given by T1(a ⊗ b) = ∆(a)(1 ⊗ b) when a, b ∈ A. For the other ones, we refer to Section 2 (see Notations 2.8). It is one of the requirements of the coproduct as used in this paper that these maps have ranges in A ⊗ A. We will use Ker(Ti) and Ran(Ti) for the kernel and the range of these maps. 5 Finally, see also the material about conventions and notations as discussed in the dictionary that we have provided in Appendix A of the paper. Various references For the theory of bialgebras and Hopf algebras, we have the standard references [Abe] and [Sw]. For the theory of weak bialgebras and weak Hopf algebras, we have [B-N-S] and [B-S]. See also [Ni], [N-V1], [N-V2] and [Va]. For the theory of multiplier Hopf algebras there is [VD1] and for integrals on multiplier Hopf algebras, the basic reference is [VD2]. Some information about the use of the Sweedler notation for multiplier Hopf algebras can be found in [VD3] and in the more recent work [VD6]. Multiplier bialgebras have been considered in [J-V]. The theory of weak multiplier Hopf algebras is developed in a series of papers [VD-W2], [VD-W3], [VD-W4] and [VD-W5]. Remark that [VD-W4] has two versions, [VD-W4.v1] and [VD-W4.v2]. The second version is more general in the sense that more results are included about not necessarily regular weak multiplier Hopf algebras. We refer to [VD4] for a study of separability idempotents in relation with the theory of weak multiplier Hopf algebras. Also here there are two versions, [VD4.v1] and [VD4.v2]. As for [VD-W4] also for this paper, the second version includes the more general non-regular case. For the use of the Sweedler notation in this context, we refer to a paper that is currently still under preparation but should soon be available on the arXiv (cf. [VD6]). The Larson-Sweedler theorem for Hopf algebras is found in the original paper [L-S]. For multiplier Hopf algebras, this result is obtained in [VD-W1]. The theorem for weak Hopf algebras is proven in [Ve]. In a forthcoming paper [VD7], some more reflections on the Larson-Sweedler theorem are found. Some general references about separable algebras and Frobenius algebras are [DM-I] and [Abr]. A categorical approach is found in [St]. In the theory of weak (multiplier) Hopf algebras, these two concepts are connected and possible references then are [Sz], [Sc] and [K-S]. Acknowledgments The first-named author (Byung-Jay Kahng) is especially grateful to his coauthor (Alfons Van Daele), for being a helpful mentor to him. The current work was mostly done during his sabbatical leave visit to the University of Leuven during 2012/2013. He is very much grateful to both Alfons Van Daele and the mathematics department at University of Leuven for their warm hospitality. He also wishes to thank Michel Enock for inspiring discussions on the topological version of the theory, on measured quantum groupoids. Finally, He wishes to thank the U.S. Fulbright Foundation for providing the financial support for his visit to Belgium. The second-named author (Alfons Van Daele) thanks Gabriella Bohm for the hospitality while he was visiting Budapest where part of this work was done and for private com- munications further on this subject. He is also grateful to Gabriella Bohm, Jos´e G´omez- Torrecillas and Esperanza L´opez-Centella for providing a preliminary version of their work on weak multiplier bialgebras and for fruitful discussions about this work. 6 1. Weak multiplier bialgebras First in this section we recall the notion of a weak multiplier bialgebra as introduced by Bohm, G´omez-Torrecillas and L´opez-Centella in [B-G-L]. We will also compare this notion here with the other (more restrictive) cases that we have formulated in the introduction. We will give the main results about these objects, needed for the proof of the Larson- Sweedler theorem as we will treat it in the next section. We will not include proofs. We refer to [B-G-L]. It should be mentioned however that our set of assumptions is slightly different from the ones used in [B-G-L]. Fortunately, this makes little difference for the proofs. We will make more comments on this difference later, in Section 2. The concept of a weak multiplier bialgebra A weak multiplier bialgebra is a pair (A, ∆) of an algebra A with a coproduct ∆ satisfying certain properties. The algebra need not be unital, but is assumed to be non-degenerate in the sense that the product as a bilinear map is non-degenerate. It is also assumed to be idempotent. This means that any element of A is the sum of products of two elements in A. The condition is written as A = A2. These two conditions are automatic if the algebra is unital. They are also automatically fulfilled if the algebra has local units. In fact, as we will see further, the other conditions imposed on the pair (A, ∆) will imply that A has to have local units. Recall also from the introduction that we only work with algebras over the field of complex numbers. By a coproduct on A we mean a homomorphism ∆ from A to the multiplier algebra M (A ⊗ A). It is assumed that all four elements of the form (1.1) (1.2) ∆(a)(1 ⊗ b) ∆(a)(b ⊗ 1) (c ⊗ 1)∆(a) (1 ⊗ c)∆(a) belong to A ⊗ A for all a, b, c ∈ A. Moreover ∆ has to be coassociative in the sense that e.g. (1.3) (c ⊗ 1 ⊗ 1)(∆ ⊗ ι)(∆(a)(1 ⊗ b)) = (ι ⊗ ∆)((c ⊗ 1)∆(a))(1 ⊗ 1 ⊗ b) is true for all a, b, c ∈ A. Remark that we use (1.1) to formulate coassociativity as in (1.3). In general, for a coprod- uct on an algebra A that is not assumed to be unital, only (1.1) and not (1.2) is required. If also (1.2) is assumed, the coproduct is called regular (see e.g. [VD-W3]). So, in this paper, we assume from the very beginning that the coproduct is regular. This is not done in [B-G-L] while on the other hand, in order to obtain the main results, eventually also in 7 [B-G-L] the coproduct is assumed to be regular. We will give some comments in Section 3 (Conclusions and further research) on the possibility to drop regularity as a condition for the results in this paper. For a regular coproduct, it is possible to formulate coassociativity with the factors on the other side as (1.4) (∆ ⊗ ι)((1 ⊗ b)∆(a))(c ⊗ 1 ⊗ 1) = (1 ⊗ 1 ⊗ b)(ι ⊗ ∆)(∆(a)(c ⊗ 1)) for all a, b, c ∈ A. Non-degeneracy of the product in A will imply that (1.4) for all a, b, c is equivalent with (1.3) for all a, b, c. Of course, if the algebra has a unit, these extra conditions are automatic and we just have the original condition of coassociativity. For algebras without a unit, this is a little more subtle. See e.g. a recent note on coassociativity for coproducts on algebras without identity [VD8]. The coproduct is also assumed to be full. Roughly speaking, this means that the legs of ∆(A) are all of A. More precisely, if V, W are subspaces of A so that ∆(A)(1 ⊗ A) ⊆ V ⊗ A and (A ⊗ 1)∆(A) ⊆ A ⊗ W, then V = A and W = A. Because we assume the product to be regular, we can also express these conditions by multiplying on the other side. It can be shown that for a full coproduct, the span of elements of the form (ω ⊗ ι)((a ⊗ 1)∆(b)) where a, b ∈ A and ω is a linear functional on A is all of A. The same is true when we put a on the right hand side, and if we act on the second leg. For more information about this notion and this property, we refer to [VD-W1] and also e.g. [VD-W2]. Again, if we compare with the original work of [B-G-L], we assume fullness of the product from the very beginning in this note. In [B-G-L] it is eventually assumed to obtain the main results. See e.g. Theorem 3.13 in [B-G-L]. For the coproduct we also have the following result. The proof is straightforward and can be found e.g. in Lemma 3.3 of [VD-W2]. 1.1 Lemma If there exists an idempotent element E ∈ M (A ⊗ A) satisfying (1.5) ∆(A)(A ⊗ A) = E(A ⊗ A) and (A ⊗ A)∆(A) = (A ⊗ A)E, this idempotent is unique. It is then the smallest idempotent E in M (A⊗A) satisfying E∆(a) = ∆(a) and ∆(a)E = ∆(a) for all a ∈ A. (cid:3) If this idempotent E exists, it is usually referred to as the canonical idempotent. If it exists, it can be shown that the coproduct ∆ has a unique extension to a homomor- phism e∆ : M (A) → M (A ⊗ A) provided we require that still e∆(m) = Ee∆(m) = e∆(m)E 8 for all m ∈ M (A). In what follows we will denote this extension also by ∆. Similarly, also the homomorphisms ∆ ⊗ ι and ι ⊗ ∆ have unique extensions to M (A ⊗ A) requiring now, using the same symbols for these extensions, that (∆ ⊗ ι)(m) = (E ⊗ 1)((∆ ⊗ ι)(m)) = ((∆ ⊗ ι)(m))(E ⊗ 1) whenever now m ∈ M (A ⊗ A). Similarly for ι ⊗ ∆. Observe that we automatically have (∆ ⊗ ι)E = (ι ⊗ ∆)E. For more information about this extension procedure and for the proofs of these properties, we refer to the appendix in [VD-W2]. See also the recent note on coassociativity [VD8]. Using this result, we can formulate the following properties. 1.2 Definition The coproduct ∆ is called weakly non-degenerate if an idempotent E ∈ M (A ⊗ A) exist, satisfying the conditions (1.5) of the lemma. We also consider the condition (1.6) (∆ ⊗ ι)E = (E ⊗ 1)(1 ⊗ E) = (1 ⊗ E)(E ⊗ 1). and we refer to it as weak comultiplicativity of the unit. (cid:3) The above terminology needs to be explained. This is done in the following remark. 1.3 Remark First, the term weak non-degeneracy of the coproduct is motivated by the theory of multiplier Hopf algebras where the condition of Lemma 1.1, with E = 1, is indeed non-degeneracy of the coproduct. In the case of a unit, where E = ∆(1), the property (1.6) is called the weak comulti- plicativity of the unit in [B-N-S] (see Definition 2.1 in that paper). And although there is no unit in our case, we will still refer to this condition as weak comultiplicativity of the unit. (cid:3) We will give more comments on the last notion later (see a remark just before Definition 1.5). Finally we come to the existence and requirements about the counit. A counit is a linear map from A to C satisfying (ε ⊗ ι)(∆(a)(1 ⊗ b)) = ab and (ι ⊗ ε)((a ⊗ 1)∆(b)) = ab for all a, b ∈ B. Because we assume that the coproduct is regular, we can equivalently require that (ε ⊗ ι)((1 ⊗ a)∆(b)) = ab and (ι ⊗ ε)(∆(a)(1 ⊗ b)) = ab for all a, b ∈ B. Because the coproduct is assumed to be full, the counit is unique. See e.g. Proposition 1.12 in [VD-W2]. 9 Now, it is not assumed that the counit is multiplicative but only weakly multiplicative as in the following definition. 1.4 Definition The counit is called weakly multiplicative if ε(abc) = (ε ⊗ ε)((a ⊗ 1)∆(b)(1 ⊗ c)) ε(abc) = (ε ⊗ ε)((1 ⊗ a)∆(b)(c ⊗ 1)) for all a, b. (cid:3) Remark that the conditions for weak multiplicativity of the counit are dual to the condi- tions of weak comultiplicativity of the unit as formulated above in Definition 1.2. To make such a statement precise, one can e.g. consider the finite-dimensional case, two algebras with a non-degenerate pairing and coproducts induced by the products. Then it is not hard to verify that the conditions of Definition 1.4 are indeed precisely the dual forms of the condition (1.6). Note that the weak multiplicativity of the counit as formulated above, is generalizing the same notion in the case of a weak bialgebra, see Definition 2.1 in [B-N-S]. And because of this duality, it is indeed also justified to call the property of the coproduct (1.6) as assumed in the Definition 1.2 weak comultiplicativity of the unit. Finally remark that the second condition in Definition 1.4 above can only be formulated for a regular coproduct. We refer again to the discussion in the last section of this paper where we consider possible future research on this subject. Now we are ready for the main definition in this section. 1.5 Definition [B-G-L] A weak multiplier bialgebra is a pair (A, ∆) of a non-degenerate idempotent algebra A with a full coproduct ∆ that is weakly non-degenerate in the sense of Definition 1.2. Weak comultiplicativity of the unit is assumed. There also is a counit that is assumed to be weakly multiplicative in the sense of Definition 1.4. (cid:3) Before we formulate the main results, let us make some more comments about the concept of a weak multiplier bialgebra as just formulated in the above definition. Also observe that the definition as given above, is more restrictive than the one originally given in [B-G-L]. 1.6 Remarks i) Assuming weak comultiplicativity of the unit is a condition about the behavior of ∆ on the legs of E, together with requiring that these legs commute. It is quite natural as is explained in Section 3 of [VD-W2]. ii) As mentioned already, the assumptions about the weak multiplicativity of the counit are dual to the assumptions about the behavior of the coproduct on the legs of E and therefore also these are quite natural. iii) Finally, the assumptions on the counit as formulated in Definition 1.4 are slightly more general then the ones originally assumed by Bohm, G´omez-Torrecillas and L´opez- Centella in [B-G-L] but the difference is not essential. Remark however that we assume regularity of the coproduct from the vey beginning. (cid:3) 10 Before we continue with formulating some of the main properties of weak multiplier bial- gebras as obtained in [B-G-L], let us compare with the notions of bialgebras, multiplier bialgebras and weak bialgebras. First consider a bialgebra. In this case, the algebra is assumed to have an identity, ∆ is assumed to be unital and we have E = 1. So, obviously, the conditions in Definition 1.2 are satisfied. In this case, the counit is assumed to be a homomorphism and then also weak multiplicativity as in Definition 1.4 is true. In the case of a weak bialgebra, still the algebra A is unital and E = ∆(1). The conditions in Definition 1.2 and for the counit in Definition 1.4 are now part of the axioms for a weak bialgebra (see e.g. [B-N-S]). The comparison with the case of a multiplier bialgebra is clear if we consider the notion as in [B-G-L] but it is not so obvious with the approach in [J-V], where this is also considered, but in a somewhat different way from the other cases. Properties of weak multiplier bialgebras In what follows, (A, ∆) is a weak multiplier bialgebra as in Definition 1.5, E is the canonical idempotent in M (A ⊗ A) and ε is the counit. We first introduce the associated counital maps. 1.7 Proposition There exist four linear maps from A to M (A), given by εs(a) = (ι ⊗ ε)((1 ⊗ a)E) εt(a) = (ε ⊗ ι)(E(a ⊗ 1)) and and ε′ s(a) = (ι ⊗ ε)(E(1 ⊗ a)) ε′ t(a) = (ε ⊗ ι)((a ⊗ 1)E). (cid:3) It is clear that e.g. εs(a) is defined as a right multiplier of A. However, it can be shown that it is actually an element in M (A). In fact, one has that (1 ⊗ a)E belongs to M (A) ⊗ A and from this, it would follow that εs(a) ∈ M (A). Similarly for the other cases. These results however are not obvious and it is somewhat remarkable that weak multiplicativity of the counit is used to prove this. Our notation is different from what is found in the original literature on weak bialgebras and weak Hopf algebras. We use the conventions as in [VD-W3]. For convenience of the reader, we have included a short appendix (Appendix A) to this paper where we give a dictionary, relating various notations in our paper with the notations as they are used e.g. in [B-N-S] and in [B-G-L]. 1.8 Proposition The ranges of the maps εs and ε′ s coincide. Similarly, the ranges of εt and ε′ t coincide. Moreover εs(A) and εt(A) are non-degenerate subalgebras of M (A) and their multiplier algebras M (εs(A)) and M (εt(A)) can be considered as sitting in M (A). (cid:3) In fact, we have the following characterization of the multiplier algebras. 1.9 Proposition Let As = M (εs(A)) and At = M (εt(A)). Then we have As = {x ∈ M (A) ∆(x) = E(1 ⊗ x)} At = {y ∈ M (A) ∆(y) = (y ⊗ 1)E}. (cid:3) 11 It can be shown that, in some sense, εs(A) is the left leg of E while εt(A) is the right leg of E. In particular these algebras commute in M (A) and the same is still true for the multiplier algebras As and At. This implies e.g. that we have similar formulas as in the proposition with the opposite product. The main result is the following. It is the one we will consider in the next section. We use B for the algebra εs(A) and C for the algebra εt(A). 1.10 Theorem [B-G-L] The canonical idempotent E belongs to the multiplier algebra M (B ⊗ C). It is a separability idempotent in the sense of [VD4.v1]. The canonical anti-isomorphisms SB : B → C and SC : C → B, characterized by the formulas E(x ⊗ 1) = E(1 ⊗ SB(x)) and (1 ⊗ y)E = (SC(y) ⊗ 1)E for x ∈ B and y ∈ C are given by SB(ε′ s(a)) = εt(a) and SC (ε′ t(a)) = εs(a) for all a ∈ A. (cid:3) Because we have assumed from the very beginning that we are in the regular case, the separability element is also regular. Remark that in [VD4.v1], only regular separability idempotents are considered while in [V4.v2] also the non-regular case is studied. For this paper, this is irrelevant and so we essentially only need reference [VD4.v1]. Remark that it follows that the algebras B and C have local units. See Proposition 1.9 in [VD4.v1]. There exist what we call left and right integrals in [VD4.v1] and distinguished linear func- tionals in [VD4.v2]. We will adopt the better terminology of [VD4.v2]. These distinguished linear functionals are the linear functional ϕB on B and the linear functional ϕC on C, characterized by the formulas (ϕB ⊗ ι)E = 1 and (ι ⊗ ϕC)E = 1 and given by ϕB(εs(a)) = ε(a) and ϕC(εt(a)) = ε(a) for all a ∈ A. The modular automorphisms of these functionals are precisely the compositions SBSC on C for ϕC and (SC SB)−1 on B for ϕB. See Section 2 of [VD4.v1]. We will be using these functionals for the construction of the counit in Section 2. Also for these notions, we are using a different terminology than what is common and again we refer to Appendix A and Appendix B for the relation with the more commonly used ones in the algebraic literature. For further details about all this, in particular for the proofs, we refer to [B-G-L]. 12 To finish, we just make a small remark about the involutive case. If A is a ∗-algebra, it is natural to assume that ∆ is a ∗-homomorphism. Regularity of the coproduct is automatic in this case. It will follow from the uniqueness of E that it is self-adjoint. And from the uniqueness of the counit, it also follows that ε is self-adjoint. Then it will not be difficult to discover the behavior of the counital maps obtained in Proposition 1.7 with respect to the involution. It will also follow that the ranges and their multiplier algebras As and At are ∗-subalgebras as expected. For the antipodal maps, we find e.g. SC(SB(x)∗)∗ = x for all x ∈ B, also as expected from the general theory developed in [VD4.v1]. 2. The Larson-Sweedler theorem In this section, we start with a non-degenerate idempotent algebra A and a regular and full coproduct ∆ on A. We assume the existence of an idempotent E ∈ M (A ⊗ A) satisfying ∆(A)(A ⊗ A) = E(A ⊗ A) and (A ⊗ A)∆(A) = (A ⊗ A)E. We know that such an idempotent is unique and it is the smallest one that satisfies E∆(a) = ∆(a)E = ∆(a) for all a ∈ A. Then we can extend ∆ ⊗ ι and ι ⊗ ∆ and we will have (∆ ⊗ ι)E = (ι ⊗ ∆)E (see e.g. the appendix in [VD-W2] or the more recent note [VD8]). It is natural to assume also (2.1) (∆ ⊗ ι)E = (1 ⊗ E)(E ⊗ 1) = (E ⊗ 1)(1 ⊗ E). This has been motivated in [VD-W2]. Finally, we will assume that E is a regular separability idempotent in the following sense. First we assume the existence of two algebras B and C sitting in M (A) in a non-degenerate way. By this we mean that BA = AB = A and CA = AC = A. It follows that the products in B and C are non-degenerate. The embeddings extend to embeddings of M (B) and M (C) in M (A). We will consider these multiplier algebras as subalgebras of M (A). Similarly B ⊗ C sits in M (A ⊗ A) non-degenerately and we can consider M (B ⊗ C) as sitting in M (A ⊗ A) as well. We will denote As = M (B) and At = M (C). We now assume further that E ∈ M (B ⊗ C) and that it is a regular separability idempotent in M (B ⊗ C) in the sense of [VD4.v1]. We have collected the main definitions and results about this notion in Appendix B. We can prove the following. 2.1 Proposition The algebras B and C, if they exist, are completely determined by E. 13 Proof: Because E is assumed to be a separability idempotent in M (B ⊗ C), it follows that C is the smallest subspace V of C so that E(b ⊗ 1) ∈ B ⊗ V for all b ∈ B. We claim that it will also be the smallest subspace of V of M (A) so that E(a ⊗ 1) ∈ A ⊗ V for all a ∈ A. This will imply that C is uniquely determined by E and the conditions imposed on E. Similarly for B. To prove the claim, first observe that E(ba ⊗ 1) ∈ BA ⊗ C ⊆ A ⊗ C for all a ∈ A. And as BA = A, it follows that E(a ⊗ 1) ∈ A ⊗ C for all a ∈ A. On the other hand, assume that V is a subspace of M (A) so that E(a ⊗ 1) ∈ A ⊗ V for all a ∈ A. We need to argue that C ⊆ V . To do this, assume that ω is a linear functional on M (A) that is zero on elements of V . Because also (b ⊗ 1)E(a ⊗ 1) ∈ A ⊗ V for all b in B and a ∈ A, we will have (ι ⊗ ω)((b ⊗ 1)E(a ⊗ 1)) = 0 for all b ∈ B and a ∈ A. Because (b⊗1)E ∈ B ⊗M (A), we can define x = (ι⊗ω)((b⊗1)E) as an element of B and we see that xa = 0 for all a. This implies x = 0. So (ι ⊗ ω)((b ⊗ 1)E) = 0 for all b. This implies that ω = 0 on C because C is the right leg of E. (cid:3) Because of this result, it makes sense to say that we require E to be a separability idempotent as we will do in the sequel of this section. Before we continue however, we want to make one more remark about the 'legs' of E. 2.2 Remark A separability idempotent E is assumed to be full. This means that the left leg of E is all of B and that the right leg of C is all of C. This is when the legs are considered of E being a separability idempotent in M (B ⊗ C). In this setting however, we have E ∈ M (A ⊗ A) and in the proof of the proposition, we see that B will also be the left leg of E as sitting in M (A ⊗ A) and similarly for C. In this case, it means e.g. that elements of the form (ι ⊗ ω(· a))E, where now a ∈ A and ω is a linear functional on A, will be in B and will span all of B. (cid:3) The reader is advised to compare with similar results about the legs of a regular coproduct as e.g. discussed in Section 1 of [VD-W1]. In fact, not only the results are similar, also the arguments are essentially the same. See also Definition 1.10 and Lemma 1.11 in [VD-W2] for some more information. In the following proposition, we will argue that the aforementioned conditions on (A, ∆) are fulfilled if it is a weak multiplier bialgebra (as defined and reviewed in the previous section). 2.3 Proposition Let (A, ∆) be a weak multiplier bialgebra as in Definition 1.5 of the previous section. Then the associated idempotent E (cf. Definition 1.2) is a regular separability idempotent as above. 14 Proof: Of course we take B = εs(A) and C = εt(A). We have seen in Proposition 1.8 that these algebras sit in M (A) in a non-degenerate way. In Theorem 1.10 it is stated that actually E is in M (εs(A) ⊗ εt(A)) and that it is a separability idempotent. (cid:3) The result is important and it shows that we could have started with the assumptions as in Section 1. On the other hand we do not from the very beginning assume that we have a counit for reasons we have explained already in the introduction. Indeed, since we are motivated by a possible development of the theory in an operator algebra framework, where the use of the counit is always avoided, we think it is important to start in this section without the assumption of the existence of a counit. See the discussion in Section 3. In fact, we will show that a counit exists when we start with an idempotent E as above and if we require the existence of sufficiently many integrals as we will do further. We will use SB and SC for the canonical anti-isomorphisms from B to C and from C to B respectively defined by E(x ⊗ 1) = E(1 ⊗ SB(x)) and (1 ⊗ y)E = (SC(y) ⊗ 1)E for x ∈ B and y ∈ C. See [VD4.v1] or Appendix B about separability in this paper. Observe that we assume regularity so that actually the maps SB and SC have range in C and B, rather than in their multiplier algebras as in the non-regular case (as studied in [VD4.v2]). For the multiplier algebras As of B and At of C we find the following property (obtained already in Proposition 1.9 in the case of a weak multiplier bialgebra). 2.4 Proposition If x ∈ As then ∆(x) = E(1 ⊗ x) = (1 ⊗ x)E. If y ∈ At then ∆(y) = (y ⊗ 1)E = E(y ⊗ 1). Proof: Using formula (2.1) and because of Remark 2.2, saying that the left leg of E is B, it follows that for elements x ∈ B we have ∆(x) = E(1 ⊗ x) = (1 ⊗ x)E. Then this also follows for the elements in the multiplier algebra As of B. Indeed, let x ∈ M (B). If b ∈ B we find ∆(b)∆(x) = ∆(bx) = E(1 ⊗ bx) = ∆(b)(1 ⊗ x). Now we can multiply from the left with ∆(a) for any a ∈ A and because it is assumed that AB = A, it follows that also ∆(a)∆(x) = ∆(a)(1 ⊗ x) for all a ∈ A. We can now multiply with p ⊗ q from the left for p, q ∈ A and use that (A ⊗ A)∆(A) = (A ⊗ A)E. Then it follows that also ∆(x) = E(1 ⊗ x). Similarly on the other side and for C and its multiplier algebra At. (cid:3) With these notions we can define left and right integrals on the pair (A, ∆). 2.5 Definition A linear functional ϕ on A is called left invariant if (ι ⊗ ϕ)∆(a) ∈ At for all a ∈ A. A non-zero left invariant linear functional is called a left integral. Similarly 15 a linear functional ψ is called right invariant if (ψ ⊗ ι)∆(a) ∈ As for all a. If it is non-zero, it is called a right integral. (cid:3) Because we assume that the coproduct is regular, the elements (ι⊗ϕ)∆(a) and (ψ ⊗ι)∆(a) are well-defined in M (A) for all a ∈ A and so the conditions make sense. Left and right integrals on regular weak multiplier Hopf algebras are studied in [VD-W5] (under preparation). We can not use the results from that paper anyway because we do not have all the properties of a weak multiplier Hopf algebra. So we will need to give some other arguments for the proofs of the following properties of the integrals. To begin with, we will need the following properties about the canonical idempotent E. They are found in Section 4 of [VD-W3] but as mentioned, we need to show that they are still valid under the weaker conditions in this setting. Remark that we are using the notations as in Proposition 4.7 in [VD-W3]. For the proof we refer to Appendix B because it is really a result about separability idempotents in general. 2.6 Proposition Denote (2.2) (2.3) F1 = (ι ⊗ SC)E F2 = (SB ⊗ ι)E and and F3 = (ι ⊗ S−1 F4 = (S−1 B )E C ⊗ ι)E. Then the elements F1 and F3 belong to M (B⊗B) while F2 and F4 belong to M (C⊗C). They satisfy (2.4) (2.5) E13(F1 ⊗ 1) = E13(1 ⊗ E) (1 ⊗ F2)E13 = (E ⊗ 1)E13 and and (F3 ⊗ 1)E13 = (1 ⊗ E)E13 E13(1 ⊗ F4) = E13(E ⊗ 1). (cid:3) Recall that the leg numbering notation, as used in this proposition, has been explained in the introduction. We can now use these multipliers and obtain the following properties of the integrals. 2.7 Proposition If ϕ is a left integral and ψ a right integral we have for all a in A that and (ι ⊗ ϕ)∆(a) = (ι ⊗ ϕ)(F2(1 ⊗ a)) (ι ⊗ ϕ)∆(a) = (ι ⊗ ϕ)((1 ⊗ a)F4) (ψ ⊗ ι)∆(a) = (ψ ⊗ ι)((a ⊗ 1)F1) (ψ ⊗ ι)∆(a) = (ψ ⊗ ι)(F3(a ⊗ 1)) Proof: Consider a left integral ϕ. Take an element a ∈ A and let y = (ι ⊗ ϕ)∆(a). By definition it is in At and so ∆(y) = E(y ⊗ 1). This means that ∆(y) = (ι ⊗ ι ⊗ ϕ)((E ⊗ 1)∆13(a)). 16 Now we use that (E ⊗ 1)E13 = (1 ⊗ F2)E13 where F2 is as in the previous proposition. This implies that also (E ⊗ 1)∆13(a) = (1 ⊗ F2)∆13(a). Then X (a) a(1) ⊗ a(2)ϕ(a(3)) = ∆(y) = (ι ⊗ ι ⊗ ϕ)((1 ⊗ F2)∆13(a)). This is true for all a and we can use the fullness of the coproduct to obtain that then also (ι ⊗ ϕ)∆(a) = (ι ⊗ ϕ)(F2(1 ⊗ a)) for all a. This proves the first formula. If we start with the equality ∆(y) = (y ⊗ 1)E we arrive at the second formula in a completely similar way. On the other hand, if ψ is right invariant, and if we use that x, defined as (ψ ⊗ ι)∆(a) belongs to As, similar arguments as above will provide the two last formulas. (cid:3) These formulas are found in [VD-W5]. They are also discussed in [VD7]. There is a nice consequence of these formulas. If follows that for any left integral ϕ we have ϕ(ya) = ϕ(aσC(y)) for all a ∈ A and y ∈ At where σC(y) = SBSC(y). Remark that SBSC is the modular automorphism for the distinguished linear functional ϕC on E (see Proposition 2.3 in [VD4.v1] and the Appendix B). Indeed, we have (SCSB ⊗ SB SC)E = E and therefore also (ι ⊗ SBSC)F2 = F4. Similarly we have for any right integral ψ that ψ(xa) = ψ(aσB(x)) for all a ∈ A and x ∈ As where σB(x) = S−1 C (y). Again this is the modular automorphism for the unique right integral ϕB on E (see also Proposition 2.3 in [VD4.v1] and again also in the appendix). We will discuss this property again in Section 3 where we draw some conclusions and formulate further remarks. B S−1 Next recall the formulas for the canonical maps T1, T2, T3 and T4, associated with a regular coproduct (cf. Section 1 in [VD-W3]). 2.8 Notations We denote T1(a ⊗ b) = ∆(a)(1 ⊗ b) T3(a ⊗ b) = (1 ⊗ b)∆(a) and and T2(c ⊗ a) = (c ⊗ 1)∆(a) T4(c ⊗ a) = ∆(a)(c ⊗ 1) for all a, b, c ∈ A. Recall that the images of these maps are in A ⊗ A because the coproduct is assumed to be regular. (cid:3) Then we have the following result. 17 2.9 Proposition Let ϕ be a left integral and let ψ be a right integral. Let a, b ∈ A. We have T1((ι ⊗ ι ⊗ ϕ)(∆13(a)∆23(b))) = E(p ⊗ 1) T3((ι ⊗ ι ⊗ ϕ)(∆23(a)∆13(b))) = (p ⊗ 1)E where p = (ι ⊗ ϕ)(∆(a)(1 ⊗ b)) where p = (ι ⊗ ϕ)((1 ⊗ a)∆(b)) and T2((ψ ⊗ ι ⊗ ι)(∆12(a)∆13(b))) = (1 ⊗ q)E T4((ψ ⊗ ι ⊗ ι)(∆13(a)∆12(b))) = E(1 ⊗ q) where q = (ψ ⊗ ι)((a ⊗ 1)∆(b)) where q = (ψ ⊗ ι)(∆(a)(b ⊗ 1)). These formulas have to be covered, i.e. we need to multiply with an element of the algebra, left or right to make these formulas meaningful. This is done in the proof below. Proof: Take a, b, c ∈ A. First observe that ∆13(a)∆23(b)(1 ⊗ c ⊗ 1) is well-defined as an element in the three fold tensor product A ⊗ A ⊗ A by the regularity of the coproduct. Then we can apply the left integral ϕ on the last factor to get elements in A ⊗ A. On such elements we can apply the canonical map T1. Then we find T1((ι ⊗ ι ⊗ ϕ)(∆13(a)∆23(b)(1 ⊗ c ⊗ 1))) (a) = (ι ⊗ ι ⊗ ϕ)(((ι ⊗ ∆)∆(a))∆23(b)(1 ⊗ c ⊗ 1)) = (ι ⊗ ι ⊗ ϕ)((ι ⊗ ∆)(∆(a)(1 ⊗ b))(1 ⊗ c ⊗ 1)) (b) = (ι ⊗ ι ⊗ ϕ)((1 ⊗ F2)(∆13(a)(1 ⊗ c ⊗ b))) (c) = (ι ⊗ ι ⊗ ϕ)((E ⊗ 1)(∆13(a)(1 ⊗ c ⊗ b))) = E((ι ⊗ ϕ)(∆(a)(1 ⊗ b)) ⊗ c) = E(p ⊗ c) with p = (ι ⊗ ϕ)(∆(a)(1 ⊗ b)). For the equality (a) we use co associativity of the co- product, for (b) we use the results from Proposition 2.7 and for (c) we use Proposition 2.6. The three other formulas are proven in the same way, using the other expressions in Proposition 2.7 and the formulas in Proposition 2.6. (cid:3) Remark that we have used the formula (E ⊗ 1)E13 = (1 ⊗ F2)E13 first to obtain the first formula in Proposition 2.7 and then again here in the proof above. The formulas in Proposition 2.9 do not explicitly involve these elements Fi. This seems to suggest that another proof is possible, not using these multipliers. We discuss this further in [VD7]. Now we assume the existence of sufficiently many integrals as follows. 2.10 Definition We say that there is a faithful set of left integrals if, given a ∈ A, we must have a = 0 if either ϕ(ab) = 0 for all b ∈ A and all left integrals or if ϕ(ba) = 0 18 for all b ∈ A and all left integrals on A. Similarly we define the notion of a faithful set of right integrals. (cid:3) As a consequence of Proposition 2.9, we now get the following. 2.11 Proposition If there is a faithful set of left integrals, we find that T1(A ⊗ A) = E(A ⊗ A) and T3(A ⊗ A) = (A ⊗ A)E. If there is a faithful set of right integrals, we find T2(A ⊗ A) = (A ⊗ A)E and T4(A ⊗ A) = E(A ⊗ A). Proof: Assume that there is a faithful set of left integrals. We claim that then A is spanned by elements of the form {(ι ⊗ ϕ)(∆(a)(1 ⊗ b)) ϕ is a left integral, a, b ∈ A}. Indeed, assume that there exists a linear functional ω on A so that ω is 0 on all such elements. This will imply that for any left integral ϕ and all elements a, b, c ∈ A we have ϕ((ω ⊗ ι)(∆(a)(1 ⊗ b))c) = 0. As this is true for all ϕ and all c, this implies that (ω ⊗ ι)(∆(a)(1 ⊗ b)) = 0 for all a, b. Then the fullness of ∆ will imply that ω = 0. It is now an immediate consequence of the first formula in Proposition 2.9 that T1(A ⊗ A) = E(A ⊗ A). A similar argument works for the three other cases. (cid:3) This gives the necessary results for the ranges of the canonical maps. We now consider the kernels of these maps. 2.12 Proposition If there is a faithful set of right integrals, the kernels of the maps T1 and T3 are given by Ker(T1) = (A ⊗ 1)(1 − F1)(1 ⊗ A) and Ker(T3) = (1 ⊗ A)(1 − F3)(A ⊗ 1). Similarly, if there is a faithful set of left integrals, the kernels of T2 and T4 are given by Ker(T2) = (A ⊗ 1)(1 − F2)(1 ⊗ A) and Ker(T4) = (1 ⊗ A)(1 − F4)(A ⊗ 1). Proof: i) First we show that T1((a ⊗ 1)F1(1 ⊗ b)) = T1(a ⊗ b) for all a, b. Because F1 = (ι ⊗ SC)E we have (∆ ⊗ ι)(F1) = (E ⊗ 1)(1 ⊗ F1). 19 Then we find T1((a ⊗ 1)F1(1 ⊗ b)) = ∆(a)E(1 ⊗ m(F1(1 ⊗ b))) where m is multiplication form A ⊗ A to A. Recall that F1 = (ι ⊗ SC)E and that it is one of the properties of a separability idempotent that mF1 = m(ι ⊗ SC)E = 1 (see [VD4.v1]). This implies that T1((a ⊗ 1)F1(1 ⊗ b)) = ∆(a)(1 ⊗ b) and this proves the claim. Remark that we do not need the integrals for this result. ii) Conversely, assume that Pi pi⊗qi ∈ Ker(T1). This means that Pi ∆(pi)(1⊗qi) = 0. Multiply with ∆(a) from the left and apply a right invariant functional ψ on the first leg. We find X (ψ ⊗ ι)(∆(api)(1 ⊗ qi)) = 0 and by the formula in Proposition 2.7 we find that i X i (ψ ⊗ ι)((api ⊗ 1)F1(1 ⊗ qi)) = 0. As this holds for all ψ and all a, it follows from the assumption that X i (pi ⊗ 1)F1(1 ⊗ qi) = 0. This gives the right expression for the kernel of T1. The other formulas are obtained in a similar way. (cid:3) Remark that we need sufficiently many left integrals to find the right form for the range of T1 and the kernel of T2 whereas we need sufficiently many right integrals to get the right form for the kernel of T1 and the range of T2. Similarly for the maps T3 and T4. We are almost ready with showing that (A, ∆) is a regular weak multiplier Hopf algebra. We just need to argue that the counit can also be constructed. This is done in the next proposition. 2.13 Proposition If (A, ∆) is as before and if there are enough left integrals, then there exists a counit. Similarly if there are enough right integrals. Proof: If ε is a counit, we must have that (2.6) ε((ι ⊗ ϕ)(∆(a)(1 ⊗ b)) = ϕ(ab) for all a, b in A and any linear functional, in particular any left integral ϕ on A. We now try to define ε with this formula. Therefore, let us now assume that we have elements ai, bi in A and left integrals ϕi so that (2.7) X (ι ⊗ ϕi)(∆(ai)(1 ⊗ bi)) = 0. i 20 We need to show that Pi ϕi(aibi) = 0. By Proposition 2.9 we find that X i (ι ⊗ ι ⊗ ϕi)(∆13(ai)∆23(bi)(1 ⊗ c ⊗ 1)) belongs to the kernel of T1 for all c. It follows from Proposition 2.12 that X i (ι ⊗ ι ⊗ ϕi)(∆13(ai)(F1 ⊗ 1)∆23(bi)(1 ⊗ c ⊗ 1)) = 0. We will now apply multiplication m on this formula. Using again that mF1 = 1 and if we cancel c, we find Pi(ι ⊗ ϕi)(∆(aibi)) = 0. Using the first formula in Proposition 2.7 we find that Pi(ι ⊗ ϕi)(F2(1 ⊗ aibi)) = 0 and as F2 = (SB ⊗ ι)E we get also Pi(ι ⊗ ϕi)(E(1 ⊗ aibi)) = 0. And if now, we apply the distinguished linear functional ϕC on C satisfying (ϕC ⊗ ι)E = 1 we finally get Pi ϕi(aibi) = 0. This proves that ε is well-defined and that it satisfies (2.6). We need enough left integrals in order to have that all elements in A are of the form as in (2.7). Then we have for all a, b, b′ and all ϕ that ε((ι ⊗ ϕ)(∆(a)(1 ⊗ bb′))) = ϕ(abb′) and as this is true for all elements b′ and all left integrals ϕ it follows that (2.8) (ε ⊗ ι)(∆(a)(1 ⊗ b)) = ab for all a, b. Now we can use coassociativity of ∆ and the fullness of the coproduct to argue that also (ι ⊗ ε)((a ⊗ 1)∆(b)) = ab for all a, b. Indeed, take a, b, c ∈ A and apply ι ⊗ ε ⊗ ι on (ι ⊗ ∆)((a ⊗ 1)∆(b))(1 ⊗ 1 ⊗ c). By (2.8) we find that this is equal to (a ⊗ 1)∆(b)(1 ⊗ c). We now apply a linear functional ω on the last leg and use coassociativity. Because the coproduct is full, we can replace (ι ⊗ ω)(∆(b)(1 ⊗ c)) by b and we find that (2.9) (ι ⊗ ε)((a ⊗ 1)∆(b)) = ab. Equation (2.8) and (2.9) show that ε is the counit. A similar proof works in the case of right integrals. (cid:3) Combining all this, we can prove the following form of the Larson-Sweedler theorem for weak multiplier Hopf algebras. 21 2.14 Theorem Assume that (A, ∆) is a pair of a non-degenerate idempotent algebra A with a full and regular coproduct ∆. Assume that there is an idempotent element E in M (A ⊗ A) satisfying the assumptions as formulated in the beginning of this section, in particular assumption (2.1). If there are a faithful set of left integrals and a faithful set of right integrals, then (A, ∆) is a regular weak multiplier Hopf algebra. Proof: First we have to verify that the pair (A, ∆) satisfies the conditions of Defi- nition 1.14 of [VD-W3]. We have a pair (A, ∆) of a non-degenerate idempotent algebra A with a full coprod- uct ∆ by assumption. We also have a counit as shown in Proposition 2.13. There exist an idempotent E ∈ M (A ⊗ A) giving the ranges of the canonical maps T1 and T2 as required. This is the content of Proposition 2.11. The idempotent is assumed to satisfy (2.1). This takes care of all the requirements in i) and ii) of Definition 1.14 in [VD-W3]. Let us now also check that the third set of assumptions in Definition 1.14 of [VD- W3] are fulfilled. For this we need to consider the maps G1 and G2 as defined in Proposition 1.11 of [VD-W3]. Using e.g. the first of formula (2.4), one verifies (G1 ⊗ ι)(∆13(a)(1 ⊗ b ⊗ c)) = ∆13(a)(1 ⊗ E)(1 ⊗ b ⊗ c) = ∆13(a)(F1 ⊗ 1)(1 ⊗ b ⊗ c) for all a, b, c ∈ A. By the fullness of ∆ we conclude that G1(a ⊗ b) = (a ⊗ 1)F1(1 ⊗ b) for all a, b. Similarly we will find that G2(a ⊗ b) = (a ⊗ 1)F2(1 ⊗ b) for all a, b. Then, the result of Proposition 2.12 provides the necessary condition required in Definition 1.14 for the kernels of the canonical maps. This shows that (A, ∆) is a weak multiplier Hopf algebra. Regularity easily follows from the fact that we can e.g. replace A by Aop since we also have the necessary properties for the ranges and the kernels of T3 and T4. (cid:3) The counit is weakly multiplicative, in the sense of Definition 1.4. This follows, as in the proof of Proposition 4.12 of [VD-W3] from the previous result. This shows that we are in fact in the situation as in the beginning of Section 1. In [VD7], some more aspects of the Larson-Sweedler theorem for weak multiplier Hopf algebras are considered. We now complete this section by formulating some properties of the antipode S of the weak multiplier Hopf algebra (A, ∆). 2.15 Proposition Assume that ϕ is a left integral. Let a, b ∈ A and define p = (ι ⊗ ϕ)(∆(a)(1 ⊗ b)). Then S(p) = (ι ⊗ ϕ)((1 ⊗ a)∆(b)). Proof: We have shown that for any left integral ϕ we have T1((ι ⊗ ι ⊗ ϕ)∆13(a)∆23(b)(1 ⊗ c ⊗ 1)) = E(p ⊗ c) 22 where a, b, c ∈ A and where p = (ι ⊗ ϕ)(∆(a)(1 ⊗ b)). When R1 is the generalized inverse of T1 giving the antipode S by the formula R1(p ⊗ c) = X (p) p(1) ⊗ S(p(2))c, we find, using that R1T1 is given by R1T1(u ⊗ v) = (u ⊗ 1)F1(1 ⊗ v) for all u, v that R1(p ⊗ c) = R1T1((ι ⊗ ι ⊗ ϕ)∆13(a)∆23(b)(1 ⊗ c ⊗ 1))) = (ι ⊗ ι ⊗ ϕ)∆13(a)(F1 ⊗ 1)∆23(b)(1 ⊗ c ⊗ 1)) = (ι ⊗ ι ⊗ ϕ)∆13(a)(1 ⊗ E)∆23(b)(1 ⊗ c ⊗ 1)) = (ι ⊗ ι ⊗ ϕ)∆13(a)∆23(b)(1 ⊗ c ⊗ 1)). As this is equal to P(p) p(1) ⊗ S(p(2))c, we get S(p)c = (ι ⊗ ϕ)(1 ⊗ a)∆(b))(c ⊗ 1). We can now cancel c and this completes the proof. (cid:3) In fact, this result can be proven without reference to the previous formulas. Indeed, it is shown to be true for any left integral on a regular weak multiplier Hopf algebra in [VD-W5]. Similarly for other cases. 2.16 Remark The above result suggests another approach to the theory. The argument presented in the proof of the previous proposition can be used to define the antipode. This idea has been used e.g. already in [VD5]. (cid:3) We finally verify that S coincides with the antipodal maps on the source and target alge- bras. 2.17 Proposition The antipode S of (A, ∆), extended to the multiplier algebra M (A) coincides with the antipodal maps SB on B and with SC on C. Proof: Let y ∈ C. For any pair a, b of elements in A and for any left integral ϕ we have (ι ⊗ ϕ)((1 ⊗ ay)∆(b)) = S((ι ⊗ ϕ)(∆(ay)(1 ⊗ b))) = S((ι ⊗ ϕ)(∆(a)(y ⊗ b))) = S((ι ⊗ ϕ)(∆(a)(1 ⊗ b))y) = S(y)S((ι ⊗ ϕ)(∆(a)(1 ⊗ b))) = S(y)(ι ⊗ ϕ)((1 ⊗ a)∆(b)). As this is true for all left integrals and all elements a, it follows that (1 ⊗ y)∆(b) = (S(y) ⊗ 1)∆(b) 23 for all b. This implies that S(y) = SC(y). Similarly, the antipode S of the weak multiplier Hopf algebra (A, ∆) coincides on B with the antipode map SB associated with E. (cid:3) Also here, we should have a quick look at the involutive case, just as we did at the end of the previous section. However, we postpone this to the next section, where discuss the possible development of a theory in the operator algebra setting. 3. Conclusions and further remarks This paper has two main sections. In the first one, Section 1, we have discussed the arguments needed to show that the (unique) canonical idempotent E in the definition of a weak multiplier bialgebra is a (regular) separability idempotent in the multiplier algebra M (B ⊗ C) where B and C are the images of the counital maps εs and εt. Also these maps are obtained from the axioms of a weak multiplier bialgebra. The result has been proven by Bohm, G´omez-Torrecillas and L´opez-Centella in [B-G-L]. The existence of a counit ε is assumed from the very beginning and ε is required to satisfy two weak multiplicativity axioms. Proofs are not given here as they are found in [B-G-L]. This section is mainly included for completeness and because the result is crucial for the interpretation of the main result of this paper. The main result of this paper is the Larson-Sweedler theorem for weak multiplier Hopf algebras. It says that, if a weak multiplier bialgebra has enough integrals, then it has an antipode and it is a (regular) weak multiplier Hopf algebra. This result is obtained by combining the results of the first section with the ones of the second section. However, in accordance with the aim of this paper, as explained in the introduction - see also further here - we start in Section 2 with a slightly different setting as the one obtained at the end of Section 1. The main difference lies in the fact that we no longer assume the existence of a counit. We start with a pair of a non-degenerate algebra A and a full and regular coproduct on A. We assume that the canonical idempotent E in M (A ⊗ A) exists and that it is a separability idempotent. Under these conditions, if there exist enough integrals, we obtain an invertible antipode, as well as a counit satisfying the weak multiplicativity as required in the first section. Recently, the theory of separability idempotents for non-degenerate algebras has been considered with less regularity constraints, see [VD4.v2]. Recall that in the first version of this paper, reference [VD4.v1], only regular separability idempotents were considered. This paper on the Larson-Sweedler theorem has been written before the more general non-regular separability idempotents were studied. This raises the question whether or not the Larson-Sweedler theorem as we treat it in this paper, can further be generalized with weaker regularity conditions. This is not at all clear as it seems that the existence of faithful integrals implies regularity in the end. Still, the problem is worth an investigation and a first attempt is found in a paper with reflections on the Larson-Sweedler theorem 24 for (weak) multiplier Hopf algebras, see [VD7]. Remark however that the non-regular case is not relevant when we talk about the involutive setting as then regularity is always automatic. The Larson-Sweedler theorem is well-known among algebraists. On the other hand, it plays an implicit role in the operator algebra approach to quantum groups. Consider e.g. the definition of a locally compact quantum group in the setting of operator algebras. In the von Neumann algebra framework, the starting point is a pair of a von Neumann algebra with a normal and unital coproduct and with the assumption that there is a left and a right Haar weight. See e.g. [K-V2] and also [VD5]. The existence of a counit, as well as of an antipode, is not part of the starting assumptions. This is just like in the case of the Larson-Sweedler theorem, as we have treated it in Section 2 of this paper. Indeed, this is precisely the situation that we start with in Section 2, but now for weak multiplier Hopf algebras. And the challenge is clear. The aim is to develop the theory as it now exists for weak multiplier Hopf algebras in the operator algebra framework, either in the C∗-algebra setting or in the von Neumann algebra context. What is done in Section 2 of this paper should be a possible source of inspiration for such a project (as it was in the case of the development of locally compact quantum groups). Of course, first one has to see what happens with the results in Section 2 in the case of a ∗-algebra. In Appendix B, where we recall the notion of a separability idempotent, we briefly mention this case while in the original paper ([VD4.v1]), this case got special attention. However not in Section 1, nor in Section 2, we have really investigated the involutive case yet. Therefore let us have a look at both the starting point and the final result in Section 2, for the involutive case and give some comments. So consider a non-degenerate idempotent ∗-algebra A and a full coproduct ∆ on A. Assume that ∆ is a ∗-map. It will follow that the canonical idempotent E is self-adjoint because it is the smallest idempotent in M (A ⊗ A) satisfying E∆(a) = ∆(a)E = ∆(a) for all a ∈ A. Remark also that the equality (∆ ⊗ ι)E = (1 ⊗ E)(E ⊗ 1) (1 ⊗ E)(E ⊗ 1) = (E ⊗ 1)(1 ⊗ E), will imply again by taking adjoints. The separability idempotent E in M (B ⊗ C), being self-adjoint, will imply that B and C are ∗-subalgebras of M (A). Then we know from the theory of separability idempotents, as developed in [VD4.v1], that B and C are direct sums of full matrix algebras. Moreover, the functionals ϕB on B and ϕC on C are positive. Furthermore, the pair of the algebra B and the faithful positive linear functional ϕB on B determine the separability idempotent completely. Recall from [VD4.v2] that in the involutive case, the separability idempotent is automatically regular. We will need the extra assumption that there are enough positive integrals on A. As these integrals are not assumed to be faithful on their own, there may not exist modular 25 automorphisms. However, they still have some 'restricted modular behavior' in the sense that e.g. ϕ(ya) = ϕ(aσC(y)) when ϕ is left invariant, a is in A and y in C. Here σC is the modular automorphism of ϕC on C. See a remark after the proof of Proposition 2.7. It might happen that, in the involutive case, if enough integrals exists, there are automatically enough positive integrals. This however is not clear. All this suggests a possible set of axioms for the generalization of the concept of a weak multiplier Hopf ∗-algebra to the operator algebra setting. In the framework of von Neu- mann algebras, the starting point will be a von Neumann algebra M and a coproduct ∆ on M satisfying the obvious conditions (without being unital). There is also the subalgebra N of M with a faithful normal semi-finite weight ν. The pair (N, ν) has to give rise to the canonical idempotent E, now a self-adjoint projection in the von Neumann algebraic tensor product of N ⊗ L where L is a von Neumann subalgebra of M , anti-isomorphic with N . Finally, there will be the requirement of having enough normal semi-finite (not necessarily faithful) weights on M that are left invariant, as well as having enough such weights that are right invariant. They will need to have some restricted modular behavior in the sense given by the modular automorphisms of ν on N . This project is carried out further in [K-VD2] and [K-VD3], see also [K-VD1]. Finally we should also say something about the relation of all of this with the existing theory of measured quantum groupoids as developed e.g. by Enock and Lesieur in [L] and [E]. The difference of their work and the theory as outlined above lies in the existence of the canonical idempotent E for the pair (N, ν). This is of the same nature as the difference between multiplier Hopf algebroids (as developed in [T-VD1]) and the weak multiplier Hopf algebras as developed in [VD-W3]. One of the main conditions for a multiplier Hopf algebroid to have an underlying weak multiplier Hopf algebra is that the base algebra is separable Frobenius, see [T-VD2]. And so it is expected that there should be some equivalent condition in the case of operator algebras. Appendix A. A dictionary In analysis, in particular in functional analysis and the theory of operator algebras, several attempts to generalize Pontryagin's duality for abelian locally compact groups to the non- abelian case have led to the interest of Hopf algebraic structures. More than a decade ago, this resulted in a theory of locally compact quantum groups, see e.g. [K-V1] and [K-V2]. At the same time, in fact even earlier, the algebraists were studying Hopf algebras and more recently quantum groups in a purely algebraic setting. Unfortunately, both fields use different terminologies, different notations and sometimes even a very different way of thinking and arguing. A similar phenomenon occurred with the introduction of the theory of quantum groupoids. In pure algebra, there is the literature that started with the study of weak Hopf algebras by G. Bohm, F. Nill & K. Szlach´anyi [B-N-S]. Also the operator algebraists were interested because of the relation with the study of subfactors (and partly also as a generalization 26 of the theory of locally compact quantum groups to locally compact quantum groupoids). And although in the very beginning, the original work was entitled: Weak Hopf algebras I. Integral theory and C∗-structure, soon after, the algebraists developed the theory of weak Hopf algebras independently of the study of these objects in the operator algebras setting. We feel that the theory of weak multiplier Hopf algebras, and in particular the theory of weak multiplier Hopf ∗-algebras with positive integrals, can be considered to be a bridge between the two approaches. Such was already the case with multiplier Hopf algebras, more precisely with the theory of multiplier Hopf ∗-algebras with positive integrals, sometimes called algebraic quantum groups. An important aspect of this last theory is the possibility to construct duals in very much the same way as is done in the original Pontryagin's duality for abelian locally compact groups. Moreover, this duality extends the duality of finite-dimensional Hopf algebras to a more general, infinite-dimensional case. The aim of this appendix is to provide a small dictionary as far as the topic of this paper is concerned, namely about weak multiplier bialgebras and weak multiplier Hopf algebras. First we look at the more general aspects. Differences and common practices First, in analysis and certainly in operator algebras, most if not all algebras that play a role, are algebras over the field of complex numbers, whereas in algebra, not only algebras over other fields are considered, but often people just work with rings. This implies that several results about e.g. tensor products, obvious for analysts, are no longer valid. In analysis one is often also interested in the case where the underlying algebras are ∗- algebras and mostly where they are operator algebras. This means that they have a ∗- representation by say bounded operators on a Hilbert space. This will imply e.g. that a∗a = 0 can only happen if a = 0. The notion of a multiplier algebra of a non-unital, but non-degenerate algebra is something familiar for people working with operator algebras. This is not so for algebraists. For algebraists, a multiplier of an algebra would be considered as a pair (λ, ρ) of module maps from A to itself, satisfying the compatibility ρ(a)b = aλ(b) for all a, b ∈ A. This is also true for operator algebraists. But they will rather look at multipliers as sitting in the multiplier algebra M (A), characterized as the largest algebra with identity in which A sits as an essential ideal. Then A is seen as a subalgebra of M (A). This is however only possible if the product in the algebra is non-degenerate and so usually, in this setting, only such algebras are considered. For a linear functional ω on an algebra A, there is the notion of faithfulness as used in analysis. It means that an element a ∈ A has to be 0 if ω(ab) = 0 for all b ∈ A and also if ω(ba) = 0 for all b ∈ A. In the case of a ∗-algebra, and when the functional is positive, this is equivalent with the property that a = 0 if ω(a∗a) = 0. The existence of a faithful functional on an algebra implies that the product is non-degenerate. The existence of a positive faithful functional on a ∗-algebra is related with the property that the algebra is an operator algebra, although this is not sufficient to have a faithful ∗-representation by bounded operators on a Hilbert space. Fortunately, in the situations we consider in Hopf 27 algebra theory, it seems to be so that e.g. positivity of the integrals give rise to bounded operators on the associated Hilbert spaces. For a finite-dimensional algebra the existence of a faithful linear functional implies not only that the algebra is unital, but also that it is Frobenius. Also the converse is true here. The Frobenius property is not familiar to operator algebraists because there, often algebras are nice ∗-algebras and the property is always satisfied. If ω is a faithful linear functional on a finite-dimensional algebra A, there automatically exists an automorphism σ satisfying ω(ab) = ω(bσ(a)) for all a, b ∈ A. This result is no longer true for infinite-dimensional algebras. Nevertheless, in the situation studied here, the integrals turn out to have this property, also in the infinite-dimensional case. In the field of functional analysis, this automorphism is called the modular automorphism. The terminology finds its origin in the theory of non-unimodular locally compact groups and by extension of this, in the modular theory of faithful positive linear functionals on a von Neumann algebra (see e.g. [T]). In algebra, this automorphism (or rather its inverse) is known as the Nakayama automorphism (see e.g. [Na]). For the identity map on the space A, we intend to use ιA or idA and often we even drop the index because most of the time it is clear from the context on which space the identity map ι acts. In algebra, people often use the same symbol for the space and for the identity map on this space. For a coproduct ∆ : A → A ⊗ A, whereas we use e.g. ∆ ⊗ ι for the map from A ⊗ A to A ⊗ A ⊗ A, obtained by applying ∆ to the first factor and the identity map on the second factor, in the algebra literature, it is common to write this map as ∆ ⊗ A. This looks quite strange to functional analysts. For algebraists, it is quite natural to consider the dual notion, namely that of a coalgebra. It is a vector space C with a coproduct ∆ : C → C ⊗ C that is coassociative and has a counit. In many situations however this notion of a coproduct is too restrictive and one needs a bigger space for the range of the coproduct. If e.g. A is a non-degenerate algebra, a natural concept of a coproduct on A is a homomorphism ∆ : A → M (A ⊗ A) and in order to be able to formulate coassociativity, there is the need of some extra assumptions about the range of ∆ in M (A ⊗ A) (see the discussion in the beginning of Section 1). There is also the notion of regularity and fullness of a coproduct that is completely irrelevant if the coproduct maps into the tensor product and if there is a counit. There seems to be no way to extend the notion of a coproduct simply on a vector space that fits our needs, without more structure. More specific comments about weak multiplier bialgebras As homomorphisms like ∆ go from one algebra into the multiplier algebra of another one, compositions of such maps are not immediately possible. One has to find a way to extend such homomorphisms to the multiplier algebra. There is a general theory about this for non-degenerate homomorphisms γ from an algebra A to the multiplier algebra M (B) of an algebra B. Recall that γ is called non-degenerate if γ(A)B = Bγ(A) = B. In that case, there exists a unique extension γ which is a unital homomorphism from M (A) to M (B). In some sense, the non-degenerate homomorphisms from A to M (B) are the 28 natural generalizations to non-unital non-degenerate algebras of unital homomorphisms in the case of unital algebras. However, this is only true for idempotent algebras. It seems to be more subtle when the algebras are not idempotent. See some remarks about this problem in [VD8]. In the theory of weak multiplier Hopf algebras, the coproduct (in general) is not a non- degenerate homomorphism from A to M (A ⊗ A) and so the above extension procedure does not work. Fortunately, there is the canonical idempotent E with its properties that allows to find such extensions. This is explained e.g. in an appendix of [VD-W2]. See also the remark in the beginning of Section 1 of this paper. It is also a common practice in the field of operator algebras, to use the same symbols for these extensions as we do in this paper as well. In Definition 1.2 we use the term weakly non-degenerate for the coproduct, whereas in the original work on weak Hopf algebras [B-N-S] this is called the weak multiplicativity of the unit. The latter is quite natural for algebras that have an identity whereas the former is perhaps more natural if the algebra has no unit because then it is indeed a weaker form of non-degeneracy and it is used in the same way to extend homomorphisms. See also some comments in Section 1 on the use of this terminology. Now we consider the counital maps. In the original paper by Bohm, Nill & Szlach´anyi, in the case of a weak Hopf algebra, the counital maps are denoted and defined as follows in terms of the antipode S. The Sweedler notation P(a) a(1) ⊗ a(2) is used for ∆(a). A.1 Notation Let (A, ∆) be a weak Hopf algebra with antipode S. Then we have the maps ⊓R(a) = X ⊓L(a) = X (a) (a) S(a(1))a(2) a(1)S(a(2)) ⊓R(a) = X ⊓L(a) = X (a) (a) a(2)S−1(a(1)) S−1(a(2))a(1) (cid:3) These counital maps can also be defined for a weak multiplier bialgebra, in terms of the counit, without reference to the (non-existing) antipode. We have the formulas in Proposition 1.7 and we recall them here. We have, for all a ∈ A, εs(a) = (ι ⊗ ε)((1 ⊗ a)E) εt(a) = (ε ⊗ ι)(E(a ⊗ 1)) ε′ s(a) = (ι ⊗ ε)(E(1 ⊗ a)) ε′ t(a) = (ε ⊗ ι)((a ⊗ 1)E). Now, here are the connections with the case of a weak (multiplier) bialgebra and the formulas as found e.g. in Section 3 of [B-G-L]. For a ∈ A we have εs(a) = ⊓R(a) εt(a) = ⊓L(a) s(a) = ⊓R(a) ε′ t(a) = ⊓L(a). ε′ 29 The notations we use for the counital maps, are also used in e.g. [N-V2]. In the next appendix, we will say more about the common conventions and differences, also differences in notations, for the notion of separability and separability idempotents. Appendix B. Separability and Frobenius algebras In this appendix we will first recall the notion of a (regular) separability idempotent as it is studied in [VD4.v1]. As we mentioned already, in the more recent version [VD4.v2], also non-regular separability idempotents are considered. However, in this note, only the regular case is studied and therefore, it is sufficient to recall here only the material that is already present in the first version of the paper on separability for multiplier algebras. We not only recall the main properties, we will also prove some new properties we need in this paper. Throughout we will give some comments, refer to the existing literature about separable algebras and continue with the dictionary of the previous appendix. We start with two non-degenerate algebras B and C and an idempotent E in M (B ⊗ C). We require that (B.1) (B.2) E(1 ⊗ c) ∈ B ⊗ C (1 ⊗ c)E ∈ B ⊗ C and and (b ⊗ 1)E ∈ B ⊗ C E(b ⊗ 1) ∈ B ⊗ C for all b ∈ B and c ∈ C. It is further assumed that the left and the right leg of E are respectively all of B and C. This means that E is full in the terminology of [VD4.v1]. Remark that in the setting of Section 2, we have the element E sitting in M (A ⊗ A) and we also have that E(1 ⊗ a) ∈ B ⊗ A (1 ⊗ a)E ∈ B ⊗ A and and (a ⊗ 1)E ∈ A ⊗ C E(a ⊗ 1) ∈ A ⊗ C for all a ∈ A. See the proof of Proposition 2.1. In what follows, we will systematically use letters b, b′ for elements in B and letters c, c′ for elements in C. We can do this here because further, we do not consider the bigger algebra A. We now recall the following definition from [VD4.v1]. B.1 Definition We call E a regular separability idempotent if also (B.3) E(B ⊗ 1) = E(1 ⊗ C) and (B ⊗ 1)E = (1 ⊗ C)E. (cid:3) Remark once more that in [VD4.v1] all separability idempotents are assumed to be regular in the sense of the generalization studied in the second version [VD4.v2]. 30 One can show that this condition implies that the algebras have local units (see Proposition 1.9 in [VD4.v1]). In particular they are idempotent algebras. Also remark that the conditions (B.1) and (B.2) follow from (B.3) and the fact that E is a multiplier of B ⊗ C (when B and C are already known to be idempotent). We have the existence of the antipodal maps SB and SC: B.2 Proposition Assume that E is a regular separability idempotent in M (B ⊗ C). Then there are anti-isomorphisms SB : B → C and SC : C → B given by E(b ⊗ 1) = E(1 ⊗ SB(b)) and (1 ⊗ c)E = (SC(c) ⊗ 1)E for all b ∈ B and c ∈ C. See Proposition 1.7 in [VD4.v1]. (cid:3) Again, from the existence of these anti-isomorphisms, the previous condition (B.3) follows. We see that an idempotent E ∈ M (B ⊗ C) will be a regular separability idempotent if the algebras B and C are idempotent, if the anti-isomorphisms SB and SC exist and if E is full. As B is anti-isomorphic with C, we could have taken C = Bop from the very beginning (using e.g. SC - see remark B.4.i below). One of the reasons for not doing this is the fact that we are also interested in the involutive case where in general SC : C → B is not a ∗-anti-isomorphism. See also some more comments on the involutive case at the end of this appendix. It is a fairly easy consequence that the following holds (see Proposition 1.8 in [VD4.v1]). B.3 Proposition If E is a regular separability idempotent in M (B ⊗ C), then mC(SB ⊗ ι)(E(1 ⊗ c)) = c and mB(ι ⊗ SC)((b ⊗ 1)E) = b for all c ∈ C and all b ∈ B where mB and mC are used to denote the multiplication maps from B ⊗ B → B and C ⊗ C → C respectively. (cid:3) We think of the formulas above as SB(E(1))E(2) = 1 and E(1)SC (E(2)) = 1 with the Sweedler type notation E = E(1) ⊗ E(2). One can also show that (SB ⊗ SC)E = ζE where ζ is the flip from B ⊗ C to C ⊗ B, extended to the multiplier algebra (Proposition 1.13 in [VD4.v1]). Before we continue, let us compare this notion with the notion of separability as it appears elsewhere in the literature (see e.g. [DM-I]). 31 B.4 Remark i) Consider the case where B and C are finite-dimensional algebras with an identity, so that in particular E ∈ B ⊗ C. The anti-isomorphism SC can be used to identify C with Bop. Define F = (ι ⊗ SC)E. Then F ∈ B ⊗ Bop and we will have mF = 1 as well as (b ⊗ 1)F = F (1 ⊗ b) for all b ∈ B. This means that F is a separability idempotent in the sense of the literature. ii) Moreover, in our case, the two legs of F are all of the algebra B. This is not necessarily true for any separability element. Consider e.g. B = Mn(C), the algebra of n × n complex matrices, and let F = X i ei1 ⊗ e1i where eij are matrix elements. Then F is a separability element in the sense of i), but the legs do not give all of the algebra. Therefore, it will not fit into the concept as we studied it in [VD4.v1] (and as we use it here). (cid:3) The separability idempotents as we use them here are all of Frobenius type. Now we again consider the general setting and any regular separability idempotent E as in Definition B.1 above. B.5 Proposition There exist unique linear functionals ϕB on B and ϕC on C so that (ϕB ⊗ ι)E = 1 and (ι ⊗ ϕC)E = 1 in M (C) and M (B) respectively. These functionals are faithful. (cid:3) It follows from the formula (SB ⊗ SC)E = ζE and the uniqueness that ϕB = ϕC ◦ SB and ϕC = ϕB ◦ SC. If we now consider again the finite-dimensional case and F = (ι ⊗ SC )E as in the remark, we see that (ϕB ⊗ ι)F = 1 and (ι ⊗ ϕB)F = 1. Because it can be shown that B is made into a coalgebra with the coproduct ∆B on B defined as ∆B(b) = (b ⊗ 1)F = F (1 ⊗ b), in the literature, ϕB is called a counit. Indeed, we clearly have (ϕB ⊗ ι)∆B(b) = b and (ι ⊗ ϕB)∆B(b) = b with this definition of ∆B. In [VD4.v1], we have called ϕB and ϕC the integrals on B and C respectively. This seems to be completely contradictory. But in fact, there are arguments to support both points of view, see e.g. [VD4.v1] where this is explained. In the more recent version of this paper 32 [VD4.v2] we no longer use this terminology, but call the linear functionals distinguished linear functionals. Again in the finite-dimensional case, the existence of a faithful linear functional is equiv- alent with the algebra being Frobenius. So if a separability idempotent exists in our terminology, this implies that the algebra is separable and Frobenius. However, the sep- arability and the Frobenius property are connected and so we should really say that we have a separable Frobenius algebra. See e.g. [Sz], [Sc] or [K-S]. The distinguished linear functionals on B and C satisfy the weak K.M.S.-property. Indeed we have ϕB(bb′) = ϕB(b′σB(b)) and ϕC(cc′) = ϕC (c′σC(c)) where σB(b) = SC SB(b) and σC(c) = S−1 C (c) (see Proposition 2.3 in [VD4.v1]). Such automorphisms automatically exist for any faithful functional on a finite-dimensional al- gebra. They are then called the Nakayama automorphisms. More precisely, the Nakayama automorphism θ for the linear functional ϕB on B is the inverse of σB. See also some of our remarks about this topic in the previous appendix. B S−1 As it turns out, these modular automorphisms also exist for the integrals on a general separability idempotent. We tend to call these automorphisms the modular automorphisms. The terminology is motivated by the modular function relating the left and right Haar measure on a non-unimodular locally compact group (and furthermore also by the modular theory in operator algebras). Finally, we consider the multipliers Fi used in Propositions 2.6 and 2.7 of this paper. We also give a proof of Proposition 2.6. B.6 Proposition There exist elements F1 and F3 in M (B⊗B) and F2 and F4 in M (C ⊗C) characterized by F1 = (ι ⊗ SC)E F2 = (SB ⊗ ι)E and and F3 = (ι ⊗ S−1 F4 = (S−1 B )E C ⊗ ι)E. Proof: Because we assume that (b ⊗ 1)E and E(b ⊗ 1) belong to B ⊗ C for all b, we can define the element F1 in M (B ⊗ B) by (b ⊗ 1)F1 = (ι ⊗ SC)((b ⊗ 1)E) and F1(b ⊗ 1) = (ι ⊗ SC )(E(b ⊗ 1). Similarly for the other cases. (cid:3) We see that actually we already have that (b ⊗ 1)F1 and F1(b ⊗ 1) belong to B ⊗ B for all b ∈ B. Also (1 ⊗ b)F1 and F1(1 ⊗ b) belong to B ⊗ B for all b ∈ B. Similarly for the other elements F2, F3 and F4. Now we can easily prove the result of Proposition 2.6 of Section 2 of this paper. 33 B.7 Proposition The elements Fi satisfy (B.4) (B.5) E13(F1 ⊗ 1) = E13(1 ⊗ E) (1 ⊗ F2)E13 = (E ⊗ 1)E13 and (F3 ⊗ 1)E13 = (1 ⊗ E)E13 and E13(1 ⊗ F4) = E13(E ⊗ 1). Proof: Take any element b in B. If we use the Sweedler type notation E(1)b ⊗ E(2) for E(b ⊗ 1) we find E13(1 ⊗ E)(1 ⊗ b ⊗ 1) = E13(S−1 B (E(2)) ⊗ E(1)b ⊗ 1). Because (SB ⊗ SC )E = ζE we have S−1 B (E(2) ⊗ E(1)b = E(1) ⊗ SC(E(2))b and this proves the first formula. The other formulas are proven in a completely similar way. (cid:3) Observe that the expressions in (B.4) are in M (B ⊗ B ⊗ C) whereas the elements in the formulas (B.5) are in M (B ⊗ C ⊗ C). We also have that the elements Fi are determined by these formulas (because E is full). We finish this appendix with some more information about the involutive case (cf. Section 3 in [VD4.v1]). In this setting, it is assumed that B and C are ∗-algebras and that E is self-adjoint. It follows that SC (SB(b∗))∗ = b and SB(SC(c)∗)∗ = c for all b ∈ B and c ∈ C. The linear functionals ϕB and ϕC are positive. We also see that the antipodal maps SB and SC are ∗-maps if and only if the modular automorphisms σB and σC are trivial, that is if ϕB and ϕC are traces on B and C respectively. Because this is not true in general, from this point of view, it is not appropriate to identify C with Bop because they are not identified as involutive algebras. References [Abe] E. Abe: Hopf algebras. Cambridge University Press (1977). [Abr] L. Abrams: Modules, comodules and cotensor products over Frobenius algebras. J. Algebra 219 (1999), 201-213. [B-N-S] G. Bohm, F. Nill & K. Szlach´anyi: Weak Hopf algebras I. Integral theory and C∗-structure. J. Algebra 221 (1999), 385-438. 34 [B-G-L] G. Bohm, J. G´omez-Torecillas and E. L´opez-Centella: Weak multiplier bialgebras. Trans. Amer. Math. Soc. 367 (2015), no. 12, 8681-872. See also arXiv: 1306.1466 [math.QA]. [B-S] G. Bohm & K. Szlach´anyi: Weak Hopf algebras II. Representation theory, dimensions and the Markov trace. J. Algebra 233 (2000), 156-212. [DM-I] F. De Meyer & E. Ingraham: Separable algebras over a commutative rings. Lecture Notes in Mathematics 181 (1971). [E] M. Enock: Measured quantum groupoids with a central basis. J. Operator Theory, 66 (2011), 3-58. [J-V] K. Janssen & J. Vercruysse: Multiplier bi- and Hopf algebras. J. Algebra Appl. 9 (2), (2010), 275-303. [K-S] L. Kadison & K. Szlach´anyi: Bialgebroid actions on depth two extensions and duality. Adv. in Math. 179 (2003), 75-121. [K-V1] J. Kustermans & S. Vaes: Locally compact quantum groups. Ann. Sci. ´Ec. Norm. Sup. 33 (2000), 837 -- 934. [K-V2] J. Kustermans & S. Vaes: Locally compact quantum groups in the von Neumann algebraic setting. Math. Scand. 92 (2003), 68-92. [K-VD1] B.-J. Kahng & A. Van Daele: Seperability idempotents in C∗-algebras . Preprint Canisius College Buffalo (USA) and University of Leuven (Belgium). [K-VD2] B.-J. Kahng & A. Van Daele: A class of C∗-algebraic locally compact quantum groupoids I . Preprint Canisius College Buffalo (USA) and University of Leuven (Belgium). [K-VD3] B.-J. Kahng & A. Van Daele: A class of C∗-algebraic locally compact quantum groupoids II . Preprint Canisius College Buffalo (USA) and University of Leuven (Belgium). [L-S] R.G. Larson & M.E. Sweedler: An associative orthogonal bilinear form for Hopf algebras. Amer. J. Math. 91 (1969), 75-93. [L] F. Lesieur: Measured quantum groupoids. M´emoirs de la Soci´et´e Mathematique de France 109 (2007), 1-122. [Na] T. Nakayama: On Frobeniusean algebras II. Ann. of Math. 42, (1941), 1-21. [Ni] D. Nikshych: On the structure of weak Hopf algebras. Adv. Math. 170 (2002), 257-286. [N-V1] D. Nikshych & L. Vainerman: Algebraic versions of a finite dimensional quantum groupoid. Lecture Notes in Pure and Applied Mathematics 209 (2000), 189-221. [N-V2] D. Nikshych & L. Vainerman: Finite quantum groupoids and their applications. In New Directions in Hopf algebras. MSRI Publications, Vol. 43 (2002), 211-262. [Sw] M. Sweedler: Hopf algebras. Benjamin, New-York (1969). [Sc] P. Schauenburg: Weak Hopf algebras and quantum groupoids. Noncommutative ge- ometry and quantum groups (Warsaw, 2001), Polish Acad. Sci., Warsaw, (2003), 171-188. [St] R. Street: Frobenius monads and pseudomonoids. J. Math. Phys. 45 (2004), 3930- 3948. [Sz] K. Szlach´anyi: Finite quantum groupoids and inclusions of finite type. Fields Inst. Comm. 30 (2001), 393-407. 35 [T] M. Takesaki: Theory of Operator Algebras II. Springer, Berlin (2003). [T-VD1] T. Timmermann & A. Van Daele: Regular multiplier Hopf algebroids. Basic theory and examples. Preprint University of Munster (Germany) and University of Leuven (Belgium). See arXiv: 1307.0769 [math.QA] [T-VD2] T. Timmermann & A. Van Daele: Multiplier Hopf algebroids arising from weak multiplier Hopf algebras. Banach Center Publications, 106 (2015), 73-110. See also arXiv: 1406.3509 [math.RA] [Va] L. Vainerman (editor): Locally compact quantum groups and groupoids. IRMA Lec- tures in Mathematics and Theoretical Physics 2, Proceedings of a meeting in Strasbourg, de Gruyter (2003). [VD1] A. Van Daele: Multiplier Hopf algebras. Trans. Am. Math. Soc. 342(2) (1994), 917-932. [VD2] A. Van Daele: An algebraic framework for group duality. Adv. in Math. 140 (1998), 323-366. [VD3] A. Van Daele: Tools for working with multiplier Hopf algebras. ASJE (The Ara- bian Journal for Science and Engineering) C - Theme-Issue 33 (2008), 505 -- 528. See also arXiv:0806.2089 [math.QA]. [VD4.v1] A. Van Daele: Separability idempotents and multiplier algebras. Preprint Uni- versity of Leuven (Belgium) (2013). See arXiv:1301.4398v1 [math.RA] [VD4.v2] A. Van Daele: Separability idempotents and multiplier algebras. Preprint Uni- versity of Leuven (Belgium) (2015). See arXiv:1301.4398v2 [math.RA] [VD5] A. Van Daele: Locally compact quantum groups. A von Neumann algebra approach. Preprint University of Leuven (Belgium) (2013). SIGMA 10 (2014), 082, 41 pages. See also arXiv:0602.212v2 [math.OA]. [VD6] A. Van Daele: The Sweedler notation for (weak) multiplier Hopf algebras. Prepint University of Leuven (Belgium) (2016). In preparation. [VD7] A. Van Daele: Reflections on the Larson-Sweedler theorem for (weak) multiplier Hopf algebras. Prepint University of Leuven (Belgium) (2016). In preparation. [VD8] A. Van Daele: A note on coassociativity. Prepint University of Leuven (Belgium) (2016). In preparation. [VD-W1] A. Van Daele & S. Wang: The Larson-Sweedler theorem for multiplier Hopf algebras. J. of Alg. 296 (2006), 75-95. [VD-W2] A. Van Daele & S. Wang: Weak multiplier Hopf algebras. Preliminaries, mo- tivation and basic examples. Operator Algebras and Quantum Groups. Banach Center Publications 98 (2012), 367-415. See also arXiv:1210.3954v1 [math.RA] [VD-W3] A. Van Daele & S. Wang: Weak multiplier Hopf algebras I. The main theory. Journal fur die reine und angewandte Mathematik (Crelles Journal) 705 (2015), 155-209, ISSN (Online) 1435-5345, ISSN (Print) 0075-4102, DOI: 10.1515/crelle-2013-0053, July 2013. See also arXiv:1210.4395v1 [math.RA] 36 [VD-W4.v1] A. Van Daele & S. Wang: Weak multiplier Hopf algebras II. The source and target algebras. Preprint University of Leuven (Belgium) and Southeast University of Nanjing (China) (2014). See arXiv:1403.7906v1 [math.RA] [VD-W4.v2] A. Van Daele & S. Wang: Weak multiplier Hopf algebras II. The source and target algebras. Preprint University of Leuven (Belgium) and Southeast University of Nanjing (China) (2015). See arXiv:1403.7906v2 [math.RA] [VD-W5] A. Van Daele & S. Wang: Weak multiplier Hopf algebras III. Integrals and duality. Preprint University of Leuven (Belgium) and Southeast University of Nanjing (China) (2016). In preparation. [Ve] P. Vecserny´es: Larson-Sweedler theorem and the role of grouplike elements in weak Hopf algebras. J. Algebra 270 (2003), 471-520. See also arXiv: 0111045v3 [math.QA] for an extended version. 37
1710.02969	1	1710	2017-10-09T07:14:40	Hochshild cohomology of the associative conformal algebra Cend_{1,x}	[ "math.RA" ]	It is established in this work that second Hochshild cohomology group of the associative conformal algebra Cend_{1,x} is zero. As a corollary, this algebra split off in each extension with a nilpotent kernel. Key words: associative conformal algebra, splitting off radical, Hochshild cohomology.	math.RA	math	Hochshild cohomology of the associative conformal algebra Cend1,x UDC 512.55 Roman A. KOZLOV Sobolev Institute of Mathematics, Novosibirsk State University, 630090, Novosibirsk, Russia E-mail address: [email protected] It is established in this work that second Hochshild cohomology group of the associative conformal algebra Cend1,x is zero. As a corollary, this algebra split off in each extension with a nilpotent kernel. Key words: associative conformal algebra, splitting off radical, Hochshild cohomology. 1 Introduction and preliminary information 1.1 Introduction The axiomatic quantum field theory was originated in 50th of the last century in A. Whiteman's articles etc. Fields with a conformal simmetry were consid- ered in [5]. It can be represented as infinite in both side formal power series whose coefficients are linear operators. A key role in the conformal field theory is played by construction of an operator product expansion (OPE) an alge- braical description of which lead us to a vertex operator algebra (VOA) con- cept. An axiomatic definition of vertex algebra was introduced by R. Borcherds in [6]. V.G. Kac established [see 9] that coefficients of a singular part of the OPE allow us to describe a commutator of formal power series responding for VOA multiplication operators. In case of an algebraical description of these structures there appears a concept of conformal Lie algebras. For these algebras the structure theory was constructed in [8] and the representations 2 Roman A. Kozlov theory respectively in [7]. During the process of studying representations of the finite type there appear natural algebraic structures playing the same role with respect to conformal Lie algebras as common associative algebras to Lie algebras. These structures are named associative conformal algebras [see 7]. Let H = F[D] be a common polynomial algebra over a field F with a zero n! Dn, n ∈ Z+ and D(n) = 0, char, Z+ are non-negative integers, D(n) is for 1 n < 0. DEFINITION. A conformal algebra is a left H-module A equipped with a countable number of F-linear maps ◦n : A ⊗ A → A, n ∈ Z+ satisfying axioms: 1. a ◦n b = 0 for sufficiently large n, 2. Da ◦n b = −na ◦n−1 b, 3. a ◦n Db = D(a ◦n b) + na ◦n−1 b, where a, b ∈ A, n ∈ Z+. First axiom allows us to determine a locality function N : A × A → Z+ by the rule: N (a, b) = min{n ∈ Z+a ◦k b = 0 ∀k ≥ n}, a, b ∈ A. A conformal algebra is called associative if for every a, b, c ∈ A holds (a ◦n b) ◦m c = Ps≥0 (−1)s(cid:0)n s(cid:1)a ◦n−s (b ◦m+s c), m, n ∈ Z+, or equivalent condition a ◦n (b ◦m c) = Pt≥0(cid:0)n t(cid:1)(a ◦n−t b) ◦m+t c, m, n ∈ Z+. Construction of simple and semisimple associative algebras of the finite type was described in [8] (see also [12]). It is established in [13] that analogue of the Wedderburn theorem about a splitting radical (in reference to Lie con- formal algebras the same statement, which is analogue of the Levi - Maltsev theorem for Lie algebras, is incorrect [4]) is fulfilled for associative conformal algebras of the finite type. Associative conformal algebras with a faithful representation of the finite type form more wide class in comparison with algebras of the finite type. All Cohomology of the conformal algebra Cend1,x 3 such algebras is subalgebras of Cendn (the algebra of conformal linear maps of free n-generated F[D]-module) Cendn = Mn(F[D, x]), where ◦m : Cendn ⊗ Cendn → Cendn is determined by a common matrix differentiation, i.e. A(x) ◦m B(x) = A(x)B(m)(x). Simple and semisimple subalgebras of Cendn were desribed in [10] and there also was proved that every such algebra C has a nilpotent ideal (radical) defining a semisimple quotient with a faithful representation of the finite type. The radical splitting problem of this class of algebras was considered in [11, see also 1, 2]. There is a definition of Hochschild cohomology for associative conformal algebras in [1, 2] and in a context of pseudoalgebras [3] (slightly different from the definition in [4]). Cohomology play a key role in the radical splitting problem. Specifically, to proof analogue of the Wedderburn radical splitting theorem for a conformal algebra A = C/R it's enough to show that for arbitrary conformal A-bimodule M second Hochschild cohomology group H 2(A, M ) is zero. It is shown in [1] that for A = Cend1 and for every bimodule M next condition H 2(A, M ) = 0 is fulfilled. Contrariwise, it follows from the example [see 11] that there exist A-bimodule M for A = Cend1,x2 = x2Cend1 such that H 2(A, M ) 6= 0. In this paper we consider an intermidiate case: A = Cend1,x = xCend1. This associative conformal algebra is the universal enveloping algebra for the simple conformal Virasoro algebra with locality 2. The main result of this paper is next THEOREM. Let C = Cend1,x and let M be arbitrary conformal bi- module over C. Then H 2(C, M ) = 0. 1.2 Modules over conformal algebras DEFINITION. A left module over an associative conformal algebra A is an H- module V , closed with respet to the left action by elements of A. Specifically, 4 Roman A. Kozlov there is determined such linear maps ◦n : A ⊗ V → V, n ∈ Z+ that for every a, b ∈ A, v ∈ V , m ∈ Z+ (1) a ◦n v = 0 for sufficiently large n, (2) Da ◦n v = −na ◦n−1 v, (3) a ◦n Dv = D(a ◦n v) + na ◦n−1 v, (4) (a ◦m b) ◦n v = Ps≥0 (−1)s(cid:0)m s(cid:1)a ◦m−s (b ◦n+s v). Hereinafter a left module over an associative conformal algebra A is a simply left A-module. A right A-module could be defined in the same way. A bimodule is a simultaneously left and right A-module with one addi- tional axiom: (a ◦m v) ◦n b = Ps≥0 (−1)s(cid:0)m s(cid:1)a ◦m−s (v ◦n+s b). 2 Main definitions and formulation of the problem 2.1 Connection with pseudoalgebras An algebra H = F[D] is Hopf's algebra with a comultiplication ∆ : H → H ⊗ H, ∆(D) = D ⊗ 1 + 1 ⊗ D, a counit ε : H → F, ε(D) = 0 and an antipode S : H → H, S(D) = −D where ∆, ε and S are homomorhpisms of algebras. Assuming ∆1 = ∆ define by induction ∆n = (∆ ⊗ id ⊗ · · · ⊗ id)∆n−1, n > 1. Because of conformal algebras are objects of the pseudotensor category of left H-modules, it could be constructed by the pseudotensor language. One can define the right action H on H ⊗n by the rule: (h1 ⊗ h2 ⊗ · · · ⊗ hn)h = (h1 ⊗ h2 ⊗ · · · ⊗ hn)∆n−1(h). Then H ⊗n is a right H-module. Let V be a left unital H-module endowed with a bilinear operation ∗ : A ⊗ A → (H ⊗ H) ⊗H A. Such H-module A is called an H-pseudoalgebra and the operation ∗ is a pseudomultiplication. Every conformal algebra A is a H-pseudoalgebra and next equality a ∗ b = Xn≥0 (−D)(n) ⊗ 1 ⊗H (a ◦n b), a, b ∈ A, Cohomology of the conformal algebra Cend1,x 5 set up a correspondence between the pseudomultiplication ∗ and operations ◦n, n ∈ Z+. The definition of a module (bimodule) over an associative conformal algebra along with the associativity property could be represented through pseudomultiplication-like operations, see datails in [1,3]. For research purposes we translate main definitions in [1](a cochain, cocycle and coboundary) from the pseudoalgebras language to the ◦n-multiplications language, n ∈ Z+. 2.2 Definition of the Hochschild cohomology group DEFINITION. An n-cochain with coefficients in V is a H-polylinear map ϕ : A⊗n → (H ⊗n) ⊗H V , i.e. ϕ(h1a1 ⊗ · · · ⊗ hnan) = (h1 ⊗ · · · ⊗ hn ⊗H 1)ϕ(a1 ⊗ · · · ⊗ an). Denote by C n(A, V ) a set of every n-cochains of the algebra A and with coefficients in the bimodule V . In this paper only 1- and 2-cochains are interesting for us. Obviously, an 1-cochain τ is arbitrary H-polylinear map τ : A → V . For 2-cochains we use the natural representation: ϕ(a, b) = Xs≥0 ((−D)(s) ⊗ 1) ⊗H ϕs(a, b), where ϕs(a, b) ∈ V. (1) Thus a 2-cochain ϕ is a collection of bilinear maps ϕs : A ⊗ A → V , s ∈ Z+, such that 1. ϕs(Da, b) = −sϕs−1(a, b), 2. ϕs(a, Db) = Dϕs(a, b) + sϕs−1(a, b), 3. ϕs(a, b) = 0 for sufficiently large s. Similarly, a 3-cochain ψ is a 2-parametric collection of linear maps ψmn : A ⊗ A ⊗ A → V , n, m ∈ Z+ with the same properties. DEFINITION. A differential is a map δn : C n(A, V ) → C n+1(A, V ) given by the rule (δnϕ)(a1, . . . , an+1) = a1 ∗ ϕ(a2, . . . , an+1) + n Pi=1 (−1)iϕ(a1, . . . , ai ∗ ai+1, . . . , an+1) + (−1)n+1ϕ(a1, . . . , an) ∗ an+1. 6 Roman A. Kozlov Translating this definition to the language of ◦n-operations, n ∈ Z+, we obtain the differential for the 1-cohain τ (δ1τ )s(a, b) = τ (a) ◦n b − τ (a ◦n b) + a ◦n τ (b), and for the 2-cohain ϕ (δ2ϕ)mn(a, b, c) = a ◦m ϕn(b, c) + ϕm(a, b ◦n c)− (m s )(ϕn+s(a ◦m−s b, c) + ϕm−s(a, b) ◦n+s c), a, b, c ∈ M, n, m ∈ Z+ Xs≥0 Say an n-cochain ϕ is called an n-cocycle if δnϕ = 0. A cocycle ϕ ∈ Z n(A, V ) is called an n-coboundary if there exists an (n-1)-cochain ψ such that ϕ = δn−1ψ. Denote by Z n(A, V ) a set of every n-cocycles and by Bn(A, V ) a set of every n-coboundaries. DEFINITION. The Hochschild cohomology group of an associative con- formal algebra A on an A-bimodule V is H n(A, V ) = Z n(A, V )/Bn(A, V ). 2.3 Extensions. Connection with second cohomology group DEFINITION. An extension of a conformal algebra A is an ordered pair (B, σ) where B is a conformal algebra and σ is a homomorphism B to A. Any extension can be identified with an exact sequence 0 → Ker σ → B → A → 0. The conformal algebra A is called splitting off in the extension (B, σ) if B = A′ ⊗ Ker σ where A′ ⊆ B is a subalgebra isomorfic to A. The extension B called singular if (Ker σ ◦n Ker σ) = 0 for every n ∈ Z+. Let M be a bimodule over the conformal algebra A and ϕ ∈ C 2(A, M ). Construct a singular extension of the algebra A by the bimodule M . Let B be equal to the H-modules direct sum A ⊕ M . The operation ∗ can be defined by the rule (a1 + u1) ∗ (a2 + u2) = a1 ∗ a2 + a1 ∗ u2 + u1 ∗ a2 + ϕ(a1, a2), where (a1 + u1), (a2 + u2) ∈ B. Then B is a conformal algebra. Denote constructed extension by (A; M, ϕ). Cohomology of the conformal algebra Cend1,x 7 Next results are proved in [1]: LEMMA. If ϕ ∈ Z 2(A, M ), then B = (A; M, ϕ) is an associative conformal algebra. THEOREM. A conformal algebra A is splitted off in a singular ex- tension (B, σ) if and only if a cocycle ϕ is trivial in H 2(A, Ker σ). 2.4 Formulation of the problem Earlier we determined the conformal algebra Cendn, n ≥ 1. Here we consider case n = 1. The conformal subalgebra Cend1,x = xCend1 ⊂ Cend1 is a free H-module with a basis given by elements xk = x ◦0 x ◦0 · · · ◦0 x (k times), k ≥ 1. A question has been investigated in the paper sounds in this way: do the conformal algebra Cend1,x split off in every singular extension? In other words, whether we can construct for every exact sequence 0 → M → C → Cend1,x → 0, (where M 2 = 0 in sense of ◦-multiplication) such cross-section Cend1,x → C which is a conformal algebras homomorphism. 3 Main result As it is shown in 2.3 the splitting off problem for the conformal algebra Cend1,x in a singular extension with the nilpotent kernel M is in a tough connection with algebra's second Hochschild cohomology group with coefficients in M . The main result of the paper is next Theorem 1. Let Cend1,x be the associative conformal algebra described back in the text, M is a bimodule over it. Then H 2(Cend1,x, M ) = 0. Proof. We prove the theorem in several steps. Lemma 1. Arbitrary 2-cocycle ϕ ∈ Z 2(Cend1,x, M ) is completely determined by elements ϕt(x, x) and ϕ0(x, xl), t, l ≥ 1 laying down in the bimodule M . 8 Roman A. Kozlov Proof. Use the 2-cocycle definition: xk◦nϕm(xl, xq)+ϕn(xk, xl◦mxq) = Xs≥0 (cid:18)n s(cid:19)(ϕm+s(xk◦n−sxl, xq)+ϕn−s(xk, xl)◦m+sxq), (2) where n, m ∈ Z+, k, l, q ≥ 1. For n = 0, k = 1: ϕm(xl+1, xq) = x ◦0 ϕm(xl, xq) + m!(cid:18) q m(cid:19)ϕ0(x, xl+q−m) − ϕ0(x, xl) ◦m xq. (3) Use induction and express every values of the cocycle through ϕm(x, xq), m ≥ 0, q ≥ 1. Consider (2) for m = 0, k = l = 1: x◦nϕ0(x, xq)+ϕn(x, xq+1) = ϕn(x2, xq)+nϕn−1(x, xq)−Xs≥0 (cid:18)n s(cid:19)ϕn−s(x, x)◦sxq. From (3) one (4) that ϕn(x2, xq) = x ◦0 ϕn(x, xq) + n(cid:1)ϕ0(x, xq−n+1) − ϕ0(x, x) ◦n xq. We substitute the result into expression can see n!(cid:0)q (4): ϕn(x, xq+1) = x ◦0 ϕn(x, xq) + n!(cid:0)q n(cid:1)ϕ0(x, xq−n+1) − ϕ0(x, x) ◦n xq + nϕn−1(x, xq) − Ps≥0(cid:0)n s(cid:1)ϕn−s(x, x) ◦s xq − x ◦n ϕ0(x, xq). Assuming n = 1 and using induction on q we obtain that ϕ1(x, xq+1) is expressed through ϕt(x, x) and ϕ0(x, xl) for q, l ≥ 1, q ≥ l. Applying sequently induction on n and on q we obtain the same expression for each ϕn(x, xq+1), n, q ≥ 1. As required. Lemma 2. Let ϕ1(x, x) = 0 and ϕ0(x, xl) = 0 for every l ≥ 1. Then ϕ = 0. Proof. Due to Lemma 1 it is enough to prove that ϕ1(x, x) = 0 implies ϕl(x, x) = 0. Hereinafter designate ϕm(x, x) = ϕm, (x ◦m ·) = Lm ∈ End(M ), (· ◦m x) = Rm ∈ End(M ). Cohomology of the conformal algebra Cend1,x 9 Applying associativity identities we've rewritten useful formulas in the new designation: n LnLm(u) = (n s )((x ◦n−s x) ◦m+s u) = L0Lm+n(u) + nLm+n−1(u), Xs=0 RnRm(u) = m Xs=0 (−1)s(m s )(u ◦m−s (x ◦n+s x)) =   (5) (6) Rm(u), n = 1, 0, n > 1, m Xs=0 RnLm(u) = for u ∈ M . (−1)s(m s )(x◦m−s (u◦n+s x)) = (−1)s(m s )Lm−sRn+s(u), (7) m Xs=0 Using 2-cocycle identity (2) obtain following relation: (x ◦1 ϕm(x, x) − x ◦0 ϕm+1(x, x)) − ϕm(x ◦1 x, x) + (ϕ1(x, x ◦m x) − ϕ0(x, x ◦m+1 x)) − ϕ1(x, x) ◦m x = 0 Since the ◦-multiplication act on xk ∈ Cend1,x like a differentiation we obtain equalities: x ◦0 ϕ2 = x ◦1 ϕ1 − ϕ1 ◦1 x, x ◦0 ϕm+1 = x ◦1 ϕm − ϕm − ϕ1 ◦m x, m > 1. (8) (9) An assumption ϕ1 = 0 implies there are neither right part in (8) nor last summand in (9). For n = 1, m = 0 formula (5) gives L1L0 = ((x◦1 x)◦0 ·)+ ((x◦0 x)◦1 ·) = L0 + L0L1. Applying it to (9) obtain next equality Lm−2 0 ϕm = (L1 − (m − 2)) . . . (L1 − 1)ϕ2. The locality axiom implies there is such N that ϕm = 0 for each m > N . Let N > 1. Then for m = N + 1: (L1 − (N − 1)) . . . (L1 − 1)ϕ2 = 0. Denoting (L1 − (N − 2)) . . . (L1 − 1)ϕ2 = get a new equalities system: N −2 Ql=1 (L1 − l)ϕ2 = χ1 ∈ M we L1χ1 = x ◦1 χ1 = (N − 1)χ1, L0χ1 = x ◦0 χ1 = 0. (10)   10 Roman A. Kozlov Second equality is obtained from (8) by the sequently left multiplication on (L1 − 2), . . . , (L1 − (N − 1)) and with the condition (L1 − 1)L0 = L0L1. Show the system implies χ1 = 0. Indeed, using (5), the associativity identity and system (10): 0 = (x◦2 x)◦0 χ1 = L2L0χ1 −2L1L1χ1 +L0L2χ1 = L0L2χ1 −2L1L1χ1 ⇒ L0L2χ1 = 2L1L1χ1 (N − 1)2χ1 = L1L1χ1 = (x ◦1 x) ◦1 χ1 + (x ◦0 x) ◦2 χ1 = L1χ1 + L0L2χ1 = (N − 1)χ1 + 2(N − 1)2χ1 ⇒ 0 = ((N − 1) + (N − 1)2)χ1 = N (N − 1)χ1. As long as N > 1 last element in the chain of equalities is zero if only χ1 = 0. N −k Let χk−1 = (L1−l)ϕ2 = (L1−(N −k))χk, k = 2, . . . , N (χN −1 = ϕ2). Suppose reckoning above as the base of induction. Let χk−1 = 0 for certain k, Ql=1 1 < k < N . Then L1χk = (N − k)χk, L0χ1 = 0. (11)   By similar actions one can obtain 0 = (N − (k − 1))(N − k)χk = (N − (k − 1))(N − k) N −(k+1) Yl=1 (L1 − l)ϕ2. hence χk = 0. Therefore χk = 0 for k = 1, . . . , N − 1 and particularly χN −1 = ϕ2 = 0. Now, with the condition ϕ2 = 0 we have (9) rewritten in next way L0ϕ3 = 0, L0ϕm+1 = (L1 − 1)ϕm and with the same reflections we get ϕ3 = 0 and so on until ϕN +1 = 0 by the locality axiom. This is it, ϕm = 0 for every m ∈ Z+. Lemma 3. Consider a cocycle ϕ ∈ Z 2(Cend1,x, M ). Assume exists such ψ1 ∈ M that ϕ1(x, x) = x ◦1 ψ1 − ψ1 + ψ1 ◦1 x. (12) Then ϕ ∈ B2(Cend1,x, M ). Cohomology of the conformal algebra Cend1,x 11 Proof. Define by induction ψl, l ≥ 2 ψl+1 = −ϕ0(x, xl) + x ◦0 ψl + ψ1 ◦0 xl, l ≥ 1. (13) Construct a 1-cochain with values ψ(xl) = ψl. Consider δψ = ϕ′ ∈ B2(Cend1,x, M ). Then (12) implies ϕ′ 0(x, xl) = ϕ0(x, xl). ϕ′ 1(x, x) = ϕ1(x, x) and from (13) follows Denote ϕ − ϕ′ = π ∈ Z 2(Cend1,x, M ). Then π1(x, x) = 0 by the hipoth- esis, π0(x, xl) = 0 by the choice of ψl and then regarding Lemma 2 we have π = 0, i.e. ϕ = δψ as required. It is remained to show existance of the element ψ1. Firstly we derive a formula for Lm 0 ϕm+1, m ≥ 1. Lemma 4. If ϕ ∈ Z 2(Cend1,x, M ) then m Lm 0 ϕm+1 = (L1−m) . . . (L1−2)L1ϕ1− (L1−m) . . . (L1−(s+1))Ls−1 0 Rsϕ1. Xs=1 (14) Proof. By induction on m. Using (8) for m = 1 we obtain L0ϕ2 = (L1−R1)ϕ1. Then it is multiplied on the left by (L1 − 2). Applying (5): (L1 − 2)(L1 − R1)ϕ1 = (L1 − 2)L0ϕ2 = L0(L1 − 1)ϕ2. Substitute the result in (9): L2 0ϕ3 = (L1 − 2)(L1 − R1)ϕ1 − L0R2ϕ1. m → m + 1: Let the formula be true for Lm−1 0 ϕm. Then 0 ϕm+1 = Lm−1 Lm 0 = (L1 − m)Lm−1 0 (L0ϕm+1) = Lm−1 0 ((L1 − 1)ϕm − Rmϕ1) ϕm − Lm−1 0 Rmϕ1 = (L1 − m)((L1 − (m − 1)) . . . (L1 − 2)L1ϕ1 − m−1 Xs=1 (L1 − (m − 1)) . . . (L1 − (s + 1))Ls−1 0 Rsϕ1) − Lm−1 0 Rmϕ1 m = (L1 − m) . . . (L1 − 2)L1ϕ1 − As required. (L1 − m) . . . (L1 − (s + 1))Ls−1 0 Rsϕ1, Xs=1 12 Roman A. Kozlov Consider in this step multiplications Lk−1 0 Lk. By induction on k we will show that the equality Lk−1 0 Lk = L1(L1 − 1) . . . (L1 − (k − 1)) (15) is fulfilled. Using (5) obtain k = 2 : L0L2 = L2 k → k + 1 : (L1 − k)Lk−1 1 − L1 = L1(L1 − 1); 0 Lk = Lk−1 0 (L1 − 1)Lk = −Lk−1 0 Lk + Lk−1 0 (Lk + L0Lk+1) = Lk 0Lk+1. By the hypothesis the formula is correct for Lk−1 0 Lk. Obviously, (L1 − k) commutates with every elements of the same form. So (15) is correct. Now it's easy to prove next result. Lemma 5. Let V ⊂ M be a finite-dimensional space invariant with respect to L1.Then there exists a decomposition in the direct sum of L1-invariant subspaces V = V0 ⊕ V1 ⊕ · · · ⊕ Vk where L1v = iv for v ∈ Vi. Proof. Since dim V < ∞ there exists only finite number of the basis elements v1, . . . , vn. Formula (15) for each vi gives: Lk 0Lk+1vi = L1(L1 − 1) . . . (L1 − k)vi. The left part of the equality is zero for sufficiently large k due to locality. Choose k = max(k1, . . . , kn). Then Pk = t(t − 1) . . . (t − k) is such polynomial that Pk(L1)V = 0, i.e. Pk(L1) is the annihilator of the space V . This polyno- mial has no multiple roots thus the minimal polinomial of the operator L1 on V has no multiple roots likewise, i.e. L1 is semisimple and every eigenvalues are laying down into Z+. Then required decomposition V = V0 ⊕ V1 ⊕ . . . ⊕ Vk is exists but generally speaking there might be zero spaces in the direct sum. Consider the algebra of endomorphisms End M and subalgebra gener- ated by operators L1 and Lm 0 Rm+1, m ≥ 0: A = sub hL1, Lm 0 Rm+1 m ≥ 0i. Cohomology of the conformal algebra Cend1,x 13 Lemma 6. Let W = Au = {au a ∈ A} be an A-module for arbitrary u ∈ M . Then dim W < ∞. Proof. It's easy to see that W ⊆ W ′=spanhLs 1 Rn+s, m, s ≥ 0, n ≥ 1i. Indeed, it follows from (5), (6), (7) that W ′ is closed with respect to the 0Lm action of generators of A. Let us show that dim W ′ < ∞. Firstly notice that the locality axiom implies dim V ′ < ∞ where V ′=spanhRn+s(u), s ≥ 0, n > 0i. After, 1 = L1 + F (L0L2, . . . , Lm−2 Lm 0 Lm−1) + Lm−1 0 Lm, 1) = L1(L1 + L0L2) = L2 where F is a linear function. Actually, using (5) obtain L2 1 = L1(L2 1 + L1L0L2 = L1 + L0L2 + L0(L1 + 1)L2 = L1 + 2L0L2 + L0L1L2 = L1 + 2L0L2 + L0L2 + L2 0L3 and so on. Hence elements L1V ′ generate the same linear subspace as L0Lk+1V ′ 0L3 = L1 + 3L0L2 + L2 1 = L1 + L0L2, L3 which are finite dimensional by virtue of the locality axiom. For end of the proof of the theorem consider the algebra A ⊆ End(M ) defined above and the A-module V = Aϕ1. Let ϕ1 ∈ M be a value of a 2-cocycle ϕ and it fulfill the formula (14). For sufficiently large m it turns to m (L1 − m) . . . (L1 − 2)L1ϕ1 = (L1 − m) . . . (L1 − (s + 1))Ls−1 0 Rsϕ1 (16) Xs=1 As it shown back in the proof the operator L1 divides the algebra V on the finite number of invariant subspaces hence ϕ1 = v0 + · · · + vm, where L1vi = ivi. Substitute that decomposition in (16). Since there is a collection of nilpotent operators in the left part of the formula in result only v1 remains: m Ps=1 ∓(m − 1)!v1 = (L1 − m) . . . (L1 − (s + 1))Ls−2 0 Rs−1ϕ1 14 Roman A. Kozlov Use the operator Rk, k ≥ 2 to both part of the equality and obtain Rkv1 = Const · Rk( (L1 − m) . . . (L1 − s)Ls−2 0 Rs−1 − Lm−1 0 Rm)ϕ1. m Ps=2 It follows from (6) and (7) that Rkv1 = 0, for k ≥ 2. Initially find such z ∈ V that (L1 − 1)z = v0 + v2 + v3 · · · + vm = ϕ1 − v1. It exists because (L1 − 1) annihilate only eigenspace V1 and is non-degenerate in other. Choose a 1-cochain ξ such that its value in a bimodule M is z, i.e. ξ(x) = z. Denote ϕ′ = ϕ − δξ ∈ Z 2(Cend1,x, M ). Since (6), (7) the cocycle ϕ′ has a property: Rkϕ′ 1 = Rk(ϕ1 −(L1+R1−1)z) = Rk(ϕ1 −(L1−1)z−R1z) = Rk(v1 −R1z) = 0 for k ≥ 2. Without loss of generality one can assume Rkϕ1 = 0, k ≥ 2. (17) Applying (5), (6), (7) to a decomposition of aϕ1 ∈ Aϕ1, a ∈ A we gather all operators Rk at the right. Obtain V = Aϕ1 ⊆ F[L1, L0]R1ϕ1. It's easy to check using (7), (17) that RkV = 0 for each k ≥ 2. Relying on this result, show there is ψ1 satisfying the equality (L1 +R1 − 1)ψ1 = ϕ1. Notice the identity [Lk, R1] = 0, k ≥ 0 is fulfilled in End(V ) ([·, ·] is a common commutator). Indeed, for k = 0 endomorphisms commutate. For other k it follows because of RkV = 0: [Lk, R1]v = x ◦k (v ◦1 x) − (x ◦k v) ◦1 x = x ◦k (v ◦1 x) − k Xl=0 (−1)l(cid:18)k l(cid:19)x ◦k−l (v ◦1+l x) (−1)l−1(cid:18)k l(cid:19)Lk−lR1+lv = 0. k Xl=1 = x ◦k (v ◦1 x) − x ◦k (v ◦1 x) + (18) Also (6) implies that R1 acts on M as the projection operator, i.e. R2 1 = R1. Hence the space V is decomposed in the direct sum of the kernel V (0) and fixed points V (1). It follows from (18) that V (0) and V (1) are closed with respect to the action of L1. Cohomology of the conformal algebra Cend1,x 15 Having the decomposition ϕ1 = ϕ(0) 1 + ϕ(1) 1 it's enough to show there exists such ψ(0), ψ(1) that (L1 + R1 − 1)ψ(0) = (L1 − 1)ψ(0) = ϕ(0) 1 , (L1 + R1 − 1)ψ(1) = L1ψ(1) = ϕ(1) 1 . (19) Values of the cocycle ϕ1 satisfies to (14) and (19) hence 0 = (L1 − m) . . . (L1 − 2)(L1 − R1)ϕ1 = (L1 − m) . . . (L1 − 2)(L1 − R1)ϕ(0) = (L1 − m) . . . (L1 − 2)L1ϕ(0) 1 + (L1 − m) . . . (L1 − 2)(L1 − R1)ϕ(1) 1 + (L1 − m) . . . (L1 − 2)(L1 − 1)ϕ(1) 1 . 1 First summand is the element of V (0), second summand is the element of V (1) and the sum might be zero if and only if each summand is zero: (L1 − m) . . . (L1 − 2)L1ϕ(0) 1 = 0, (L1 − m) . . . (L1 − 2)(L1 − 1)ϕ(1) 1 = 0. Therefore ϕ(0) 1 ∈ V0 ⊕ V2 ⊕ · · · ⊕ Vk and the operator L1 − 1 is non-degenerate in this direct sum of spaces. It implies there exists such ψ(0) ∈ V that (L1 − 1)ψ(0) = ϕ(0) 1 . By the same way, ϕ(1) 1 ∈ V1 ⊕ · · · ⊕ Vk and L1 is non-degenerate there and hence exists ψ(1) ∈ V such that L1ψ(1) = ϕ(1) 1 . Hereby it is shown existance of the element ψ1 = ψ(0) + ψ(1) which is required for the Lemma 3. The theorem is proved. Bibliography: 1. Dolguntseva I. A. Hochschild cohomologies for associative conformal al- gebras // Algebra and logic. 2007. V. 46, N.6. P. 688–706. 2. Dolguntseva I. A. Triviality of the second cohomology group of the con- formal algebras Cendn and Curn // St. Petersburg Math. J. 2010. V.21, P. 53-63. 16 Roman A. Kozlov 3. Bakalov B., D'Andrea A., Kac V. G. Theory of finite pseudoalgebras // Adv. Math. 2001. V.162, N.1. P. 1-140. 4. Bakalov B., Kac V. G., Voronov A. Cohomology of conformal algebras // Comm. Math. Phys. 1999. V. 200, N.3. P.561-598. 5. Belavin A. A., Polyakov A. M., Zamolodchikov A. B. Infinite conformal symmetry in two-dimansional quantum field theory // Nucl. Phis. B. 1984. V. 241. P. 333-380. 6. Borcherds R. E. Vertex algebras, Kac - Moody algebras, and the Monster // Proc. Nat. Acad. Sci. U.S.A. 1986. V. 83. P. 3068-3071. 7. Cheng S.-J., Kac V. G. Conformal modules // Asian J. Math. 1997. V. 1. P. 181-193. 8. D'Andrea A., Kac V. G. Structure theory of finite conformal alhebras // Selecta Math. New. Ser.1998. V. 4. P. 377-418. 9. Kac V. G. Vertex algebras for beginners. Second edition. Providence, RI: AMS,1998. (University Lecture Series, vol. 10). 10. Kolesnikov P. S. Associative conformal algebras with finite faithful rep- resentation // Adv. Math. 2006. V. 202, N.1. P. 602-637. 11. Kolesnikov P. S. On the Wedderburn principal theorem in conformal algebras // Journal of Algebra and Its Applications. 2007. V. 6, N.1. P. 119-134. 12. Zelmanov E. I. On the structure of conformal algebras // Proc. / In- tern. Conf. on Combinatorial and Computational Algebra. Hong Kong, China. May 24-29, 1999. Providence, RI: AMS, 2000. P. 139-153. (Cont. Math., vol. 264). 13. Zelmanov E. I. Idempotentns in conformal algebras // Proc. / Third Intern. Alg. Conf. Tainan, Taiwan. June 16-July 1, 2002. Ed. by Y. Fong et al. Dordrecht: Kluwen Acad. Publ., 2003. P. 257-266.
1706.07988	1	1706	2017-06-24T18:25:04	A note on multiplicative commutators of division rings	[ "math.RA" ]	We give an example of a division ring $D$ whose multiplicative commutator subgroup does not generate $D$ as a vector space over its centre, thus disproving the conjecture posed in the paper "Vector space generated by the multiplicative commutators of a division ring, J. Algebra Appl. 12 (2013), no. 8".	math.RA	math	A NOTE ON MULTIPLICATIVE COMMUTATORS OF DIVISION RINGS ROOZBEH HAZRAT Abstract. We give an example of a division ring D whose multiplicative commutator subgroup does not generate D as a vector space over its centre, thus disproving the conjecture posed in [1]. Let D be a division ring. Denote by F := Z(D), the centre of D and D′ its multiplicative commutator subgroup, i.e., the group generated by the set of multiplicative commutators (cid:8)xyx−1y−1 x, y ∈ D\{0}(cid:9). There are classical results due to Herstein, Kaplansky and Scott, among others, showing that the group D′ is "dense" in D (see for example [3, §13]). In [1] the authors study the F -vector space T (D) generated by the set of multiplicative commutators. They prove that if T (D) is radical over F , then D = F , and if dimF T (D) < ∞, then dimF D < ∞. Furthermore, they prove that T (D) contains all separable elements of D and thus if D is an algebraic division ring over its centre with char(D) = 0, then T (D) = D. They then conjecture [1, Abstract and Conjecture 1] that a division ring is generated by all multi- plicative commutators as a vector space over its centre, i.e., T (D) = D for any arbitrary division ring. Here we give a counterexample to this conjecture. In fact we show that the multiplicative subgroup D′ can not recover D as a vector space and [D : T (D)] = ∞. We recall the Hilbert classical construction of division rings (see [2, §1]). Let L be a field, σ ∈ Aut(L) and F be the fixed field of σ. Let D = L((x, σ)) be the division ring of formal Laurent series, consisting of elements P∞ i=n aiti, where ai ∈ L and n ∈ Z, addition defined component-wise and multiplication by ∞ ∞ ∞ (cid:16) X i=m aiti(cid:17)(cid:16) X j=n bjtj(cid:17) = X r=n+m (cid:16) X i+j=r aiσi(bj)(cid:17)tr. (1) By [2, §1, Lemma 4] if the order of σ is infinite then Z(D) = F and [D : F ] = ∞ and if the order is a finite number n, then Z(D) = F ((tn)) and [D : Z(D)] = n2. It is easy to see that D = L((x, σ)) is a valued division ring with value group Z as follows v : D∗ −→ Z, ∞ X i=n aiti 7−→ n. Therefore we have D′ ⊆ ker(v) = nL + X i>0 aitio. (2) Now if we choose an automorphism σ of infinite order, since Z(D) = F , Equation 2 shows that T (D) ⊆ nL + Pi>0 aitio, and thus [D : T (D)] = ∞. References [1] Aghabali, M., Akbari, S., Ariannejad, M., Madadi, A. Vector space generated by the multiplicative commutators of a division ring, J. Algebra Appl. 12 (2013), no. 8, 7 pp. 1 Key words and phrases. Division ring, Multiplicative commutator. The author acknowledges Australian Research Council grant DP160101481. This work was done at the University of Munster, where the author was a Humboldt Fellow. 1 2 ROOZBEH HAZRAT [2] Draxl, P.K, Skew fields, London Mathematical Society Lecture Note Series, 81, Cambridge University Press, Cambridge, 1983. 1 [3] Lam, T.Y, A first course in noncommutative rings, volume 13, Graduate Texts in Mathematics. Springer-Verlag, New York, Second edition, 2001. 1 Western Sydney University, Australia E-mail address: [email protected]
1711.05970	1	1711	2017-11-16T07:53:55	On homological smoothness of generalized Weyl algebras over polynomial algebras in two variables	[ "math.RA" ]	Homological smoothness and twisted Calabi-Yau property of generalized Weyl algebras over polynomial algebras in two variables is studied. A necessary and sufficient condition to be homologically smooth is given. The Nakayama automorphisms of such algebras are also computed in terms of the Jacobian determinants of defining automorphisms.	math.RA	math	ON HOMOLOGICAL SMOOTHNESS OF GENERALIZED WEYL ALGEBRAS OVER POLYNOMIAL ALGEBRAS IN TWO VARIABLES LIYU LIU Abstract. Homological smoothness and twisted Calabi-Yau property of gen- eralized Weyl algebras over polynomial algebras in two variables is studied. A necessary and sufficient condition to be homologically smooth is given. The Nakayama automorphisms of such algebras are also computed in terms of the Jacobian determinants of defining automorphisms. 1. Introduction Motivated by the study of algebras analogous with the classical Weyl algebra A1(k) over a field k, Bavula introduced in [2] the notion of (degree one) generalized Weyl algebras over the polynomial algebra k[z]. Later on, generalized Weyl algebras over any algebra B were defined in [3]. Roughly speaking, a generalized Weyl algebra over B is an extension of B by two formal variables x, y, parameterized by an automorphism σ on B and a central element ϕ ∈ B, denoted by B(σ, ϕ). Many people have intensively studied the situation B = k[z], and such generalized Weyl algebras are denoted by W(1) in this paper. It is illustrated in [2] that the global dimensions of W(1) are equal to 1, 2, or ∞. and the latter occurs if and only if the defining polynomial ϕ admits a multiple root (see also [3], [18], [35]). Their Hochschild homology and cohomology were computed in [15], [35]. In particular, a remarkable result in [15] is that to assure a duality between its Hochschild homology and cohomology (with coefficients in W(1) itself), W(1) should have finite projective dimension as a bimodule over itself, or equivalently, be homologically smooth. To answer [25, Question 2] (see also [28]), the author explained in [27] when W(1) is homologically smooth; a necessary and sufficient condition was given, under which a duality between its Hochschild homology and cohomology with coefficients in any bimodule was established. Global dimensions of generalized Weyl algebras W(n) over B = k[z1, . . . , zn] were also examined by Bavula [1]. Among the results mentioned above and others, B is always required to be com- mutative. This is not incidental. From an algebraic point of view, noncommutative rings are more rigid than commutative ones, namely, a commutative ring possesses "more" automorphisms and central elements than a noncommutative ring. A typi- cal example is: if B is the generic quantum 2-plane kq[z1, z2], then σ is necessarily given by σ(z1) = az1, σ(z2) = bz2 for some nonzero scalars a, b ∈ k, and the central element ϕ must be in k. One concludes that in this case any generalized Weyl 2010 Mathematics Subject Classification. Primary 16S38, 16E10, 16E40. Key words and phrases. Generalized Weyl algebra, Homological smoothness, Twisted Calabi- Yau algebra. The author acknowledges the supports of the Natural Science Foundation of China No. 11501492, the Natural Science Foundation of Jiangsu Province No. BK20150435 and the Nat- ural Science Foundation for Universities in Jiangsu Province No. 15KJB110022. He is grateful to the referee for the valuable comments. He would like to send his thanks to Xuefeng Mao, Quanshui Wu, and James Zhang for helpful conversations, special thanks to Shengyun Jiao for her Master research that is related to §5.3. 1 2 LIYU LIU algebra over kq[z1, z2] is isomorphic to a localization of a quantum 3-space. Hence the noncommutativity of B usually makes the research of generalized Weyl algebras trivial. So generalized Weyl algebras over commutative rings are more interesting. According to papers on this topic, especially the papers by Bavula, two spaces are extremely important in researching generalized Weyl algebras. One is the space MaxSpec(B)/hσi of orbits where hσi is the cyclic group generated by σ acting on the space MaxSpec(B) of maximum spectrum naturally; the other is the σ-stable space {σn(ϕ) n ∈ Z}. Normally, many properties, such as global dimensions and irreducible representations, depend on both spaces simultaneously. But there is an exception. It is illustrated in [27] that whether a generalized Weyl algebra W(1) is homologically smooth depends only on ϕ, also, its Nakayama automorphism depends only on σ. This paper is a sequel to [27], in which we plan to study the homological smooth- ness of generalized Weyl algebras W(2) over the polynomial algebra k[z1, z2]. Ho- mological smoothness, which is a noncommutative generalization of smoothness for commutative algebras, plays an important role in homological algebra, mathemat- ical physics, etc. An algebra A is called homologically smooth if A admits a bound resolution by finitely generated projective Ae-modules (where Ae := A ⊗k Aop is the enveloping algebra), or equivalently, A is isomorphic to a perfect complex in the derived category Db(Ae-Mod). Recall that in [27] our strategy was to construct a free W e (1)-module resolution of W(1) and then to compute cohomology by it. We will follow the idea for W(2) in this paper. Since the notion of homotopy double complex introduced in [27] is useful to construct a free resolution of W(2), we re- view it briefly in this paper. After that, noncommutative differential 1-forms and derivations on k[z1, z2] are introduced, and a noncommutative version of Jacobian determinant is defined accordingly. All of these appear in §2. After the preliminaries, we begin §3 by constructing a free resolution of W(2). The resolution is given by Proposition 3.3. Using the resolution, we compute the Hochschild cohomology H ∗(W(2), M ) with coefficients in any bimodule M . Notice that ϕ under consideration is a polynomial ϕ(z1, z2). Let ϕ1, ϕ2 be the two formal partial derivatives of ϕ with respect to z1, z2, and (ϕ, ϕ1, ϕ2) be the ideal of k[z1, z2] generated by the three elements. Our result is that H 4(W(2), M ) = 0 if (ϕ, ϕ1, ϕ2) = k[z1, z2]. On the other hand, inspired by works of Bavula, we prove that W(2) has infinite global dimension if (ϕ, ϕ1, ϕ2) is a proper ideal. Therefore, we obtain the main theorem (Theorem 3.1): Theorem 1.1. For any σ ∈ Aut(k[z1, z2]), W(2) is homologically smooth if and only if (ϕ, ϕ1, ϕ2) = k[z1, z2]. Observe that W(n) is commutative if σ = id, and in this case we write it as W (n). We remark that if k is of characteristic zero, then the condition (ϕ, ϕ1, ϕ2) = k[z1, z2] is exactly equivalent to that W (2) is smooth in the commutative sense. Coincidentally, it is the same case with W (1), which has been explained in [27] using deformation theory. We regard W(2) as a deformation of W (2) by the automorphism σ. The smoothness of W (2) is equivalent to the homological smoothness of W(2). See Remark 3.1. The Nakayama automorphism of an algebra is an important invariant. It is re- lated with the study of rigid dualizing complexes, twisted Poincar´e duality, Hopf algebra actions on Artin-Schelter regular algebras, and so forth. Examples of alge- bras that have Nakayama automorphisms are: noetherian Artin-Schelter Gorenstein algebras, many noetherian Hopf algebras, some (co)invariant subalgebras of Artin- Schelter regular algebras, PBW deformations of some graded algebras. In general, the computation of Nakayama automorphisms is not easy. We refer to [6], [7], [22], ON HOMOLOGICAL SMOOTHNESS OF GENERALIZED WEYL ALGEBRAS 3 [34], [30], [31], [33], [40] and the references therein for the progress on this topic during the past years. The goal in §4 is to compute the Nakayama automorphism of W(2). By virtue of the free resolution constructed in §3, we compute the group Ext3 (2)). An explicit formula for Nakayama automorphism is hence obtained. As a consequence, we have (Theorem 4.2, Corollaries 4.3, 4.5) (W(2), W e W e (2) Theorem 1.2. Let W(2) = k[z1, z2](σ, ϕ) be a generalized Weyl algebra, and J be the Jacobian determinant of σ. Then Exti W e (2) (W(2), W e (2)) ∼=(0, W ν (2), i 6= 3, i = 3. where ν ∈ Aut(W(2)) is the Nakayama automorphism, given by ν(x) = Jx, ν(y) = J −1y, ν(z1) = z1, ν(z2) = z2. In particular, if (ϕ, ϕ1, ϕ2) = k[z1, z2], then (1) W(2) is twisted 3-Calabi-Yau; (2) W(2) is Calabi-Yau if and only if J = 1; (3) the Hochoschild cohomology HH •(W(2)) has a Batalin-Vilkovisky structure. In the final section §5, we study the homological smoothness and Nakayama automorphisms of some concrete algebras, by using the two previous theorems. These algebras contain the quantum groups Oq(SL2), U (sl2), noetherian down-up algebras, quantum lens space, and others. In each case our main theorems give alternative proofs of results from the literature. 2. Preliminaries Throughout, k is a field, and all algebras are over k unless stated otherwise. Unadorned ⊗ means ⊗k. Let A be an algebra and M an A-bimodule. The group of algebra automorphisms of A is denoted by Aut(A). For any σ ∈ Aut(A), denote by σM (resp. M σ) the left A-module (resp. right A-module) whose ground k-module is the same with M and whose left (resp. right) A-action is twisted by σ, that is, a ⊲ m = σ(a)m (resp. m ⊳ a = mσ(a)) for any a ∈ A, m ∈ M . Let Aop be the opposite algebra of A, and Ae = A ⊗ Aop the enveloping algebra of A. An A-bimodule can be viewed as a left Ae-module in a natural way. Recall that A is homologically smooth if A as a left (or equivalently, right) Ae-module, admits a finitely generated projective resolution of finite length. 2.1. Generalized Weyl algebras. We recall the definition of generalized Weyl algebras given by Bavula in [2], [3]. Definition 2.1. Suppose B is an algebra. For a central element ϕ ∈ B and an algebra automorphism σ ∈ Aut(B), the associated (degree one) generalized Weyl algebra (GWA for short) W is by definition generated by two variables x and y over B subject to xb = σ(b)x, yb = σ−1(b)y, ∀b ∈ B, yx = ϕ, xy = σ(ϕ). The algebra is written as W = B(σ, ϕ). We adopt the convention about the super/sub-scripts of cochain/chain complexes as follows: · · · −→ C i−1 di−1 −−−→ C i di −−→ C i+1 −→ · · · · · · −→ Ci+1 di−−→ Ci di−1−−−→ Ci−1 −→ · · · 4 LIYU LIU and call a chain complex C• in an abelian category A alternate in [n, +∞) if di = di+2 as morphisms in A for all i ≥ n. So necessarily, Ci = Ci+2 = Ci+4 = · · · for all i ≥ n. A similar definition is given for cochain complexes. Proposition 2.1. [27] Let W = B(σ, ϕ) be a GWA as above. Suppose further that ϕ is a regular element. Then W as a left W e-module can be represented by an alternate complex C• ∈ Ch≥0(W e-Mod) in [1, +∞). Concretely, C0 = W ⊗B W , C1 = (W σ ⊗B W ) ⊕ (W ⊗B σW ), C2 = (W ⊗B W ) ⊕ (W σ ⊗B σW ), and dC dC 0 (1 ⊗ 1, 0) = 1 ⊗ x − x ⊗ 1, 0 (0, 1 ⊗ 1) = 1 ⊗ y − y ⊗ 1, dC dC 1 (0, 1 ⊗ 1) = (1 ⊗ y, x ⊗ 1), 1 (1 ⊗ 1, 0) = (y ⊗ 1, 1 ⊗ x), 2 (1 ⊗ 1, 0) = (−x ⊗ 1, 1 ⊗ x), dC dC 2 (0, 1 ⊗ 1) = (1 ⊗ y, −y ⊗ 1). The augmentation C0 → W is given by the multiplication map. Remark 2.1. Although the second summand of C2 can be simplified to W ⊗B W as the author wrote in [27], we insist on the expression with double σ so as to make the computation easier in the following part. 2.2. Noncommutative differential forms and partial derivations. Since in this paper we mainly focus on GWAs over the polynomial algebra in two vari- ables, a noncommutative version of differential forms and partial derivations will be introduced first of all. From now on, let B = k[z1, z2]. For any polyno- mial g = g(z1, z2) ∈ B, the noncommutative differential 1-form dg is defined as g ⊗ 1 − 1 ⊗ g. The noncommutative partial derivations with respect to z1, z2 are defined as k-linear maps ∆1 : B −→ B ⊗ B ∆2 : B −→ B ⊗ B i1 Xj=1 zi1 1 zi2 2 7−→ We have (2.1) zi1−j 1 ⊗ zj−1 1 zi2 2 , zi1 1 zi2 2 7−→ 1 zi2−j zi1 2 ⊗ zj−1 2 . i2 Xj=1 dg = ∆1(g)dz1 + ∆2(g)dz2 which is a noncommutative analogy of the total derivative formula in calculus. Let µ be the multiplication of B. Then it is easy to check that µ∆1 = ∂/∂z1 and µ∆2 = ∂/∂z2. Suppose that σ : B → B is an endomorphism which is determined by σ(z1) = f1(z1, z2), σ(z2) = f2(z1, z2). We call the determinant the noncommutative Jacobian determinant of σ. If we take the image of each entry by µ, it becomes the usual Jacobian determinant of σ, ∆1(f1) ∆1(f2) Jnc =(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ∆2(f1) ∆2(f2)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) J =(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ∂f1 ∂z1 ∂f1 ∂z2 ∂f2 ∂z1 ∂f2 ∂z2 . Let u, v be any k-linear maps of B, we denote u∆v i = (u ⊗ v) ◦ ∆i and udvg = (u ⊗ v)(dg) = u(g) ⊗ 1 − 1 ⊗ v(g). By convention, u or v is usually omitted if it is the identity map. Lemma 2.2. Let σ be an endomorphism of B, and f1, f2 as above. We have σdσzi = ∆1(fi)dz1 + ∆2(fi)dz2 ON HOMOLOGICAL SMOOTHNESS OF GENERALIZED WEYL ALGEBRAS 5 for i = 1, 2. Proof. Directly from (2.1). (cid:3) Remark 2.2. The elements dzi, ∆i(g), etc. can be regarded as in W e via the em- bedding B ⊗ B ֒→ W ⊗ W op. The reader will not confuse them. 2.3. Homotopy double complexes. In [27] the author introduced the notion of homotopy double complexes in order to present GWAs W(1). Let us recall the definition. Definition 2.2. Suppose that A is an abelian category. Let {C pq}p,q∈Z be a family of objects in A together with morphisms dv, dh, t of degrees (0, 1), (1, 0), (2, −1) respectively. The 4-tuple (C ••, dv, dh, t) is called a homotopy double cochain complex if the following conditions are fulfilled: (2.2) (2.3) (2.4) (2.5) (2.6) d2 v = 0, dhdv + dvdh = 0, d2 h + dvt + tdv = 0, dht + tdh = 0, t2 = 0. A homotopy double chain complex is defined in a similar way. The associated total complex (Tot C ••, d) is defined by (Tot C ••)n = Mp+q=n C pq and d = dv + dh + t, which is a generalization of the usual total complex of a usual double complex. 3. Homological smoothness of a GWA over k[z1, z2] Throughout this section, B = k[z1, z2], W(2) = B(σ, ϕ) is a GWA over B. We will construction a homotopy double complex for W(2) whose total complex is a free W e (2)-resolution of W(2). Using this resolution, a necessary and sufficient condition under which W(2) is homologically smooth is discussed. Recall that ϕ = ϕ(z1, z2). Write ϕi = ∂ϕ/∂zi, i = 1, 2, and let (ϕ, ϕ1, ϕ2) be the ideal in B generated by ϕ, ϕ1, ϕ2. In this section we will prove Theorem 3.1. For any σ ∈ Aut(B), W(2) is homologically smooth if and only if (ϕ, ϕ1, ϕ2) = B. Remark 3.1. Notice that W (2) := B(id, ϕ) is commutative. When char k = 0, W (2) is smooth (in the commutative sense) if and only if (ϕ, ϕ1, ϕ2) = B. To put it another way: W(2) is homologically smooth if and only if W (2) is smooth. (1) A similar phenomenon exists for W(1): W(1) is homologically smooth if and only if W (1) := k[z](σ, ϕ) is smooth. This has been explained in deformation theory [27]. Since an automorphism σ : k[z] → k[z] is necessarily given by σ(z) = λz + η for some λ ∈ k× and η ∈ k, one can regard W(1) as a deformation of W (1), following Van den Bergh [36]. (2) Back to B = k[z1, z2], one does not know the expressions of σ(z1), σ(z2) for an arbitrary σ ∈ Aut(B), although the van der Kulk theorem reveals the structure of the group Aut(B), i.e., it decomposes into a coproduct of two subgroups [14], [32], [39]. Thus one cannot say that W(2) is a deformation of W (2), unless σ is affine. However, the phenomenon can be summarized in this sentence: σ preserves the (non)smoothness of W(n) for n = 1, 2. 6 LIYU LIU Remark 3.2. Differential smoothness is another noncommutative generalization of smoothness. Brzezi´nski discussed noncommutative calculi for a class of differentially smooth GWA W(1) and W(2) whose defining automorphisms σ are affine [9]. Two kinds of smoothness are compared, and examples of algebras that are differentially but not necessarily homologically smooth are given. A relationship between the two forms of smoothness has not yet been understood. 3.1. Construction of homotopy double complex. Since B admits the follow- ing Koszul complex 0 −→ B ⊗ B −−−−−−−→ (B ⊗ B)2 (cid:16) dz1 dz2 (cid:17) (dz2 −dz1) −−−−→ B ⊗ B as a B-bimodule resolution via B ⊗ B of W(2) ⊗B W(2), W σ (2) ⊗B W(2), W(2) ⊗B µ −−→ B, we obtain left W e (2)-free resolutions σW(2), and W σ (2) ⊗B σW(2) as follows: 0 −→ W e (2) (dz2 −dz1) −−−−−−−→ (W e −−−−→ W e (2) −→ W(2) ⊗B W(2) −→ 0, (σdz2 −σdz1) −−−−−−−−→ (W e −−−−−→ W e (2) −→ W σ (2) ⊗B W(2) −→ 0, 0 −→ W e (2) 0 −→ W e (2) (2))2 (cid:16) dz1 dz2 (cid:17) σdz1 (2))2 (cid:16) σdz2 (cid:17) (2))2 (cid:16) dσz1 dσz2 (cid:17) σdσz1 (2))2 (cid:16) σdσz2 (cid:17) (dσz2 −dσz1) −−−−−−−−→ (W e −−−−−→ W e (2) −→ W(2) ⊗B 0 −→ W e (2) (σdσz2 −σdσz1) −−−−−−−−−−→ (W e −−−−−→ W e (2) −→ W σ (2) ⊗B σW(2) −→ 0, σW(2) −→ 0. We will construct a double complex (P••, dh, dv, t) in the next step. To be more intuitive, we draw a diagram to illustrate our construction P12 h❘❘❘❘❘❘❘ P11 h❘❘❘❘❘❘❘ P22 P21 h❘❘❘❘❘❘❘ ❘❘❘❘❘❘❘ h❘❘❘❘❘❘❘ ❘❘❘❘❘❘❘ P10 P20 ❘❘❘❘❘❘❘ ❘❘❘❘❘❘❘ P32 P31 P30 P02 P01 P00 ② ② µ ② C0 C1 C2 C3 ② W(2) · · · · · · · · · · · · where C• is the complex given in Proposition 2.1, Pi• is a projective resolution of Ci for each i, and the dashed arrow P00 → W(2) is the composition P00 → C0 → W(2), equal to the multiplication map µ. Based on the alternate complex C• in Proposition 2.1, we erect the four resolu- tions, and then obtain the embryo of a homotopy double complex: (2), P00 = P02 = W e P01 = P10 = P12 = P20 = P22 = · · · = (W e P11 = P21 = P31 = · · · = (W e (2))4, (2))2, and all other entries are zero. Moreover, the morphisms dv are expressed by dv 00 =(cid:18)dz1 dz2(cid:19) , dv 01 =(cid:0)−dz2 dz1(cid:1) , o o o o o o o o o o o o h o o h o o o o o o h o o h o o o o o o o o o o o o ON HOMOLOGICAL SMOOTHNESS OF GENERALIZED WEYL ALGEBRAS 7 dv dv σdz1 σdz2 0 0 dz1 dz2 0 0 10 =  20 =  0 0 dσz1 dσz2 0 0 σdσz1 σdσz2     , dv 11 =(cid:18)−σdz2 0 , dv 21 =(cid:18)−dz2 0 σdz1 0 0 −dσz2 0 dσz1(cid:19) , 0 dz1 0 −σdσz2 0 σdσz1(cid:19) , and the rest are hence known according to the alternating feature. Next we add appropriate morphisms dh, t making P•• into a homotopy double complex. The morphisms are given as follows: −(1 ⊗ x)∆2(f1) x ⊗ 1 − (1 ⊗ x)∆2(f2) (y ⊗ 1)∆2(f1) −1 ⊗ y + (y ⊗ 1)∆2(f2) ,   −1 ⊗ x 0 −x ⊗ 1 −1 ⊗ x 0 −x ⊗ 1 0 −1 ⊗ x 0 y ⊗ 1     , , ∆2(ϕ) −1 ⊗ y + (y ⊗ 1)∆1(f1) 0 0 0 0 dh dh dh dh dh dh dh dh dh 0 0 1 ⊗ x 1 ⊗ x 1 ⊗ x −1 ⊗ y −1 ⊗ y −y ⊗ 1 −y ⊗ 1 x ⊗ 1 − (1 ⊗ x)∆1(f1) (y ⊗ 1)∆1(f2) −(1 ⊗ x)∆1(f2) 00 =(cid:18)1 ⊗ x − x ⊗ 1 1 ⊗ y − y ⊗ 1(cid:19) , 01 =  1 ⊗ y − (y ⊗ 1)Jnc (cid:19) , 02 =(cid:18)−x ⊗ 1 + (1 ⊗ x)Jnc 10 =(cid:18)y ⊗ 1 1 ⊗ y x ⊗ 1(cid:19) , 11 =  12 =(cid:18)y ⊗ 1 1 ⊗ y x ⊗ 1(cid:19) , 20 =(cid:18)−x ⊗ 1 1 ⊗ y −y ⊗ 1(cid:19) , 21 =  22 =(cid:18)−x ⊗ 1 t01 =(cid:18) t02 =  t11 =(cid:18)σ∆1(ϕ) t12 =  1 (ϕ)∆1(f1) + σ∆σ −∆2(ϕ) ∆1(ϕ) σ∆σ σ∆σ 0 0 −∆σ 2 (ϕ) ∆σ 1 (ϕ) 1 ⊗ x 1 ⊗ y −y ⊗ 1(cid:19) , −σ∆2(ϕ) σ∆1(ϕ) 2 (ϕ) 1 (ϕ)     −Jnc Jnc σ∆2(ϕ) −1 ⊗ y −1 ⊗ y ∆1(ϕ) x ⊗ 1 x ⊗ 1 σ∆σ 0 0 0 0 0 0 0 0 0 0 , , −1 ⊗ x y ⊗ 1 2 (ϕ)∆1(f2) σ∆σ 1 (ϕ)∆2(f1) + σ∆σ 2 (ϕ)∆2(f2)(cid:19) , 0 0 1 (ϕ) ∆σ ∆σ 2 (ϕ)(cid:19) , LIYU LIU 0 σ∆σ 2 (ϕ)(cid:19) , 8 0 σ∆σ 1 (ϕ) 0 0 t21 =(cid:18)∆1(ϕ) ∆2(ϕ) t22 =  −∆2(ϕ) ∆1(ϕ) 0 0 0 0 −σ∆σ 2 (ϕ) σ∆σ 1 (ϕ) .   Proposition 3.2. The 4-tuple (P••, dv, dh, t) is a homotopy double cochain com- plex. Proof. Notice that (2.2) is clearly satisfied, and so (2.3)-(2.6) are to be verified. All the verifications are translated into anti-multiplications of matrices over W e (2). We confine ourselves to the proof of (3.1) (3.2) dh 00dh dh 01dh 11 + dv 10 + dv 00t01 = 0, 01t02 + t01dv 20 = 0, leaving others to the reader. We have dh 00dh 10 =(cid:18)y ⊗ 1 1 ⊗ y x ⊗ 1(cid:19)(cid:18)1 ⊗ x − x ⊗ 1 1 ⊗ y − y ⊗ 1(cid:19) =(cid:18) 1 ⊗ ϕ − ϕ ⊗ 1 1 ⊗ σ(ϕ) − σ(ϕ) ⊗ 1(cid:19) , 1 ⊗ x and dv 00t01 equals 2 (ϕ)∆1(f2) σ∆σ 1 (ϕ)∆2(f1) + σ∆σ ∆2(ϕ) dz2(cid:19) 2 (ϕ)∆2(f2)(cid:19)(cid:18)dz1 ∆1(ϕ)dz1 + ∆2(ϕ)dz2 2 (ϕ)∆1(f2))dz1 + (σ∆σ 1 (ϕ)∆2(f1) + σ∆σ 2 (ϕ)∆2(f2))dz2(cid:19) 2 (ϕ)σdσz2(cid:19) (by Lemma 2.2) ∆1(ϕ) 1 (ϕ)∆1(f1) + σ∆σ σ∆σ (σ∆σ (cid:18) =(cid:18) =(cid:18) =(cid:18) ϕ ⊗ 1 − 1 ⊗ ϕ σ∆σ σ(ϕ) ⊗ 1 − 1 ⊗ σ(ϕ)(cid:19) . 1 (ϕ)∆1(f1) + σ∆σ ϕ ⊗ 1 − 1 ⊗ ϕ 1 (ϕ)σdσz1 + σ∆σ Hence (3.1) is proven. For (3.2), we have dh 01dh −y ⊗ 1 0 −1 ⊗ x 0 0 −y ⊗ 1 0 −1 ⊗ x −1 ⊗ y 0 −x ⊗ 1 0 0 −1 ⊗ y 0 −x ⊗ 1   x ⊗ 1 − (1 ⊗ x)∆1(f1) −(1 ⊗ x)∆2(f1) −(1 ⊗ x)∆1(f2) x ⊗ 1 − (1 ⊗ x)∆2(f2) −1 ⊗ y + (y ⊗ 1)∆1(f1) (y ⊗ 1)∆2(f1) (y ⊗ 1)∆1(f2) −yx ⊗ 1 + 1 ⊗ yx −1 ⊗ y + (y ⊗ 1)∆2(f2) 0 0 −yx ⊗ 1 + 1 ⊗ yx (1 ⊗ xy − xy ⊗ 1)∆1(f1) (1 ⊗ xy − xy ⊗ 1)∆1(f2) (1 ⊗ xy − xy ⊗ 1)∆2(f1) (1 ⊗ xy − xy ⊗ 1)∆2(f2)     1 ⊗ ϕ − ϕ ⊗ 1 0 0 1 ⊗ ϕ − ϕ ⊗ 1 (1 ⊗ σ(ϕ) − σ(ϕ) ⊗ 1)∆1(f1) (1 ⊗ σ(ϕ) − σ(ϕ) ⊗ 1)∆1(f2) (1 ⊗ σ(ϕ) − σ(ϕ) ⊗ 1)∆2(f1) (1 ⊗ σ(ϕ) − σ(ϕ) ⊗ 1)∆2(f2) ,   11 =  ·  =  =  ON HOMOLOGICAL SMOOTHNESS OF GENERALIZED WEYL ALGEBRAS 9 dv 01t02 =  −∆2(ϕ) ∆1(ϕ) −J σ∆σ J σ∆σ 2 (ϕ) 1 (ϕ)  (cid:0)−dz2 dz1(cid:1) =  ∆2(ϕ)dz2 −∆1(ϕ)dz2 −∆2(ϕ)dz1 ∆1(ϕ)dz1 Jnc −Jnc σ∆σ σ∆σ 2 (ϕ)dz2 −Jnc 1 (ϕ)dz2 Jnc σ∆σ σ∆σ 2 (ϕ)dz1 1 (ϕ)dz1 ,   and t01dv 20 is equal to ∆1(ϕ)dz1 ∆1(ϕ)dz2 ∆2(ϕ)dz1 ∆2(ϕ)dz2 (σ∆σ (σ∆σ 1 (ϕ)∆1(f1) + σ∆σ 1 (ϕ)∆1(f1) + σ∆σ 2 (ϕ)∆1(f2))σdσz1 2 (ϕ)∆1(f2))σdσz2 (σ∆σ (σ∆σ 1 (ϕ)∆2(f1) + σ∆σ 1 (ϕ)∆2(f1) + σ∆σ 2 (ϕ)∆2(f2))σdσz1 2 (ϕ)∆2(f2))σdσz2   .   It follows that the (1, 1)-, (1, 2)-, (2, 1)-, (2, 2)-entries of dh all zero. The (3, 1)-entry is 01dh 11 + dv 01t02 + t01dv 20 are σ∆σ σ∆σ 2 (ϕ)dz2 + (σ∆σ 2 (ϕ)dz2 + σ∆σ 1 (ϕ)∆1(f1) + σ∆σ 1 (ϕ)∆1(f1)σdσz1 + σ∆σ 1 (ϕ)σdσz1 − σdσϕ)∆1(f1) + Jncdz2 + ∆1(f2)σdσz1)σ∆σ −σdσϕ ∆1(f1) + Jnc = −σdσϕ ∆1(f1) + Jnc = (σ∆σ = (−σ∆σ = (Jncdz2 − ∆1(f1)σdσz2 + ∆1(f2)σdσz1)σ∆σ = (Jncdz2 − ∆1(f1)(∆1(f2)dz1 + ∆2(f2)dz2) + ∆1(f2)(∆1(f1)dz1 2 (ϕ)σdσz2)∆1(f1) + Jncdz2 + ∆1(f2)σdσz1)σ∆σ 2 (ϕ) 2 (ϕ) 2 (ϕ) 2 (ϕ)∆1(f2))σdσz1 2 (ϕ)∆1(f2)σdσz1 + ∆2(f1)dz2))σ∆σ 2 (ϕ) = (Jnc − ∆1(f1)∆2(f2) + ∆1(f2)∆2(f1))dz2 = 0. σ∆σ 2 (ϕ) Similarly, the (3, 2)-, (4, 1)-, (4, 2)-entries are zero. Thus (3.2) is also proven. (cid:3) Remark 3.3. When constructing homotopy double complex for W(1) in [27], the verification of (2.5), (2.6) is trivial. But for W(2), this is not so easy. Proposition 3.3. If ϕ 6= 0, Tot P•• is a resolution of W(2) by finitely generated free W e (2)-modules via µ. Moreover, the complex Tot P•• is alternate in [3, +∞). Proof. This follows from spectral sequence argument. See [27] for the details. (cid:3) 3.2. Proof of sufficiency. Suppose (ϕ, ϕ1, ϕ2) = B. Let us prove that W(2) is homologically smooth in this case. First of all, notice that ϕ 6= 0 is automatically satisfied. Hence by Proposition 3.3, Tot P•• is a free resolution of W(2), and so we can compute Hochschild cohomol- ogy H ∗(W(2), M ) = Ext∗ (2)-module M by Tot P••. Next pq, M ), dpq let Q•• pq, M ), M = HomW e and tpq = HomW e M , dv, dh, t) is a homotopy double cochain complex, and (W(2), M ) for any W e (dv (tpq, M ). Clearly, (Q•• h = HomW e v = HomW e (P••, M ), dpq (dh W e (2) (2) (2) (2) (2) H ∗(Tot Q•• M ) ∼= H ∗(HomW e (Tot P••, M )) ∼= H ∗(W(2), M ). 10 LIYU LIU We write Q• M schematically, as follows. d32 h / M 2 / · · · M d01 v M 2 d00 v M d02 h M 2 ◗◗◗◗◗◗◗◗◗◗ d11 v d01 h M 4 ◗◗◗◗◗◗◗◗◗◗ d10 v M 2 d12 h d22 h v t02 t12 d11 d31 d21 M 2 ◗◗◗◗◗◗◗◗◗◗ (◗◗◗◗◗◗◗◗◗◗ M 4 ◗◗◗◗◗◗◗◗◗◗ (◗◗◗◗◗◗◗◗◗◗ M 2 ◗◗◗◗◗◗◗◗◗◗ (◗◗◗◗◗◗◗◗◗◗ M 4 ◗◗◗◗◗◗◗◗◗◗ (◗◗◗◗◗◗◗◗◗◗ d21 d20 d30 t11 t01 h h v v v M 2 M 2 t22 (◗◗◗◗◗◗◗◗◗◗ d31 h d41 v M 4 t21 (◗◗◗◗◗◗◗◗◗◗ d40 v M 2 d00 h d10 h d20 h d30 h · · · · · · Proposition 3.4. One has H 4(Tot Q•• quently, W(2) is homologically smooth if (ϕ, ϕ1, ϕ2) = B. M ) = 0 for all M if (ϕ, ϕ1, ϕ2) = B. Conse- Proof. There exist polynomials α, β1, β2 such that αϕ + β1ϕ1 + β2ϕ2 = 1. We write d• Q for the differentials of Tot Q•• M . 3 , m31 2 , m31 1 , m22 1 , m31 2 , m31 Let m = (m22 4 , m40 1 , m40 2 ) be a 4-cocycle of Tot Q•• M . We have t22(m22 1 , m22 d22 h (m22 2 ) + d31 h (m31 t31(m31 1 , m22 1 , m31 1 , m31 2 ) + d31 2 , m31 2 , m31 3 , m31 3 , m31 1 , m31 4 ) + d40 4 ) + d40 2 , m31 v (m40 h (m40 3 , m31 1 , m40 1 , m40 4 ) = 0 2 ) = 0 2 ) = 0 v (m31   which is equivalent to the following eight equations −(x ⊗ 1)m22 (1 ⊗ y)m22 1 + (1 ⊗ x)m22 1 − (y ⊗ 1)m22 1 − (y ⊗ 1)m31 1 − (y ⊗ 1)m31 −∆2(ϕ)m22 ∆1(ϕ)m22 2 (ϕ)m22 1 (ϕ)m22 −σ∆σ σ∆σ 2 − (1 ⊗ y)m31 2 − (1 ⊗ y)m31 1 − (x ⊗ 1)m31 2 − (x ⊗ 1)m31 σ∆1(ϕ)m31 1 (ϕ)m31 1 + σ∆2(ϕ)m31 2 (ϕ)m31 3 + ∆σ 2 − (x ⊗ 1)m40 4 + (1 ⊗ y)m40 ∆σ 2 − σdz2m31 2 − dσz2m31 1 − (1 ⊗ x)m31 2 − (1 ⊗ x)m31 1 + σdz1m31 3 + dσz1m31 3 + dz1m40 4 + dz2m40 3 + σdσz1m40 4 + σdσz2m40 1 + (1 ⊗ x)m40 1 − (y ⊗ 1)m40 2 = 0, 4 = 0, 1 = 0, 1 = 0, 2 = 0, 2 = 0, 2 = 0, 2 = 0. (3.3) (3.4) (3.5) (3.6) (3.7) (3.8) (3.9) (3.10) Define n12 1 = (1 ⊗ β1)m31 2 = (1 ⊗ σ(β1))m31 n12 1 = (1 ⊗ β1)m40 n21 n21 2 = (1 ⊗ β2)m40 3 = (1 ⊗ σ(β1))m40 n21 n21 4 = (1 ⊗ σ(β2))m40 1 = −(1 ⊗ β1)σ∆∂1 n30 2 = (1 ⊗ αy)m40 n30 2 − (1 ⊗ β2)m31 1 , 3 + (1 ⊗ αy)m22 4 − (1 ⊗ σ(β2))m31 1 , 2 (ϕ)m22 1 , 1 (ϕ)m22 1 , 1 + (1 ⊗ β2)∆∂2 1 − (1 ⊗ β1)∆∂1 2 − (1 ⊗ αy)m31 2 − (1 ⊗ αy)m31 1 (ϕ)m31 1 + (1 ⊗ σ(β2))σ∆σ∂2 2 − (1 ⊗ σ(β1))σ∆σ∂1 2 1 (ϕ)m22 2 , (ϕ)m22 2 , 1 − (1 ⊗ β2)σ∆∂2 (ϕ)m31 2 (ϕ)m31 2 , 3 − (1 ⊗ σ(β2))∆σ∂2 2 1 − (1 ⊗ σ(β1))∆σ∂1 1 (ϕ)m31 4 . / / ( / / ( / / ( / / / / ( O O / / ( O O / / ( O O / / O O / / O O / / O O / / O O / / O O / / O O / / O O ON HOMOLOGICAL SMOOTHNESS OF GENERALIZED WEYL ALGEBRAS 11 ∗ constitute a 3-cochain n. Let us prove d3 Q(n) = m. The following three These n∗∗ equations are to be checked: (3.11) (cid:18)m22 1 m22 2 (cid:19) =(cid:18)y ⊗ 1 1 ⊗ y x ⊗ 1(cid:19)(cid:18)n12 1 n12 1 ⊗ x 2 (cid:19) +(cid:18)−dz2 0 0 dz1 0 −σdσz2 0 σdσz1(cid:19)  n21 1 n21 2 n21 3 n21 4 ,   0 x ⊗ 1 0 −1 ⊗ y 0 x ⊗ 1 0 0 −1 ⊗ y −1 ⊗ x 0 y ⊗ 1 0 −1 ⊗ x 0 y ⊗ 1   (3.12) m31 1 m31 2 m31 3 m31 4   (3.13) (cid:18)m40 1 m40 −σ∆2(ϕ) σ∆1(ϕ)   0 0 n21 1 n21 2 n21 3 n21 4 =   ·   2 (cid:19) =(cid:18)σ∆1(ϕ) 0 0 0 −∆σ 2 (ϕ) ∆σ 1 (ϕ) σdz1 σdz2   0 0 0 0 dσz1 dσz2 1 n12 2 (cid:19) + (cid:18)n12   (cid:18)n30 2 (cid:19) ,  1 n30 +  σ∆2(ϕ) 0 0 0 1 (ϕ) σ∆σ σ∆σ 2 (ϕ)(cid:19)  n21 1 n21 2 n21 3 n21 4 1 ⊗ x  1 ⊗ y x ⊗ 1(cid:19)(cid:18)n30 +(cid:18)y ⊗ 1 2 (cid:19) .  1 n30 There are eight equalities in total to be verified and the verification is tediously long. So we divide the whole proof into four lemmas. The sufficiency follows from them. (cid:3) Lemma 3.5. For i = 1, 2, we have i (ϕ)dzi = 1 ⊗ ϕi, (1) ∆i(ϕ) − ∆∂i (2) σ∆i(ϕ) − σ∆∂i i (ϕ) − ∆σ∂i (3) ∆σ i (ϕ) − σ∆σ∂i (4) σ∆σ i i i (ϕ)σdzi = 1 ⊗ ϕi, (ϕ)dσzi = 1 ⊗ σ(ϕi), (ϕ)σdσzi = 1 ⊗ σ(ϕi). Proof. Clearly, (2), (3), (4) follow from (1). So let us prove (1) for i = 1. The case i = 2 is left to the reader. Suppose ϕ =Pi,j λij zi 1zj 2. Then ∆1(ϕ) =Xi,j λij zi−k 1 ⊗ zk−1 1 zj 2 i Xk=1 and so i ∆∂1 1 (ϕ)dz1 =Xi,j =Xi,j =Xi,j =Xi,j i i Xk=1 Xk=1 Xk=1 Xk=0 i−1 λij zi−k 1 ⊗ ∂1(zk−1 1 zj 2)(z1 ⊗ 1 − 1 ⊗ z1) λij zi−k 1 ⊗ (k − 1)zk−2 1 zj 2(z1 ⊗ 1 − 1 ⊗ z1) λij (k − 1)zi−k+1 1 ⊗ zk−2 1 zj λij kzi−k 1 ⊗ zk−1 1 zj 2 −Xi,j λij (k − 1)zi−k 1 ⊗ zk−1 1 zj 2 i Xk=1 λij(k − 1)zi−k 1 ⊗ zk−1 1 zj 2 i 2 −Xi,j Xk=1 12 LIYU LIU i λij zi−k 1 ⊗ zk−1 1 =Xi,j Xk=1 = ∆1(ϕ) − 1 ⊗ ϕ1. λij i ⊗ zi−1 1 zj 2 zj 2 −Xi,j (cid:3) Lemma 3.6. Eq. (3.11) holds true. Proof. We have (y ⊗ 1)n12 1 + (1 ⊗ x)n12 2 − dz2n21 1 + dz1n21 2 = (y ⊗ 1)[(1 ⊗ β1)m31 + (1 ⊗ αy)m22 − (1 ⊗ β1)∆∂1 2 − (1 ⊗ β2)m31 1 ] − dz2[(1 ⊗ β1)m40 1 (ϕ)m22 1 ] 1 ] + (1 ⊗ x)[(1 ⊗ σ(β1))m31 1 + (1 ⊗ β2)∆∂2 2 (ϕ)m22 4 − (1 ⊗ σ(β2))m31 3 1 ] + dz1[(1 ⊗ β2)m40 1 = (1 ⊗ β1)[(y ⊗ 1)m31 2 + (1 ⊗ x)m31 4 − dz2m40 1 − ∆∂1 1 (ϕ)dz1m22 · [−(y ⊗ 1)m31 1 − (1 ⊗ x)m31 †1= (1 ⊗ β1)[∆1(ϕ)m22 1 − ∆∂1 3 − ∆∂2 1 (ϕ)dz1m22 1 + dz1m40 2 (ϕ)dz2m22 1 ] + (1 ⊗ β2)[∆2(ϕ)m22 1 ] + (1 ⊗ β2) 1 ] + (1 ⊗ αyx)m22 1 1 − ∆∂2 2 (ϕ)dz2m22 1 ] + (1 ⊗ αϕ)m22 1 †2= (1 ⊗ β1)(1 ⊗ ϕ1)m22 = m22 1 1 + (1 ⊗ β2)(1 ⊗ ϕ2)m22 1 + (1 ⊗ αϕ)m22 1 where †1 follows from (3.6), (3.5) , and †2 from Lemma 3.5 (1). Also, (1 ⊗ y)n12 1 + (x ⊗ 1)n12 = (1 ⊗ y)[(1 ⊗ β1)m31 2 − σdσz2n21 2 − (1 ⊗ β2)m31 3 + σdσz1n21 4 1 ] + (x ⊗ 1)[(1 ⊗ σ(β1))m31 4 − (1 ⊗ σ(β2))m31 3 + (1 ⊗ αy)m22 1 ] − σdσz2[(1 ⊗ σ(β1))m40 · σ∆σ∂2 2 ] + σdσz1[(1 ⊗ σ(β2))m40 (ϕ)m22 (ϕ)m22 · σ∆σ∂1 2 ] 2 1 2 − (1 ⊗ αy)m31 2 − (1 ⊗ αy)m31 1 + (1 ⊗ σ(β2)) 2 − (1 ⊗ σ(β1)) = (1 ⊗ σ(β1))[(1 ⊗ y)m31 2 + (x ⊗ 1)m31 2 − σ∆σ∂1 1 (ϕ)σdσz1m22 2 ] 4 − σdσz2m40 3 − σ∆σ∂2 2 1 − (x ⊗ 1)m31 + (1 ⊗ σ(β2))[−(1 ⊗ y)m31 + (1 ⊗ αy)[(x ⊗ 1)m22 †3= (1 ⊗ σ(β1))[σ∆σ 1 (ϕ)m22 − σ∆σ∂2 2 (ϕ)σdσz2m22 1 + σdz2m31 2 − σ∆σ∂1 1 − σdz1m31 2 ] (ϕ)σdσz1m22 2 ] + (1 ⊗ αy)(1 ⊗ x)m22 2 1 (ϕ)σdσz2m22 2 + σdσz1m40 2 ] 2 ] + (1 ⊗ σ(β2))[σ∆σ 2 (ϕ)m22 2 †4= (1 ⊗ σ(β1))(1 ⊗ σ(ϕ1))m22 = m22 2 2 + (1 ⊗ σ(β2))(1 ⊗ σ(ϕ2))m22 2 + (1 ⊗ σ(αϕ))m22 2 where †3 follows from (3.8), (3.7), (3.3), and †4 from Lemma 3.5 (4). (cid:3) Lemma 3.7. Eq. (3.12) holds true. Proof. We have −σ∆2(ϕ)n12 1 + (x ⊗ 1)n21 1 − (1 ⊗ x)n21 3 + σdz1n30 1 = −σ∆2(ϕ)[(1 ⊗ β1)m31 2 (ϕ)m22 2 ] + σdz1[−(1 ⊗ β1)σ∆∂1 + (1 ⊗ β2)∆∂2 · σ∆σ∂2 (ϕ)m22 2 − (1 ⊗ β2)m31 1 ] − (1 ⊗ x)[(1 ⊗ σ(β1))m40 1 (ϕ)m31 2 1 ] + (x ⊗ 1)[(1 ⊗ β1)m40 1 = (1 ⊗ β1)[−σ∆2(ϕ)m31 2 + (x ⊗ 1)m40 1 − (1 ⊗ x)m40 2 − σ∆∂1 1 − (1 ⊗ β2)σ∆∂2 2 (ϕ)m31 2 ] 1 (ϕ)σdz1m31 1 ] 2 − (1 ⊗ αy)m31 1 + (1 ⊗ σ(β2)) ON HOMOLOGICAL SMOOTHNESS OF GENERALIZED WEYL ALGEBRAS 13 1 + σ∆∂2 2 (ϕ)(x ⊗ 1)m22 1 − σ∆∂2 2 (ϕ)(1 ⊗ x)m22 2 1 ] + (1 ⊗ β2)[σ∆2(ϕ)m31 1 + (1 ⊗ β2)[σ∆2(ϕ)m31 − σ∆∂2 2 (ϕ)σdz1m31 2 ] + (1 ⊗ αyx)m31 1 †1= (1 ⊗ β1)[σ∆1(ϕ)m31 1 − σ∆∂1 2 (ϕ)σdz2m31 1 ] 1 (ϕ)σdz1m31 − σ∆∂2 + (1 ⊗ αϕ)m31 1 †2= (1 ⊗ β1)(1 ⊗ ϕ1)m31 = m31 1 1 + (1 ⊗ β2)(1 ⊗ ϕ2)m31 1 + (1 ⊗ αϕ)m31 1 where †1 follows from (3.9), (3.3), and †2 from Lemma 3.5 (2). Next we have σ∆1(ϕ)n12 1 + (x ⊗ 1)n21 2 − (1 ⊗ x)n21 4 + σdz2n30 1 = σ∆1(ϕ)[(1 ⊗ β1)m31 2 −(1 ⊗ β2)m31 1 ] + (x ⊗ 1)[(1 ⊗ β2)m40 1 −(1 ⊗ β1)∆∂1 1 (ϕ)m22 1 ] − (1 ⊗ x)[(1 ⊗ σ(β2))m40 + σdz2[−(1 ⊗ β1)σ∆∂1 1 (ϕ)m31 1 − (1 ⊗ β2)σ∆∂2 2 (ϕ)m31 2 ] 2 − (1 ⊗ αy)m31 2 − (1 ⊗ σ(β1))σ∆σ∂1 1 (ϕ)m22 2 ] = (1 ⊗ β1)[σ∆1(ϕ)m31 2 − σ∆∂1 1 (ϕ)(x ⊗ 1)m22 · σdz2m31 · σdz2m31 1 ] + (1 ⊗ β2)[−σ∆1(ϕ)m31 2 ] + (1 ⊗ αyx)m31 2 2 − σ∆∂1 1 (ϕ)σdz1m31 †3= (1 ⊗ β1)[σ∆1(ϕ)m31 1 + σ∆∂1 1 + (x ⊗ 1)m40 1 (ϕ)(1 ⊗ x)m22 1 − (1 ⊗ x)m40 2 − σ∆∂1 2 − σ∆∂2 1 (ϕ) 2 (ϕ) 2 ]+(1 ⊗ β2)[σ∆2(ϕ)m31 2 − σ∆∂2 2 (ϕ)σdz2m31 2 ] + (1 ⊗ αϕ)m31 2 †4= (1 ⊗ β1)(1 ⊗ ϕ1)m31 = m31 2 2 + (1 ⊗ β2)(1 ⊗ ϕ2)m31 2 + (1 ⊗ αϕ)m31 2 where †3, †4 again follow from (3.3), (3.9), and Lemma 3.5 (2) respectively. For the third, −∆σ = −∆σ 2 (ϕ)n12 2 − (1 ⊗ y)n21 2 (ϕ)[(1 ⊗ σ(β1))m31 1 + (y ⊗ 1)n21 4 − (1 ⊗ σ(β2))m31 3 + dσz1n30 2 3 + (1 ⊗ αy)m22 1 ] − (1 ⊗ y)[(1 ⊗ β1)m40 1 + (1 ⊗ β2)∆∂2 2 (ϕ)m22 · σ∆σ∂2 2 ] + dσz1[(1 ⊗ αy)m40 (ϕ)m22 2 · ∆σ∂2 (ϕ)m31 4 ] 2 1 ] + (y ⊗ 1)[(1 ⊗ σ(β1))m40 2 − (1 ⊗ αy)m31 1 + (1 ⊗ σ(β2)) 1 − (1 ⊗ σ(β1))∆σ∂1 1 (ϕ)m31 3 − (1 ⊗ σ(β2)) = (1 ⊗ σ(β1))[−∆σ + (1 ⊗ σ(β2))[∆σ − ∆σ∂2 4 − (1 ⊗ y)m40 3 − ∆σ∂2 2 (ϕ)m31 2 (ϕ)m31 4 ] + (1 ⊗ αy)[−∆2(ϕ)m22 (ϕ)dσz1m31 2 (ϕ)(1 ⊗ y)m22 2 1 + ∆σ∂2 2 1 − (y ⊗ 1)m31 1 + (y ⊗ 1)m40 2 − ∆σ∂1 †5= (1 ⊗ σ(β1))[∆σ 1 (ϕ)m31 3 − ∆σ∂1 1 (ϕ)dσz1m31 3 ] + (1 ⊗ σ(β2))[∆σ (ϕ)dσz1m31 3 ] 1 (ϕ)(y ⊗ 1)m22 2 1 + dz1m40 1 ] 2 (ϕ)m31 3 − ∆σ∂2 2 (ϕ) · dσz2m31 3 ] + (1 ⊗ αy)(1 ⊗ x)m31 3 †6= (1 ⊗ σ(β1))(1 ⊗ σ(ϕ1))m31 = m31 3 3 + (1 ⊗ σ(β2))(1 ⊗ σ(ϕ2))m31 3 + (1 ⊗ σ(αϕ))m31 3 where †5 follows from (3.10), (3.4), (3.5), and †6 from Lemma 3.5 (3). For the last, ∆σ = ∆σ 2 − (1 ⊗ y)n21 1 (ϕ)n12 1 (ϕ)[(1 ⊗ σ(β1))m31 2 + (y ⊗ 1)n21 4 − (1 ⊗ σ(β2))m31 4 + dσz2n30 2 3 + (1 ⊗ αy)m22 1 ] − (1 ⊗ y)[(1 ⊗ β2)m40 1 1 ] + (y ⊗ 1)[(1 ⊗ σ(β2))m40 2 − (1 ⊗ αy)m31 2 − (1 ⊗ σ(β1)) 1 − (1 ⊗ σ(β1))∆σ∂1 1 (ϕ)m31 3 − (1 ⊗ σ(β2)) (ϕ)(1 ⊗ y)m22 1 (ϕ)m31 1 + ∆σ∂1 1 1 + (y ⊗ 1)m40 3 − (1 ⊗ y)m40 (ϕ)(y ⊗ 1)m22 1 2 − ∆σ∂2 2 − ∆σ∂1 2 (ϕ) (ϕ) 14 LIYU LIU 1 (ϕ)m22 − (1 ⊗ β1)∆∂1 · σ∆σ∂1 (ϕ)m22 2 ] + dσz2[(1 ⊗ αy)m40 1 (ϕ)m31 · ∆σ∂2 4 ] = (1 ⊗ σ(β1))[∆σ 2 1 (ϕ)m31 4 − ∆σ∂1 3 ] + (1 ⊗ σ(β2))[−∆σ 4 ] + (1 ⊗ αy)[∆1(ϕ)m22 1 · dσz2m31 · dσz2m31 1 − (y ⊗ 1)m31 (ϕ)dσz1m31 2 + dz2m40 1 ] †7= (1 ⊗ σ(β1))[∆σ 1 (ϕ)m31 4 − ∆σ∂1 1 4 ] + (1 ⊗ σ(β2))[∆σ 2 (ϕ)m31 4 − ∆σ∂2 2 (ϕ) · dσz2m31 4 ] + (1 ⊗ αϕ)(1 ⊗ x)m31 4 †8= (1 ⊗ σ(β1))(1 ⊗ σ(ϕ1))m31 = m31 4 4 + (1 ⊗ σ(β2))(1 ⊗ σ(ϕ2))m31 4 + (1 ⊗ σ(αϕ))m31 4 where †7 follows from (3.4), (3.10), (3.6), and †8 from Lemma 3.5 (3). (cid:3) Lemma 3.8. Eq. (3.13) holds true. Proof. We have ∆1(ϕ)n21 1 + ∆2(ϕ)n21 = ∆1(ϕ)[(1 ⊗ β1)m40 2 + (y ⊗ 1)n30 1 + (1 ⊗ β2)∆∂2 1 + (1 ⊗ x)n30 2 2 (ϕ)m22 1 ] + ∆2(ϕ)[(1 ⊗ β2)m40 1 − (1 ⊗ β1) 1 ] + (y ⊗ 1)[−(1 ⊗ β1)σ∆∂1 1 (ϕ)m31 1 (ϕ)m22 · ∆∂1 + (1 ⊗ x)[(1 ⊗ αy)m40 1 − (1 ⊗ σ(β1))∆σ∂1 1 1 (ϕ)∆2(ϕ)m22 1 − ∆∂1 = (1 ⊗ β1)[∆1(ϕ)m40 1 − ∆∂1 (ϕ)m31 1 − (1 ⊗ β2)σ∆∂2 2 (ϕ)m31 2 ] 3 − (1 ⊗ σ(β2))∆σ∂2 2 1 − ∆∂1 1 (ϕ) 1 (ϕ)(y ⊗ 1)m31 (ϕ)m31 4 ] 3 ] + (1 ⊗ β2)[∆∂2 2 (ϕ)∆1(ϕ)m22 1 + ∆2(ϕ)m40 1 − ∆∂2 2 (ϕ)(y ⊗ 1)m31 2 · (1 ⊗ x)m31 − ∆∂2 2 (ϕ)(1 ⊗ x)m31 4 ] + (1 ⊗ αyx)m40 1 †1= (1 ⊗ β1)[∆1(ϕ)m40 1 − ∆∂1 1 (ϕ)dz1m40 1 ] + (1 ⊗ β2)[∆2(ϕ)m40 1 − ∆∂2 2 (ϕ)dz2m40 1 ] + (1 ⊗ αϕ)m40 1 †2= (1 ⊗ β1)(1 ⊗ ϕ1)m40 = m40 1 1 + (1 ⊗ β2)(1 ⊗ ϕ2)m40 1 + (1 ⊗ αϕ)m40 1 where †1 follows from (3.5), (3.6), and †2 from Lemma 3.5 (1). Furthermore, σ∆σ = σ∆σ 3 + σ∆σ 2 (ϕ)n21 1 (ϕ)n21 1 (ϕ)[(1 ⊗ σ(β1))m40 4 + (1 ⊗ y)n30 2 − (1 ⊗ αy)m31 1 + (x ⊗ 1)n30 2 1 + (1 ⊗ σ(β2))σ∆σ∂2 2 (ϕ)m22 2 ] 2 − (1 ⊗ σ(β1))σ∆σ∂1 1 (ϕ)m22 2 ] = (1 ⊗ σ(β1))[σ∆σ 2 (ϕ)[(1 ⊗ σ(β2))m40 + σ∆σ + (1 ⊗ y)[−(1 ⊗ β1)σ∆∂1 + (x ⊗ 1)[(1 ⊗ αy)m40 1 (ϕ)m40 (ϕ)(x ⊗ 1)m31 (ϕ)(1 ⊗ y)m31 − σ∆σ∂1 − σ∆σ∂2 − σ∆2(ϕ)m31 1 2 2 + (x ⊗ 1)m40 1 ] 2 − (1 ⊗ αy)m31 1 (ϕ)m31 1 − (1 ⊗ β2)σ∆∂2 (ϕ)m31 1 2 (ϕ)m22 (ϕ)σ∆σ 1 − (1 ⊗ σ(β1))∆σ∂1 2 − σ∆σ∂1 3 ] + (1 ⊗ σ(β2))[σ∆σ∂2 2 − σ∆σ∂2 (ϕ)(x ⊗ 1)m31 1 2 2 2 (ϕ)m31 2 ] 3 − (1 ⊗ σ(β2))∆σ∂2 (ϕ)(1 ⊗ y)m31 1 2 − σ∆σ∂1 2 (ϕ)m40 (ϕ)σ∆σ 2 4 ] + (1 ⊗ αy)[−σ∆1(ϕ)m31 1 2 + σ∆σ 1 (ϕ)m22 2 1 (ϕ)m31 4 ] †3= (1 ⊗ σ(β1))[σ∆σ 1 (ϕ)m40 2 − σ∆σ∂1 1 (ϕ)σdσz1m40 2 ] + (1 ⊗ σ(β2))[σ∆σ 2 (ϕ)m40 2 − σ∆σ∂2 2 (ϕ)σdσz2m40 2 ] + (1 ⊗ σ(αϕ))m40 2 ON HOMOLOGICAL SMOOTHNESS OF GENERALIZED WEYL ALGEBRAS 15 †4= (1 ⊗ σ(β1))(1 ⊗ σ(ϕ1))m40 = m40 2 2 + (1 ⊗ σ(β2))(1 ⊗ σ(ϕ2))m40 2 + (1 ⊗ σ(αϕ))m40 2 where †3 follows from (3.7), (3.8), (3.9), and †4 from Lemma 3.5 (4). (cid:3) 3.3. Proof of necessity. Now suppose (ϕ, ϕ1, ϕ2) 6= B. Let us first consider the situation that k is algebraically closed. The arguments presented below are mainly inspired by [1]. Lemma 3.9. [1] If ϕ = 0, then then W(2) has infinity global dimension. Lemma 3.10. Suppose that k is algebraically closed and (ϕ, ϕ1, ϕ2) 6= B. If ϕ 6= 0 then W(2) has infinity global dimension. Proof. Recall from Proposition 3.3 that Tot P•• is alternate. For any right W(2)- W(2) module M and any left W(2)-module N , we have Tor (M, N ) = 4 W(2) Tor 8 W(2) Tor 4 Since (ϕ, ϕ1, ϕ2) is proper and k is algebraically closed, there exists a maximal ideal m = (z1 − λ1, z2 − λ2) in B containing (ϕ, ϕ1, ϕ2). Let εm : B → B/m be the projection. We claim that (εm ⊗ εm)(∆1(ϕ)) = (εm ⊗ εm)(∆2(ϕ)) = 0. In fact, we (M, N ) = · · · . So, it suffices to show that there exist M and N such that (M, N ) 6= 0. W(2) (M, N ) = Tor 6 write ∆1(ϕ) =P b′ ⊗ b′′ using Sweedler's notation, then (εm ⊗ εm)(∆1(ϕ)) =X εm(b′) ⊗ εm(b′′) =X εm(b′)εm(b′′) ⊗ ¯1 = εm(cid:18)X b′b′′(cid:19) ⊗ ¯1 = εm(ϕ1) ⊗ ¯1 = 0 and similarly for ∆2(ϕ). Denote by Ir and Il respectively the right ideal in W(2) generated by y, z1 − λ1, z2 − λ2, and the left ideal in W(2) generated by x, z1 − λ1, z2 − λ2. By analyzing the basis elements of W(2), we know that as k-modules, Ir =Mi≥1 Il =Mi≥0 Byi ⊕Mj≥0 myi ⊕Mj≥1 mxj, Bxj, which are both proper. Hence the right W(2)-module M = W(2)/Ir and the left W(2)-module N = W(2)/Il are nonzero. Moreover, by abuse of notations, we also refer to εm the map B → M or B → N by identifying B/m with the summand of M or N . W(2) Let us consider Tor 4 (M, N ), which is the fourth homology group of M ⊗W(2) (Tot P••) ⊗W(2) N . The following is a part of M ⊗W(2) P•• ⊗W(2) N : (M ⊗ N )4 j❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱❱ t21 (M ⊗ N )4 dv 40 (M ⊗ N )2 dh 30 (M ⊗ N )2 (M ⊗ N )2 dh 40 with M ⊗W(2) P40 ⊗W(2) N displayed in box. Consider the element (¯1 ⊗ ¯1, 0) in M ⊗W(2) P40 ⊗W(2) N . We have t21(¯1 ⊗ ¯1, 0) = ((εm ⊗ εm)(∆1(ϕ)), (εm ⊗ εm)(∆2(ϕ)), 0, 0) = 0, 30(¯1 ⊗ ¯1, 0) = (¯y ⊗ ¯1, ¯1 ⊗ ¯x) = 0, dh o o j o o 16 LIYU LIU and thus (¯1 ⊗ ¯1, 0) gives rise to a 4-cycle in M ⊗W(2) (Tot P••) ⊗W(2) N . Were (¯1 ⊗ ¯1, 0) a boundary, it would be of the form dv 40(b1, b2). It follows that 40(a1, a2, a3, a4) + dh 1 ⊳ z1 ⊗ a′′ 1 − a′ 1 ⊗ z1 ⊲ a′′ 2 ⊳ z2 ⊗ a′′ 2 − a′ 2 ⊗ z2 ⊲ a′′ 2 ) ¯1 ⊗ ¯1 =X(a′ −X b′ 1 ⊳ x ⊗ b′′ 1 +X b′ 1 ) +X(a′ 2 ⊗ y ⊲ b′′ 2 . Since W(2) is a Z-graded algebra by setting x = 1 and y = −1, M , N are graded W(2)-modules with M = M≥0 = ⊕j≥0kxj and N = N≤0 = ⊕i≥0kyi. By taking degree into account, we may assume a1, a2 ∈ M0 ⊗ N0, b1 ∈ M−1 ⊗ N0, and b2 ∈ M0 ⊗ N1. This forces b1 = b2 = 0, and a′ 2 ∈ B/m. Thus the above equality is simplified into 1, a′′ 2, a′′ 1 , a′ ¯1 ⊗ ¯1 =X(a′ a contradiction. 1 ) +X(a′ 1λ1 ⊗ a′′ 1 − a′ 1 ⊗ λ1a′′ 2λ2 ⊗ a′′ 2 − a′ 2 ⊗ λ2a′′ 2 ) = 0, W(2) Thus we catch a nontrivial 4-class and hence Tor 4 (M, N ) 6= 0, as desired. (cid:3) Proposition 3.11. If (ϕ, ϕ1, ϕ2) 6= B, then W(2) has infinite global dimension. Accordingly, W(2) is not homologically smooth. For the general case, let K be the algebraic closure of k and W K Note that W K proper. According to our argument above, the K-algebra W K Proof. By Lemmas 3.9, 3.10, the proposition is true when k is algebraically closed. (2) = K ⊗ W(2). (2) is a GWA over BK = K[z1, z2] and the ideal (ϕ, ϕ1, ϕ2) in BK is (2) has infinite global (M, N ) ∼= (Mn, Nn) 6= (cid:3) W K (2) Tor ∗ 0 for all n ∈ N. Therefore we finish the proof in the general case. W(2) dimension. Since the Tor groups respect localization, i.e., K ⊗ Tor ∗ W(2) (M K, N K) for all M , N , there exist Mn and Nn such that Tor n 4. Nakayama automorphisms In [37], Van den Bergh proved the existence of a duality between Hochschild homology and cohomology for a class of Gorenstein algebras A under the homolog- ical smoothness condition. Namely, there exists an invertible A-bimodule U and a positive integer d such that H i(A, M ) ∼= Hd−i(A, U ⊗A M ) naturally holds for all Ae-modules M and all integers i. In particular, if the invert- ible bimodule U is of the form Aν for some ν ∈ Aut(A), then the duality becomes (4.1) H i(A, M ) ∼= Hd−i(A, M ν). This is usually called twisted Poincar´e duality in the literature. Recall that an algebra A enjoys twisted Poincar´e duality if A is a twisted d-Calabi-Yau algebra, namely, A is homologically smooth and the condition (4.2) Exti Ae(A, Ae) ∼=(0, Aν, i 6= d, i = d is fulfilled for some d ∈ N and ν ∈ Aut(A). The number d is called the Hochschild cohomology dimension of A and respectively ν the Nakayama automorphism of A, which coincide the d, ν in (4.1). Many classes of algebras arising from noncommutative algebraic geometry or quantum group are twisted Calabi-Yau. We refer to [6], [7], [11], [16], [25], [29], [38], [40] and the references therein for more information and in particular plenty of examples. ON HOMOLOGICAL SMOOTHNESS OF GENERALIZED WEYL ALGEBRAS 17 Remark 4.1. (1) In (4.2), the second variable Ae in the Ext group has left and right Ae-module structures. The left is used for computing Ext, and the right is survival, inducing the A-bimodule structure on the Ext group. (2) An algebra A is said to have a Nakayama automorphism ν if (4.2) is sat- isfied, even if A is not homologically smooth. The automorphism ν is unique up to inner automorphism. In [30], ν is proven to be central in Aut(A)/Inn(A). (3) When ν is inner, the algebra A is Calabi-Yau in the sense of Ginzburg [16]. Let us focus on Nakayama automorphisms of GWAs. Recall that a GWA W(1) is proven to have a Nakayama automorphism (with d = 2) whose explicit expression is determined in [27]. By adapting the proof in loc. cit. one can conclude that W(2) also has a Nakayama automorphism (with d = 3). However, one does not know the expressions of σ(z1), σ(z2) for an arbitrary σ ∈ Aut(B), as is mentioned in Remark 3.1. Due to the indeterminacy of σ, we are not able to capture the Nakayama automorphism ν as we did in [27]. Instead, we will deduce the expression of ν by spectral sequence argument. It is illustrated in [27] that the filtration by column of a homotopy double complex gives rise to a spectral sequence. If the homotopy double complex sits in the first quadrant, then the spectral sequence converges to the (co)homology of the associated total complex. So let us apply it to the homotopy double complex Q•• W e . Denote by Epq • Since W(2) the induced spectral sequence. ∼= B(N) as a left B-module, we have W e (2) ∼= (Be)(N×N) ∼= (Be)(N) as a left Be-module. Observe that B is 2-Calabi-Yau and so E0q 1 = H q(Q0• W e (2) ) = Extq W e (2) (W e (2) ⊗Be B, W e (2)) ∼= Extq Be (B, Be)(N) is zero unless q = 2. By a similar manner, we conclude that for all p ≥ 1, Epq 1 nonzero only if q = 2. is Notice that E02 1 is a quotient of E02 0 = W e (2). It is easy to prove that E02 1 =nX[w′ ⊗ w′′](cid:12)(cid:12)(cid:12) w′ ∈ W(2) and w′′ is a power of x or yo for all p ≥ 1. Hence, the as a k-module, and similarly, we have Ep2 differentials E02 1 1 ⊕ E02 1 1 are respectively given by d0 −→ E12 1 1 = E02 d1 −→ E2 d0 =(cid:18)−x ⊗ 1 + J ⊗ x 1 ⊗ y − Jy ⊗ 1(cid:19) and d1 =(cid:18)y ⊗ 1 1 ⊗ x 1 ⊗ y x ⊗ 1(cid:19) . Here we remind the reader that J is the Jacobian determinant of σ and that d0 is induced by dh 02 given in §3.1, and their matrix representations are in fact equal up to im d02 v . We begin to compute E12 2 . Suppose x := (x1, x2)T ∈ ker d2 with x1 = Xi,j≥1 x2 = Xi,j≥0 [a1 [b1 ij xi ⊗ xj ] + Xi≥1,j≥0 ijxi ⊗ xj] + Xi≥0,j≥1 [b2 [a2 ij xi ⊗ yj] + Xi≥0,j≥1 ijxi ⊗ yj] + Xi≥1,j≥0 [a3 ij yi ⊗ xj] + Xi,j≥0 ijyi ⊗ xj] + Xi,j≥1 [b3 [a4 ijyi ⊗ yj], [b4 ijyi ⊗ yj]. Then by a direct computation, we have ij = −ϕσ−1(a1 b1 ij = −σ−1(a3 b3 i+1,j+1), b2 i−1,j+1), a4 ij = −σ−1(a2 ij = −σ(ϕ)σ(b4 i+1,j−1), i+1,j+1) 18 and hence LIYU LIU x = Xi≥1,j≥1(cid:18) + Xi≥0,j≥1(cid:18) [a1 [−ϕσ−1(a1 ijxi ⊗ xj] ijxi ⊗ yj] [a2 [−σ−1(a2 ij)xi−1 ⊗ xj−1](cid:19) + Xi≥1,j≥0(cid:18) ij)yi+1 ⊗ xj−1](cid:19) + Xi≥0,j≥0(cid:18)[−σ(ϕ)σ(b4 ij )xi−1 ⊗ yj+1](cid:19) ijyi+1 ⊗ yj+1] (cid:19) ij )yi ⊗ yj] ij yi ⊗ xj] [b4 [a3 [−σ−1(a3 i+1,j+1. On the other hand, it is routine to check ijxi ⊗ xj] ij )xi+1 ⊗ xj−1] where b4 ij = b4 [a1 [−ϕσ−1(a1 [a2 [−σ−1(a2 [a3 [−σ−1(a3 (cid:18) (cid:18) (cid:18) [−Jϕσ−2(a2 [−J −1ϕa1 ij yi ⊗ xj] ij xi ⊗ yj] ij )xi−1 ⊗ xj−1](cid:19) + im d0 =(cid:18)[J −1σ(a1 ij)xi−1 ⊗ yj+1](cid:19) + im d0 =(cid:18)[Jσ(ϕ)σ−1(a2 ij)yi+1 ⊗ xj−1](cid:19) + im d0 =(cid:18)[J −1σ(ϕ)σ(a3 ijyi+1 ⊗ yj+1] (cid:19) + im d0 =(cid:18)[−Jσ(ϕ)b4 (cid:18)[−σ(ϕ)σ(b4 y =Xi≥1(cid:18) i )xi−1 ⊗ y](cid:19) +Xi≥1(cid:18) ij )yi ⊗ yj] [−J −1ϕa3 [Jσ−1(b4 [−σ−1(c1 i xi ⊗ 1] [b4 [c1 [c2 (4.3) Henceforth, x + im d0 can be uniquely expressed as y + im d0 with ij)xi−1 ⊗ yj−1] ijxi ⊗ xj−2] (cid:19) + im d0, ij )xi−2 ⊗ yj] (cid:19) + im d0, (cid:19) + im d0, ij )yi−1 ⊗ xj−1] ij yi ⊗ xj−2] ij yi+1 ⊗ yj−1] ij)yi+2 ⊗ yj] (cid:19) + im d0. [−σ−1(c2 i yi ⊗ x] i )yi+1 ⊗ 1](cid:19) . Since Ext3 W e (2) (W(2), W e (2)) ∼= E12 2 which is induced by the right regular module structure on W e 2 , we have to study the W(2)-bimodule structure (2).1 Let us for any 2 → W(2) of k-module explicitly as follows: on E12 write an isomorphism Φ : E12 class ¯y := y + im d0 ∈ E12 2 as above, define Φ(¯y) =Xi≥1 J iσ−1(c1 i )xi−1 +Xi≥1 J −ic2 i yi. Lemma 4.1. Let J be the Jacobian determinant of σ, and ν : W(2) → W(2) be the automorphism defined by (4.4) Then Φ : E12 ν(x) = Jx, ν(y) = J −1y, ν(z1) = z1, ν(z2) = z2. 2 → W ν (2) is an isomorphism of W(2)-bimodules. Proof. Let us prove that Φ is an isomorphism of right W(2)-modules. The left modules case is similar. Any class in E12 2 can be represented by some y as in (4.3). So we need to verify Φ(¯y ⊳ w) = Φ(¯y) ⊳ w for w ∈ {z1, z2, x, y}. Let z = z1 or z2. We have [−σ−1(c1 [−σ−1(c1 [c1 [c1 i xiz ⊗ 1] i )xi−1z ⊗ y](cid:19) +Xi≥1(cid:18) i σi(z))xi−1 ⊗ y](cid:19)+Xi≥1(cid:18) i σi(z)xi ⊗ 1] [c2 i yiz ⊗ x] [−σ−1(c2 i )yi+1z ⊗ 1](cid:19)  i σ−i(z))yi+1 ⊗ 1](cid:19)  i σ−i(z)yi ⊗ x] [c2 [−σ−1(c2 Φ(¯y ⊳ z) = Φ Xi≥1(cid:18) = Φ Xi≥1(cid:18) =Xi≥1 =Xi≥1 J −ic2 i σ−i(z)yi J iσ−1(c1 J iσ−1(c1 i σi(z))xi−1 +Xi≥1 i )xi−1z +Xi≥1 J −ic2 i yiz 1This is also called the inner bimodule structure on W(2) ⊗ W(2). ON HOMOLOGICAL SMOOTHNESS OF GENERALIZED WEYL ALGEBRAS 19 = Φ(¯y) ⊳ z. Also, we have i yix ⊗ x] i )yi+1x ⊗ 1](cid:19)  [−σ−1(c2 i σ1−i(ϕ)yi−1 ⊗ x] i σ1−i(ϕ))yi ⊗ 1](cid:19)  1ϕ)y ⊗ 1](cid:19) 1ϕ ⊗ x] 1ϕ)x ⊗ 1] 1ϕ ⊗ y] (cid:19) Φ(¯y ⊳ x) = Φ Xi≥1(cid:18) [c1 = Φ Xi≥2(cid:18) = Φ Xi≥2(cid:18) + Xi≥1(cid:18) = Φ Xi≥2(cid:18) + Xi≥1(cid:18) =Xi≥2 =Xi≥1 =Xi≥1 J iσ−1(c1 J iσ−1(c1 J i+1σ−1(c1 = Φ(¯y) ⊳ x, [c2 i+1σ−i(ϕ)yi ⊗ x] [−σ−1(c2 i xi+1 ⊗ 1] [−σ−1(c1 [c1 i−1xi ⊗ 1] [−σ−1(c1 [c1 i−1xi ⊗ 1] [−σ−1(c1 [c2 [c2 [−σ−1(c2 [−σ−1(c2 i )xi ⊗ y](cid:19) +Xi≥1(cid:18) i−1)xi−1 ⊗ y](cid:19) +Xi≥1(cid:18) [c2 i−1)xi−1 ⊗ y](cid:19) +(cid:18) i+1σ−i(ϕ))yi+1 ⊗ 1](cid:19)  i−1)xi−1 ⊗ y](cid:19) +(cid:18)[J −1σ(c2 i+1σ−i(ϕ))yi+1 ⊗ 1](cid:19)  1ϕ)) +Xi≥1 [−J −1c2 [c1 i−1xi ⊗ 1] [−σ−1(c1 [c2 i+1σ−i(ϕ)yi ⊗ x] [−σ−1(c2 i )xi + c2 J −ic2 i+1yiϕ 1ϕ +Xi≥1 i )xi−1Jx +Xi≥1 J −ic2 i yiJx i−1)xi−1 + Jσ−1(J −1σ(c2 J −ic2 i+1σ−i(ϕ)yi and [c2 i−1yi ⊗ x] i yi+1 ⊗ x] i )yi+2 ⊗ 1](cid:19)  i−1)yi+1 ⊗ 1](cid:19)  1)y ⊗ y](cid:19) [c1 [−σ−1(c1 Φ(¯y ⊳ y) = Φ Xi≥1(cid:18) = Φ Xi≥0(cid:18) [c1 = Φ Xi≥1(cid:18) [c1 + Xi≥2(cid:18) = Φ Xi≥1(cid:18) + Xi≥2(cid:18) i+1xi+1y ⊗ 1] [−σ−1(c1 i+1xi+1y ⊗ 1] [−σ−1(c1 [c2 i−1yi ⊗ x] [−σ−1(c2 [c2 i xiy ⊗ 1] 1xy ⊗ 1] [−σ−1(c1 [−σ−1(c2 [−σ−1(c2 i )xi−1y ⊗ y](cid:19) +Xi≥1(cid:18) i+1)xiy ⊗ y](cid:19) +Xi≥2(cid:18) i+1)xiy ⊗ y](cid:19) +(cid:18) [c1 i−1)yi+1 ⊗ 1](cid:19)  i+1σi+1(ϕ))xi−1 ⊗ y](cid:19) +(cid:18)[Jσ−1(c1 i−1)yi+1 ⊗ 1](cid:19)  i+1σi+1(ϕ)xi ⊗ 1] [−Jc1 [c2 i−1yi ⊗ x] [−σ−1(c2 [c1 [−σ−1(c1 1)y ⊗ x] 1y2 ⊗ 1] (cid:19) 20 LIYU LIU J iσ−1(c1 i+1σi+1(ϕ))xi−1 + J −1Jσ−1(c1 J −ic2 i−1yi 1)y +Xi≥2 J −ic2 i−1yi =Xi≥1 =Xi≥1 =Xi≥0 =Xi≥1 J iσ−1(c1 i+1)xi−1σ(ϕ) + σ−1(c1 1)y +Xi≥2 J −i−1c2 i yi+1 J iσ−1(c1 J iσ−1(c1 i+1)xiy +Xi≥1 i )xi−1J −1y +Xi≥1 = Φ(¯y) ⊳ y. J −ic2 i yiJ −1y (cid:3) Thus we obtain Theorem 4.2. Any GWA W(2) has a Nakayama automorphism ν given by (4.4), namely, Exti W e (2) (W(2), W e (2)) ∼=(0, W ν (2), i 6= 3, i = 3. In particular, if W(2) is homologically smooth, then W(2) is twisted 3-Calabi-Yau. Remark 4.2. Since W(2) is noetherian and has (left and right) injective dimension 3, an equivalent statement of Theorem 4.2 is: the rigid dualizing complex over W(2) is ν W(2)[3]. Corollary 4.3. A GWA W(2) = B(σ, ϕ) is 3-Calabi-Yau if and only if (ϕ, ϕ1, ϕ2) = B and J = 1. Since Hochschild established the cohomology theory for associative algebras A [17], the theory has been studied by many mathematicians. Amongst the devel- opments, a structure on HH •(A) := ⊕n∈NHH n(A) which is a differential graded version of Poisson algebra was found by Gerstenhaber for all algebras A. The structure is nowadays called Gerstenhaber algebra. Batalin-Vilkovisky algebras are a subclass of Gerstenhaber algebras arising from theoretical physics. A remarkable relationship between Batalin-Vilkovisky structure and Hochschild cohomology was illustrated by Ginzburg [16], saying that HH •(A) is a Batalin-Vilkovisky algebra for all Calali-Yau algebras A. Later on, this result was generalized for some twisted Calabi-Yau algebras [24], that is, Theorem 4.4. [24, Thm. 1.7] If A is a twisted Calabi-Yau algebra with semisimple (namely, diagonalizable) Nakayama automorphism, then the Hochschild cohomology HH •(A) of A is a Batalin-Vilkovisky algebra. According to the expression (4.4), the Nakayama automorphism ν of W is semisimple. Hence we have Corollary 4.5. The Hochschild cohomology HH •(W(2)) of W(2) is a Batalin- Vilkovisky algebra if W(2) is homologically smooth. Remark 4.3. Note that the defining automorphism σ ∈ Aut(k[z]) of W(1) is nec- essarily determined by σ(z) = λz + η for some λ ∈ k×, η ∈ k. One of the results in the author's previous paper [27] is that the Nakayama automorphism of W(1) is given by x 7→ λx, y 7→ λ−1y, z 7→ z. Obviously we have J = λ in this case. It seems reasonable to conjecture that the analogy exists for all GWA W(n) if n is any positive integer. This will be our future work. ON HOMOLOGICAL SMOOTHNESS OF GENERALIZED WEYL ALGEBRAS 21 5. Examples In this section, we apply Theorem 3.1 to concrete algebras, judging them smooth or not. Most results are known, obtained by other people in different manners. 5.1. Quantum groups Oq(SL2) and U (sl2). The definitions of these well-known algebras can be found, for example, in [21] and it is well-known [3] that they are GWA as discussed in this paper. Thus Theorems 3.1 and 4.2 show that they are homologically smooth and determine their Nakayama automorphisms. These theorems therefore reproduce [7, §6] where the detailed formulas can be found. 5.2. Noetherian down-up algebras. Motivated by the study of posets, Benkart and Roby [4] introduced the notion of a down-up algebra A(α, β, γ). Down-up algebras have been intensively studied in for example [5], [13], [23], [26] among many other articles. It is shown in [23] that A(α, β, γ) is right (or left) noetherian if and only if β 6= 0. Also in [23], a noetherian down-up algebra A(α, β, γ) is a GWA. Thus Theorems 3.1 and 4.2 show that A(α, β, γ) is homologically smooth and determine the Nakayama automorphism, reproducing the formula shown in [30], [34]. 5.3. A quotient algebra of M (p, q). In [12] a noncommutative and noncocom- mutative bialgebra M (p, q) for two parameters p and q is constructed. The algebra is generated by four elements a, b, c, d satisfying some relations similar to those of the quantum matrix algebra Mq(2). Concretely, these relations are: ba = qab, dc = qcd, ca = qac, db = qbd, bc = cb, da − qad = p(1 − bc). The element u = da − p(qbc + 1)/(1 − q) is normal regular in M (p, q). It is not hard to check that the quotient algebra N (p, q) = M (p, q)/(u) is realized as a GWA over k[b, c]. By Theorem 3.1 N (p, q) is homologically smooth if and only if p 6= 0. Remark 5.1. The homological smoothness of N (1, q) is studied by Shengyun Jiao, and the related results appear in her Master Thesis [20], under the direction of the author. 5.4. Quantum lens space and quantum Seifert manifold. Let us consider two algebras which can be regarded as coinvariant of Hopf algebras. They are the quantum lens space Oq(L(l; 1, l)) where l is a positive integer, and the quantum Seifert manifold Oq(Σ3). For their background, we refer to [19] and [10] respectively. We should remark that both algebras were defined as C∗-algebras originally; but here we adapt to an arbitrary base field k. Both of them are GWA as discussed in this paper. So they are homologically smooth by theorem 3.1. This fact was first obtained by Brzezi´nski in [8], using a completely different manner. References [1] V. Bavula, Global dimension of generalized Weyl algebras, Representation theory of algebras (Cocoyoc, 1994), CMS Conf. Proc., vol. 18, Amer. Math. Soc., Providence, RI, 1996, pp. 81– 107. [2] V. Bavula, Generalized Weyl algebras and their representations, Algebra i Analiz 4 (1992), 75–97. [3] V. Bavula, Tensor homological minimal algebras, global dimension of the tensor product of algebras and of generalized Weyl algebras, Bull. Sci. Math. 120 (1996), 293–335. [4] G. Benkart and T. Roby, Down-up algebras, J. Algebra 209 (1998), 305–344. [5] G. Benkart and S. Witherspoon, A Hopf structure for down-up algebras, Math. Z. 238 (2001), 523–553. [6] R. Bocklandt, T. Schedler, and M. Wemyss, Superpotentials and higher order derivations, J. Pure Appl. Algebra 214 (2010), 1501–1522. 22 LIYU LIU [7] K.A. Brown and J.J. Zhang, Dualising complexes and twisted Hochschild (co)homology for Noetherian Hopf algebras, J. Algebra 320 (2008), 1814–1850. [8] T. Brzezi´nski, On the smoothness of the noncommutative pillow and quantum teardrops, SIGMA Symmetry Integrability Geom. Methods Appl. 10 (2014), Paper 015, 8. [9] T. Brzezi´nski, Noncommutative differential geometry of generalized Weyl algebras, SIGMA Symmetry Integrability Geom. Methods Appl. 12 (2016), Paper 059, 18. [10] T. Brzezi´nski and B. Zieli´nski, Quantum principal bundles over quantum real projective spaces, J. Geom. Phys. 62 (2012), 1097–1107. [11] K. Chan, C. Walton, and J. Zhang, Hopf actions and Nakayama automorphisms, J. Algebra 409 (2014), 26–53. [12] H.-X. Chen, A class of noncommutative and noncocommutative Hopf algebras: the quantum version, Comm. Algebra 27 (1999), 5011–5032. [13] S. Chouhy, E. Herscovich, and A. Solotar, Hochschild homology and cohomology of down-up algebras, preprint, arXiv:1609.09809, 2016. [14] W. Dicks, Automorphisms of the polynomial ring in two variables, Publ. Sec. Mat. Univ. Aut`onoma Barcelona 27 (1983), 155–162. [15] M.A. Farinati, A. Solotar, and M. Su´arez- ´Alvarez, Hochschild homology and cohomology of generalized Weyl algebras, Ann. Inst. Fourier (Grenoble) 53 (2003), 465–488. [16] V. Ginzburg, Calabi-Yau algebras, preprint, arXiv:math/0612139, 2006. [17] G. Hochschild, On the cohomology groups of an associative algebra, Ann. of Math. (2) 46 (1945), 58–67. [18] T.J. Hodges, Noncommutative deformations of type-A Kleinian singularities, J. Algebra 161 (1993), 271–290. [19] J.H. Hong and W. Szyma´nski, Quantum lens spaces and graph algebras, Pacific J. Math. 211 (2003), 249–263. [20] S. Jiao, Homological smoothness of a class generalized Weyl algebras of Gelfand-Kirilov dimension three (in Chinese), Master Thesis, Yangzhou University, 2017. [21] C. Kassel, Quantum Groups, Graduate Texts in Mathematics vol. 155, Springer, New York, 1995. [22] E. Kirkman, J. Kuzmannovich, and J.J. Zhang, Nakayama automorphism and rigidity of dual reflection group coactions, J. Algebra 487 (2017), 60–92. [23] E. Kirkman, I.M. Musson, and D.S. Passman, Noetherian down-up algebras, Proc. Amer. Math. Soc. 127 (1999), 3161–3167. [24] N. Kowalzig and U. Krahmer, Batalin-Vilkovisky structures on Ext and Tor, J. Reine Angew. Math. 697 (2014), 159–219. [25] U. Krahmer, On the Hochschild (co)homology of quantum homogeneous spaces, Israel J. Math. 189 (2012), 237–266. [26] R.S. Kulkarni, Down-up algebras and their representations, J. Algebra 245 (2001), 431–462. [27] L. Liu, Homological smoothness and deformations of generalized Weyl algebras, Israel J. Math. 209 (2015), 949–992. [28] L. Liu, Y. Shen, and Q. Wu, Homological properties of Podle´s quantum spheres, Sci. China Math. 57 (2014), 69–80. [29] L.-Y. Liu and Q.-S. Wu, Rigid dualizing complexes over quantum homogeneous spaces, J. Algebra 353 (2012), 121–141. [30] J.-F. Lu, X.-F. Mao, and J.J. Zhang, Nakayama automorphism and applications, Tran. Amer. Math. Soc. 369 (2017), 2425–2460. [31] J.-F. Lu, X.-F. Mao, and J.J. Zhang, Nakayama automorphisms of a class of graded algebras, Israel J. Math. 219 (2017), 707–725. [32] J.H. McKay and S.S. Wang, An elementary proof of the automorphism theorem for the polynomial ring in two variables, J. Pure Appl. Algebra 52 (1988), 91–102. [33] M. Reyes, D. Rogalski, and J.J. Zhang, Skew Calabi-Yau algebras and homological identities, Adv. Math. 264 (2014), 308–354. [34] Y. Shen and D. Lu, Nakayama automorphisms of PBW deformations and Hopf actions, Sci. China Math. 59 (2016), 661–672. [35] A. Solotar, M. Su´arez- ´Alvarez, and Q. Vivas, Hochschild homology and cohomology of gen- eralized Weyl algebras: the quantum case, Ann. Inst. Fourier (Grenoble) 63 (2013), 923–956. [36] M. Van den Bergh, Noncommutative homology of some three-dimensional quantum spaces, K-Theory 8 (1994), 213–230. [37] M. Van den Bergh, A relation between Hochschild homology and cohomology for Gorenstein rings, Proc. Amer. Math. Soc. 126 (1998), 1345–1348. Erratum ibid. Proc. Amer. Math. Soc. 130 (2002), 2809–2810. [38] M. Van den Bergh, Calabi-Yau algebras and superpotentials, Selecta Math. (N.S.) 21 (2015), 555–603. ON HOMOLOGICAL SMOOTHNESS OF GENERALIZED WEYL ALGEBRAS 23 [39] W. van der Kulk, On polynomial rings in two variables, Nieuw Arch. Wiskunde (3) 1 (1953), 33–41. [40] A. Yekutieli, The rigid dualizing complex of a universal enveloping algebra, J. Pure Appl. Algebra 150 (2000), 85–93. School of Mathematical Science, Yangzhou University, No. 180 Siwangting Road, 225002 Yangzhou, Jiangsu, China E-mail address: [email protected]
1209.2627	1	1209	2012-09-12T14:48:59	The center of a Kumjian-Pask algebra	[ "math.RA" ]	The Kumjian-Pask algebras are path algebras associated to higher-rank graphs, and generalize the Leavitt path algebras. We study the center of simple Kumjian-Pask algebras and characterize commutative Kumjian-Pask algebras.	math.RA	math	THE CENTER OF A KUMJIAN-PASK ALGEBRA JONATHAN H. BROWN AND ASTRID AN HUEF Abstract. The Kumjian-Pask algebras are path algebras associated to higher-rank graphs, and generalize the Leavitt path algebras. We study the center of simple Kumjian-Pask algebras and characterize commutative Kumjian-Pask algebras. 1. Introduction Let E be a directed graph E and let F be a field. The Leavitt path algebras LF(E) of E over F were first introduced in [1] and [2], and have been widely studied since then. Many of the properties of Leavitt path algebras can be inferred from properties of the graph, and for this reason provide a convenient way to construct examples of algebras with a particular set of attributes. The Leavitt path algebras are the algebraic analogues of the graph C ∗-algebras associated to E. In [10], Tomforde constructed an analogous Leavitt path algebra LR(E) over a commutative ring R with 1, and introduced more techniques from the graph C ∗-algebra setting to study it. In [3], Aranda Pino, Clark, an Huef and Raeburn generalized Tomforde's construction and associated to a higher-rank graph Λ a graded algebra KPR(Λ) called the Kumjian- Pask algebra. Example 7.1 of [3] shows that even the class of commutative Kumjian- Pask algebras over a field is strictly larger than the class of Leavitt path algebras over that field. In this paper we study the center of Kumjian-Pask algebras. In §3 we work over C and show how the embedding of KPC(Λ) in the C ∗-algebra C ∗(Λ) can be used together with the Dauns-Hofmann theorem to deduce that the center of a simple Kumjian-Pask algebra is either {0} or isomorphic to C. More generally, it follows from Theorem 4.7, that the center of a "basically simple" (see page 7) Kumjian-Pask algebra KPR(Λ) is either zero or is isomorphic to the un- derlying ring R. Thus our Theorem 4.7 generalizes the analogous theorem for Leavitt path algebras over a field [4, Theorem 4.2], but our proof techniques are very different and more informative. Indeed, the Kumjian-Pask algebra is basically simple if and only if the graph Λ is cofinal and aperiodic, and our proofs show explicitly which of these graph properties are needed to infer various properties of elements in the center. In Proposition 5.3 we show that a Kumjian-Pask algebra of a k-graph Λ is commuta- tive if and only it is a direct sum of rings of Laurant polynomials in k-indeterminates, if and only if Λ is a disjoint union of copies of the category Nk. This generalizes Proposition 2.7 of [4]. 2. Preliminaries We view Nk as a category with one object and composition given by addition. We call a countable category Λ = (Λ0, Λ, r, s) a k-graph if there exists a functor d : Λ → Nk for all λ ∈ Λ, d(λ) = m + n implies there with the unique factorization property: exist unique µ, ν ∈ Λ such that d(µ) = m, d(ν) = n and λ = µν. Using the unique Date: 12 September, 2012. 1 2 J. H. BROWN AND A. AN HUEF factorization property, we identify the set of objects Λ0 with the set of morphisms of degree 0. Then, for n ∈ Nk, we set Λn := d−1(n), and call Λn the paths of shape n in Λ and Λ0 the vertices of Λ. A path λ ∈ Λ is closed if r(λ) = s(λ). For V, W ⊂ Λ0, we set V Λ := {λ ∈ Λ : r(λ) ∈ V }, ΛW := {λ ∈ Λ : s(λ) ∈ W } and V ΛW := V Λ ∩ ΛW ; the sets V Λn, ΛnW and V ΛnW are defined similarly. For simplicity we write vΛ for {v}Λ. A k-graph Λ is row-finite if vΛn < ∞ for all v ∈ Λ0 and n ∈ Nk and has no sources if vΛn 6= ∅ for all v ∈ Λ0 and n ∈ Nk. We assume throughout that Λ is a row-finite k-graph with no sources. Following [9, Lemma 3.2(iv)], we say that a k-graph Λ is aperiodic if for every v ∈ Λ0 and m 6= n ∈ Nk there exists λ ∈ vΛ such that d(λ) ≥ m ∨ n and λ(m, m + d(λ) − (m ∨ n)) 6= λ(n, n + d(λ) − (m ∨ n)). This formulation of aperiodicity is equivalent to the original one from [7, Definition 4.3] when Λ is a row-finite graph with no sources, but is often more convenient since it only involves finite paths. Let Ωk := {(m, n) ∈ Nk : m ≤ n}. As in [7, Definition 2.1] we define an infinite path in Λ to be a degree-preserving functor x : Ωk → Λ, and denote the set of infinite paths by Λ∞. As in [7, Definition 4.1] we say Λ is cofinal if for every infinite path x and every vertex v there exists m ∈ Nk such that vΛx(m) 6= ∅. For each λ ∈ Λ we introduce a ghost path λ∗; for v ∈ Λ0 we take v∗ = v. We write G(Λ) for the set of ghost paths and G(Λ6=0) if we exclude the vertices. Let R be a commutative ring with 1. Following [3, Definition 3.1], a Kumjian- Pask Λ-family (P, S) in an R-algebra A consists of functions P : Λ0 → A and S : Λ6=0 ∪ G(Λ6=0) → A such that (KP1) {Pv : v ∈ Λ0} is a set of mutually orthogonal idempotents; (KP2) for λ, µ ∈ Λ6=0 with r(µ) = s(λ), SλSµ = Sλµ, Sµ∗Sλ∗ = S(λµ)∗ , Pr(λ)Sλ = Sλ = SλPs(λ), Ps(λ)Sλ∗ = Sλ∗ = Sλ∗Pr(λ); (KP3) for all λ, µ ∈ Λ6=0 with d(λ) = d(µ), we have Sλ∗Sµ = δλ,µPs(λ); (KP4) for all v ∈ Λ0 and n ∈ Nk \ {0}, we have Pv = Pλ∈vΛn SλSλ∗. By [3, Theorem 3.4] there is an R-algebra KPR(Λ), generated by a nonzero Kumjian- Pask Λ-family (p, s), with the following universal property: whenever (Q, T ) is a Kumjian-Pask Λ-family in an R-algebra A, then there is a unique R-algebra homo- morphism πQ,T : KPR(Λ) → A such that πQ,T (pv) = Qv, πQ,T (sλ) = Tλ and πQ,T (sµ∗) = Tµ∗ for v ∈ Λ0 and λ, µ ∈ Λ6=0. Also by Theorem 3.4 of [3], the subgroups KPR(Λ)n := spanR{sλsµ∗ : λ, µ ∈ Λ and d(λ) − d(µ) = n} (n ∈ Zk) give a Zk-grading of KPR(Λ). Let S be a Zk-graded ring; then by the graded-uniqueness theorem [3, Theorem 4.1], a graded homomorphism π : KPR(Λ) → S such that π(rpv) 6= 0 for nonzero r ∈ R is injective. We will often write elements a ∈ KPR(Λ) \ {0} in the normal form of [3, Lemma 4.2]: there exists m ∈ Nk and a finite F ⊂ Λ × Λm such that a = P(α,β)∈F rα,βsαsβ ∗ where rα,β ∈ R \ {0} and s(α) = s(β). THE CENTER OF A KUMJIAN-PASK ALGEBRA 3 3. Motivation When A is a simple C ∗-algebra (over C, of course), it follows from the Dauns-Hofmann Theorem (see, for example, [8, Theorem A.34]) that the center Z(A) of A is isomorphic to C if A has an identity and is {0} otherwise. Let Λ be a row-finite k-graph without sources. In this short section we deduce that the center of a simple Kumjian-Pask algebra KPC(Λ) is either isomorphic to C or is {0}. Lemma 3.1. Suppose A is a simple C ∗-algebra. If A has an identity, then z 7→ z1A is an isomorphism of C onto the center Z(A) of A. If A has no identity, then Z(A) = {0}. Proof. Since A is simple, Prim A = {⋆}, and f 7→ f (⋆) is an isomorphism of Cb(Prim A) onto C. By the Dauns-Hofmann Theorem, Cb(Prim A) is isomorphic to the center Z(M(A)) of the multiplier algebra M(A) of A. Putting the two isomorphisms together gives an isomorphism z 7→ z1M (A) of C onto Z(M(A)). Now suppose that A has an identity. Then M(A) = A, and it follows from the first paragraph that Z(A) is isomorphic to C. Next suppose that A does not have an identity. Let a ∈ Z(A), and let uλ be an approximate identity in A and m ∈ M(A). Then ma = lim(muλ)a = a lim(muλ) = am. Thus Z(A) ⊂ Z(M(A)). Now Z(A) ⊂ Z(M(A))∩A = {z1M (A) : z ∈ C}∩A = {0}. (cid:3) Lemma 3.2. Let D be a dense subalgebra of a C ∗-algebra A. Then Z(A) ∩ D = Z(D). Proof. Trivially, Z(A) ∩ D ⊂ Z(D). To see the reverse inclusion, let a ∈ Z(D). Let b ∈ A and choose {dλ} ⊂ D such that dλ → b. Then ba = limλ dλa = limλ adλ = ab. Now a ∈ Z(A) ∩ Z(D) ⊂ Z(A) ∩ D, and hence Z(A) ∩ D = Z(D). (cid:3) By [3, Theorem 6.1], KPC(Λ) is simple if and only if Λ is cofinal and aperiodic, so in the next corollary KPC(Λ) is simple. Also, KPC(Λ) has an identity if and only if Λ0 is finite (see Lemma 4.6 below). Corollary 3.3. Suppose that Λ is a row-finite, cofinal, aperiodic k-graph with no sources. If Λ0 is finite, then z 7→ z1KPC(Λ) is an isomorphism of C onto the center Z(KPC(Λ)) of KPC(Λ). If Λ0 is infinite, then KPC(Λ) = {0}. Proof. Let (p, s) be a generating Kumjian-Pask Λ-family for KPC(Λ) and (q, t) a gen- erating Cuntz-Krieger Λ-family for C ∗(Λ). Then (q, t) is a Kumjian-Pask Λ-family in C ∗(Λ), and the universal property of KPC(Λ) gives a ∗-homomorphism πq,t : KPC(Λ) → C ∗(Λ) which takes sµsν ∗ to tµt∗ ν. It follows from the graded-uniqueness theorem that πq,t is a ∗-isomorphism onto a dense ∗-subalgebra of C ∗(Λ) (see Proposition 7.3 of [3]). Since Λ is aperiodic and cofinal, C ∗(Λ) is simple by [9, Theorem 3.1]. Now suppose that Λ0 is finite. Then KPC(Λ) has identity 1KPC(Λ) = Pv∈Λ0 pv and C ∗(Λ) has identity 1C ∗(Λ) = Pv∈Λ0 qv, and πq,t is unital. By Lemma 3.1, Z(C ∗(Λ)) = {z1C ∗(Λ) : z ∈ C}. By Lemma 3.2, Z(πq,t(KPC(Λ))) = Z(C ∗(Λ)) ∩ πq,t(KPC(Λ)) = {z1C ∗(Λ) : z ∈ C}. Since πq,t is unital, Z(KPC(Λ)) is isomorphic to C as claimed. Next suppose that Λ0 is infinite. Then Z(C ∗(Λ)) = {0} and Z(πq,t(KPC(Λ))) = (cid:3) Z(C ∗(Λ)) ∩ πq,t(KPC(Λ)) = {0}, giving KPC(Λ) = {0}. 4. The center of a Kumjian-Pask algebra Our goal is to extend Corollary 3.3 to Kumjian-Pask algebras over arbitrary rings. Throughout R is a commutative ring with 1 and Λ is a row-finite k-graph with no sources. 4 J. H. BROWN AND A. AN HUEF We will need Lemma 4.1 several times. For notational convenience, for v ∈ Λ0, sv or sv∗ means pv. So when m = 0 Lemma 4.1, says that the set {sα : α ∈ Λ} is linearly independent. Lemma 4.1. Let m ∈ Nk. Then {sαsβ ∗ : s(α) = s(β) and d(β) = m} is a linearly independent subset of KPR(Λ). Proof. Let F be a finite subset of {(α, β) ∈ Λ × Λm : s(α) = s(β)}, and suppose that P(α,β)∈F rα,βsαsβ ∗ = 0. Fix (σ, τ ) ∈ F . Since all the β have degree m, using (KP3) twice we obtain 0 = sσ∗(cid:16) X(α,β)∈F rα,βsαsβ ∗(cid:17)sτ = rσ,τ ps(σ) + X(α,β)∈F \{(σ,τ )} = rσ,τ ps(σ) + X(α,τ )∈F rα,τ sσ∗sα. rα,τ sσ∗sαsβ ∗sτ (1) α6=σ If d(σ) 6= d(α) then, by [3, If d(σ) = d(α) and σ 6= α, then sσ∗sα = 0 by (KP3). Lemma 3.1], sσ∗sα is a linear combination of sµsν ∗ where d(µ) − d(ν) = d(α) − d(σ). It follows that the 0-graded component of (1) is rσ,τ ps(σ). Thus 0 = rσ,τ ps(σ). But ps(σ) 6= 0 by Theorem 3.4 of [3]. Hence rσ,τ = 0. Since (σ, τ ) ∈ F was arbitrary, it follows that {sαsβ ∗ : s(α) = s(β) and d(β) = m} is linearly independent. (cid:3) The next lemma describes properties of elements in the center of KPR(Λ). Lemma 4.2. Let a ∈ Z(KPR(Λ)) \ {0} be in normal form P(α,β)∈F rα,βsαsβ ∗. (1) If (σ, τ ) ∈ F , then r(σ) = r(τ ). (2) Let W = {v ∈ Λ0 : ∃(α, β) ∈ F with v = r(β)}. If µ ∈ ΛW , then r(µ) ∈ W . (3) If (σ, τ ) ∈ F , then there exists (α, β) ∈ F such that r(α) = r(β) = s(σ) = s(τ ). (4) There exists l ∈ N \ {0} and {(αi, βi)}l i=1 ⊂ F such that β1 · · · βl is a closed path in Λ. Proof. (1) Let (σ, τ ) ∈ F . By [5, Lemma 2.3] we have 0 6= sσ∗asτ . Since a ∈ Z(KPR(Λ)) 0 6= sσ∗pr(σ)apr(τ )sτ = sσ∗apr(σ)pr(τ )sτ = δr(σ),r(τ )sσ∗asτ . Hence r(σ) = r(τ ). (2) By way of contradiction, assume there exists µ ∈ ΛW such that r(µ) /∈ W . Then pvpr(µ) = 0 for all v ∈ W . Thus apr(µ) = X(α,β)∈F rα,βsαsβ ∗pr(β)pr(µ) = 0. Since a ∈ Z(KPR(Λ)) we get sµa = asµ = apr(µ)sµ = 0. Since s(µ) ∈ W , there exist (α′, β′) ∈ F with r(β′) = s(µ). Then r(α′) = s(µ) also by (1). Thus S := {(α, β) ∈ F : s(µ) = r(α)} is non-empty, and 0 = sµa = X(α,β)∈S rα,βsµαsβ ∗. But {sµαsβ ∗ : (α, β) ∈ S} is linearly independent by Lemma 4.1, and hence rα,β = 0 for all (α, β) ∈ S. This contradicts the given normal form for a. THE CENTER OF A KUMJIAN-PASK ALGEBRA 5 (3) Let (σ, τ ) ∈ F . Then s(σ) = s(τ ) by definition of normal form. By Lemma 2.3 in [5] we have sσ∗asτ 6= 0. Since a ∈ Z(KPR(Λ)), 0 6= sσ∗asτ = asσ∗sτ = X(α,β)∈F rα,βsαsβ ∗sσ∗sτ = X(α,β)∈F rα,βsαs(σβ)∗sτ . (2) r(β)=s(σ) In particular, the set {(α, β) ∈ F : r(β) = s(σ)} is nonempty. So there exists (α′, β′) ∈ F such that r(β′) = s(σ). Since r(α′) = r(β′) from (1), we are done. (4) Let M = F +1. Using (3) there exists a path β1 . . . βM such that, for 1 ≤ i ≤ M, there exists αi ∈ Λ with (αi, βi) ∈ F . Since M > F , there exists i < j ∈ {1, . . . , M} such that βi = βj. Then βi . . . βj−1 is a closed path. (cid:3) The next corollary follows from Lemma 4.2(4). Corollary 4.3. Let Λ be a row-finite k-graph with no sources and R a commutative ring with 1. If Λ has no closed paths then the center Z(KPR(Λ)) = {0}. The next lemma provides a description of elements of the center of KPR(Λ) when Λ is cofinal. Lemma 4.4. Suppose that Λ is cofinal. Let a ∈ Z(KPR(Λ)) \ {0} be in normal form in normal form, F ⊂ Λ × Λm for some m ∈ Nk. Let n = m ∨ (1, 1, . . . , 1). By (KP4), for P(α,β)∈F rα,βsαsβ ∗. Then {v ∈ Λ0 : ∃(α, β) ∈ F with v = r(β)} = Λ0. Proof. Write W := {v ∈ Λ0 : ∃(α, β) ∈ F with v = r(β)}. Since P(α,β)∈F rα,βsαsβ ∗ is each (α, β) ∈ F , we have sαsβ ∗ = Pµ∈s(α)Λn−m s(αµ)s(βµ)∗. By "reshaping" each pair of paths in F in this way, collecting like terms and dropping those with zero coefficients, we see that there exists G ⊂ Λ×Λn and r′ γ,ηsγsη∗ is also in normal form. By construction, W ′ = {v ∈ Λ0 : ∃(γ, η) ∈ G with v = r(η)} ⊂ W . i=1 ⊂ G such that η1 · · · ηl is a closed path. Since d(ηi) ≥ (1, 1, . . . , 1) for all i, x := η1 · · · ηlη1 · · · ηlη1 · · · is an infinite path. By cofinality, there exist q ∈ Nk and ν ∈ vΛx(q). By the definition of x, there exist q′ ≥ q and j such that x(q′) = r(ηj). Let λ = x(q, q′). Then νλ ∈ vΛW ′. By Lemma 4.2(2), v = r(νλ) ∈ W ′ as well. Thus W ′ = Λ0, and since W ′ ⊂ W we have W = Λ0. (cid:3) γ,η ∈ R\{0} such that a = P(γ,η)∈G r′ Let v ∈ Λ0. Using Lemma 4.2(4), there exists {(γi, ηi)}l The next lemma provides a description of elements of the center of KPR(Λ) when Λ is aperiodic. Lemma 4.5. Suppose that Λ is aperiodic and a ∈ Z(KPR(Λ)) \ {0}. Then there exist n ∈ Nk and G ⊂ Λn such that a = Pα∈G rαsαsα∗ is in normal form. Proof. Suppose a ∈ Z(KPR(Λ)) \ {0} with a = P(α,β)∈F rα,βsαsβ ∗ in normal form. Let (σ, τ ) ∈ F. From Lemma 2.3 in [5] we know that sσ∗asτ 6= 0. Let m = ∨(α,β)∈F (d(α) ∨ d(β)). Since Λ is aperiodic, by [6, Lemma 6.2], there exist λ ∈ s(σ)Λ with d(λ) ≥ m such that α, β ∈ Λs(σ), d(α), d(β) ≤ m and αλ(0, d(λ)) = βλ(0, d(λ)) (cid:27) =⇒ α = β. (3) The same argument as in [3, Proposition 4.9] now shows that sλ∗sσ∗asτ sλ 6= 0. Since a ∈ Z(KPR(Λ)), 0 6= sλ∗sσ∗asτ sλ = asλ∗sσ∗sτ sλ = as(σλ)∗ sτ λ. Thus 0 6= s(σλ)∗ sτ λ = sσλ(d(λ),d(λ)+d(σ))∗ sσλ(0,d(λ))∗ sτ λ(0,d(λ))sτ λ(d(λ),d(λ)+d(τ )) = δσ,τ sσλ(d(λ),d(λ)+d(σ))∗ sτ λ(d(λ),d(λ)+d(τ )) using (3). 6 J. H. BROWN AND A. AN HUEF Thus σ = τ . Since (σ, τ ) ∈ F was arbitrary we have α = β for all (α, β) ∈ F . Let G = {α ∈ Λ : (α, α) ∈ F } and write rα for rα,α. Note G ⊂ Λn because F ⊂ Λ × Λn for some. Thus (cid:3) a = Pα∈G rαsαsα∗ in normal form as desired. Our main theorem (Theorem 4.7) has two cases: Λ0 finite and infinite. Lemma 4.6. KPR(Λ) has an identity if and only if Λ0 is finite. that KPR(Λ) has an identity 1KPR(Λ). By way of contradiction, suppose that Λ0 is Proof. If Λ0 is finite, then Pv∈Λ0 pv is an identity for KPR(Λ). Conversely, assume infinite. Write 1KPR(Λ) in normal form P(α,β)∈F rα,βsαsβ ∗. Since F is finite, so is pw = 1KPR(Λ)pw = P(α,β)∈F rα,βsαsβ ∗pw = 0 because w 6= r(β) for any of the β. This W := {v ∈ Λ0 : ∃(α, β) ∈ F with v = r(β)}. Thus there exists w ∈ Λ0 \ W . Now contradiction shows that Λ0 must be finite. (cid:3) Theorem 4.7. Let Λ be a row-finite k-graph with no sources and R a commutative ring with 1. (1) Suppose Λ is aperiodic and cofinal, and that Λ0 is finite. Then Z(KPR(Λ)) = R1KPR(Λ). (2) Suppose that Λ is cofinal and Λ0 is infinite. Then Z(KPR(Λ)) = {0}. Proof. (1) Suppose Λ is aperiodic and cofinal, and that Λ0 is finite. Let a ∈ Z(KPR(Λ))\ {0}. Since Λ is aperiodic, by Lemma 4.5, there exist G ⊂ Λn such that a = Pα∈G rαsαsα∗ is in normal form. Since Λ is row-finite and Λ0 is finite, Λn is finite. We claim that G = Λn. By way of contradiction, suppose that G 6= Λn, and let λ ∈ Λn \ G. Then asλ = 0 by (KP3). But since a ∈ Z(KPR(Λ)), 0 = asλ = sλa = Xα∈G r(α)=s(λ) rαsλαsα∗. Since Λ is confinal, {r(α) : α ∈ G} = Λ0 by Lemma 4.4. Thus S = {α ∈ G : r(α) = s(λ)} 6= ∅. But {sλαsα∗ : α ∈ S} is linearly independent by Lemma 4.1. Thus rα = 0 for α ∈ S, contradicting our choice of {rα}. It follows that G = Λn as claimed, and that a = Xα∈Λn rαsαsα∗. Next we claim that rµ = rν for all µ, ν ∈ Λn. Let µ, ν ∈ Λn. Let x ∈ s(µ)Λ∞. Since Λ is cofinal, there exists m ∈ Nk and γ ∈ s(ν)Λs(x(m)). Set η = x(0, m). Now rµsνγs(µη)∗ = rµsνγsη∗sµ∗ = sνγsη∗ Xα∈Λn rµsµ∗sαsα∗ rαsαsα∗ = sνγs(µη)∗ a = sνγsη∗sµ∗ Xα∈Λn = asνγs(µη)∗ = Xα∈Λn rαsαsα∗sνsγs(µη)∗ = rνsνγs(µη)∗ . Since sνγs(µη)∗ 6= 0 this implies rµ = rν. Let r = rµ. Now a = Xα∈Λn rsαsα∗ = Xv∈Λ0 Xα∈vΛn rsαsα∗ = r Xv∈Λ0 pv = r1KPR(Λ) as desired. THE CENTER OF A KUMJIAN-PASK ALGEBRA 7 (2) Suppose there exists a ∈ Z(KPR(Λ)) \ {0}. Write a = P(α,β)∈F rα,βsαsβ ∗ in normal form. Then Lemma 4.4 gives that {v ∈ Λ0 : ∃(α, β) ∈ F with v = r(β)} = Λ0, contradicting that F is finite. (cid:3) Simplicity of C ∗(Λ) played an important role in §3. To reconcile this with Theo- rem 4.7, recall from [10] that an ideal I ∈ KPR(Λ) is basic if rpv ∈ I for r ∈ R \ {0} then pv ∈ I, and that KPR(Λ) is basically simple if its only basic ideals are {0} and KPR(Λ). By [3, Theorem 5.14], KPR(Λ) is basically simple if and only if Λ is cofinal and aperiodic (and by [3, Theorem 6.1], KPR(Λ) is simple if and only if R is a field and Λ is cofinal and aperiodic). Thus Theorem 4.7 is in the spirit of Corollary 3.3. 5. Commutative Kumjian-Pask algebras We view Nk as a category with one object ⋆ and composition given by addition, and i=1 to denote the standard basis of Nk. use {ei}k Example 5.1. Let d : Nk → Nk be the identity map. Then (Nk, d) is a k-graph. By [3, Example 7.1], KPR(Nk) is commutative with identity p⋆, and KPR(Nk) is isomorphic to the ring of Laurent polynomials R[x1, x−1 k ] in k commuting indeterminates. Lemma 5.2. Suppose Λ = Λ1F Λ2 is a disjoint union of two k-graphs. Then KPR(Λ) = KPR(Λ1) ⊕ KPR(Λ2). 1 , . . . , xk, x−1 let (qi, ti) be the generating Kumjian-Pask Λi-family of Proof. For each i = 1, 2, KPR(Λi), and let (p, s) be the generating Kumjian-Pask Λ-family of KPR(Λ). Re- stricting (p, s) to Λi gives a Λi-family in KPR(Λ), and hence the universal property for KPR(Λi) gives a homomorphism πi p,s ◦ (qi, ti) = (p, s). Each πi p,s is graded, and the graded uniqueness theorem ([3, Theorem 4.1]) implies that πi p,s is injective. p,s : KPR(Λi) → KPR(Λ) such that πi We now identify KPR(Λi) with its image in KPR(Λ). If µ ∈ Λ1 and λ ∈ Λ2, then sµsλ = sµps(µ)pr(λ)sλ = 0. Similarly sλsµ, sµ∗sλ∗, sλ∗sµ∗, sλsµ∗, sµ∗sλ, sλ∗sµ, sµsλ∗ are all zero. Thus KPR(Λ1) KPR(Λ2) = 0 = KPR(Λ2) KPR(Λ1), and the internal direct sum KPR(Λ1) ⊕ KPR(Λ2) is a subalgebra of KPR(Λ). Finally, KPR(Λ1) ⊕ KPR(Λ2) is all of KPR(Λ) since the former contains all the generators of later. This gives the result. (cid:3) Proposition 5.3. Let Λ be a row-finite k-graph with no sources and R a commutative ring with 1. Then the following conditions are equivalent: (1) KPR(Λ) is commutative; (2) r = s on Λ and rΛn is injective; (3) Λ ∼= Fv∈Λ0 Nk; (4) KPR(Λ) ∼= Lv∈Λ0 R[x1, x−1 1 , . . . , xk, x−1 k ]. Proof. (1) ⇒ (2). Suppose that KPR(Λ) is commutative. By way of contradiction, suppose there exists λ ∈ Λ such that s(λ) 6= r(λ). Then sλ∗sλ = sλsλ∗, and ps(λ) = p2 s(λ) = ps(λ)sλ∗sλ = ps(λ)sλsλ∗ = ps(λ)pr(λ)sλsλ∗ = 0. But pv 6= 0 for all v ∈ Λ0 by [3, Theorem 3.4]. This contradiction gives r = s. Next, supppose λ, µ ∈ Λn with λ 6= µ. By way of contradiction, suppose that r(λ) = r(µ). Since r = s, r(λ) = s(λ) = s(µ) = r(µ). Then sλ = pr(λ)sλ = ps(λ)sλ = ps(µ)sλ = sµ∗sµsλ = sµ∗sλsµ = 0 by (KP3). Now ps(λ) = 0, contradicting that pv 6= 0 for all v ∈ Λ0 by [3, Theorem 3.4]. Thus r is injective on Λn. 8 J. H. BROWN AND A. AN HUEF (2) ⇒ (3) Assume that r = s on Λ and that rΛn is injective. Since r = s, {vΛv}v∈Λ0 is a partition of Λ. Since r is injective on Λei, the subgraph vΛeiv has a single vertex 7→ ei defines a graph isomorphism vΛv → Nk. Hence v and single edge f v i . Thus f v i Λ = Fv∈Λ0 vΛv ∼= Fv∈Λ0 Nk. (3) ⇒ (4) Assume that Λ ∼= Fv∈Λ0 Nk. By Lemma 5.2, KPR(Λ) is isomorphic to L KPR(Nk), and by Example 5.1 each KPR(Nk) is isomorphic to R[x1, x−1 (4) ⇒ (1) Follows since L R[x1, x−1 1 , . . . , xk, x−1 k ]. (cid:3) k ] is commutative. 1 , . . . , xk, x−1 References [1] G. Abrams and G. Aranda Pino, The Leavitt path algebra of a graph, J. Algebra 293 (2005), 319 -- 334. [2] P. Ara, M.A. Moreno and E. Pardo, Nonstable K-theory for graph algebras, Algebr. Represent. Theory 10 (2007), 157 -- 178. [3] G. Aranda Pino, John Clark, A. an Huef and I. Raeburn, Kumjian-Pask algebras of higher-rank graphs, to appear in Trans. Amer. Math. Soc. [4] G. Aranda Pino and K. Crow, The center of a Leavitt path algebra, Rev. Mat. Iberoam. 27 (2011), 621 -- 644. [5] J.H. Brown and A. an Huef, The socle and semisimplicity of a Kumjian-Pask algebra, arXiv:1201.1888. [6] R. Hazlewood, I. Raeburn, A. Sims and S.B.G. Webster, On some fundamental results about higher- rank graphs and their C ∗-algebras, to appear in Proc. Edinburgh. Math. Soc. [7] A. Kumjian and D. Pask, Higher rank graph C ∗-algebras, New York J. Math. 6 (2000), 1 -- 20. [8] I. Raeburn and D.P. Williams, Morita equivalence and continuous-trace C ∗-algebras, Mathematical Surveys and Monographs, vol. 60, American Mathematical Society, Providence, RI, 1998. [9] D.I. Robertson and A. Sims, Simplicity of C ∗-algebras associated to higher-rank graphs, Bull. Lond. Math. Soc. 39 (2007), 337 -- 344. [10] M. Tomforde, Leavitt path algebras with coefficients in a commutative ring, J. Pure Appl. Algebra 215 (2011), 471 -- 484. Department of Mathematics, Kansas State University, 138 Cardwell Hall, Manhat- tan, KS 66506-2602, USA. E-mail address: [email protected] Department of Mathematics and Statistics, University of Otago, Dunedin 9054, New Zealand. E-mail address: [email protected]
1801.02088	1	1801	2018-01-06T20:50:21	Mobi algebra as an abstraction to the unit interval and its comparison to rings	[ "math.RA" ]	We begin by introducing an algebraic structure with three constants and one ternary operation to which we call mobi algebra. This structure has been designed to capture the most relevant properties of the unit interval that are needed in the study of geodesic paths. Another algebraic structure, called involutive medial monoid (IMM), can be derived from a mobi algebra. We prove several results on the interplay between mobi algebras, IMM algebras and unitary rings. It turns out that every unitary ring with one half uniquely determines and is uniquely determined by a mobi algebra with one double. This paper is the second of a planned series of papers dedicated to the study of geodesic paths from an algebraic point of view.	math.RA	math	MOBI ALGEBRA AS AN ABSTRACTION TO THE UNIT INTERVAL AND ITS COMPARISON TO RINGS J. P. FATELO AND N. MARTINS-FERREIRA Abstract. We begin by introducing an algebraic structure with three constants and one ternary operation to which we call mobi algebra. This structure has been designed to capture the most relevant properties of the unit interval that are needed in the study of geodesic paths. Another algebraic structure, called involutive medial monoid (IMM), can be derived from a mobi algebra. We prove several results on the interplay between mobi algebras, IMM algebras and unitary rings. It turns out that every unitary ring with one half uniquely determines and is uniquely determined by a mobi algebra with one double. This paper is the second of a planned series of papers dedicated to the study of geodesic paths from an algebraic point of view, the first paper in the series is [2]. 1. Introduction This paper is part of an ongoing work which aims at an axiomatic characterization of spaces with unique geodesic paths between any two of its points. Our interest in the study of geodesic paths from an alge- braic point of view has started with an observation on weakly Mal'cev categories [4]. The observation is that a certain ternary operation, say p = p(x, y, z), which arises in the context of weakly Mal'tsev quasi- varieties [5], could be used to describe the path of a particle moving in space along a geodesic curve from a point x to a point z at an instant y. By fixing y to a value that positions the particle at half way from x to z, a binary operation is obtained. The study of this binary operation was a first step in our investigation [2]. Allowing the variable y to range over the unit interval we are able to capture the whole movement of a particle in a geodesic path. The main purpose of this paper is to communicate the structure of a mobi algebra (Definition 2.1), which has arisen as a candidate for an algebraic version of the unit interval. It may be argued that this study could have been avoided if we would have decided to simply work within the unit 2010 Mathematics Subject Classification. Primary 08C15; Secondary 20N02. Key words and phrases. Mobi algebra, involutive medial monoid, midpoint al- gebras, unitary rings, unit interval, ternary operation, geodesic path. This research work was supported by the Portuguese Foundation for Science and Technology (FCT) through the Project reference UID/ Multi/04044/2013, post doctoral grant SFRH/BPD/43216/2008 at CMUC, and also by CDRSP and ESTG from the Polytechnic Institute of Leiria. 1 2 J. P. FATELO AND N. MARTINS-FERREIRA interval. The reason why we have chosen to take this study further has to do with the fact that the axioms of Definition 2.1 can be naturally extended into higher dimensions. This means that a mobi algebra is at the same time a model for the unit interval and an instance of a one- dimensional space with geodesics. However, this is just an intermediate step in our study. The notion which we are aiming for in the future is the one of a mobi space. In some sense, a mobi space is to a mobi algebra in the same way as a module over a ring is to the ring of its scalars. For the moment we concentrate our attention on the unit interval as an abstract range of scalars. Here the slogan is the unit interval is to a mobi algebra in the same way as the real line is to a unitary ring. The idea of abstracting the unit interval is not new and has been considered extensively in the literature (see e.g. [1] and the references therein). After some attempts and different experiments [3], it now appears to us that the structure which we are calling mobi algebra is a suitable abstraction for the unit interval capturing the main features we are interested in. It consists of a set A equipped with a ternary operation p together with three constants 0, 1⁄2, 1 ∈ A, and satisfying the eight axioms of Definition 2.1. One of the reasons that have con- vinced us about this structure's significance is its deep connection with unitary rings (see diagram (41) at the end). In summary, every unitary ring in which the element 2 is invertible determines a unique structure of mobi algebra. In addition, a mobi algebra in which the element 1⁄2 is invertible (in a suitable sense) determines a unique unitary ring structure. As expected, the prototyping example of a mobi algebra is the unit interval of the real numbers with the three constants 0, 1 2, 1, and the ternary operation p(x, y, z) = (1 − y)x + yz. More generally, if R is a unitary ring in which the element 2 is invertible then any subset A ⊆ R, containing 0, 2−1, 1 and closed under the formula x + yz − yx, gives rise to a mobi algebra structure on the set A. This paper is organized as follows. In Section 2 we give the definition of a mobi algebra and present some basic properties. Examples are given in Section 3 and in the Appendix. In Section 4, we observe that a mobi algebra gives rise to several other structures. Namely, two monoid structures (A,◦, 0), (A,·, 1), dual to each other, one midpoint algebra (A,⊕) and an involution () : A → A. These derived operations satisfy some axioms and form what we call an involutive medial monoid, IMM for short. The importance of this structure is highlighted in Section 5 where its relation to unitary rings is shown. In Section 6 we characterize those IMM that are obtained from a mobi algebra. Our main result is Theorem 7.3 stating that every unitary ring with 2−1 uniquely determines and is uniquely determined by a mobi algebra with 1⁄2 −1. 3 2. Definition and basic properties A mobi algebra consists of three constants and one ternary operation. The ternary operation p(x, y, z) is thought of as being the point, in a path linking x to z, which lies at a location y in-between. This image might provide some intuition and the reader is invited to keep in mind either one of the two formulas p = (1−y)x+yz or p = x+y(z−x). The motivating example is the unit internal [0, 1], with the three constants 0, 1 2 , 1 and the formula p(x, y, z) = x + yz − yx. Note that we will consistently distinguish between the element 1⁄2, which is only used as a symbol, and the real number 1 2 ∈ [0, 1]. Definition 2.1. A mobi algebra is a system (A, p, 0, 1⁄2, 1), in which A is a set, p is a ternary operation and 0, 1⁄2 and 1 are elements of A, that satisfies the following axioms: (A1) p(1, 1⁄2, 0) = 1⁄2 (A2) p(0, a, 1) = a (A3) p(a, b, a) = a (A4) p(a, 0, b) = a (A5) p(a, 1, b) = b (A6) p(a, 1⁄2, b) = p(a′, 1⁄2, b) =⇒ a = a′ (A7) p(a, p(c1, c2, c3), b) = p(p(a, c1, b), c2, p(a, c3, b)) (A8) p(p(a1, c, b1), 1⁄2, p(a2, c, b2)) = p(p(a1, 1⁄2, a2), c, p(b1, 1⁄2, b2)) The ternary operation p is not associative, not even partially. Nev- ertheless, it verifies several properties that involve an interaction of p with itself, as exemplified in the next Proposition. Proposition 2.1. Let (A, p, 0, 1⁄2, 1) be a mobi algebra. It follows that: p(a, p(0, c, d), b) = p(a, c, p(a, d, b)) p(a, p(1, c, d), b) = p(b, c, p(a, d, b)) p(a, p(c, d, 0), b) = p(p(a, c, b), d, a) p(a, p(c, d, 1), b) = p(p(a, c, b), d, b) (1) (2) (3) (4) Proof. These properties are obtained directly from axiom(A7) and the use of (A4) or (A5). (cid:3) Fixing some other elements in the previous relations, we get further properties. For example, from (2), we get the following result. Corollary 2.2. If (A, p, 1, 1⁄2, 0) is a mobi algebra, then: p(a, p(1, c, 0), b) = p(b, c, a) p(a, 1⁄2, b) = p(b, 1⁄2, a) p(1, p(1, c, 0), 0) = c (5) (6) (7) Proof. These properties are an immediate consequence of (2) and, re- spectively, axioms(A4), (A1), and (A2). (cid:3) 4 J. P. FATELO AND N. MARTINS-FERREIRA We finish this section with three more properties of a mobi algebra structure. Proposition 2.3. Let (A, p, 0, 1⁄2, 1) be a mobi algebra. It follows that: p(p(a1, c, b1), 1⁄2, p(a2, c, b2)) = p(p(a2, c, b1), 1⁄2, p(a1, c, b2)) (8) (9) (10) p(p(a, p(1, c, 0), b), 1⁄2, p(a, c, d)) = p(a, 1⁄2, p(b, c, d)). p(p(a, c, b), 1⁄2, p(b, c, a)) = p(a, 1⁄2, b) Proof. Using (A8), (6) and again (A8), we get: p(p(a1, c, b1), 1⁄2, p(a2, c, b2)) = p(p(a1, 1⁄2, a2), c, p(b1, 1⁄2, b2)) = p(p(a2, 1⁄2, a1), c, p(b1, 1⁄2, b2)) = p(p(a2, c, b1), 1⁄2, p(a1, c, b2)). Property (9) is just a particular case of (8) if we consider (A3). Using (A7),(A4), (A5), (A8), (6) and (A3), we get: p(p(a, p(1, c, 0), b), 1⁄2, p(a, c, d)) = p(p(p(a, 1, b), c, p(a, 0, b)), 1⁄2, p(a, c, d)) = p(p(b, c, a), 1⁄2, p(a, c, d)) = p(p(b, 1⁄2, a), c, p(a, 1⁄2, d)) = p(p(a, 1⁄2, b), c, p(a, 1⁄2, d)) = p(p(a, c, a), 1⁄2, p(b, c, d)) = p(a, 1⁄2, p(b, c, d)). (cid:3) Some of these basic properties will be used to derive the structure introduced in section 4 (involutive medial monoid) as an intermediate step in the comparison with rings. For the moment we observe some examples. 3. Examples In this section we give examples of mobi algebras by presenting a set A, a ternary operation p(x, y, z) ∈ A, for all x, y, z ∈ A, and three constants 0, 1⁄2, 1 in A. Our prototyping example is clearly the unit interval. We observe that it can be defined with the usual ternary operation p(a, b, c) = (1−b)a+bc with one half as the element 1⁄2 but it can also be considered in other forms like those of Examples 2 to 5. Note that Examples 1 and 2 are isomorphic via the bijection x 7→ x 2−x . More in general, the (α−1)+(2−α)x induces an automorphism leading to 1⁄2 = 1 bijection x 7→ α , α ∈]1, +∞[ (see Example 3). Example 4 is isomorphic, via the bijective correspondence x 7→ 2x−1, to Example 1. Via x 7→ 1 x we easily observe that Example 5 is isomorphic to Example 1. x All examples have a mobi algebra structure (A, p, 0, 1⁄2, 1), as given in Definition 2.1, where the symbols 0, 1⁄2 and 1 are explicit. Example 1: (A, p, 0, 1 2 , 1), with A = [0, 1] and p(a, b, c) = (1 − b)a + bc 5 Example 2: (A, p, 0, 1 3 , 1), with A = [0, 1] and for all a, b, c ∈ A p(a, b, c) = a − ab + ac + 2bc + abc 1 + b + c + 2ab − bc Example 3: For any value of α ∈]1, +∞[ we have (A, p, 0, 1 α , 1), with A = for all a, b, c ∈ A [0, 1] and p(a, b, c) = a − ab + (α − 2)ac + (α − 1)bc + (α − 2)2abc 1 + (α − 2)(b + c − bc) + (α − 1)(α − 2)ab for all a, b, c ∈ A Example 4: (A, p,−1, 0, 1), with A = [−1, 1] and p(a, b, c) = a(1 − b) + c(1 + b) 2 for all a, b, c ∈ A Example 5: (A, p, +∞, 2, 1), with A = [1, +∞] and p(a, b, c) = abc a − c + bc for all a, b, c ∈ A Example 6: Excluding the trivial case, where 0 = 1⁄2 = 1, the smallest mobi algebra is the set of constants A = {0, 1⁄2, 1} with the operation p defined as in the following tables. p(-,1,-) 0 1⁄2 1 p(-,0,-) 0 1⁄2 1 0 1⁄2 1 0 1⁄2 1 Isomorphically, we can consider the correspondence (0, 1⁄2, 1) with (−1, 0, 1), (0, 1 2 , 1), (0, 1, 2) or (0, 2, 1) obtaining thus some natural structures on the given sets. 1 1⁄2 0 1 0 0 1⁄2 1 1⁄2 1 1⁄2 0 1⁄2 0 1⁄2 1 1 0 1⁄2 1 0 0 1 1⁄2 p(-,1⁄2,-) 0 1⁄2 1 0 1⁄2 1 0 1⁄2 1 Example 7: Considering the correspondence (0, 1⁄2, 1) 7→ (0, 2, 1), the previ- ous example may be generalized as follows. For every number n ∈ N, we have (A, p, 0, n + 1, 1) with A = {0, 1, . . . , 2n} and p(a, b, c) = a − ba + bc mod 2n + 1. Example 8: The dyadic numbers as a subset of the unit interval, with the same structure as in Example 1. Clearly, if a, b, c are dyadic numbers then p(a, b, c) is dyadic. Example 9: (A, p, 0, 1 2 , 1) with A as Q, R or C, and p(x, y, z) = (1−y)x+yz. 6 J. P. FATELO AND N. MARTINS-FERREIRA Example 10: An example of a mobi algebra in R2 is (A, p, 0, 1⁄2, 1) with A = (cid:8)(x, y) ∈ R2 : y ≤ x ≤ 1 − y(cid:9) p(a, b, c) = (a1 − b1a1 − b2a2 + b1c1 + b2c2, a2 − b1a2 − b2a1 + b1c2 + b2c1) Example 11: The previous example may be generalized, for any K ∈ R, by and the three constants 1⁄2 = (cid:0) 1 2 , 0(cid:1); 1 = (1, 0); 0 = (0, 0). defining p(a, b, c) = ((1 − b1)a1 + b1c1 + Kb2(c2 − a2), (1 − b1)a2 + b1c2 + b2(c1 − a1)). 2 , 0(cid:1), 1 = (1, 0) and 0 = (0, 0), (A, p, 0, 1⁄2, 1) is a With 1⁄2 = (cid:0) 1 mobi algebra on A = R2 and, for K ∈ R+ A = n(x, y) ∈ R2 : √Ky ≤ x ≤ 1 − √Kyo . 0 , on This example is obtained by identifying the plane with the ring of 2 by 2 matrices of the form (x, y) 7→ (cid:18) x k2 y k1 y x (cid:19) , letting K = k1 k2 and computing p(a, b, c) as (1 − b)a + bc. Example 12: Any unitary ring R in which the element 2 is invertible gives an example of a mobi algebra (R, p, 0, 2−1, 1) with p(x, y, z) = (1 − y)x + yz (Theorem 7.2). Example 13: Any subset S of a unitary ring (in which 2 is an invertible element) that is closed under the formula p(x, y, z) = (1− y)x + yz and contains the three constants 0, 2−1 and 1 gives a mobi algebra (S, p, 0, 2−1, 1). Example 14: Let (A, +,·, 0, 1) be a semiring such that 2 is invertible (i.e. ∃ 2−1 : 2 · 2−1 = 1 = 2−1 · 2) and it exists B ⊆ A with the following properties: p + b · a = a + b · c, i) 1, 2−1, 0 ∈ B; ii) ∀a, b, c ∈ B,∃ 1! solution p = p(a, b, c) ∈ B to the equation then the system (B, p, 0, 2−1, 1) is a mobi Algebra. The same arguments that are used in the case of rings (Theorem 7.2) are valid here by rearranging the terms in order to avoid negative terms. Example 15: Every finite mobi is uniquely determined by (and uniquely de- termines) a unitary ring in which 2 is invertible. Indeed, let (A, p, 0, 1⁄2, 1) be a finite mobi algebra and consider the function h : A → A such that h(x) = p(0, 1⁄2, x). By axiom (A6), h is injective. Now, as A is a finite set, h is also surjective. So h is a bijection and, in particular, it exists h−1(1) which is a solution 7 to the equation p(0, 1⁄2, x) = 1. In other words, h−1(1) is 1⁄2 and so Theorem 7.3 holds. −1 Note that Example 7 above illustrates the fact that every finite mobi must have an odd number of elements. 4. Derived operations and IMM algebras A closer look to the propositions of Section 2 suggests that some properties of mobi algebras can be suitably expressed in terms of a unary operation "()" and binary operations "·", "◦" and "⊕" defined as follows. Definition 4.1. Let (A, p, 0, 1⁄2, 1) be a mobi algebra. We define: (11) (12) (13) (14) In the first example of previous section, with p(a, b, c) = (1− b)a + bc a = p(1, a, 0) a · b = p(0, a, b) a ⊕ b = p(a, 1⁄2, b) a ◦ b = p(a, b, 1). on A = [0, 1], these operations have the following form: a = 1 − a a + b a · b = ab a ⊕ b = 2 a ◦ b = a + b − ab. In Example 11, the derived operations are: (x, y) = (1 − x,−y) y + y′ ) , (x, y) · (x′, y′) = (xx′ + Kyy′, xy′ + yx′) (x, y) ⊕ (x′, y′) = ( (x, y) ◦ (x′, y′) = (x + x′ − x′x − K y′y, y + y′ − x′y − y′x). x + x′ 2 2 In particular, complex multiplication is obtained as ·, if letting A = From property (5) and (7), as well as axioms (A5) and (A7), we R2 ∼= C and K = −1. immediately find that: a = a 1 = 0 p(b, c, a) = p(a, c, b) p(a, c, b) = p(a, c, b). (15) (16) (17) (18) 8 J. P. FATELO AND N. MARTINS-FERREIRA These relations show, in particular, that · and ◦ are dual operations in the sense that: a · b = b ◦ a a ◦ b = b · a. (20) This is why, we will leave out the operation ◦ in the rest of the article, except for noting that (8) gives the following relation between the three binary operations: (19) (a ◦ b) ⊕ (b · a) = b ⊕ a. It is also worth noting that, (A2) implies that 1⁄2 = 1 ⊕ 1. (21) (22) (23) Using (10), we can relate the ternary operation p of any mobi algebra with the derived operations through the relation: 1⁄2 · p(a, b, c) = (b · a) ⊕ (b · c). This property will be at the bottom line of Section 7 for comparing a mobi algebra with rings. Before that, we show, in Proposition 4.1 below, that every mobi algebra gives rise to a new structure, that we call involutive medial monoid (IMM), presented in the following Definition . Definition 4.2. An IMM algebra is a system (B, (),⊕,·, 1), in which B is a set, () is an unary operation, ⊕ and · are binary operations and 1 is an element of B, that satisfies the following axioms: (B1) a ⊕ a = a (B2) a ⊕ b = b ⊕ a (B3) (a ⊕ b) ⊕ (c ⊕ d) = (a ⊕ c) ⊕ (b ⊕ d) (B4) a · (b · c) = (a · b) · c (B5) a · 1 = a = 1 · a (B6) a · (b ⊕ c) = (a · b) ⊕ (a · c), (B7) a = a (B8) a ⊕ b = a ⊕ b (B9) a · 1 = 1 = 1 · a (B10) a ⊕ a = 1 ⊕ 1 (a ⊕ b) · c = (a · c) ⊕ (b · c) The name IMM is chosen to highlight the existence of an involution, the presence of the medial law (B3)[6] satisfied by ⊕ and the fact that (B,·, 1) is a monoid. Proposition 4.1. If (A, p, 0, 1⁄2, 1) is a mobi algebra and if (), ⊕, and · are defined as in (11), (12) and (13), then (A, (),⊕,·, 1) is an IMM algebra in which 1 = 0 and 1 ⊕ 1 = 1⁄2. Proof. All axioms of an IMM are easily proved using the axioms of a mobi algebra and the properties presented in Section 2. Indeed: (B1) is a particular case of (A3); (B2) is just a rewriting of (6); (B3) is a consequence of (A8); (B4) follows from (A4) and (A7), like (1). The first equality in (B5) is (A2) and the second comes from (A5). Left-distributivity in (B6) is deduced from (A3) and (A8) and right- distributivity from (A7), as follows: 9 a · (b ⊕ c) = p(0, a, p(b, 1⁄2, c)) = p(p(0, 1⁄2, 0), a, p(b, 1⁄2, c)) = p(p(0, a, b), 1⁄2, p(0, a, c)) = (a · b) ⊕ (a · c); (a ⊕ b) · c = p(0, p(a, 1⁄2, b), c) = p(p(0, a, c), 1⁄2, p(0, b, c)) = (a · c) ⊕ (b · c). (B7) is just a rewriting of (7); (B8) is a consequence of (A7), while (B9) can be proved through (A3), (A4) and (A5): a · 1 = p(0, a, p(1, 1, 0) = p(0, a, 0) = 0 = 1 1 · a = p(0, p(1, 1, 0), a) = p(0, 0, a) = 0 = 1. Using (A1), (A2), (A3) and (A8), we can prove (B10): a ⊕ a = p(p(1, a, 0), 1⁄2, a) = p(p(1, a, 0), 1⁄2, p(0, a, 1)) = p(p(1, 1⁄2, 0), a, p(0, 1⁄2, 1)) = p(1⁄2, a, 1⁄2) = 1⁄2. (cid:3) A finite example of an IMM algebra which is not obtained from a mobi algebra is (A, (),⊕,·, 1) with A = {0, 1⁄2, 1} and the operations ⊕ and · as in the following tables. 1 1⁄2 1⁄2 1⁄2 ⊕ 0 0 0 1⁄2 1⁄2 1 1⁄2 1⁄2 1⁄2 1⁄2 1 0 1⁄2 · 0 0 0 1⁄2 0 1⁄2 1 0 1⁄2 1 0 1⁄2 1 It is clear that ⊕ is not cancellative. We finish this section with some properties of an IMM algebra. Proposition 4.2. Let (B, (),⊕,·, 1) be a IMM algebra. It follows that: (24) a ⊕ (b ⊕ c) = (a ⊕ b) ⊕ (a ⊕ c) 1 ⊕ 1 = 1 ⊕ 1 a = a ⇒ a = 1 ⊕ 1 (1 ⊕ 1) · a = 1 ⊕ a (1 ⊕ 1) · a = a · (1 ⊕ 1). Proof. (24) is a direct consequence of (B1) and (B3); (25) of (B2), (B7) and (B8); and (26) is a consequence of (B1) and (B10). To prove (27) and (28), we use (B6), (B5) and (B9): (1 ⊕ 1) · a = (1 · a) ⊕ (1 · a) = 1 ⊕ a a · (1 ⊕ 1) = (a · 1) ⊕ (a · 1) = 1 ⊕ a. (25) (26) (27) (28) (cid:3) 10 J. P. FATELO AND N. MARTINS-FERREIRA These are the main properties that will be used in the following sections. 5. IMM algebras and unitary rings We are now going to see how, in an IMM algebra, the operation ⊕, under the existence of an inverse (in the sense of ·) to the element 1⊕1, gives rise to the additive structure of a unitary ring with one half. Let's begin by recalling that, in a monoid (A,·, 1), if an element admits an inverse, the inverse is unique. Indeed, suppose that x · a = 1 = a · x and a · x′ = 1 = x′ · a, then x = x · 1 = x · (a · x′) = (x · a) · x′ = 1 · x′ = x′. As usual, the inverse of a is denoted a−1. So, in an IMM algebra (A, (),⊕,·, 1), when the equation (1 ⊕ 1) · x = 1 has a solution, its unique solution is denoted (1 ⊕ 1)−1. The fact that 1 ⊕ 1 is central in the monoid (A,·, 1), as expressed in (28), implies that its inverse, when it exists, is also central: x · (1 ⊕ 1)−1 = (1 ⊕ 1)−1 · x, (29) Indeed, if x·a = a·x then x·a−1 = a·a−1·x·a−1 = a−1·x·a−1·a = a−1·x. As we will see, the following proposition is essential to find the sym- metric elements in the induced ring. ∀x ∈ A. Proposition 5.1. If (A, (),⊕,·, 1) is an IMM algebra and if 1 ⊕ 1 admits an inverse, then: Proof. This Property follows directly from (27). 1 ⊕ (1 ⊕ 1)−1 = 1. (30) (cid:3) Before presenting the main results of this section through the Theo- rems 5.2 and 5.3, we enumerate the axioms of a ring in Definition 5.1 below so we can refer to them in the subsequent demonstrations. Definition 5.1. A unitary ring is a system (R, +,·, 0, 1) that satisfies the following axioms: (R1) (a + b) + c = a + (b + c) (R2) a + b = b + a (R3) a + 0 = a (R4) ∀a,∃ − a : −a + a = 0 (R5) a · (b · c) = (a · b) · c (R6) a · 1 = a = 1 · a (R7) a · (b + c) = (a · b) + (a · c), Immediate consequences of the axioms are the following properties: (a + b) · c = (a · c) + (b · c) a · 0 = 0 = 0 · a a + b = a′ + b ⇒ a = a′. (31) (32) 11 The main result of this section shows the connection between an IMM algebra and a ring. It claims that an IMM algebra (A, (),⊕,·, 1) determines a unique structure of a ring on its underlying set, if, and only if, its element (1 ⊕ 1) is invertible, i.e. that (1 ⊕ 1)−1 exists. Theorem 5.2. Let (A, (),⊕,·, 1) be an IMM algebra. The following three affirmations are equivalent. (i) The equation (1 ⊕ 1) · x = 1 has a solution in A; (ii) There is a unique unitary ring (A, +,·, 1, 1) such that: a + b = (1 + 1) · (a ⊕ b) (iii) There is a unique unitary ring (A, +,·, 1, 1) such that: a ⊕ b = (1 ⊕ 1) · (a + b). Proof. In order to present this proof in a concise way, we will use the following notations: 1⁄2 = 1 ⊕ 1 2 = (1 ⊕ 1)−1. To prove that (i) implies (ii), we observe that 1⁄2 · 2 = 1 and define a + b = 2 · (a ⊕ b) −a = 2 · a. To prove that the structure (A, +,·, 1, 1) is a ring, we will prove axioms (R1) to (R7) above. For (R1), we begin with a particular case of the IMM axiom (B3) and, using also (B2), (B5), (B6) and property (27), we get: (a ⊕ b) ⊕ (1 ⊕ c) = (a ⊕ 1) ⊕ (b ⊕ c) =⇒ (a ⊕ b) ⊕ (1⁄2 · c) = (1⁄2 · a) ⊕ (b ⊕ c) =⇒ (1⁄2 · 2 · (a ⊕ b)) ⊕ (1⁄2 · c) = (1⁄2 · a) ⊕ (1⁄2 · 2 · (b ⊕ c)) =⇒ 1⁄2 · (2 · (a ⊕ b) ⊕ c) = 1⁄2 · (a ⊕ (2 · (b ⊕ c))) =⇒ 2 · 2 · 1⁄2 · ((a + b) ⊕ c) = 2 · 2 · 1⁄2 · (a ⊕ (b + c)) =⇒ (a + b) + c = a + (b + c). (R2) is a direct consequence of (B2), and (R3) follows from the state- ment 0 = 1 and the use of property (27). To prove (R4), we use (B6) and (B9), after remarking that (30), together with (B8), im- plies 1 ⊕ 2 = 1: −a + a = 2 · ((2 · a) ⊕ a) = 2 · (2 ⊕ 1) · a = 2 · 1 · a = 1. (R5) and (R6) are guaranteed by (B4) and (B5). Finally (R7) is deduced from (B6) with the use of (29). This proves existence of a ring induced by an IMM when (1 ⊕ 1)−1 exists. To prove uniqueness, we just need to show that in any ring 12 J. P. FATELO AND N. MARTINS-FERREIRA (A, +′,·, 1, 1) such that a +′ b = (1 +′ 1) · (a ⊕ b), we have (1 +′ 1) = (1 ⊕ 1)−1. Indeed: a +′ b = (1 +′ 1) · (a ⊕ b) =⇒ 1 +′ 1 = (1 +′ 1) · (1 ⊕ 1) =⇒ 1 = (1 +′ 1) · (1 ⊕ 1) =⇒ 1 +′ 1 = (1 ⊕ 1)−1. This also shows directly that (ii) implies (iii) because the inverse of 1 + 1 is 1 ⊕ 1. Now, as 1 + 1 exists in any ring and 1 ⊕ 1 = 1, we also have that (iii) implies (i). (cid:3) The previous proposition creates the question of characterizing those rings that come from an IMM algebra. Theorem 5.3 tell us that a ring (R, +,·, 0, 1) is determined by an IMM algebra structure if and only if its element (1 + 1) is invertible. Theorem 5.3. Let (R, +,·, 0, 1) be a unitary ring. The following three affirmations are equivalent. (i) The element 1 + 1 admits an inverse in R; (ii) There is a unique IMM algebra (R, (),⊕,·, 1) such that: a = 1 − a a ⊕ b = (1 ⊕ 1) · (a + b). (iii) There is a unique IMM algebra (R, (),⊕,·, 1) such that: a = 1 − a a + b = (1 + 1) · (a ⊕ b). Proof. To prove that (i) implies (ii), we define a ⊕ b = (1 + 1)−1 · (a + b). To prove that the structure (A, (),⊕,·, 1) is an IMM algebra, we will (B1) is satisfied deduce axioms (B1) to (B10) of Definition 4.2. because a + a = (1 + 1)· a. (B2) is a consequence of the commutativity of +. The medial law (B3) may be proved using the associativity of + and the distributivity of · over +: (a ⊕ b) ⊕ (c ⊕ d) = (1 + 1)−1 · ((1 + 1)−1(a + b) + (1 + 1)−1(c + d)) = (1 + 1)−1 · (1 + 1)−1 · ((a + b) + (c + d)) = (1 + 1)−1 · (1 + 1)−1 · ((a + c) + (b + d)) = (a ⊕ c) ⊕ (b ⊕ d). (B4) and (B5) are guaranteed by (R5) and (R6) while (B6) is guar- anteed by (R7) because (1 + 1)−1 commutes with all the elements of the ring. (B7) is a consequence of 1 − (1 − a) = a. A proof of (B8) goes like this: 13 a ⊕ b = 1 − (1 + 1)−1(a + b) = (1 + 1)−1(1 + 1) − (1 + 1)−1(a + b) = (1 + 1)−1((1 + 1) − (a + b)) = (1 + 1)−1((1 − a) + (1 − b)) = a ⊕ b. (B9) is obvious because 1 = 0. Finally, to prove (B10), we observe that a ⊕ a = (1 + 1)−1(1 − a + a) = (1 + 1)−1 And consequently 1 ⊕ 1 = (1 + 1)−1. This proves existence of an IMM induced by a ring when (1 + 1)−1 exists. To prove uniqueness, we just need to show that in any IMM (A, (),⊕′,·, 1) such that a ⊕′ b = (1 ⊕′ 1) · (a + b), we have (1 ⊕′ 1) = (1 + 1)−1. Indeed: a ⊕′ b = (1 ⊕′ 1) · (a + b) =⇒ 1 ⊕′ 1 = (1 ⊕′ 1) · (1 + 1) =⇒ 1 = (1 ⊕′ 1) · (1 + 1) =⇒ 1 ⊕′ 1 = (1 + 1)−1. This also shows directly that (ii) implies (iii) because the inverse of 1 ⊕ 1 is 1 + 1. Now, as 1 ⊕ 1 exists in any IMM algebra and 1 + 1 = 1, we also have that (iii) implies (i). (cid:3) 6. Mobi algebras and IMM algebras We have seen the comparison between IMM algebras and rings. We have also seen that a mobi algebra gives rise to an IMM algebra (Propo- sition 4.1). It remains to answer the question on whether an IMM algebra is obtained from a mobi algebra. Due to axiom (A6), the sub- algebra (A,⊕), of an IMM algebra which is induced by a mobi algebra, is a midpoint algebra [1, 2]. In other words, the operation ⊕ is can- cellative. We present in the Appendix two examples of a IMM algebra in which ⊕ is not cancellative (IMM 2 and IMM 3). Note that the existence of (1 ⊕ 1)−1 is sufficient to imply that ⊕ is cancellative but it is not necessary. An IMM algebra in which the operation ⊕ is cancellative will be called an IMM* algebra. Therefore, the question, that will be answered in Theorem 6.2 below, is to determine what extra conditions on an IMM* algebra are needed to certify that it comes from a mobi algebra. On an IMM* algebra, some axioms of IMM algebras may be deduced from the others using the cancellation of ⊕, thus we decided to present it as an independent algebraic structure. 14 J. P. FATELO AND N. MARTINS-FERREIRA Definition 6.1. An IMM* algebra is a system (C, (),⊕,·, 1), in which C is a set, () is an unary operation, ⊕ and · are binary operations and 1 is an element of C, that satisfies the following axioms: (C1) a ⊕ a = a (C2) a ⊕ b = b ⊕ a (C3) a ⊕ b = a′ ⊕ b =⇒ a = a′ (C4) (a ⊕ b) ⊕ (c ⊕ d) = (a ⊕ c) ⊕ (b ⊕ d) (C5) a · (b · c) = (a · b) · c (C6) a · 1 = a = 1 · a (C7) a · (b ⊕ c) = (a · b) ⊕ (a · c), (C8) a · 1 = 1 = 1 · a (C9) a ⊕ a = 1 ⊕ 1 We now observe that, in particular, every IMM* algebra is a IMM (a ⊕ b) · c = (a · c) ⊕ (b · c) algebra. Proposition 6.1. Let (C, (),⊕,·, 1) be a IMM* algebra. It follows that: (33) = a a ⊕ b a a ⊕ b b ⊕ a = 1 ⊕ 1 =⇒ b = a = (34) (35) 1 ⊕ x = (b · a) ⊕ (b · c) =⇒ 1 ⊕ x = (b · a) ⊕ (b · c). (36) Proof. Using (C2) and (C9), we have a ⊕ a = 1 ⊕ 1 and a ⊕ a = 1 ⊕ 1 which, by (C3), implies (33). Using (C4), (C9) and (C1) we find that (a ⊕ b) ⊕ (a ⊕ b) = (a ⊕ a) ⊕ (b ⊕ b) = (1 ⊕ 1) ⊕ (1 ⊕ 1) = 1 ⊕ 1 which implies (34) by cancellation, because (a ⊕ b) ⊕ (a ⊕ b) = 1 ⊕ 1. (35) is again a consequence of (C3) and (C9). To prove (36), suppose that 1 ⊕ x = (b · a) ⊕ (b · c), 1 ⊕ y = (b · a) ⊕ (b · c). and Then, we have: 1 ⊕ (x ⊕ y) = (1 ⊕ 1) ⊕ (x ⊕ y) = (1 ⊕ x) ⊕ (1 ⊕ y) = ((b · a) ⊕ (b · c)) ⊕ ((b · a) ⊕ (b · c)) = ((b · a) ⊕ (b · a)) ⊕ ((b · c) ⊕ (b · c)) = (b · (a ⊕ a)) ⊕ (b · (c ⊕ c)) = (b · (1 ⊕ 1)) ⊕ (b · (1 ⊕ 1)) = (b ⊕ b) · (1 ⊕ 1) = (1 ⊕ 1) · (1 ⊕ 1) = 1 ⊕ (1 ⊕ 1). 15 So, from (C3), we conclude that x ⊕ y = (1 ⊕ 1) which means, using (35), that y = x. (cid:3) We can now state the main result of this section. Theorem 6.2. Let (A, (),⊕,·, 1) be an IMM* algebra. The following two affirmations are equivalent. (i) For each a, b, c ∈ A, the equation 1 ⊕ x = (b · a) ⊕ (b · c) has a solution x in A; (ii) There is a unique mobi algebra (A, p, 1, 1 ⊕ 1, 1) such that: 1 ⊕ p(a, b, c) = (b · a) ⊕ (b · c). (37) (38) Proof. (ii) implies (i) because, for each a, b, c ∈ A, p(a, b, c) exists if (ii) is true and is therefore the solution of the equation (37). To prove that (i) implies (ii), we will deduce axioms (A1) to (A8) of Definition 2.1. This is facilitated by the fact that ⊕ is cancellative which means that if we find an element d of A such that 1 ⊕ d = (b · a) ⊕ (b · c), we can conclude that p(a, b, c) = d. To prove (A1), we observe that ((1 ⊕ 1) · 1) ⊕ ((1 ⊕ 1) · 1) = (1 ⊕ 1) ⊕ 1 = 1 ⊕ (1 ⊕ 1) which means that p(1, 1⊕ 1, 1) = 1⊕ 1. In a similar way, (A2) to (A5) are satisfied: (a · 1) ⊕ (a · 1) = 1 ⊕ a =⇒ p(1, a, 1) = a (b · a) ⊕ (b · a) = (b ⊕ b) · a = 1 ⊕ a =⇒ p(a, b, a) = a (1 · a) ⊕ (1 · b) = (1 · a) ⊕ 1 = 1 ⊕ a =⇒ p(a, 1, b) = a (1 · a) ⊕ (1 · b) = 1 ⊕ b =⇒ p(a, 1, b) = b. (A6) is guaranteed by (C3). To prove (A7), we first observe that and, using (36): 1 ⊕ p(c1, c2, c3) = c2 · c1 ⊕ c2 · c3 1 ⊕ p(a, c1, b) = c1 · a ⊕ c1 · b 1 ⊕ p(a, c3, b) = c3 · a ⊕ c3 · b 1 ⊕ p(c1, c2, c3) = c2 · c1 ⊕ c2 · c3. 16 J. P. FATELO AND N. MARTINS-FERREIRA Then, we use these relations, as well as the axioms of an IMM, to transform an obvious identity into (A7): (c2 · c1 · a ⊕ c2 · c3 · a) ⊕ (c2 · c1 · b ⊕ c2 · c3 · b) = (c2 · c1 · a ⊕ c2 · c3 · a) ⊕ (c2 · c1 · b ⊕ c2 · c3 · b) ⇔ (c2 · c1 · a ⊕ c2 · c1 · b) ⊕ (c2 · c3 · a ⊕ c2 · c3 · b) = (c2 · c1 · a ⊕ c2 · c3 · a) ⊕ (c2 · c1 · b ⊕ c2 · c3 · b) ⇔ c2 · (c1 · a ⊕ c1 · b) ⊕ c2 · (c3 · a ⊕ c3 · b) = (c2 · c1 ⊕ c2 · c3) · a ⊕ (c2 · c1 ⊕ c2 · c3) · b ⇔ c2 · (1 ⊕ p(a, c1, b)) ⊕ c2 · (1 ⊕ p(a, c3, b)) = (1 ⊕ p(c1, c2, c3)) · a ⊕ (1 ⊕ p(c1, c2, c3)) · b ⇔ (1 ⊕ c2 · p(a, c1, b)) ⊕ (1 ⊕ c2 · p(a, c3, b)) = (1 ⊕ p(c1, c2, c3) · a) ⊕ (1 ⊕ p(c1, c2, c3) · b) ⇔ 1 ⊕ (c2 · p(a, c1, b) ⊕ c2 · p(a, c3, b)) = 1 ⊕ (p(c1, c2, c3) · a ⊕ p(c1, c2, c3) · b) ⇔ p(p(a, c1, b), c2, p(a, c3, b)) = p(a, p(c1, c2, c3), b). The proof of (A8) is similar. To write the proof in a concise way, let's use the notation 1 ⊕ 1 = 1⁄2 and recall that (28) reads 1⁄2 · a = a · 1⁄2 for all a and (25) means 1⁄2 = 1⁄2. (c · 1⁄2 · a1 ⊕ c · 1⁄2 · b1) ⊕ (c · 1⁄2 · a2 ⊕ c · 1⁄2 · b2) = (1⁄2 · c · a1 ⊕ 1⁄2 · c · b1) ⊕ (1⁄2 · c · a2 ⊕ 1⁄2 · c · b2) ⇔ (c · 1⁄2 · a1 ⊕ c · 1⁄2 · a2) ⊕ (c · 1⁄2 · b1 ⊕ c · 1⁄2 · b2) = (1⁄2 · c · a1 ⊕ 1⁄2 · c · b1) ⊕ (1⁄2 · c · a2 ⊕ 1⁄2 · c · b2) ⇔ c · (1⁄2 · a1 ⊕ 1⁄2 · a2) ⊕ c · (1⁄2 · b1 ⊕ 1⁄2 · b2) = 1⁄2 · (c · a1 ⊕ c · b1) ⊕ 1⁄2 · (c · a2 ⊕ c · b2) ⇔ c · (1 ⊕ p(a1, 1⁄2, a2)) ⊕ c · (1 ⊕ p(b1, 1⁄2, b2)) = 1⁄2 · (1 ⊕ p(a1, c, b1)) ⊕ 1⁄2 · (1 ⊕ p(a2, c, b2)) ⇔ 1 ⊕ (c · p(a1, 1⁄2, a2) ⊕ c · p(b1, 1⁄2, b2)) = 1 ⊕ (1⁄2 · p(a1, c, b1) ⊕ 1⁄2 · p(a2, c, b2)) ⇔ p(p(a1, 1⁄2, a2), c, p(b1, 1⁄2, b2)) = p(p(a1, c, b1), 1⁄2, p(a2, c, b2)). (cid:3) An IMM algebra in which 1 ⊕ 1 is invertible is an IMM algebra. Moreover, x = (1 ⊕ 1)−1 · ((b · a) ⊕ (b · c)) is a solution to (37). Hence it gives rise to a mobi algebra. Corollary 6.3. Let (A, (),⊕,·, 1) be an IMM algebra (or an IMM* algebra). If the element 1 ⊕ 1 is invertible, then there exists a unique mobi algebra (A, p, 1, 1 ⊕ 1, 1) such that: p(a, b, c) = (1 ⊕ 1)−1 · ((b · a) ⊕ (b · c)). 17 (39) Proof. It is a consequence of Theorem 6.2 and property (27). (cid:3) The previous results show the connection between mobi algebras and IMM algebras, or IMM* algebras. The case when the monoid part (A,·, 1) of an IMM algebra is a commutative monoid has an interesting reflection on the axiom (A8). Instead of having it restricted to the element 1⁄2, p(p(a1, c, b1), 1⁄2, p(a2, c, b2)) = p(p(a1, 1⁄2, a2), c, p(b1, 1⁄2, b2)) it holds for an arbitrary element as shown in the following proposition. Proposition 6.4. Let (A, (),⊕,·, 1) be an IMM* algebra satisfying con- dition (i) of Theorem 6.2 and suppose that (A, p, 1, 1⊕ 1, 1) is its corre- sponding mobi algebra. The monoid (A,·, 1) is a commutative monoid if and only if p(p(a1, c, b1), d, p(a2, c, b2)) = p(p(a1, d, a2), c, p(b1, d, b2)) (40) for all a1, b1, a2, b2, c, d ∈ A. Proof. If (40) is a property of the mobi algebra, then: a · b = p(0, a, b) = p(p(0, b, 0), a, p(0, b, 1)) = p(p(0, a, 0), b, p(0, a, 1)) = p(0, b, a) = b · a. 18 J. P. FATELO AND N. MARTINS-FERREIRA Conversely and considering (38), if a · b = b · a, for all a, b ∈ A in the IMM, then: ((d · c · a1) ⊕ (d · c · b1)) ⊕ ((d · c · a2) ⊕ (d · c · b2)) = ((c · d · a1) ⊕ (c · d · b1)) ⊕ ((c · d · a2) ⊕ (c · d · b2)) =⇒ ((d · c · a1) ⊕ (d · c · b1)) ⊕ ((d · c · a2) ⊕ (d · c · b2)) = ((c · d · a1) ⊕ (c · d · a2)) ⊕ ((c · d · b1) ⊕ (c · d · b2)) =⇒ (d · ((c · a1) ⊕ (c · b1))) ⊕ (d · ((c · a2) ⊕ (c · b2))) = (c · ((d · a1) ⊕ (d · a2))) ⊕ (c · ((d · b1) ⊕ (d · b2))) =⇒ (d · (1 ⊕ p(a1, c, b1))) ⊕ (d · (1 ⊕ p(a2, c, b2))) = (c · (1 ⊕ p(a1, d, a2))) ⊕ (c · (1 ⊕ p(b1, d, b2))) =⇒ (1 ⊕ (d · p(a1, c, b1))) ⊕ (1 ⊕ (d · p(a2, c, b2))) = (1 ⊕ (c · p(a1, d, a2))) ⊕ (1 ⊕ (c · p(b1, d, b2))) =⇒ 1 ⊕ ((d · p(a1, c, b1)) ⊕ (d · p(a2, c, b2))) = 1 ⊕ ((c · p(a1, d, a2)) ⊕ (c · p(b1, d, b2))) =⇒ 1 ⊕ (1 ⊕ p(p(a1, c, b1), d, p(a2, c, b2))) = 1 ⊕ (1 ⊕ p(p(a1, d, a2), c, p(b1, d, b2))) =⇒ p(p(a1, c, b1), d, p(a2, c, b2)) = p(p(a1, d, a2), c, p(b1, d, b2)). (cid:3) 7. Mobi algebras and unitary rings with one half We have seen the passage from mobi algebras to IMM() algebras and back (Proposition 4.1 and Theorem 6.2), as well as the passage from IMM algebras to unitary rings and back (Theorem 5.2 and Theorem 5.3). Here we make explicit the fact that there is a straightforward connection between mobi algebras in which 1⁄2 is invertible and unitary rings in which 2 is invertible. Theorem 7.1. Let (A, p, 0, 1⁄2, 1) be a mobi algebra on the set A and let 2 ∈ A be a distinguished element on that set. The structure (A, +,·, 0, 1), with a·b = p(0, a, b) and a+b = 2·p(a, 1⁄2, b), is a unitary ring if and only if 2 is the inverse of 1⁄2. Proof. If (A, +,·, 0, 1) is a unitary ring, with a + b = 2· p(a, 1⁄2, b), then, in particular, 0 + 1 = 2 · p(0, 1⁄2, 1) which implies 1 = 2 · 1⁄2 proving that 2 is the inverse of 1⁄2. Conversely, we begin by using Proposition 4.1 to obtain, from the mobi, an IMM structure (A, (),⊕,·, 1) in which 1⁄2 = 1⊕1, a· b = p(0, a, b) and a⊕b = p(a, 1⁄2, b). Within this structure, if 2 is the inverse of 1⁄2, we have, by Theorem 5.2, that a⊕ b = 1⁄2· (a+ b) which is equivalent to a + b = 2 · p(a, 1⁄2, b). Theorem 7.2. Let (A, +,·, 0, 1) be a unitary ring with a distinguished element 1⁄2 ∈ A. (cid:3) 19 The structure (A, p, 0, 1⁄2, 1), with p(a, b, c) = a + bc − ba, is a mobi algebra if and only if 1⁄2 is the inverse of 1 + 1. Proof. If (A, p, 0, 1⁄2, 1) is a mobi algebra, with p(a, b, c) = a + bc − ba, then axiom (A1), p(1, 1⁄2, 0) = 1⁄2, reads 1− 1⁄2 = 1⁄2, i.e. (1 + 1)· 1⁄2 = 1. On the other hand, if 1⁄2 is the inverse of 1 + 1, Theorem 5.3 gives us an IMM algebra (A, (),⊕,·, 1) in which b = 1−b and a+b = (1+1)·(a⊕b). This implies, in particular, that 1 = (1 + 1) · (1 ⊕ 1). Hence, through Corollary 6.3, we conclude that (A, p, 0, 1⁄2, 1) is a mobi algebra, with p(a, b, c) = (1 ⊕ 1)−1 · (((1 − b) · a) ⊕ (b · c)) = (a − ba) + bc . (cid:3) Theorem 7.3. There is a bijective correspondence between unitary rings containing the element 2−1 and mobi algebras containing 2. Proof. Let (A, +,·, 0, 1) be a unitary ring such that 2−1 ∈ A, with 2 = 1 + 1. Then, by Theorem 7.2, the system (A, p, 0, 2−1, 1) where p(a, b, c) = a + bc − ba is a mobi algebra. This mobi algebra contains 2 (the inverse of 2−1) and, consequently, by Theorem 7.1, it determines a unitary ring (A, +′,·′, 0, 1). This ring is identical to the initial ring (A, +,·, 0, 1) because: a ·′ b = p(0, a, b) = 0 + a · b − a · 0 = a · b 1 a +′ b = 2 ·′ p(a, 2 · a) = a + b. , b) = 2 · (a + 1 2 1 2 · b − Conversely, let (A, p, 0, 1⁄2, 1) be a mobi algebra such that p(0, 1⁄2, 2) = 1 with 2 ∈ A. Then, by Theorem 7.1, (A, +,·, 0, 1) with a · b = p(0, a, b) and a+b = 2·p(a, 1⁄2, b) is a unitary ring. This ring contains 1⁄2 and, con- sequently, by Theorem 7.2, it determines a mobi algebra (A, p′, 0, 1⁄2, 1). This mobi algebra is identical to the initial one. Indeed, by definition of p′, we have p′(a, b, c) = (1 − b) · a + b · c = 2 · p((1 − b) · a, 1⁄2, b · c). Then, as shown in the proof of Proposition 4.1, we get p(b, 1⁄2, b) = 1⁄2. When 2 exists, this equality may be written as b + b = 1, i.e., 1− b = b. Therefore p′(a, b, c) = 2 · p(b · a, 1⁄2, b · c) which, using property (23), implies that p′(a, b, c) = p(a, b, c). (cid:3) The finite case is of particular interest because every finite mobi is such that its element 1⁄2 is invertible, and hence it is uniquely deter- mined by a unitary ring structure in which 2 is invertible (see Example 15 in Section 3). 8. conclusion We conclude with a schematic diagram relating the algebraic struc- tures considered here and the results that relate them. We use an 20 J. P. FATELO AND N. MARTINS-FERREIRA arrow labelled with the number of the Theorem, Proposition or Corol- lary where the result is proved on the direction indicated by the arrow. For example, the arrow labelled P.4.1, with source Mobi and target IMM, simply means that the Proposition 4.1 establishes an effective passage from the algebraic structure of a mobi algebra to the algebraic structure of an IMM algebra. Moreover, we use the name IMM** to designate an IMM* algebra in which condition (i) of Theorem 6.2 holds. This structure, by Corollary 6.3, is clearly in between IMM algebras, in which 1⁄2 is invertible (that we are denoting by IMM2), and IMM* algebras. Following the same line, we denote by Ring1⁄2 the rings in which 2 is invertible and by Mobi2 the mobi algebras in which 1⁄2 is invertible. Ring1⁄2 T.7.1 :ttttttttt zttttttttt T.7.2 T.5.3 %❑❑❑❑❑❑❑❑❑ e❑❑❑❑❑❑❑❑❑ T.5.2 (41) Mobi2 Mobi P.4.1 IMM C.6.3 T.6.2 IMM2 IMM** IMM* We observe that in the proof of property P.4.1, the axiom (A6) is not used which suggests that we could also consider a mobi without this axiom. Representing such a structure by Mobi†, we get the following additional diagram. Mobi / Mobi† (42) $❍❍❍❍❍❍❍❍❍ z✉✉✉✉✉✉✉✉✉ P.4.1 IMM Furthermore, the inclusion Mobi ⊂ Mobi† is strict. Indeed, the example IMM 2 (see Appendix) is an IMM algebra that is obtained from a Mobi† that is not a Mobi (axiom (A6) is not satisfied). Moreover, there are IMM algebras which are not obtained from Mobi† as it is shown by example IMM 3 in the Appendix. Our last comment is that the connection between IMM algebras and rings can be lifted to the more general case of semi-rings. Indeed, if in the definition of IMM algebra we remove the unary operation () while keeping the existence of the element 1 such that 1 · x = 1 then, in Theorem 5.2, we can replace rings by semi-rings. z % : / / e o o / / o o o o $ / z Appendix 21 In this appendix, we present some examples of finite IMM algebras with 5 elements. Let A = {α, 0, 1⁄2, 1, β} be a set with 5 elements. In the three examples bellow, the unary operation () is defined by α = β, 0 = 1, 1 = 0, 1⁄2 = 1⁄2, β = α. IMM 1: The system (A, (),⊕,·, 1), with ⊕ and · defined as follows, is an IMM algebra. 1⁄2 1 ⊕ α 0 1 0 α α β 0 0 α 1⁄2 β 1 α 1⁄2 β 1⁄2 0 1 β 1⁄2 β 1⁄2 1 0 1 α 0 α β 1⁄2 β 1 1 · α 0 1⁄2 β 0 β α 1⁄2 α 1 0 0 0 0 0 0 1 1⁄2 β 0 α 1⁄2 1 1 α 0 1⁄2 β β 1⁄2 0 1 β α Defining p(a, b, c) = β · ((b · a) ⊕ (b · c)), it can be checked that (A, p, 0, 1⁄2, 1) is a mobi algebra. IMM 2: The system (A, (),⊕,·, 1), with ⊕ and · defined as follows, is an IMM algebra. ⊕ α 0 α α 1⁄2 0 0 1⁄2 1⁄2 1 α 1⁄2 1⁄2 β 1⁄2 β 1⁄2 1⁄2 1⁄2 1 1⁄2 β 1⁄2 α 1⁄2 1⁄2 β 1⁄2 1⁄2 1⁄2 1 1⁄2 1⁄2 β 1⁄2 0 · α 0 1⁄2 1 β α α 0 1⁄2 α β 0 0 0 0 0 1⁄2 0 1⁄2 1⁄2 β 1⁄2 1 α 0 1⁄2 1 β β 0 β β 0 β It is obvious that the operation ⊕ is not cancellative. Nevethe- less, the equation 1⁄2 · p = (b · a) ⊕ (b · c) can be solved for all a, b, c ∈ A. It can be shown that there are solutions p of that equation for which (A, p, 0, 1⁄2, 1) is a Mobi† (a mobi algebra without axiom (A6)). IMM 3: The system (A, (),⊕,·, 1), with ⊕ and · defined as follows, is an IMM algebra. 1⁄2 ⊕ α 0 α α 1⁄2 0 0 1⁄2 1⁄2 1 1⁄2 β 1⁄2 1⁄2 1⁄2 1⁄2 1⁄2 1⁄2 1⁄2 1⁄2 1⁄2 1⁄2 1 1⁄2 1⁄2 β 1⁄2 1⁄2 1⁄2 1⁄2 1 1⁄2 1⁄2 β 0 1 · α 0 1⁄2 β α β 0 1⁄2 α 1 0 0 0 0 1⁄2 0 1⁄2 1⁄2 1⁄2 1 α 0 1⁄2 β 0 1⁄2 β α β 0 1⁄2 1 1 This IMM cannot be induced by a mobi algebra because the equation 1⁄2 · p = (b · a) ⊕ (b · c) doesn't have a solution p for all a, b, c ∈ A, in contradiction with (23). Indeed, for example, the equation 1⁄2· p = (β · β)⊕ (β · α) is equivalent to 1⁄2· p = 1 which does not have a solution. 22 J. P. FATELO AND N. MARTINS-FERREIRA References [1] M. Escard´o and A. Simpson, A universal characterisation of the closed Eu- clidean interval, in: Proceedings of 16th Annual IEEE Symposium on Logic in Computer Science (2001) 115 -- 125 [2] J. P. Fatelo and N. Martins-Ferreira, Internal monoids and groups in the cat- egory of commutative cancellative medial magmas, Portugaliae Mathematica, Vol. 73, Fasc. 3 (2016) 219-245 . [3] N. Martins-Ferreira and J. P. Fatelo, On Finite Mobi Algebras, CDRSP- IPLeiria Technical Report, GTLab(Mobi-2) 59 (2015) 1-30. [4] N. Martins-Ferreira, Weakly Mal'cev categories, Theory Appl. Categ. 21 (6) (2008) 91 -- 117. [5] N. Martins-Ferreira, New wide classes of weakly Mal'tsev categories, Appl. Categ. Struct. 23 (5) (2015) 741 -- 751. [6] J. Jezek and T. Kepka, Medial groupoids, Rozpravy CSAV, Rada mat. a prir. ved 93-2, Academia Praha (1983). IPL E-mail address: [email protected]
1008.0286	2	1008	2011-06-01T12:32:15	A topological approach to leading monomial ideals	[ "math.RA", "math.AC", "math.AG" ]	We introduce a very natural topology on the set of total orderings of monomials of any algebra having a countable basis over a field. This topological space and some notable subspaces are compact. This topological framework allows us to deduce some finiteness results about leading monomial ideals of any fixed ideal, namely: (1) the number of minimal leading monomial ideals with respect to total orderings is finite; (2) the number of leading monomial ideals with respect to degree orderings is finite; (3) the number of leading monomial ideals with respect to admissible orderings is finite under some multiplicativity assumptions on the considered algebra. Finally we are able to infer the existence of universal Groebner bases from the topological properties of degree and admissible orderings in a class of algebras that includes at least the algebras of solvable type. These existence results turn out to be independent from the finiteness results mentioned above, in contrast to the typical situation that occurs with "classical" more combinatorial proofs.	math.RA	math	A TOPOLOGICAL APPROCH TO LEADING MONOMIAL IDEALS ROBERTO BOLDINI Abstract. We define a very natural topology on the set of total orderings of monomials of any algebra having a countable basis over a field. This topological space and some notable subspaces are compact. This topological framework allows us to deduce some finiteness results about leading monomial ideals of any fixed ideal, namely: (1) the number of minimal leading monomial ideals with respect to total orderings is finite; (2) the number of leading monomial ideals with respect to degree orderings is finite; (3) the number of leading monomial ideals with respect to admissible orderings is finite under some multiplicativity assumptions on the considered algebra. Finally we are able to infer the existence of universal Grobner bases from the topological properties of degree and admissible orderings in a class of algebras that includes at least the algebras of solvable type. These existence results turn out to be independent from the finiteness results mentioned above, in contrast to the typical situation that occurs with "classical" more combinatorial proofs. Introduction In this paper we deal with leading monomial ideals of ideals in some classes of algebras over a field with respect to several sorts of total orderings on their bases, whose elements we call monomials. We introduce a topology on the set of all total orderings of monomials. It turns out that the so obtained topological space is compact and, in the case of countable bases, this topology is precisely the one induced by a very natural metric on such total orderings. In virtue of this fact, after showing that certain kinds of total orderings build closed subsets and hence are compact subspaces, and by considering certain quotient spaces (with respect to an appropriate equivalence relation) which turn out to be discrete, we are able to prove some finiteness results about leading monomial ideals of such algebras, namely: if A is an algebra over a field K such that A has a countable basis as a free K-module, and if H is any subset of A, then: (1) the number of minimal leading monomial ideals of H with respect to total orderings of monomials of A is finite, see Theorem 4.6, Date: 19th May 2010. 2010 Mathematics Subject Classification. Primary 13C05 13C13 16D25. I am very grateful to Prof. Dr. Markus Brodmann of the University of Zurich, Switzerland, for his precious comments and his gracious encouragement during the writing of this paper. My acknowledgements also to Prof. Dr. Matthias Aschenbrenner of the University of California, Los Angeles, for his kind communication of the proof of Theorem 1.5. 1 2 ROBERTO BOLDINI (2) the number of leading monomial ideals of H with respect to degree orderings of monomials of A is finite, see Theorem 4.9, (3) the number of leading monomial ideals of H with respect to admissible orderings of monomials of A is finite whenever H is a (left, right, or two-sided) ideal of A and A satisfies two multiplicativity conditions, namely, A is a domain and A behaves multiplicatively on taking leading monomials with respect to admissible orderings, see Theorem 8.4. Carrying on with this topological approach, generalizing [8] and [9], we prove that every (left, right, or two-sided) ideal J of A admits a T-universal Grobner basis U , that is, U is a Grobner basis of J with respect to each (cid:22) ∈ T, where T is a closed subset of the set of all degree orderings or of all admissible orderings of monomials of A. Statements about the existence of universal Grobner bases, for instance in the context of commutative polynomial rings over a field, are usually infered from a finiteness result similar to (3) and from the availability of a division algorithm by which one can construct reduced Grobner bases, a selected finite union of which is then a universal Grobner basis, see [10]. We shall see that, actually, the topological properties of the considered spaces of total orderings of monomials, above all compactness, are sufficient to prove the existence of universal Grobner bases, even in the more general context treated here. The algebras on which these results can be applied comprehend at least the algeb- ras of solvable type and the enveloping algebras of finite-dimensional Lie algebras. Some of our results, such as (3) and the existence of universal Grobner bases in the just mentioned classes of algebras, are not new, see [11] for instance. New are, in our knowledge, (1) and (2). Through (1) one gains a new insight why there exist only finitely many leading monomial ideals of a given ideal with respect to admissible orderings (Theorem 8.4). Indeed, there exist at most finitely many minimal such ideals at all with respect to any closed subset of total orderings (Theorem 4.6), and the admissible orderings form a closed subset (Proposition 6.8) and force leading monomial ideals to be minimal (Corollary 5.3 of the Macaulay Basis Theorem 5.2). Through (2) one gets a deeper intuition why one finds only finitely many leading monomial ideals of a given ideal with respect to degree-compatible orderings (Re- mark 7.6). Indeed, degree preservation on taking leading monomials alone without the compatibility axiom already implies this behaviour (Theorem 4.9). Our intention has been also to push the topological methods introduced in [8] and [9] to the case of some further orderings than only admissible ones and of some noncommutative algebras. Beside the mentioned finiteness results, we have obtained A TOPOLOGICAL APPROCH TO LEADING MONOMIAL IDEALS 3 a sort of topological framework for orderings of monomials, which we were able to successfully apply to the study of leading monomial ideals and universal Grobner bases. Furthermore, some relations among different kinds of orderings was put to evidence. Beside those already mentioned, two further topological phenomena came to light: (4) there exist "few" degree-compatible orderings, that is, precisely, the degree- compatible orderings are nowhere dense among the degree orderings, clearly ex- cept for the case of univariate polynomials, see Proposition 7.3 and Remark 7.4, (5) there is a relation between topological density and the possibility to find a universal Grobner basis, see Remark 10.6, Lemma 10.7 and Example 10.8. We conclude by saying that remarkable benefits of the topological approach are, in our opinion, the high level of generality and the simplicity of the argumentations. A drawback, at least at first sight, is the nonconstructivity of the proofs. But who knows? See 10.6. R´esum´e In [9], for semigroups S, Sikora introduced a natural topology U(S) on the set TO(S) of the total orderings on S and proved that TO(S) is compact with respect to U(S). This can be done actually for any set S. We start with a polynomial ring K[X] = K[X1, . . . , Xt] over a field K, where 0} of t ∈ N, and with several sorts of total orderings on the set M = {X ν ν ∈ Nt the monomials of K[X], namely, we consider the following subsets of TO(M ): (1) the set WO(M ) of the total well-orderings on M ; (2) the set FO1(M ) = {≤ ∈ TO(M ) m ∈ M ⇒ 1 ≤ m} of the 1-founded orderings on M ; (3) the set CO(M ) = {≤ ∈ TO(M ) X υ ≤ X ν ⇒ X υ+γ ≤ X ν+γ} of the compat- ible orderings, or semigroup orderings, on M ; (4) the set DO(M ) = {≤ ∈ TO(M ) p ∈ K[X] ⇒ deg(p) = deg(LM≤(p))} of the degree orderings on M ; (5) the set AO(M ) = FO1(M ) ∩ CO(M ) of the admissible orderings, or monoid orderings, on M ; (6) the set DCO(M ) = DO(M ) ∩ CO(M ) of the degree-compatible orderings on M . Then we have the following results: (1) FO1(M ) is closed in TO(M ); (2) CO(M ) is closed in TO(M ); (3) DO(M ) is closed in TO(M ) and DO(M ) ⊆ WO(M ) ∩ FO1(M ); (4) AO(M ) is closed in TO(M ) and AO(M ) = WO(M ) ∩ CO(M ); (5) DCO(M ) is closed in TO(M ); 4 ROBERTO BOLDINI (6) DCO(M ) is nowhere dense in DO(M ) if t > 1, otherwise DCO(M ) = DO(M ). The Venn diagram in Figure 1 sketches the situation. DO(M ) DCO(M ) WO(M ) FO1(M ) AO(M ) CO(M ) TO(M ) Figure 1. Subspaces of total orderings of monomials After these preliminaries, given any S ⊆ TO(M ) and any E ⊆ K[X], we con- sider the set „‡S(E) = {LM≤(E) ≤ ∈ S} of the leading monomial ideals LM≤(E) of E with respect to the total orderings ≤ ∈ S and the set i‡S(E) of the min- imal elements of „‡S(E) with respect to the inclusion relation ⊆, and show that i‡S(E) is finite if S is closed in TO(M ). The proof goes as follows. The set minE(S) of the elements ≤ ∈ S such that LM≤(E) is ⊆-minimal in „‡S(E) is closed in S, and hence minE(S) is compact under our hypothesis on S. Thus the quotient space minE(S) / ∼E of minE(S), where ≤ ∼E ≤′ if and only if LM≤(E) = LM≤′(E), is compact. Since minE(S) / ∼E is also discrete, it follows that minE(S) / ∼E is finite. Of course, there exists a canonical bijection between minE(S) / ∼E and i‡S(E). Now we turn our attention to degree orderings. DO(M ) and DO(M ) / ∼E are compact. We show by means of Hilbert functions that DO(M ) / ∼E is discrete and hence finite. Thus „‡DO(M)(E) is finite, that is, there exist at most finitely many leading monomial ideals of E from degree orderings. The idea of applying Hilbert functions in such "topological contexts" was already used in a similar manner by Schwartz in [8] in the case of admissible orderings. A TOPOLOGICAL APPROCH TO LEADING MONOMIAL IDEALS 5 When considering closed subsets S of AO(M ), we obtain a similar and well- known finiteness result. Indeed, in this case, if I is an ideal of K[X], the Macaulay Basis Theorem holds and comes to our aid as it implies that „‡S(I) = i‡S(I), which we already know to be finite. Next let Φ be a K-module isomorphism of V in K[X] and consider the K-basis N = Φ−1(M ) of V . Then Φ induces a homeomorphism φ of TO(N ) in TO(M ). Now, given a total ordering (cid:22) on N , we may speak of the (cid:22)-leading component lm(cid:22)(v) ∈ N in the unique representation v = Pn∈N cnn with cn ∈ K r {0} of any element v ∈ V as a K-linear combination over N . Further, given H ⊆ V , we consider the ideal LM(cid:22)(H) = hΦ(lm(cid:22)(h)) h ∈ Hi = hLMφ((cid:22))(Φ(h)) h ∈ Hi of K[X]. For all H ⊆ V , E ⊆ K[X], (cid:22) ∈ TO(N ), ≤ ∈ TO(M ), T ⊆ TO(N ), S ⊆ TO(M ) we have: (1) LM(cid:22)(H) = LMφ((cid:22))(Φ(H)) and LM≤(E) = LMφ−1(≤)(Φ−1(E)); (2) „‡T(H) = „‡φ(T)(Φ(H)) and „‡S(E) = „‡φ−1(S)(Φ−1(E)); (3) i‡T(H) = i‡φ(T)(Φ(H)) and i‡S(E) = i‡φ−1(S)(Φ−1(E)). Thus what we have said above about K[X] easily translates to V . With one excep- tion: assuming that T is closed in AO(N ), the equality „‡T(H) = i‡T(H) holds so far only under the hypothesis that H = Φ−1(I) for some ideal I of K[X]. Therefore, when considering the set AO(N ) = φ−1(AO(M )) of the admissible orderings on N , we replace the K-module V by an associative but not necessar- ily commutative K-algebra A that is a domain and is isomorphic to K[X] as a K-module. Assuming similar multiplicativity properties of A on taking leading monomials as in the case of K[X], we prove a generalized version of the Macaulay Basis Theorem, which then implies the equality „‡T(J) = i‡T(J) for each closed T ⊆ AO(N ) and each (left, right, two-sided) ideal J ⊆ A. Finally, for a K-algebra A isomorphic to K[X] as a K-module, following this topological approach and applying the results obtained so far, we show that every (left, right, two-sided) ideal of A admits a T-universal Grobner basis, where T is any closed subset of DO(N ). To prove a similar result for closed subsets T of AO(N ), we have to require that A is a domain and is multiplicative on taking leading monomials over T. As mentioned before, our proofs of theorems about universal Grobner bases do not rely on the finiteness of the total number of leading monomial ideals. Indeed, the statements about universal Grobner bases as well as the finiteness results both descend directly from some of the topological properties of total orderings and, partly, from the generalized Macaulay Basis Theorem. 6 ROBERTO BOLDINI General remark In this paper all the statements involving ideals of noncommutative rings are proved only for left ideals. These statements translate word by word to right and two-sided ideals, too. 1. Topological spaces of total orderings on sets In this section, let S be a set. Definition 1.1. A total ordering on S is a binary relation (cid:22) on S such that it holds antisymmetry: a (cid:22) b ∧ b (cid:22) a ⇒ a = b, transitivity: a (cid:22) b ∧ b (cid:22) c ⇒ a (cid:22) c, totality: a (cid:22) b ∨ b (cid:22) a, for all a, b, c ∈ S. Totality implies reflexivity: a (cid:22) a for all a ∈ S. The nonempty set of all total orderings on S is denoted TO(S). Given any ordered pair (a, b) ∈ S × S, let U(a,b) be the set of all total orderings (cid:22) on S for which a (cid:22) b. Let U(S) be the coarsest topology of S for which all the sets U(a,b) are open. This is the topology for which {U(a,b) (a, b) ∈ S × S} is a subbasis, that is, the open sets in U(S) are precisely the unions of finite intersections of sets of the form U(a,b). Observe that U(a,a) = TO(S) and that U(a,b) = TO(S) r U(b,a) if a 6= b, so that the sets U(a,b) are also closed. Let S be any filtration of S, that is, S = (Si)i∈N0 is a family of subsets Si of S such that (a) S0 = ∅, (b) Si ⊆ Si+1 for all i ∈ N0, (c) S = Si∈N0 Si. Let us define the function dS : TO(S) × TO(S) → R by the rule dS((cid:22)′, (cid:22)′′) = 2−r where r = sup {i ∈ N0 (cid:22)′↾Si = (cid:22)′′↾Si}. Here ↾ denotes restriction. First of all, we have {0} ⊆ Im(dS) ⊆ [0, 1]. Because S is exhaustive by (c), it holds dS((cid:22)′, (cid:22)′′) = 0 if and only if (cid:22)′ = (cid:22)′′. Further, dS((cid:22)′, (cid:22)′′) = dS((cid:22)′′, (cid:22)′). Finally, dS((cid:22)′, (cid:22)′′′) ≤ dS((cid:22)′, (cid:22)′′) + dS((cid:22)′′, (cid:22)′′′), since dS((cid:22)′, (cid:22)′′′) ≤ max {dS((cid:22)′, (cid:22)′′), dS((cid:22)′′, (cid:22)′′′)}. Thus dS is a metric on TO(S), dependent on the choice of the filtration S of S. Theorem 1.2. Assume that there exists a filtration S = (Si)i∈N0 of S such that each of the sets Si is finite. Let N (S) be the topology of S induced by the metric dS, that is more precisely, N ∈ N (S) if and only if N is a union of finite intersections of sets of the form Nr((cid:22)) = {(cid:22)′ ∈ TO(S) dS((cid:22), (cid:22)′) < 2−r} where r ∈ N0 and (cid:22) ∈ TO(S). Then it holds N (S) = U(S), in particular the topology N (S) is independent of the choice of S, and the topology U(S) is Hausdorff. Proof. Let r ∈ N0 and (cid:22) ∈ TO(S). We claim that Nr((cid:22)) ∈ U(S). Indeed, let U = T(a,b) U(a,b), where the intersection is taken over all ordered pairs (a, b) in Sr+1 × Sr+1 with a (cid:22) b. Then (cid:22) ∈ U ∈ U(S). Hence (cid:22)′ ∈ Nr((cid:22)) if and only if (cid:22)′↾Sr+1 = (cid:22)↾Sr+1, and this is the case if and only if it holds a (cid:22)′ b ⇔ a (cid:22) b for all (a, b) ∈ Sr+1 × Sr+1, which is true if and only if (cid:22)′ ∈ U. Thus Nr((cid:22)) = U, and this shows that N (S) ⊆ U(S). A TOPOLOGICAL APPROCH TO LEADING MONOMIAL IDEALS 7 On the other hand, let (a, b) ∈ S × S be any ordered pair. We claim that the set U(a,b) is open with respect to the metric dS. Let (cid:22) ∈ U(a,b), so that a (cid:22) b. We find r ∈ N0 such that (a, b) ∈ Sr+1 × Sr+1. If (cid:22)′ ∈ Nr((cid:22)), then (cid:22)′↾Sr+1 = (cid:22)↾Sr+1, in particular a (cid:22)′ b, so that (cid:22)′ ∈ U(a,b), thus Nr((cid:22)) ⊆ U(a,b). Hence U(a,b) is open with respect to N (S), and we conclude that U(S) ⊆ N (S). (cid:3) Convention 1.3. Henceforth, unless otherwise stated, whenever we refer to to- pological properties of TO(S), we always intend that TO(S) is provided with the topology U(S). Subsets of TO(S) are tacitly furnished with their relative topology with respect to U(S). Quotient sets of TO(S) by equivalence relations are equipped with their quotient topology with respect to U(S). Definition 1.4. A filter over a set X is a subset F of the power set P(X) of X that enjoys the properties (a) X ∈ F , (b) ∅ /∈ F , (c) A ⊆ B ⊆ X ∧ A ∈ F ⇒ B ∈ F , (d) A ∈ F ∧ B ∈ F ⇒ A ∩ B ∈ F . An ultrafilter over X is a filter L over X that fulfills the further property (e) A ⊆ X ⇒ A ∈ L ∨ X r A ∈ L. The disjunction in (e) is exclusive by (d) and (b). Equivalently, an ultrafilter over X is a maximal filter over X with respect to inclusion. Theorem 1.5. TO(S) is compact. Proof. Suppose by contradiction that TO(S) is not compact. Then we find an infinite index set I and families (ai)i∈I and (bi)i∈I of elements ai, bi ∈ S such that (U(ai,bi))i∈I is a covering of TO(S) which admits no finite subcovering. Thus for each finite subset s ⊆ I there exists (cid:22)s ∈ TO(S) such that (cid:22)s /∈ Si∈s U(ai,bi), that is, for all i ∈ s it holds ai ≻s bi. Let I ∗ be the set of all nonempty finite subsets of I. For each s ∈ I ∗ let us define s∗ = {t ∈ I ∗ s ⊆ t}. Since s ∈ s∗ for all s ∈ I ∗ and s∗ 2 = (s1 ∪ s2)∗ for all s1, s2 ∈ I ∗, the set S = {s∗ s ∈ I ∗} has the finite intersection property, that is to say, any finite intersection of elements of S is nonempty. Therefore F = {Y ∈ P(I ∗) ∃ n ∈ N ∃ Z1, . . . , Zn ∈ S : Z1 ∩ . . . ∩ Z1 ⊆ Y } is a filter over I ∗ that extends S. Hence, by the Ultrafilter Lemma, which descends from Zorn's Lemma, there exists an ultrafilter L over I ∗ that extends F , so that s∗ ∈ L for all s ∈ I ∗. 1 ∩ s∗ We fix a family ((cid:22)s)s∈I ∗ of total ordering (cid:22)s on S as above and define a binary relation (cid:22) on S by x (cid:22) y ⇔ {s ∈ I ∗ x (cid:22)s y} ∈ L. By axioms (d) and (b) of 1.4, (cid:22) is antisymmetric. By axioms (d) and (c) of 1.4, (cid:22) is transitive. By axioms (e) and (c) of 1.4, (cid:22) is total. So (cid:22) ∈ TO(S). On the other hand, by our choice of the orderings (cid:22)s, it holds ai ≻ bi for all i ∈ I, thus (cid:22) /∈ Si∈I U(ai,bi) = TO(S), a contradiction. (cid:3) 8 ROBERTO BOLDINI Definition 1.6. For each a ∈ S let FOa(S) = {(cid:22) ∈ TO(S) ∀ b ∈ S : a (cid:22) b}, the set of all a-founded orderings on S. Corollary 1.7. For each a ∈ S the set FOa(S) is closed in TO(S), and hence FOa(S) is a compact subspace of TO(S). Proof. It holds FOa(S) = Tb∈S U(a,b), thus FOa(S) is closed in TO(S) as each U(a,b) is closed in TO(S) as observed in 1.1. If S is countable, then TO(S) is compact by 1.5, and hence, as a closed subset of a compact set, FOa(S) equipped with its relative topology is compact. (cid:3) 2. Leading monomial ideals from total orderings Let t ∈ N, let K be a field, and let K[X] denote the commutative polynomial ring K[X1, . . . , Xt]. Reminder & Definition 2.1. The countable set M = {X ν ν ∈ Nt 0} of the monomials of K[X] is a basis of the K-module K[X], often referred to as the canonical K-basis of K[X]. We fix once for all this K-basis M of K[X]. Thus each p ∈ K[X] can be written in canonical form as Pν∈supp(p) ανX ν for 0 such that αν ∈ K r {0} for all a uniquely determined finite subset supp(p) of Nt ν ∈ supp(p). Notice that supp(p) = ∅ if and only if p = 0. For each p ∈ K[X] let us define the subset Supp(p) = {X ν ν ∈ supp(p)} of M , which we call the support of p. Clearly, Supp(p) = ∅ if and only if p = 0. We also put Supp(E) = Se∈E Supp(e) for each subset E of K[X]. For each p ∈ K[X]r{0} and each ≤ ∈ TO(M ) we denote by LM≤(p) the uniquely determined maximal element of Supp(p) with respect to ≤ and call LM≤(p) the leading monomial of p with respect to ≤. In this situation, there exists a unique α ∈ K r {0} such that either p − α LM≤(p) = 0 or LM≤(p − α LM≤(p)) < LM≤(p). Such element α is denoted LC≤(p) and called the leading coefficient of p with respect to ≤. For each E ⊆ K[X] and each ≤ ∈ TO(M ) we denote by LM≤(E) the monomial ideal hLM≤(e) e ∈ E r {0}i of K[X], and we call LM≤(E) the leading monomial ideal of E with respect to ≤. Finally, let „‡S(E) = {LM≤(E) ≤ ∈ S}, for E ⊆ K[X] and S ⊆ TO(M ), be the set of all leading monomial ideals of E from S. Remark 2.2. We shall, almost always tacitly, make use of the following well-known results, see [3, II.4.2 & II.4.4]. Let N ⊆ Nt 0. Then a monomial X υ of K[X] lies in the ideal hX ν ν ∈ N i of K[X] if and only if there exists γ ∈ N such that X γ divides X υ. A TOPOLOGICAL APPROCH TO LEADING MONOMIAL IDEALS 9 From this it follows that two monomials ideals are equal if and only if they contain the same monomials. Remark 2.3. If p ∈ K[X] and ≤, ≤′ ∈ TO(M ) are such that ≤ and ≤′ agree on Supp(p), then clearly LM≤(p) = LM≤′ (p). Hence, if ≤, ≤′ ∈ TO(M ) and F ⊆ K[X] are such that ≤ and ≤′ agree on Supp(F ), then LM≤(F ) = hLM≤(f ) f ∈ F i = hLM≤′(f ) f ∈ F i = LM≤′(F ). In this situation, if in addition we have F ⊆ E ⊆ K[X] and LM≤(F ) = LM≤(E), then clearly LM≤(E) ⊆ LM≤′(E). Definition 2.4. Let E ⊆ K[X] and let S ⊆ TO(M ). We say that ≤′ ∈ S is a minimalizer of E in S if the condition LM≤(E) ⊆ LM≤′(E) already implies LM≤(E) = LM≤′ (E) for all ≤ ∈ S, that is, if LM≤′(E) is a minimal element of „‡S(E) with respect to ⊆. We denote the set of all minimalizers of E in S by minE(S). We write i‡S(E) for the set „‡minE (S)(E) = {LM≤(E) ≤ ∈ minE(S)} of all minimal leading monomial ideals of E from S. Lemma 2.5. Let E ⊆ K[X] and S ⊆ TO(M ). Then minE(S) is a closed subset of S. Hence, if S is closed in TO(M ), then minE(S) is compact. Proof. We may choose a filtration (Si)i∈N0 of M consisting of finite subsets Si of S. Let ≤ ∈ S be any accumulation point of minE(S). Thus for each r ∈ N0 there exists ≤r ∈ minE(S) ∩ Nr(≤) r {≤}. Since K[X] is noetherian, there exists a finite set F ⊆ E such that LM≤(E) = LM≤(F ). We can find r ∈ N0 such that Supp(F ) ⊆ Sr+1. We fix then ≤r ∈ minE(S) ∩ Nr(≤) r {≤}. Thus ≤ and ≤r agree on Sr+1 and in particular on Supp(F ). From 2.3 it follows LM≤(E) ⊆ LM≤r (E). As ≤ ∈ S and ≤r ∈ minE(S), it follows LM≤(E) = LM≤r (E). Hence LM≤(E) is a minimal element of „‡S(E) with respect to ⊆, that is, ≤ ∈ minE(S). Therefore minE(S) contains all its accumulation points in S, and hence minE(S) is closed in S. The statement about compactness follows now from 1.5. (cid:3) Definition 2.6. Let E ⊆ K[X] and S ⊆ TO(M ). We define an equivalence relation ∼E on minE(S) by ≤ ∼E ≤′ ⇔ LM≤(E) = LM≤′(E). We also provide the set minE(S) / ∼E of the equivalence classes of minE(S) with respect to ∼E with its quotient topology. Remark 2.7. Let E ⊆ K[X] and S ⊆ TO(M ). By 2.5, minE(S) / ∼E is compact whenever S is closed in TO(M ). Theorem 2.8. Let E ⊆ K[X] and S ⊆ TO(M ). Then minE(S) / ∼E is discrete. Hence, if S is closed in TO(M ), then minE(S) / ∼E is finite. 10 ROBERTO BOLDINI Proof. Let πE : minE(S) → minE(S) / ∼E be the natural projection that maps each ≤ to its equivalence class [≤] with respect to ∼E. Let ≤ ∈ minE(S). It is enough to show that {[≤]} is open in minE(S) / ∼E. Put U = π−1 E ([≤]). By definition, {[≤]} is open in minE(S) / ∼E if and only if U is open in minE(S). We may assume that U 6= ∅. Let ≤′ ∈ U. We aim to find an open subset V of minE(S) such that ≤′ ∈ V ⊆ U. As K[X] is noetherian, there exists a finite subset F of E with LM≤′ (F ) = LM≤′ (E). Let (Si)i∈N0 be a filtration of M by finite sets Si. As the set Supp(F ) is finite, we find r ∈ N0 such that Supp(F ) ⊆ Sr+1. Put V = Nr(≤′) ∩ minE(S). Of course, V is open in minE(S) and ≤′ ∈ V. We claim that V ⊆ U. Let ≤′′ ∈ V. Then ≤′ and ≤′′ agree on Sr+1 and hence on Supp(F ). It follows LM≤′ (E) ⊆ LM≤′′ (E), as we have already observed in 2.3. Because ≤′′ ∈ minE(S) and ≤′ ∈ S, we obtain LM≤′(E) = LM≤′′ (E). Thus [≤′′] = [≤′] = [≤], that is, ≤′′ ∈ U. Hence V ⊆ U, so U is open in minE(S). We have proved that minE(S) / ∼E is discrete. If S is closed in TO(M ), then minE(S) / ∼E is also compact by 2.7, and hence finite. (cid:3) Corollary 2.9. For each E ⊆ K[X] and each closed S ⊆ TO(M ) the set i‡S(E) is finite, that is, there exist at most finitely many distinct minimal leading monomial ideals of E from S. Proof. The statement follows from 2.8 as clearly there exists a bijection between the sets i‡S(E) and minE(S) / ∼E given by LM≤(E) 7→ [≤] for all ≤ ∈ minE(S). (cid:3) 3. Leading monomial ideals from degree orderings We keep the notation of the previous section. Definition 3.1. For all s ∈ N0 we denote by K[X]≤s the K-submodule of K[X] of finite length consisting of all polynomials of total degree less than or equal to s. Given any subset E of K[X], we put E≤s = K[X]≤s ∩ E for all s ∈ N0. Let I be an ideal of K[X]. Then I≤s is a K-submodule of K[X]≤s. Therefore, as in [3, IX.3.2], we may define the Hilbert function HFI : N0 → N0 of I by the assignment s 7→ lenK K[X]≤s / I≤s. By [3, IX.3.3(a)], if I is a monomial ideal, then HFI (s) equals the cardinality of the set M≤s r I≤s. Moreover, by [3, IX.2.4 & IX.3.3(b)], there exists a uniquely determined uni- variate polynomial HPI with rational coefficients and at most of degree t with the property that HPI (s) = HFI (s) for s ≫ 0, the Hilbert polynomial of I. We may thus define (I) = min {s0 ∈ N0 ∀ s ≥ s0 : HFI (s) = HPI (s)} ∈ N0, the index of regularity of I. A TOPOLOGICAL APPROCH TO LEADING MONOMIAL IDEALS 11 Lemma 3.2. If I and J are monomial ideals of K[X] such that I ⊆ J , then (I) ≥ (J). Proof. This follows from [3, IX.2.5 & IX.3.3]. See also the proof of [3, IX.2.6]. (cid:3) Lemma 3.3. If I and J are monomial ideals of K[X] with I ⊆ J and HFI = HFJ , then I = J . Proof. If there existed a monomial m ∈ J r I, then with s = deg(m) it would hold I≤s ( J≤s, thus HFI (s) = M≤s r I≤s > M≤s r J≤s = HFJ (s), a contradiction. Hence I ∩ M = J ∩ M , whence I = J as these are monomial ideals, see also 2.2. (cid:3) Definition 3.4. One clearly has deg(LM≤(p)) ≤ deg(p) for all ≤ ∈ TO(M ) and all p ∈ K[X] r {0}, where deg(−) denotes the total degree function on K[X]. A degree ordering on M or of K[X] is a total ordering ≤ on M such that it holds deg(LM≤(p)) = deg(p) for all p ∈ K[X] r {0}. The set of all degree orderings on M is denoted DO(M ). Example 3.5. For each ≤ ∈ TO(M ) the binary relation ≤deg on M defined by m ≤deg m′ ⇔ deg(m) < deg(m′) ∨ (deg(m) = deg(m′) ∧ m ≤ m′) is a degree ordering of K[X]. Proposition 3.6. It holds DO(M ) ⊆ FO1(M ). Proof. Let ≤ ∈ DO(M ). Suppose ≤ /∈ FO1(M ). Then there exists m ∈ M such that 1 6≤ m. So m < 1 by totality. It follows LM≤(m+1) = 1, thus deg(LM≤(m+1)) = 0. But m is a monomial different than 1, hence deg(m + 1) > 0, a contradiction. (cid:3) Reminder 3.7. Let S be a set. We recall that a partial ordering on S is a reflexive, transitive, and antisymmetric binary relation on S, and that a partial ordering (cid:22) on S is said a well-ordering on S if each nonempty subset T of S admits a minimal element with respect to (cid:22), that is, for each T ⊆ S with T 6= ∅ there exists t′ ∈ T such that for each t ∈ T it holds the implication t (cid:22) t′ ⇒ t = t′. If (cid:22) is a total ordering of S, then (cid:22) is a well-ordering on S precisely when each nonempty subset T of S admits a minimum, that is, for each T ⊆ S with T 6= ∅ there exists t′ ∈ T such that for each t ∈ T it holds t′ (cid:22) t. Notation 3.8. For each set S we denote by WO(S) the set of all total orderings on S that are also well-orderings on S. Proposition 3.9. It holds DO(M ) ⊆ WO(M ). Proof. Let ≤ ∈ DO(M ). Let ∅ 6= T ⊆ M . Suppose that there exists no minimum in T with respect to ≤. Let t0 ∈ T . We find t1 ∈ T such that t1 < t0, and then 12 ROBERTO BOLDINI find t2 ∈ T such that t2 < t1, and then. . . Thus there exists in T an infinite strictly descending chain . . . < t2 < t1 < t0. For each k ∈ N0 it holds deg(tk) ≥ deg(tk+1). Indeed, let k ∈ N0 and consider the polynomial tk +tk+1. We have LM≤(tk +tk+1) = tk as tk > tk+1. Since ≤ ∈ DO(M ), it follows deg(tk + tk+1) = deg(tk). Hence deg(tk) ≥ deg(tk+1). Therefore we can write . . . ≤ deg(t2) ≤ deg(t1) ≤ deg(t0). Now, for each d ∈ N0 there exist only finitely many distinct monomials of degree d. Hence we can find a sequence (ki)i∈N0 of integers ki with k0 = 0 and ki < ki+1 with the property that the strict descending chain . . . < deg(tk2 ) < deg(tk1 ) < deg(tk0 ) in N0 is infinite, and this is absurd. (cid:3) Lemma 3.10. DO(M ) is a closed subset of TO(M ) and hence compact. Proof. Let (Si)i∈N0 be a filtration of M consisting of finite sets Si. Let ≤ ∈ TO(M ) be an accumulation point of DO(M ). For each r ∈ N0 we find ≤r in DO(M )∩Nr(≤) with ≤r 6= ≤, so that ≤ and ≤r agree on Sr+1. Let p ∈ K[X] r {0}. We find r ∈ N0 such that Supp(p) ⊆ Sr+1. We choose ≤r as above, and so LM≤(p) = LM≤r (p), thus deg(LM≤(p)) = deg(LM≤r (p)) = deg(p) as ≤r is a degree ordering. Hence ≤ ∈ DO(M ). Therefore DO(M ) contains all its accumulation points in TO(M ) and so is closed in TO(M ). Since TO(M ) is compact by 1.5, it follows that DO(M ) is compact. (cid:3) Definition 3.11. Let E ⊆ K[X] and S ⊆ DO(M ). Analogously as in 2.6, we define an equivalence relation ∼E on S by ≤ ∼E ≤′ ⇔ LM≤(E) = LM≤′ (E). We also provide the set S with its relative topology and the set S / ∼E of the equivalence classes of S with respect to ∼E with its quotient topology. Remark 3.12. Let E ⊆ K[X] and S ⊆ DO(M ). From 3.10 it follows that S / ∼E is compact whenever S is closed in DO(M ). By 3.10 it is also clear that S is closed in DO(M ) if and only if S is closed in TO(M ). Lemma 3.13. Let E ⊆ K[X] and ≤ ∈ DO(M ). There exists an open neighbour- hood U of ≤ in DO(M ) such that LM≤(E) = LM≤′ (E) for all ≤′ ∈ U. Proof. Fix a filtration (Si)i∈N0 of M by finite sets Si. As K[X] is noetherian, there exists a finite subset F of E such that LM≤(F ) = LM≤(E). Put s0 = (LM≤(E)) and recall that t is the number of indeterminates of our polynomial ring K[X]. As Si and as the sets Supp(F ) and M≤s0+t are finite, we find r ∈ N0 such M = Si∈N0 that Supp(F ) ∪ M≤s0+t ⊆ Sr+1. Trivially U = Nr(≤) ∩ DO(M ) is open in DO(M ), and clearly ≤ ∈ U. Let ≤′ ∈ U. Since ≤ and ≤′ agree on Sr+1 and hence on Supp(F ), by 2.3 we get (a) LM≤(E) ⊆ LM≤′(E). Similarly, ≤ and ≤′ agree on M≤s0+t, and because ≤ and A TOPOLOGICAL APPROCH TO LEADING MONOMIAL IDEALS 13 ≤′ are degree orderings, we obtain LM≤(E)≤s = LM≤′ (E)≤s for 0 ≤ s ≤ s0 + t, and hence M≤s r LM≤(E)≤s = M≤s r LM≤′ (E)≤s for 0 ≤ s ≤ s0 + t, and therefore we have (b) HFLM≤(E)(s) = HFLM≤′ (E)(s) for 0 ≤ s ≤ s0 + t. By (a) and 3.2 it holds (LM≤(E)) ≥ (LM≤′ (E)). It follows HPLM≤(E)(s) = HPLM≤′ (E)(s) for s0 ≤ s ≤ s0 + t. As the polynomials HPLM≤(E) and HPLM≤′ (E) have at most degree t and as they agree on t + 1 points, it follows (c) HPLM≤(E) = HPLM≤′ (E). By (b) and (c) we get HFLM≤(E) = HFLM≤′ (E). Hence, by 3.3, LM≤(E) = LM≤′(E). (cid:3) Theorem 3.14. Let E ⊆ K[X] and S ⊆ DO(M ). Then S / ∼E is discrete. Hence, if S is closed in DO(M ), then S / ∼E is finite. Proof. Let πE : S → S / ∼E be the natural projection that maps each ≤ to its equivalence class [≤] with respect to ∼E. Let ≤ ∈ S. It is enough to show that {[≤]} is open in S / ∼E. Put U = π−1 E ([≤]). By definition, {[≤]} is open in S / ∼E if and only if U is open in S. We may assume that U 6= ∅. Let ≤′ ∈ U. We aim to find an open subset W of S such that ≤′ ∈ W ⊆ U. By 3.13, we find an open subset V of DO(M ) with ≤′ ∈ V such that for all (cid:22)′′ ∈ V it holds [(cid:22)′′] = [(cid:22)′] = [(cid:22)]. Thus, putting W = V ∩ S, we have that W is open in S and (cid:22)′ ∈ W ⊆ U. Therefore U is open in S. We have proved that S / ∼E is discrete. If S is closed in DO(M ), then S and thus S / ∼E are also compact by 3.10, and hence S / ∼E is finite. (cid:3) Corollary 3.15. For each E ⊆ K[X] and each S ⊆ DO(M ) the set „‡S(E) is finite, that is, there exist at most finitely many distinct leading monomial ideals of E from S. Proof. Let E ⊆ K[X]. By 3.14, DO(M ) / ∼E is finite. We have a bijection between the sets „‡DO(M)(E) and DO(M ) / ∼E given by LM≤(E) 7→ [≤] for all ≤ ∈ DO(M ), thus „‡DO(M)(E) is finite. Now, if S ⊆ DO(M ), then „‡S(E) ⊆ „‡DO(M)(E). (cid:3) 4. Action of K-module isomorphisms We keep the notation of the previous section. Further, let V be a K-module such that there exists a K-module isomorphism Φ of V in K[X], and put N = Φ−1(M ), so that N is a countable K-basis of V . Sometimes we denote the inverse of Φ by Ψ . Remark 4.1. We have a map φ : TO(N ) → TO(M ) such that for any given (cid:22) ∈ TO(N ) it holds Φ(n) φ((cid:22)) Φ(n′) ⇔ n (cid:22) n′ for all n, n′ ∈ N . Indeed, fixed any (cid:22) ∈ TO(N ), simply define m φ((cid:22)) m′ ⇔ Φ−1(m) (cid:22) Φ−1(m′) for all m, m′ ∈ M . Then φ((cid:22)) is uniquely determined by (cid:22) as Φ−1 is surjective, and φ((cid:22)) is total and hence reflexive and is transitive as (cid:22) is. The antisymmetry of φ((cid:22)) follows immediately from the injectivity of Φ−1. 14 ROBERTO BOLDINI In a similar way, there exists a map ψ : TO(M ) → TO(N ) such that for any given ≤ ∈ TO(M ) it holds Ψ (m) ψ(≤) Ψ (m′) ⇔ m ≤ m′ for all m, m′ ∈ M . The maps φ and ψ are inverse of each other, thus they are isomorphisms of sets. Indeed, they are more, as the following theorem asserts. Theorem 4.2. The bijection φ of 4.1 is a homeomorphism of TO(N ) in TO(M ). Proof. We only have to show that φ is continuous and open. Since φ is bijective, it is enough to check this for one choice of subbases of TO(N ) and TO(M ). For each (n, n′) ∈ N × N one has φ(U(n,n′)) = U(Φ(n),Φ(n′)), thus φ is open. For each (m, m′) ∈ M × M it holds φ−1(U(m,m′)) = U(Φ−1(m),Φ−1(m′)), hence φ is continuous. (cid:3) Definition & Remark 4.3. Each v ∈ V can be written in canonical form as a sum Pn∈Supp(v) αnn for a uniquely determined finite subset Supp(v) of N such that αn ∈ K r {0} for all n ∈ Supp(v). We call Supp(v) the support of v. For each subset H of V let Supp(H) = Sh∈H Supp(h). In the notation of 2.1, one has Supp(Φ(v)) = Φ(Supp(v)) for all v ∈ V , and hence Supp(Φ(H)) = Φ(Supp(H)) for all H ⊆ V . Conversely, Supp(Ψ (p)) = Ψ (Supp(p)) for all p ∈ K[X], and hence Supp(Ψ (E)) = Ψ (Supp(E)) for all E ⊆ K[X]. Given any (cid:22) ∈ TO(N ), for each v ∈ V r {0} we denote by lm(cid:22)(v) the uniquely determined maximal element of Supp(v) with respect to (cid:22). In the notation of 4.1, one has LMφ((cid:22))(Φ(v)) = Φ(lm(cid:22)(v)) for all v ∈ V r {0}. For each v ∈ V r {0} we write LM(cid:22)(v) for LMφ((cid:22))(Φ(v)), and with abuse of language we call LM(cid:22)(v) the leading monomial of v with respect to (cid:22). In this situation, we denote LCφ((cid:22))(Φ(v)) by LC(cid:22)(v) or lc(cid:22)(v), and with abuse of language we call LC(cid:22)(v) alias lc(cid:22)(v) the leading coefficient of v with respect to (cid:22). Observe that either v − lc(cid:22)(v) lm(cid:22)(v) = 0 or lm(cid:22)(v − lc(cid:22)(v) lm(cid:22)(v)) ≺ lm(cid:22)(v). For each (cid:22) ∈ TO(N ) and each H ⊆ V we denote by LM(cid:22)(H) the monomial ideal hLM(cid:22)(h) h ∈ H r {0}i of K[X], and again with abuse of language we call LM(cid:22)(H) the leading monomial ideal of H with respect to (cid:22). Further, for each H ⊆ V and each T ⊆ TO(N ) let „‡T(H) = {LM(cid:22)(H) (cid:22) ∈ T} be the set of all leading monomial ideals of H from T. Similarly as in 2.4, given H ⊆ V and T ⊆ TO(N ), we say that (cid:22) ∈ TO(N ) is a minimalizer of H in T if LM(cid:22)(H) is a minimal element of „‡T(H) with respect to ⊆. We denote the set of all minimalizers of H in T by minH (T). We write i‡T(H) for the set „‡minH (T)(H) = {LM(cid:22)(H) (cid:22) ∈ minH (T)} of all minimal leading monomial ideals of H from T. A TOPOLOGICAL APPROCH TO LEADING MONOMIAL IDEALS 15 Remark 4.4. Let T ⊆ TO(N ) and H ⊆ V . The homeomorphism φ↾T : T → φ(T) induces a homeomorphism φ↾T : T/∼H → φ(T)/∼Φ(H) with πΦ(H) ◦φ↾T = φ↾T ◦πH , where ∼H is the equivalence relation on T given by (cid:22) ∼H (cid:22)′ if and only if LM(cid:22)(H) = LM(cid:22)′ (H), and ∼Φ(H) is the equivalence relation on φ(T) defined as in 3.11, and πH and πΦ(H) are the respective natural projections. Remark 4.5. Given any H ⊆ V and T ⊆ TO(N ), it follows immediately from the definitions that LM(cid:22)(H) = LMφ((cid:22))(Φ(H)) for all (cid:22) ∈ T. Conversely, given any E ⊆ K[X] and S ⊆ TO(M ), one has LM≤(E) = LMψ(≤)(Ψ (E)) for all ≤ ∈ S. It immediately follows that „‡T(H) = „‡φ(T)(Φ(H)) and „‡S(E) = „‡ψ(S)(Ψ (E)), and even that i‡T(H) = i‡φ(T)(Φ(H)) and i‡S(E) = i‡ψ(S)(Ψ (E)). Theorem 4.6. Let H ⊆ V and let T ⊆ TO(N ) be closed. Then i‡T(H) is finite, that is, there exist at most finitely many distinct minimal leading monomial ideals of H from T. Proof. Clear by 4.5, 4.2, and 2.9. (cid:3) Definition 4.7. We put DO(N ) = φ−1(DO(M )), and call DO(N ) the set of all degree orderings on N . Remark 4.8. FOΦ−1(1)(N ) = φ−1(FO1(M )) and WO(N ) = φ−1(WO(M )). Hence DO(N ) ⊆ FOΦ−1(1)(N )∩WO(N ) by 3.6 and 3.9. Moreover, by 4.2 and 3.10, DO(N ) is closed in TO(N ) and compact. Theorem 4.9. For each H ⊆ V and each T ⊆ DO(N ) the set „‡T(H) is finite, that is, there exist at most finitely many distinct leading monomial ideals of H from T. Proof. Clear by 4.5 and 3.15. (cid:3) 5. T-multiplicative algebras of countable type We keep the notation of the previous section. Definition 5.1. An algebra of countable type is a quadruple At,Φ K = (A, K, t, Φ) consisting of an associative, not necessarily commutative algebra A over a field K, a nonnegative integer t, and a K-module isomorphism Φ of A in K[X1, . . . , Xt]. If At,Φ K is an algebra of countable type and if M is the canonical K-basis of 0, then N = Φ−1(M ) is a K[X1, . . . , Xt] consisting of all monomials X ν, ν ∈ Nt countable K-basis of A, which we call the canonical basis of At,Φ K . Given any subset T of the set TO(N ) of all total orderings on N , we say that At,Φ K or simply A is multiplicative on T or T-multiplicative if A is a domain and in the notation of 4.3 it holds LM(cid:22)(ab) = LM(cid:22)(a) LM(cid:22)(b) for all a, b ∈ A r {0} and all (cid:22) ∈ T. 16 ROBERTO BOLDINI Henceforth in this section, let At,Φ K be an algebra of countable type. We write K[X] for K[X1, . . . , Xt] and fix the canonical K-bases M and N of K[X] and At,Φ K , respectively. Now we may make use of the notation introduced in 4.3. And yet another. . . Macaulay Basis Theorem, that is, a slight generalization of a classical result. Theorem 5.2. Let (cid:22) ∈ WO(N ), assume that At,Φ be a left ideal of A, put B = M r LM(cid:22)(L), and let K is multiplicative on {(cid:22)}, let L : K[X] → K[X] / Φ(L) be the residue class epimorphism of K-modules. Then the image B of B under is a K-basis of K[X] / Φ(L). Proof. We first show that B generates K[X] / Φ(L) over K. Suppose it is not the case. Let W = Pb∈B Kb. Then the set P = {p ∈ K[X] r {0} p /∈ W } is non- empty. Thus, with ≤ = φ((cid:22)), the subset Q = {LM≤(p) p ∈ P } of M is nonempty. As φ((cid:22)) ∈ WO(M ), see 4.8, we may choose p ∈ P such that LM≤(p) is minimal in Q with respect to ≤. It holds Supp(p) r {LM≤(p)} ⊆ W . Indeed, if there existed m ∈ Supp(p) r {LM≤(p)} such that m /∈ W , then we would have m ∈ P and hence m = LM≤(m) ∈ Q, and this would contradict the minimality of LM≤(p) as clearly m < LM≤(p). It follows LM≤(p) /∈ W as otherwise we would have Supp(p) ⊆ W and hence the contradiction p ∈ W . Therefore LM≤(p) ∈ LM(cid:22)(L) as otherwise we would have LM≤(p) ∈ B and hence the contradiction LM≤(p) ∈ B ⊆ W . Thus we find x ∈ L r {0} such that LM(cid:22)(x) LM≤(p), see 2.2. So we find n ∈ N with LM≤(p) = Φ(n) LM(cid:22)(x) = LM(cid:22)(n) LM(cid:22)(x) = LM(cid:22)(nx), where this last equality holds by multiplicativity of At,Φ K on {(cid:22)}. With q = LC≤(p) LC≤(Φ(nx))−1Φ(nx) we obtain q ∈ Φ(L) as L is a left ideal and Φ(L) is a K-module, and of course we have LM≤(p) = LM≤(q) and LC≤(p) = LC≤(q). Now we consider r = p − q. It holds r = p. Thus r /∈ W . But then in particular r 6= 0, and hence clearly LM≤(r) < LM≤(p), thus r /∈ P by the minimality of LM≤(p), so that r ∈ W , a contradiction. Next we show that B is linearly independent over K. Suppose to the con- trary that there exist r ∈ N and α1, . . . , αr ∈ K r {0} and pairwise distinct b1, . . . , br ∈ B such that α1b1 + . . . + αrbr = 0. Then any respective represent- atives b1, . . . , br ∈ B of b1, . . . , br are pairwise distinct and α1b1 + . . . + αrbr = Φ(y) for some y ∈ L. Of course, y 6= 0 as the monomials b1, . . . , br are linearly in- dependent over K. It follows LM≤(Φ(y)) = bi ∈ B for some i ∈ {1, . . . , r}. There- fore LM≤(Φ(y)) ∈ B ∩ LM≤(Φ(L)), that is, LM(cid:22)(y) ∈ B ∩ LM(cid:22)(L) by 4.5. But, by definition, B ∩ LM(cid:22)(L) = ∅, a contradiction. (cid:3) Corollary 5.3. Let (cid:22), (cid:22)′ ∈ WO(N ), assume that At,Φ K is multiplicative on {(cid:22), (cid:22)′}, and let L be a left ideal of A with LM(cid:22)(L) ⊆ LM(cid:22)′ (L). Then LM(cid:22)(L) = LM(cid:22)′ (L). A TOPOLOGICAL APPROCH TO LEADING MONOMIAL IDEALS 17 Proof. Put B = M rLM(cid:22)(L) and B′ = M rLM(cid:22)′ (L). Let be the residue class homomorphism (of K-modules). Suppose by contradiction that LM(cid:22)(L) ( LM(cid:22)′(L). Then B ) B′, hence B ⊇ B′. : K[X] → K[X] / Φ(L) If it held B = B′, then we would find b ∈ B r B′ and b′ ∈ B′ such that b = b′, hence b − b′ ∈ Φ(L), thus LMφ((cid:22))(b − b′) ∈ LMφ((cid:22))(Φ(L)) = LM(cid:22)(L); on the other hand, either LMφ((cid:22))(b − b′) = b or LMφ((cid:22))(b − b′) = b′, in any case LMφ((cid:22))(b − b′) ∈ B, a contradiction. Thus B ) B′. But, by 5.2, B and B′ are K-bases of K[X] / Φ(L), hence the one cannot strictly contain the other, a contradiction. (cid:3) Corollary 5.4. Let T ⊆ WO(N ) such that T is closed in TO(N ), assume that At,Φ K is multiplicative on T, and let L be a left ideal of A. Then „‡T(L) = i‡T(L). In particular, „‡T(L) is finite, that is, L admits at most finitely many distinct leading monomial ideals from T. Proof. By 5.3, T = minL(T), thus „‡T(L) = i‡T(L), which is finite by 4.6. (cid:3) 6. Admissible orderings We keep the notation of the previous section. Definition 6.1. A compatible ordering on M or of K[X] is a total ordering ≤ on 0 it holds compatibility: X υ ≤ X ν ⇒ X υ+γ ≤ X ν+γ. M such that for all υ, ν, γ ∈ Nt Compatible orderings are also known as semigroup orderings. The set of all compatible orderings of K[X] is denoted by CO(M ). We also consider the set of compatible orderings on N or of At,Φ K or simply of A defined as CO(N ) = φ−1(CO(M )). Proposition 6.2. CO(M ) and CO(N ) are closed in TO(M ) and TO(N ), respect- ively, and hence compact. Proof. Let (Si)i∈N0 be a filtration of M consisting of finite sets Si. Let ≤ ∈ TO(M ) be an accumulation point of CO(M ). Thus, by definition, for each r ∈ N0 there exists ≤r ∈ CO(M ) ∩ Nr(≤) r {≤}, so that ≤r and ≤ agree on Sr+1. Choose any υ, ν ∈ Nt 0. Then we find r ∈ N0 such that Sr+1 contains the monomials X υ, X ν, X υ+γ, X ν+γ. There exists ≤r as above that agrees with ≤ on Sr+1, so that X υ ≤r X ν. Since ≤r is a compatible ordering of K[X], it follows X υ+γ ≤r X ν+γ. Therefore X υ+γ ≤ X ν+γ. Hence ≤ ∈ CO(M ). 0 and assume that X υ ≤ X ν, say. Let γ ∈ Nt Thus CO(M ) contains all its accumulation points in TO(M ) and hence CO(M ) is closed in TO(M ). Since TO(M ) is compact by 1.5, CO(M ) is compact. Since φ is a homeomorphism by 4.2, also CO(N ) is closed in TO(N ) and compact. (cid:3) 18 ROBERTO BOLDINI Definition 6.3. AO(M ) = FO1(M ) ∩ CO(M ) is the set of all admissible orderings on M or of K[X], and AO(N ) = FOΦ−1(1)(N ) ∩ CO(N ) is the set of all admissible orderings on N or of At,Φ K or simply of A. Observe that φ−1(AO(M )) = AO(N ). Admissible orderings are also known as monoid orderings. Remark 6.4. One sees that this definition of admissible ordering on M and on N is equivalent to the one given in [5], and it is equivalent to the notion of term orderings given in [7] in the case of Weyl algebras under the assumption that Φ(1) = 1. Remark 6.5. An admissible ordering of K[X] is a total ordering ≤ on M such that it holds well-foundedness: 1 ≤ X ν, and compatibility: X υ ≤ X ν ⇒ X υ+γ ≤ X ν+γ. Since M is a K-basis of K[X], these axioms are equivalent to: 1 < X ν whenever ν 6= 0, and X υ < X ν ⇒ X υ+γ < X ν+γ. Example 6.6. The lexicographical ordering ≤lex on M defined by X υ ≤lex X ν :⇔ (υ = ν) ∨ (υ 6= ν ∧ υm(υ,ν) < νm(υ,ν)) for all υ, ν ∈ Nt α, β ∈ Nt 0 with α 6= β, is an admissible ordering of K[X]. 0, where we put m(α, β) = min { k 1 ≤ k ≤ t ∧ αk 6= βk } for all Example 6.7. Fixed any ≤ ∈ AO(M ), for all ω ∈ Nt ≤-ordering ≤ω by 0 one can define the ω-graded X υ ≤ω X ν :⇔ (ω · υ < ω · ν) ∨ (ω · υ = ω · ν ∧ X ν ≤ Y υ) for all υ, ν ∈ Nt 0, and one has that ≤ω is an admissible ordering of K[X], see Exercise 12 in [3, II.4] Proposition 6.8. AO(M ) and AO(N ) are closed in TO(M ) and TO(N ), respect- ively, and hence compact. Proof. Clear by 6.2, 1.7, and 1.5. (cid:3) Proposition 6.9. AO(M ) = WO(M ) ∩ CO(M ) and AO(N ) = WO(N ) ∩ CO(N ). Proof. By [3, II.4.6] one has FO1(M ) ∩ CO(M ) = WO(M ) ∩ CO(M ). Since φ−1 is injective and since φ−1(CO(M )) = CO(N ) and φ−1(FO1(M )) = FOΦ−1(1)(N ) and φ−1(WO(M )) = WO(N ), the second claim follows. (cid:3) 7. Degree-compatible orderings We keep the notation of the previous section. Example 7.1. It holds DO(M ) * CO(M ) and hence DO(N ) * CO(N ). Indeed, any degree ordering ≤ of K[Y, Z] such that 1 < Y < Z < Y Z < Y 2 < Z 2 < . . . is not compatible because compatibility would force Y 2 < Y Z from Y < Z. A TOPOLOGICAL APPROCH TO LEADING MONOMIAL IDEALS 19 Also it holds CO(M ) * DO(M ) and hence CO(N ) * DO(N ). For instance, the lexicographic ordering ≤lex of K[Y, Z] induced by Y <lex Z is compatible but is not a degree ordering as deg(LM≤(Y + Z 2)) = deg(Y ) = 1 6= 2 = deg(Y + Z 2). Remark & Definition 7.2. It is not to expect that there exist interesting K- algebras of countable type that are multiplicative on DO(M ) since even K[X] is not. For a degree ordering ≤ of K[Y, Z] with 1 < Y < Z < Y 2 < Z 2 < Y Z < . . . for instance, it holds LM≤((Y + Z)2) = Y Z 6= Z 2 = LM≤(Y + Z) LM≤(Y + Z). Therefore we shall consider the set DCO(M ) = DO(M ) ∩ CO(M ) of the degree- compatible orderings on M or of K[X] and the set DCO(N ) = DO(N ) ∩ CO(N ) of the degree-compatible orderings on N or of At,Φ K or simply of A. Of course, it holds DCO(N ) = φ−1(DCO(M )). Moreover, DCO(M ) ⊆ AO(M ) by 3.6, and hence DCO(N ) ⊆ AO(N ). Finally, by 3.10 and 4.8 and by 6.2, DCO(M ) and DCO(N ) are closed in TO(M ) and TO(N ), respectively, and compact. Proposition 7.3. If t > 1, where t is the number of indeterminates, then DCO(M ) is nowhere dense in DO(M ), and so is DCO(N ) in DO(N ). Proof. Consider the filtration (Si)i∈N0 of M given by Si = {m ∈ M deg(m) < i}. Suppose that some ordering ≤ lies in the interior DCO(M )◦ of the closed subset DCO(M ) of DO(M ). Then we find a neighbourhood of ≤ open in DO(M ) contained in DCO(M )◦, that is, we find r ∈ N0 such that Nr(≤) ∩ DO(M ) ⊆ DCO(M ). Since S1 = {1}, we have N0(≤) = TO(M ). As DCO(M ) ( DO(M ), it follows r ≥ 1. Assume that X1 < X2, say. Then X r+2 1 X2 by compatibility. Let ≤′ be the total ordering on M given by X r+1 and m ≤′ m′ ⇔ m ≤ m′ whenever (m, m′) ∈ M × M r {(X r+1 )}. Then ≤′ ∈ Nr(≤) ∩ DO(M ), so that ≤′ ∈ DCO(M ). As r ≥ 1, we have that ≤ and ≤′ agree on S2, thus X1 <′ X2. By compatibility it follows X r+2 1 X2, a contradiction. Now we conclude (cid:3) 1 1 X2 <′ X r+2 <′ X r+1 1 X2, X r+2 1 < X r+1 1 by 4.2. 1 Remark 7.4. If t = 1, then DO(M ) = DCO(M ) = 1 = DCO(N ) = DO(N ), thus DCO(M ) = DO(M ) and DCO(N ) = DO(N ). Example 7.5. For each ≤ ∈ AO(M ) the binary relation ≤deg on M defined by m ≤deg m′ ⇔ deg(m) < deg(m′) ∨ (deg(m) = deg(m′) ∧ m ≤ m′). is a degree-compatible ordering of K[X]. More generally, the admissible orderings of Example 6.7 are degree-compatible orderings whenever ω 6= 0 or ≤ ∈ DCO(M ). Remark 7.6. By 5.4, for each H ⊆ A and each T ⊆ DCO(N ) the set „‡T(H) is finite. In particular, by 6.6, 6.7, and 7.5, the set „‡DCO(N )(H) is nonempty and finite. 20 ROBERTO BOLDINI 8. T-admissible algebras We keep the notation of the previous section. Definition 8.1. Let T ⊆ AO(N ). We say that At,Φ if At,Φ is AO(N )-admissible. We say that At,Φ K is multiplicative on T. We say that At,Φ K or simply A is T-admissible K or simply A is admissible if At,Φ K K is K or simply A is degree-compatible if At,Φ DCO(N )-admissible. Example 8.2. In the terminology of [5], every K-algebra that is of solvable type with respect to all admissible orderings is admissible. This follows indeed from [5, 1.5]. For instance, if K has characteristic 0, then every Weyl algebra W over K is isomorphic as a K-module to a commutative polynomial ring over K, see [2, I.2.1], and W clearly fulfills the axioms [5, 1.2] of an algebra of solvable type for all admissible orderings on its canonical K-basis, so that W is multiplicative on these orderings by [5, 1.5]. Example 8.3. If K has characteristic 0, then the universal enveloping algebra U (g) of any Lie algebra g of finite length over K is degree-compatible. Indeed, let X = {x1, . . . , xr} be a finite K-basis of g. By the Poincar´e -- Birkhoff -- Witt Theorem, see 2.13, 2.14, 2.22 of [6, II], there exist then a canonical K-module monomorphism h : g ֒→ U (g) and a countable K-basis Y = {yν1 0} of U (g) with yi = h(xi) such that [yj, yk] = P1≤i≤r cijkyi for some cijk ∈ K. Thus, U (g) is isomorphic as a K-module to the commutative polynomial ring K[X1, . . . Xr] by an isomorphism that maps yi to Xi, and the relations ykyj = yjyk − P1≤i≤r cijkyi imply by [5, 1.2 & 1.5] that U (g) is multiplicative on DCO(Y ). (ν1, . . . , νr) ∈ Nr 1 · · · yνr r Theorem 8.4. Let T ⊆ AO(N ) be a closed subset. Assume that At,Φ K is T-admissible. Let L be a left ideal of A. Then „‡T(L) is finite, that is, L admits only finitely many distinct leading monomial ideals from T. In particular, if At,Φ K is admissible, then the nonempty set „‡AO(N )(L) is finite. Proof. It is all clear by 5.4, 6.8, 6.9, and by 4.2, 6.6, 6.7, 8.2. (cid:3) Remark 8.5. Notice that by 7.6 we already know this result for subspaces T of DCO(N ) without having to assume that A be multiplicative on T nor that L be a left ideal. 9. Grobner bases We keep the notation of the previous section. A TOPOLOGICAL APPROCH TO LEADING MONOMIAL IDEALS 21 Definition 9.1. Let At,Φ K be an algebra of countable type, L be a left ideal of A, N denote the canonical K-basis of At,Φ K , and (cid:22) be a total ordering on N . A Grobner basis of L with respect to (cid:22) is a finite subset G of L such that L = Pg∈G Ag and LM(cid:22)(L) = LM(cid:22)(G). Remark 9.2. The definition of Grobner basis given here is equivalent to the one given in [5] if one restricts to admissible orderings and algebras of solvable type, see [5, 3.8]. This definition is also equivalent to the one given in [7] when further restricting to Weyl algebras. By [4, II.4.2] it is less general than the one given in [4, II.3.2(ii)], but it is equi- valent to the definition given in [4, III.1.1] when restricting to admissible orderings and free K-algebras KhXλ λ ∈ Λi, Λ any index set. Definition 9.3. Let At,Φ A, and let N denote the canonical K-basis of At,Φ K . K be an algebra of countable type, let L be a left ideal of Given any T ⊆ TO(N ), we say that a finite subset U of L is a T-universal Grobner basis of L if U is a Grobner basis of L with respect to all elements of T. In the following we call the T-universal Grobner bases in T-admissible algebras simply universal Grobner bases. We fix here an algebra At,Φ K of countable type and as usually denote its canonical K-basis by N . Theorem 9.4. Assume that A is left noetherian, let L be a left ideal of A, and let (cid:22) be a total ordering on N . Then L admits a Grobner basis with respect to (cid:22). Proof. Suppose that L admits no Grobner basis with respect to (cid:22). Since A is left noetherian, there exists a finite subset F0 of L such that L = AF0. It holds LM(cid:22)(F0) ( LM(cid:22)(L) as F0 is not a Grobner basis. Thus there exists x1 ∈ L r {0} with LM(cid:22)(x1) /∈ LM(cid:22)(F0). Put F1 = F0 ∪ {x1}. Again LM(cid:22)(F1) ( LM(cid:22)(L) as F1 is not a Grobner basis. Thus there exists x2 ∈ L r {0} with LM(cid:22)(x2) /∈ LM(cid:22)(F1). Put F2 = F1 ∪ {x2}. Again LM(cid:22)(F2) ( LM(cid:22)(L) as F2 is not a Grobner basis. . . We construct in this way an infinite chain LM(cid:22)(F0) ( LM(cid:22)(F1) ( LM(cid:22)(F2) ( . . . of ideals of K[X], in contradiction to the noetherianity of K[X]. (cid:3) Theorem 9.5. Assume that there exists (cid:22) ∈ WO(N ) with the property that At,Φ K is multiplicative on {(cid:22)}. Let L be a left ideal of A and F be a finite subset of L such that LM(cid:22)(L) = LM(cid:22)(F ). Then L = Pf ∈F Af . Proof. Trivially, we have Pf ∈F Af ⊆ L. Suppose that Pf ∈F Af ( L. Then the set U = {LM(cid:22)(l) l ∈ L r Pf ∈F Af } is nonempty. We have ≤ = φ((cid:22)) ∈ WO(M ), and 22 ROBERTO BOLDINI so there exists l ∈ LrPf ∈F Af such that u = LM(cid:22)(l) is minimal in U with respect to ≤. Since u ∈ LM(cid:22)(L) = LM(cid:22)(F ), we can write u = Pf ∈F r{0} pf LM(cid:22)(f ) for some family (pf )f ∈F r{0} of polynomials. As u ∈ M and M is a K-basis of K[X], we find m ∈ Sf ∈F r{0} Supp(pf ) ⊆ M and g ∈ F r {0} such that u = m LM(cid:22)(g). Put n = Φ−1(m). As n ∈ N , clearly n 6= 0. Since A is a domain, it follows ng 6= 0. Now put h = l − lc(cid:22)(l) lc(cid:22)(ng)−1ng. Then h ∈ L r Pf ∈F Af , thus LM(cid:22)(h) ∈ U . On the other hand, LM(cid:22)(ng) = LM(cid:22)(n) LM(cid:22)(g) = m LM(cid:22)(g) = u = LM(cid:22)(l), so that LM(cid:22)(h) < LM(cid:22)(l), a contradiction. (cid:3) Corollary 9.6. Assume that there exists (cid:22) ∈ WO(N ) such that At,Φ K is multiplic- ative on {(cid:22)}. Then A is left noetherian. Proof. Let L be a left ideal of A. As K[X] is noetherian, we find a finite subset F of L such that LM(cid:22)(F ) = LM(cid:22)(L). By 9.5, F is a generating set of L. Thus every left ideal of A is finitely generated. (cid:3) Corollary 9.7. Assume that there exists (cid:22) ∈ WO(N ) such that At,Φ K is multiplic- ative on {(cid:22)}. Then for each left ideal L of A and each total ordering (cid:22)′ on N there exists a Grobner basis of L with respect to (cid:22)′. Proof. Clear by 9.4 and 9.6. (cid:3) 10. Universal Grobner bases in admissible algebras We keep the notation of the previous section. Lemma 10.1. Let (cid:22), (cid:22)′ ∈ WO(N ) such that At,Φ K is multiplicative on {(cid:22), (cid:22)′}. Let L be a left ideal of A and G be a Grobner basis of L with respect to (cid:22). If (cid:22) and (cid:22)′ agree on Supp(G), then LM(cid:22)(L) = LM(cid:22)′ (L) and G is a Grobner basis of L with respect to (cid:22)′. Proof. Because (cid:22) and (cid:22)′ agree on Supp(G), it follows that φ((cid:22)) and φ((cid:22)′) agree on Φ(Supp(G)) = Supp(Φ(G)). Hence LMφ((cid:22))(Φ(G)) = LMφ((cid:22)′)(Φ(G)) by 2.3. From 4.5 it follows LM(cid:22)(L) = LM(cid:22)(G) = LM(cid:22)′ (G) ⊆ LM(cid:22)′(L). As TO(N ) is a Hausdorff space, see 1.2, points are closed, so {(cid:22), (cid:22)′} is closed in TO(N ). Thus „‡{(cid:22),(cid:22)′}(L) = i‡{(cid:22),(cid:22)′}(L) by 5.4, and hence LM(cid:22)(L) = LM(cid:22)′ (L), and therefore LM(cid:22)′(G) = LM(cid:22)′(L). (cid:3) Lemma 10.2. Let T ⊆ WO(N ) such that At,Φ K is multiplicative on T. Let L be a left ideal of A and let F be a finite subset of L. Then the set UL(F ) of all (cid:22) ∈ T such that F is a Grobner basis of L with respect to (cid:22) is open in T. Proof. Let (Si)i∈N0 be a filtration of N consisting of finite sets Si. There exists r ∈ N0 such that the finite subset Supp(F ) of N lies in Sr+1. We may assume A TOPOLOGICAL APPROCH TO LEADING MONOMIAL IDEALS 23 that UL(F ) 6= ∅, so that T 6= ∅. Let (cid:22) ∈ UL(F ). Thus F is a Grobner basis of L with respect to (cid:22). Consider the open neighbourhood Nr((cid:22)) ∩ T of (cid:22) in T and let (cid:22)′ ∈ Nr((cid:22)) ∩ T. Then (cid:22) and (cid:22)′ agree on Sr+1 and in particular on Supp(F ). By 10.1, F is a Grobner basis of L with respect to (cid:22)′, that is, (cid:22)′ ∈ UL(F ). Hence (cid:22) ∈ Nr((cid:22)) ∩ T ⊆ UL(F ), and UL(F ) is open in T. (cid:3) Remark 10.3. Let ∅ 6= T ⊆ WO(N ) such that At,Φ K is multiplicative on T. Let L be a left ideal of A. Then, by 9.7, for each (cid:22) ∈ T there exists a Grobner basis G(cid:22) of L with respect to (cid:22). Thus, in the notation of 10.2, clearly (cid:22) ∈ UL(G(cid:22)) for each (cid:22) ∈ T. Hence, by 10.2, S(cid:22)∈T UL(G(cid:22)) is an open covering of T. Theorem 10.4. Let ∅ 6= T ⊆ WO(N ) such that T is closed in TO(N ) and At,Φ K is multiplicative on T. Let L be a left ideal of A. Then L admits a T-universal Grobner basis. Proof. In the notation of 10.3, S(cid:22)∈T UL(G(cid:22)) is an open covering of T, where each G(cid:22) is a Grobner basis of L with respect to (cid:22). As TO(N ) is compact and T is closed in TO(N ), T is compact. Hence we can find s ∈ N and (cid:22)1, . . . , (cid:22)s ∈ T such that S1≤j≤s UL(G(cid:22)j ) is a finite open covering of T. We claim that U = S1≤j≤s G(cid:22)j is a T-universal Grobner basis of L. Indeed, let (cid:22)0 ∈ T. Then there exists j ∈ {1, . . . , s} such that (cid:22)0 ∈ UL(G(cid:22)j ). Thus G(cid:22)j is a Grobner basis of L with respect to (cid:22)0. As G(cid:22)j ⊆ U , of course also U is a Grobner basis of L with respect to (cid:22)0. Since the choice of (cid:22)0 in T was arbitrary, we conclude that U is a T-universal Grobner basis of L. (cid:3) Corollary 10.5. Let T be a nonempty closed subset of AO(N ) such that At,Φ K is T- admissible. Then for each left ideal L of A there exists a T-universal Grobner basis of L. In particular, every left ideal of an admissible or degree-compatible algebra has a universal Grobner basis. (cid:3) Remark 10.6. To effectively compute a T-universal Grobner basis, one should start walking among the orderings in T and pick some ones that allow to cover T as in 10.3. But how to pluck the right flowers in that vast meadow? The following Lemma 10.7 might be of help. Once one thinks to have located a suitable kind of orderings, that is, an appropriate subset D of T, if one is able to show that D is dense in T, then one can indeed restrict the own search to D. This fact might be the first step toward the construction of a "topological algorithm" that computes a T-universal Grobner basis. Lemma 10.7. In the hypotheses of 10.4, let D be a dense subset of T. Then we can find finitely many (cid:22)1, . . . , (cid:22)s in D and respective Grobner bases G1, . . . , Gs of L such that S1≤j≤s Gj is a T-universal Grobner basis of L. 24 ROBERTO BOLDINI Proof. Because T is compact, we can find finitely many (cid:22)′ s ∈ T such that T = S1≤j≤s UL(Gj), where each Gj is a Grobner basis of L with respect to (cid:22)′ j. Then S1≤j≤s Gj is a T-universal Grobner basis of L. 1, . . . , (cid:22)′ Because D is dense in T and each UL(Gj ) is an open neighbourhood of (cid:22)′ j in T, for 1 ≤ j ≤ s we find (cid:22)j ∈ D ∩ UL(Gj). Thus each Gj is a Grobner basis of L with respect to (cid:22)j. (cid:3) Example 10.8. The orderings (cid:22) given by Φ−1(X υ) (cid:22) Φ−1(X ν) ⇔ X Γ υ ≤lex X Γ ν with Γ a t × t-matrix with entries in N0 constitute a dense subset of AO(N ). This follows easily from [1, p. 6]. Definition 10.9. Let (X, d) be a metric space and let ε ∈ R with ε > 0. We say that Y ⊆ X is ε-dense in X if for all x ∈ X there exists y ∈ Y such that d(x, y) < ε. Lemma 10.10. In the hypotheses of 10.4, assume that there exists r ∈ N0 such that for all (cid:22) ∈ T and all Grobner bases G(cid:22) of L with respect to (cid:22) and all g ∈ G(cid:22) it holds deg(Φ(g)) ≤ r. Let S = (Si)i∈N0 be the filtration of N with Si = Φ−1(M≤i−1). Let D be a 1 r -dense subset of T with respect to the metric dS↾T induced by S. Then we can find finitely many (cid:22)1, . . . , (cid:22)s in D and respective Grobner bases G1, . . . , Gs of L such that S1≤j≤s Gj is a T-universal Grobner basis of L. 1, . . . , (cid:22)′ Proof. We find s ∈ N and (cid:22)′ a (cid:22)′ s ∈ T and G1, . . . , Gs ⊆ L such that each Gj is j-Grobner basis of L and U = S1≤j≤s Gj is a T-universal Grobner basis of L. It holds Supp(U ) ⊆ Sr+1. Because D is 1 r -dense in T, for 1 ≤ j ≤ s there exists j and (cid:22)j agree on Supp(U ) and hence on Supp(Gj), by (cid:22)j ∈ D ∩ Nr((cid:22)′ 10.1 Gj is a Grobner basis of L with respect to (cid:22)j. j). Since (cid:22)′ (cid:3) Remark 10.11. Assume that At,Φ K is a quadric algebra of solvable type, this means, Φ−1(Xi)Φ−1(Xj) = Φ−1(Xj)Φ−1(Xi) + Φ−1(pij ) for polynomials pij ∈ K[X] at most of degree 2. Assume further that L can be generated by finitely many elements x1, . . . , xq such that deg(Φ(xh)) ≤ d for 1 ≤ h ≤ q. As proved in [1], for each (cid:22) ∈ AO(N ) there exists a Grobner basis G(cid:22) of L with respect to (cid:22) such that deg(Φ(g)) ≤ 2( d2+2d Grobner basis U of L such that deg(Φ(u)) ≤ 2( d2+2d for all g ∈ G(cid:22). Therefore there exists a T-universal for all u ∈ U , for one can )2t−1 )2t−1 2 2 construct U as a union of (finitely many) such Grobner bases G(cid:22). Remark 10.12. An alternative, "classical" proof of 10.5 involves a division and a reduction algorithm: (i) Assume that At,Φ K is multiplicative on {(cid:22)} for some (cid:22) ∈ WO(N ). Let a ∈ A, F ⊆ L be finite, and ≤ = φ((cid:22)). Then there exist r ∈ A and (qf )f ∈F ∈ A⊕F such that: A TOPOLOGICAL APPROCH TO LEADING MONOMIAL IDEALS 25 (a) a = Pf ∈F qf + r, (b) ∀ f ∈ F : (f 6= 0 ⇒ ∀ s ∈ Supp(r) : LM(cid:22)(f ) ∤ Φ(s)), (c) a 6= 0 ⇒ (∀ f ∈ F : (qf f 6= 0 ⇒ LM(cid:22)(qf f ) ≤ LM(cid:22)(a))). (ii) Let (cid:22) ∈ AO(N ) such that At,Φ K is multiplicative on {(cid:22)}. Let L be a left ideal of A. Let G be a Grobner basis of L with respect to (cid:22). One can then transform G by applying repeatedly the following procedures: (a) If there exists g ∈ G r {0} such that LM(cid:22)(g) ∈ LM(cid:22)(G r {g}), then replace G by G r {g}. (b) If there exist g ∈ G r {0} and n ∈ Supp(g) r {LM(cid:22)(g)} such that n ∈ LM(cid:22)(G r {g}), then divide g by G r {g} as in (i), so that it holds g = Pf ∈Gr{g} qf f + r, and replace G by ({r} ∪ G) r {g}, which is equal to {r} ∪ (G r {g}) in this case. After finitely many steps both conditions become false, and the process halts with a reduced Grobner basis G of L with respect to (cid:22), that is, for each g ∈ G and each n ∈ Supp(g) it holds n /∈ LM(cid:22)(G r {g}). (iii) Let T be a closed subset of AO(N ) such that At,Φ K is T-admissible. Let L be a left ideal of A. Then there exist at most finitely many leading monomial ideals of L from T, thus we find a finite subset U of T such that „‡U(L) = „‡T(L). For each (cid:22) ∈ U we may choose a reduced Grobner basis G(cid:22) of L with respect to (cid:22). Then S(cid:22) ∈ U G(cid:22) is a T-universal Grobner basis of L. 11. Universal Grobner bases from degree orderings We keep the notation of the previous section. Lemma 11.1. Let L be a left ideal of A, let F be a finite subset of L, and let T be a subspace of DO(N ). Then the set UL(F ) of all (cid:22) ∈ T such that F is a Grobner basis of L with respect to (cid:22) is open in T. Proof. We may assume that UL(F ) 6= ∅. Let (cid:22) ∈ UL(F ). Thus F is a Grobner basis of L with respect to (cid:22), that is, it holds L = Pf ∈F Af and LM(cid:22)(F ) = LM(cid:22)(L). Put ≤ = φ((cid:22)) and E = Φ(F ) and J = Φ(L). Of course, ≤ ∈ DO(M ). Hence, by 3.13, we can find open neighbourhoods VE and VJ of ≤ in DO(M ) such that LM≤′(E) = LM≤(E) for all ≤′ ∈ VE and LM≤′ (J) = LM≤(J) for all ≤′ ∈ VJ . By 4.5 it follows LM≤′(E) = LM≤(E) = LM(cid:22)(F ) = LM(cid:22)(L) = LM≤(J) = LM≤′ (J) for all ≤′ ∈ V, where V = VE ∩ VJ . Put W = φ−1(V) ∩ T. By 4.2, W is an open subset of T such that (cid:22) ∈ W. Again by 4.5 we obtain LM(cid:22)′ (F ) = LMφ((cid:22)′)(E) = LMφ((cid:22)′)(J) = LM(cid:22)′ (L) for all (cid:22)′ ∈ W. Thus W ⊆ UL(F ). Hence W is an open neighbourhood of (cid:22) in UL(F ). (cid:3) 26 ROBERTO BOLDINI Remark 11.2. Assume that A is left noetherian, let L be a left ideal of A, and let T be a subset of DO(N ). Then, by 9.4, for each (cid:22) ∈ T there exists a Grobner basis G(cid:22) of L with respect to (cid:22). Of course, in the notation of 11.1, for each (cid:22) ∈ T it holds (cid:22) ∈ UL(G(cid:22)), and thus S(cid:22)∈T UL(G(cid:22)) is an open covering of T. Theorem 11.3. Assume that A is left noetherian, let L be a left ideal of A, and let T be a closed subset of DO(N ). Then L admits a T-universal Grobner basis. Proof. In the notation of 11.2, S(cid:22)∈T UL(G(cid:22)) is an open covering of T, where each G(cid:22) is a Grobner basis of L with respect to (cid:22). As DO(N ) is compact and T is closed in DO(N ), T is compact. Hence we can find s ∈ N and (cid:22)1, . . . , (cid:22)s ∈ T such that S1≤j≤s UL(G(cid:22)j ) is a finite open covering of T. We claim that U = S1≤j≤s G(cid:22)j is a T-universal Grobner basis of L. Indeed, let (cid:22)0 ∈ T. Then there exists j ∈ {1, . . . , s} such that (cid:22)0 ∈ UL(G(cid:22)j ). Thus G(cid:22)j is a Grobner basis of L with respect to (cid:22)0. Hence, clearly, also U is a Grobner basis of L with respect to (cid:22)0. As the choice of (cid:22)0 in T was arbitrary, we conclude that U is a T-universal Grobner basis of L. (cid:3) We have obtained another proof of the result of 10.5 about degree-compatible al- gebras, this time without appealing to the Macaulay Basis Theorem. Corollary 11.4. Left ideals of a degree-compatible algebra always admit a universal Grobner basis. (cid:3) References [1] M. Aschenbrenner, A. Leykin, Degree Bounds for Grobner Bases in Algebras of Solvable Type, J. Pure Appl. Algebra 213 (2009), 1578 -- 1605. [2] S. C. Coutinho, A Primer of Algebraic D-modules, London Math. Soc. Student Texts 33, Cambridge University Press, 1995. [3] D. Cox, J. Little, D. O'Shea, Ideals, Varieties, and Algorithms, Undergraduate Texts in Mathematics, 2nd ed., Springer, 1997. [4] H. Li, Noncommutative Grobner Bases and Filtered-Graded Transfer, Lecture Notes in Math- ematics 1795, Springer, 2002. [5] A. Kandri-Rody, V. Weispfenning, Noncommutative Grobner Bases in Algebras of Solvable Type, J. Symb. Comp. 9 (1990), 1 -- 26. [6] V. Mazorchuk, Lectures on sl2(C)-modules, Imperial College Press, 2010. [7] M. Saito, B. Sturmfels, N. Takayama, Grobner Deformations of Hypergeometric Differential Equations, Algorithms and Computation in Mathematics 6, Springer, 2000. [8] N. Schwartz, Stability of Grobner Bases, J. Pure Appl. Algebra 53 (1988), 171 -- 186. [9] A. S. Sikora, Topology on the Spaces of Orderings of Groups, Bull. London Math. Soc. 36 (2004), 519 -- 526. [10] B. Sturmfels, Grobner Bases and Convex Polytopes, University Lecture Series 8, Amer. Math. Soc., 1996. [11] V. Weispfenning, Constructing Universal Grobner Bases, appeared in Lecture Notes in Com- puter Sciences 356, Springer, 1989. Universitat Zurich, Institut fur Mathematik, Winterthurerstr. 190, CH-8057 Zurich E-mail address: [email protected]
1808.09248	2	1808	2018-08-31T20:08:45	Tribonacci and Tribonacci-Lucas Sedenions	[ "math.RA" ]	The sedenions form a 16-dimensional Cayley-Dickson algebra. In this paper, we introduce the Tribonacci and Tribonacci-Lucas sedenions. Furthermore, we present some properties of these sedenions and derive relationships between them.	math.RA	math	Tribonacci and Tribonacci-Lucas Sedenions Y UKSEL SOYKAN Zonguldak Bulent Ecevit University, Department of Mathematics, Art and Science Faculty, 67100, Zonguldak, Turkey e-mail: yuksel [email protected] Abstract. The sedenions form a 16-dimensional Cayley-Dickson algebra. In this paper, we introduce the Tribonacci and Tribonacci-Lucas sedenions. Furthermore, we present some properties of these sedenions and derive relationships between them. 2010 Mathematics Subject Classification. 11B39, 11B83, 17A45, 05A15. Keywords. Tribonacci numbers, sedenions, Tribonacci sedenions, Tribonacci-Lucas sedenions. 1. Introduction Tribonacci sequence {Tn}n≥0 and Tribonacci-Lucas sequence {Kn}n≥0 are defined by the third-order recurrence relations (1.1) and (1.2) Tn = Tn−1 + Tn−2 + Tn−3, T0 = 0, T1 = 1, T2 = 1, Kn = Kn−1 + Kn−2 + Kn−3, K0 = 3, K1 = 1, K2 = 3 respectively. Tribonacci concept was introduced by 14 year old student M. Feinberg [18] in 1963. Basic properties of it is given in [4], [35], [36], [39] and Binet formula for the nth number is given in [37]. See also [6],[13],[17], [34], [33], [38], [40]. The sequences {Tn}n≥0 and {Kn}n≥0 can be extended to negative subscripts by defining and T−n = −T−(n−1) − T−(n−2) + T−(n−3) K−n = −K−(n−1) − K−(n−2) + K−(n−3) for n = 1, 2, 3, ... respectively. Therefore, recurrences (1.1) and (1.2) hold for all integer n. 1 2 Y UKSEL SOYKAN It is well known that usual Tribonacci and Tribonacci-Lucas numbers can be expressed using Binet's formulas (1.3) and Tn = αn+1 (α − β)(α − γ) + βn+1 (β − α)(β − γ) + γn+1 (γ − α)(γ − β) Kn = αn + βn + γn respectively, where α, β and γ are the roots of the cubic equation x3 − x2 − x − 1 = 0. Moreover, α = β = γ = where , , 3 1 + 3p19 + 3√33 + 3p19 − 3√33 1 + ω 3p19 + 3√33 + ω2 3p19 − 3√33 1 + ω2 3p19 + 3√33 + ω 3p19 − 3√33 ω = −1 + i√3 = exp(2πi/3), 3 3 2 is a primitive cube root of unity. Note that we have the following identities α + β + γ = 1, αβ + αγ + βγ = −1, αβγ = 1. We can give Binet's formulas of the Tribonacci and Tribonacci-Lucas numbers for the negative subscripts: for n = 1, 2, 3, ... we have T−n = α−n+1 (α − β)(α − γ) + β−n+1 (β − α)(β − γ) + γ−n+1 (γ − α)(γ − β) K−n = α−n + β−n + γ−n. The generating functions for the Tribonacci sequence {Tn}n≥0 and Tribonacci-Lucas sequence {Kn}n≥0 and are ∞Xn=0 Tnxn = x 1 − x − x2 − x3 and ∞Xn=0 Knxn = 3 − 2x − x2 1 − x − x2 − x3 . In this paper, we define Tribonacci and Tribonacci-Lucas sedenions in the next section and give some properties of them. Before giving their definition, we present some information on Cayley-Dickson algebras. The algebras C (complex numbers), H (quaternions), and O (octonions) are real division algebras ob- tained from the real numbers R by a doubling procedure called the Cayley-Dickson Process (Construction). By doubling R (dim 20 = 1), we obtain the complex numbers C (dim 21 = 2); then C yields the quaternions H (dim 22 = 4); and H produces octonions O (dim 23 = 8). The next doubling process applied to O then TRIBONACCI AND TRIBONACCI-LUCAS SEDENIONS 3 produces an algebra S (dim 24 = 16) called the sedenions. This doubling process can be extended beyond the sedenions to form what are known as the 2n-ions (see for example [23], [29], [2]). Next, we explain this doubling process. The Cayley-Dickson algebras are a sequence A0, A1, ... of non-associative R-algebras with involution. The term "conjugation" can be used to refer to the involution because it generalizes the usual conjugation on the complex numbers. A full explanation of the basic properties of Cayley-Dickson algebras, see [2]. Cayley-Dickson algebras are defined inductively. We begin by defining A0 to be R. Given An−1, the algebra An is defined additively to be An−1 × An−1. Conjugation in An is defined by and multiplication is defined by (a, b) = (a,−b) (a, b)(c, d) = (ac − db, da + bc) and addition is defined by componentwise as (a, b) + (c, d) = (a + c, b + d). Note that An has dimension 2n as an R−vector space. If we set, as usual, kxk =pRe(xx) for x ∈ An then xx = xx = kxk2 . Now, suppose that B16 = {ei ∈ S : i = 0, 1, 2, ..., 15} is the basis for S, where e0 is the identity (or unit) and e1, e2, ..., e15 are called imaginaries. Then a sedenion S ∈ S can be written as S = 15Xi=0 aiei = a0 + aiei 15Xi=1 where a0, a1, ..., a15 are all real numbers. Here a0 is called the real part of S and P15 imaginary part. i=1 aiei is called its Addition of sedenions is defined as componentwise and multiplication is defined as follows: if S1, S2 ∈ S then we have (1.4) S1S2 = 15Xi=0 aiei! 15Xi=0 biei! = 15Xi,j=0 aibj(eiej). By setting i ≡ ei where i = 0, 1, 2, ..., 15, the multiplication rule of the base elements ei ∈ B16 can be summarized as in the following Table (see [8] and [3]). From the above table, we can see that: e0ei = eie0 = ei; eiei = −e0 for i 6= 0; eiej = −ejei for i 6= j and i, j 6= 0. The operations requiring for the multiplication in (1.4) are quite a lot. The computation of a sedenion multiplication (product) using the naive method requires 256 multiplications and 240 additions, while an algorithm which is given in [5] can compute the same result in only 122 multiplications (or multipliers -- in hardware implementation case) and 298 additions, for details see [5]. 4 Y UKSEL SOYKAN Figure 1. Multiplication table for sedenions imaginary units The problem with Cayley-Dickson Process is that each step of the doubling process results in a progressive loss of structure. R is an ordered field and it has all the nice properties we are so familiar with in dealing with numbers like: the associative property, commutative property, division property, self-conjugate property, etc. When we double R to have C; C loses the self-conjugate property (and is no longer an ordered field), next H loses the commutative property, and O loses the associative property. When we double O to obtain S; S loses the division property. It means that S is non-commutative, non-associative, and have a multiplicative identity element e0 and multiplicative inverses but it is not a division algebra because it has zero divisors; this means that two non-zero sedenions can be multiplied to obtain zero: an example is (e3 + e10)(e6 − e15) = 0 and the other example is (e2 − e14)(e3 + e15) = 0, see [8]. The algebras beyond the complex numbers go by the generic name hypercomplex number. All hyper- complex number systems after sedenions that are based on the Cayley -- Dickson construction contain zero divisors. Note that there is another type of sedenions which is called conic sedenions or sedenions of Charles Muses, as they are also known, see [26], [27], [30] for more information. The term sedenion is also used for other 16-dimensional algebraic structures, such as a tensor product of two copies of the biquaternions, or the algebra of 4 by 4 matrices over the reals. In the past, non-associative algebras and related structures with zero divisors have not been given much attention because they did not appear to have any useful applications in most mathematical subjects. Recently, however, a lot of attention has been centred by theoretical physicists on the Cayley-Dickson algebras O (octonions) and S (sedenions) because of their increasing usefulness in formulating many of the new theories of elementary particles. In particular, the octonions O (which is the only non-associative normed division algebra over the reals; see for example [1] and [31]) has been found to be involved in so many unexpected TRIBONACCI AND TRIBONACCI-LUCAS SEDENIONS 5 areas (such as topology, quantum theory, Clifford algebras, etc.) and sedenions appear in many areas of science like linear gravity and electromagnetic theory. Briefly S, the algebra of sedenions, have the following properties: • S is a 16 dimensional non-associative and non-commutative (Carley-Dickson) algebra over the reals, • S is not a composition algebra or division algebra because of its zero divisors, • S is a non-alternative algebra, i.e., if S1 and S2 are sedenions the rules S2 1 S2 = S1(S1S2) and S1S2 2 = (S1S2)S2 do not always hold, • S is a power-associative algebra, i.e., if S is an sedenion then SnSm = Sn+m. 2. The Tribonacci and Tribonacci-Lucas Sedenions and Their Generating Functions and Binet Formulas In this section we define Tribonacci and Tribonacci-Lucas sedenions and give generating functions and Binet formulas for them. First, we give some information about quaternion sequences, octonion sequences and sedenion sequences from literature. Horadam [22] introduced nth Fibonacci and nth Lucas quaternions as and Qn = Fn + Fn+1e1 + Fn+2e2 + Fn+3e3 = Rn = Ln + Ln+1e1 + Ln+2e2 + Ln+3e3 = Fn+ses Ln+ses 3Xs=0 3Xs=0 respectively, where Fn and Ln are the nth Fibonacci and Lucas numbers respectively. He also defined generalized Fibonacci quaternion as Pn = Hn + Hn+1e1 + Hn+2e2 + Hn+3e3 = Hn+ses 3Xs=0 where Hn is the nth generalized Fibonacci number (which is now called Horadam number) by the recursive relation H1 = p, H2 = p + q, Hn = Hn−1 + Hn−2 (p and q are arbitrary integers). Many other generalization of Fibonacci quaternions has been given, see for example Halici and Karata¸s [21], and Polatlı [32]. Cerda-Morales [9] defined and studied the generalized Tribonacci quaternion sequence that includes the previously introduced Tribonacci, Padovan, Narayana and third order Jacobsthal quaternion sequences. She defined generalized Tribonacci quaternion as Qv,n = Vn + Vn+1e1 + Vn+2e2 + Vn+3e3 = Vn+ses 3Xs=0 where Vn is the nth generalized Tribonacci number defined by the third-order recurrance relations here V0 = a, V1 = b, V2 = c are arbitrary integers and r, s, t are real numbers. Vn = rVn−1 + sVn−2 + tVn−3, 6 Y UKSEL SOYKAN Various families of octonion number sequences (such as Fibonacci octonion, Pell octonion, Jacobsthal octonion; and third order Jacobsthal octonion) have been defined and studied by a number of authors in many different ways. For example, Ke¸cilioglu and Akku¸s [24] introduced the Fibonacci and Lucas octonions as and Qn = Rn = Fn+ses Ln+ses 7Xs=0 7Xs=0 respectively, where Fn and Ln are the nth Fibonacci and Lucas numbers respectively. In [15], C¸ imen and Ipek introduced Jacobsthal octonions and Jacobsthal-Lucas octonions. In [10], Cerda-Morales introduced third order Jacobsthal octonions and also in [12], she defined and studied tribonacci-type octonions. A number of authors have been defined and studied sedenion number sequences (such as second order sedenions: Fibonacci sedenion, k-Pell and k-Pell -- Lucas sedenions, Jacobsthal and Jacobsthal-Lucas sede- nions). For example, Bilgici, Toke¸ser and Unal [3] introduced the Fibonacci and Lucas sedenions as and 15Xs=0 15Xs=0 bFn = bLn = Fn+ses Ln+ses respectively, where Fn and Ln are the nth Fibonacci and Lucas numbers respectively. In [7], Catarino introduced k-Pell and k-Pell -- Lucas sedenions. In [14], C¸ imen and Ipek introduced Jacobsthal and Jacobsthal- Lucas sedenions. Gul [20] introduced the k-Fibonacci and k-Lucas trigintaduonions as and T Fk,n = T Lk,n = Fk,n+ses Lk,n+ses 31Xs=0 31Xs=0 respectively, where Fk,n and Lk,n are the nth k-Fibonacci and k-Lucas numbers respectively. We now define Tribonacci and Tribonacci-Lucas sedenions over the sedenion algebra S. The nth Tri- bonacci sedenion is (2.1) and the nth Tribonacci-Lucas sedenion is 15Xs=0 bTn = 15Xs=0 bKn = Tn+ses = Tn + Kn+ses = Kn + Tn+ses Kn+ses. 15Xs=1 15Xs=1 TRIBONACCI AND TRIBONACCI-LUCAS SEDENIONS 7 For all non-negative integer n, it can be easily shown that The sequences {bTn}n≥0 and {bKn}n≥0 can be defined for negative values of n by using the recurrences (2.2) and (2.3) to extend the sequence backwards, or equivalently, by using the recurrences bTn = bTn−1 + bTn−2 + bTn−3 bKn = bKn−1 + bKn−2 + bKn−3. bT−n = −bT−(n−1) − bT−(n−2) + bT−(n−3) bK−n = −bK−(n−1) − bK−(n−2) + bK−(n−3), respectively. Thus, the recurrences (2.2) and (2.3) holds for all integer n. The conjugate of bTn and bKn are defined by Tn+ses = Tn − Tn+1e1 − Tn+2e2 − ... − Tn+15e15 (2.2) and (2.3) and and and respectively. respectively. The norms of nth Tribonacci and Tribonacci-Lucas sedenions are Kn+ses = Kn − Kn+1e1 − Kn+2e2 − ... − Kn+15e15 15Xs=1 bTn = Tn − 15Xs=1 2 bKn = Kn − (cid:13)(cid:13)(cid:13)bTn(cid:13)(cid:13)(cid:13) = N 2(bTn) = bTnbTn = bTnbTn = T 2 (cid:13)(cid:13)(cid:13)bKn(cid:13)(cid:13)(cid:13) = N 2(bKn) = bKnbKn = bKnbKn = K 2 2 n + T 2 n+1 + ... + T 2 n+15 n + K 2 n+1 + ... + K 2 n+15 Now, we will state Binet's formula for the Tribonacci and Tribonacci-Lucas sedenions and in the rest of the paper we fixed the following notations. αses, βses, γses. 15Xs=0 15Xs=0 15Xs=0 bα = bβ = bγ = 8 Y UKSEL SOYKAN Theorem 2.1. For any integer n, the nth Tribonacci sedenion is bααn+1 (α − β)(α − γ) + bββn+1 (β − α)(β − γ) + bγγn+1 (γ − α)(γ − β) bTn = and the nth Tribonacci-Lucas sedenion is (2.4) (2.5) bKn =bααn +bββn +bγγn. Proof. Repeated use of (1.3) in (2.1) enable us to write for bα = P15 bγ =P15 s=0 γses : βn+1+ses 15Xs=0 bTn = = Tn+ses = bααn+1 (α − β)(α − γ) (α − β)(α − γ) 15Xs=0(cid:18) αn+1+ses bββn+1 (β − α)(β − γ) + + (β − α)(β − γ) + bγγn+1 (γ − α)(γ − β) . Similarly, we can obtain (2.5). s=0 βses and s=0 αses, bβ = P15 (γ − α)(γ − β)(cid:19) γn+1+ses + We can give Binet's formula of the Tribonacci and Tribonacci-Lucas sedenions for the negative subscripts: for n = 1, 2, 3, ... we have bT−n = and respectively. bγγ−n+1 (γ − α)(γ − β) (α − β)(α − γ) + + bββ−n+1 (β − α)(β − γ) bαα−n+1 bK−n =bαα−n +bββ−n +bγγ−n, The next theorem gives us an alternatif proof of the Binet's formula for the Tribonacci and Tribonacci- Lucas sedenions. For this, we need the quadratik approximation of {Tn}n≥0 and {Kn}n≥0: Lemma 2.2. For all integer n, we have (a): (b): ααn+2 = Tn+2α2 + (Tn+1 + Tn)α + Tn+1, ββn+2 = Tn+2β2 + (Tn+1 + Tn)β + Tn+1, γγn+2 = Tn+2γ2 + (Tn+1 + Tn)γ + Tn+1. P αn+2 = Kn+2α2 + (Kn+1 + Kn)α + Kn+1, Qβn+2 = Kn+2β2 + (Kn+1 + Kn)β + Kn+1, Rγn+2 = Kn+2γ2 + (Kn+1 + Kn)γ + Kn+1, TRIBONACCI AND TRIBONACCI-LUCAS SEDENIONS 9 where P = 3 − (β + γ) + 3βγ, Q = 3 − (α + γ) + 3αγ, R = 3 − (α + βγ) + 3αβ. Proof. See [11] or [12]. Alternatif Proof of Theorem 2.1: Note that α2bTn+2 + α(bTn+1 + bTn) + bTn+1 = α2(Tn+2 + Tn+3e1 + ... + Tn+17e15) +α((Tn+1 + Tn) + (Tn+2 + Tn+1)e1 + ... + (Tn+16 + Tn+15)e15) +(Tn+1 + Tn+2e1 + ... + Tn+16e15) = α2Tn+2 + α(Tn+1 + Tn) + Tn+1 + (α2Tn+3 + (Tn+2 + Tn+1) + Tn+2)e1 +(α2Tn+4 + (Tn+3 + Tn+2) + Tn+3)e2 ... +(α2Tn+17 + (Tn+16 + Tn+15) + Tn+16)e15. From the identity αn+3 = Tn+2α2 + (Tn+1 + Tn)α + Tn+1 for n-th Tribonacci number Tn, we have (2.6) Similarly, we obtain (2.7) and (2.8) (2.9) (2.10) Substracting (2.7) from (2.6), we have Similarly, substracting (2.8) from (2.6), we obtain α2bTn+2 + α(bTn+1 + bTn) + bTn+1 =bααn+3. β2bTn+2 + β(bTn+1 + bTn) + bTn+1 =bββn+3 γ2bTn+2 + γ(bTn+1 + bTn) + bTn+1 =bγγn+3. (α + β)bTn+2 + (bTn+1 + bTn) = bααn+3 −bββn+3 (α + γ)bTn+2 + (bTn+1 + bTn) = bααn+3 −bγγn+3 α − β α − γ . . 10 Y UKSEL SOYKAN Finally, substracting (2.10) from (2.9), we get bTn+2 = = = 1 α − γ ! α − β bααn+3 −bββn+3 − bααn+3 −bγγn+3 α − β bββn+3 bγγn+3 bααn+3 (α − β)(α − γ) − bββn+3 bααn+3 bγγn+3 (β − α)(β − γ) (α − β)(β − γ) + + + (γ − α)(γ − β) (γ − α)(γ − β) . (α − β)(α − γ) So this proves (2.4). Similarly we obtain (2.5). Next, we present generating functions. Theorem 2.3. The generating functions for the Tribonacci and Tribonacci-Lucas sedenions are (2.11) and (2.12) respectively. g(x) = g(x) = 1 − x − x2 − x3 ∞Xn=0bTnxn = bT0 + (bT1 − bT0)x + bT−1x2 ∞Xn=0 bKnxn = bK0 + (bK1 − bK0)x + bK−1x2 1 − x − x2 − x3 Proof. Define g(x) =P∞ n=0 bTnxn. Note that x2g(x) = xg(x) = g(x) = bT0+ bT1x+ bT2x2+ bT3x3+ bT4x4+ bT5x5 + ...+ bTnxn + ... bT0x+ bT1x2+ bT2x3+ bT3x4+ bT4x5 + ...+ bTn−1xn + ... bT0x2+ bT1x3+ bT2x4+ bT3x5 + ...+ bTn−2xn + ... bT0x3+ bT1x4+ bT2x5 + ...+ bTn−3xn + ... x3g(x) = Using above table and the recurans bTn = bTn−1 + bTn−2 + bTn−3 we have g(x) − xg(x) − x2g(x) − x3g(x) = bT0 + (bT1 − bT0)x + (bT2 − bT1 − bT0)x2 + (bT3 − bT2 − bT1 − bT0)x3 + (bT4 − bT3 − bT2 − bT1)x4 + ... + (bTn − bTn−1 − bTn−2 − bTn−3+)xn + ... = bT0 + (bT1 − bT0)x + (bT2 − bT1 − bT0)x2. It follows that g(x) = bT0 + (bT1 − bT0)x + (bT2 − bT1 − bT0)x2 1 − x − x2 − x3 . Since bT2 − bT1 − bT0 = bT−1, the generating functions for the Tribonacci sedenion is g(x) = bT0 + (bT1 − bT0)x + bT−1x2 1 − x − x2 − x3 . Similarly, we can obtain (2.11). In the following theorem we present another forms of Binet formulas for the Tribonacci and Tribonacci- Lucas sedenions using generating functions. TRIBONACCI AND TRIBONACCI-LUCAS SEDENIONS 11 Theorem 2.4. For any integer n, the nth Tribonacci sedenion is bTn = (α − γ) (α − β) ((α2 − α)bT0 + αbT1 + bT−1)αn ((γ2 − γ)bT0 + γbT1 + bT−1)γn (γ − β) (γ − α) + + and the nth Tribonacci-Lucas sedenion is bKn = (α − γ) (α − β) ((α2 − α)bK0 + αbK1 + bK−1)αn ((γ2 − γ)bK0 + γbK1 + bK−1)γn (γ − β) (γ − α) + + . ((β2 − β)bT0 + βbT1 + bT−1)βn (β − γ) (β − α) ((β2 − β)bK0 + βbK1 + bK−1)βn (β − γ) (β − α) Proof. We can use generating functions. Since the roots of the equation 1−x−x2−x3 = 0 are αβ, βγ, αγ and 1 − x − x2 − x3 = (1 − αx)(1 − βx)(1 − γx) we can write the generating function of bTn as = bT0 + (bT1 − bT0)x + bT−1x2 1 − x − x2 − x3 A g(x) B + + = bT0 + (bT1 − bT0)x + bT−1x2 (1 − αx)(1 − βx)(1 − γx) C = = = (1 − αx) A(1 − βx)(1 − γx) + B(1 − αx)(1 − γx) + C(1 − αx)(1 − βx) (1 − βx) (1 − γx) (1 − αx)(1 − βx)(1 − γx) (A + B + C) + (−Aβ − Aγ − Bα − Bγ − Cα − Cβ)x + (Aβγ + Bαγ + Cαβ)x2 . (1 − αx)(1 − βx)(1 − γx) We need to find A, B and C, so the following system of equations should be solved: A + B + C = bT0 −Aβ − Aγ − Bα − Bγ − Cα − Cβ = bT1 − bT0 Aβγ + Bαγ + Cαβ = bT−1 We find that α2 − αβ − αγ + βγ A = −αbT0 + αbT1 + bT−1 + α2bT0 B = −βbT0 + βbT1 + bT−1 + β2bT0 β2 − αβ + αγ − βγ C = −γbT0 + γbT1 + bT−1 + γ2bT0 γ2 + αβ − αγ − βγ = = = , , (α − γ) (α − β) ((α2 − α)bT0 + αbT1 + bT−1) ((β2 − β)bT0 + βbT1 + bT−1) ((γ2 − γ)bT0 + γbT1 + bT−1) (β − γ) (β − α) (γ − β) (γ − α) . 12 and Y UKSEL SOYKAN g(x) = = + (α − γ) (α − β) ∞Xn=0 ((α2 − α)bT0 + αbT1 + bT−1) (−γbT0 + γbT1 + bT−1 + γ2bT0)  ∞Xn=0 (γ − β) (γ − α) ((α2−α) bT0+α bT1+ bT−1)αn (α−γ)(α−β) αnxn + γnxn ∞Xn=0 (β − γ) (β − α) ((β2 − β)bT0 + βbT1 + bT−1)  xn. + ((β 2−β) bT0+β bT1+ bT−1)βn (β−γ)(β−α) + ((γ 2−γ) bT0+γ bT1+ bT−1)γn (γ−β)(γ−α) βnxn ∞Xn=0 Thus Binet formula of Tribonacci sedenion is bTn = (α − γ) (α − β) ((α2 − α)bT0 + αbT1 + bT−1)αn ((γ2 − γ)bT0 + γbT1 + bT−1)γn (γ − β) (γ − α) + + . ((β2 − β)bT0 + βbT1 + bT−1)βn (β − γ) (β − α) Similarly, we can obtain Binet formula of the Tribonacci-Lucas sedenion. If we compare Theorem 2.1 and Theorem 2.4 and use the definition of bTn, bKn, we have the following Remark showing relations between bT−1,bT0,bT1; bK−1, bK0, bK1 andbα,bβ,bγ. We obtain (b) and (d) after solving the system of the equations in (a) and (b) respectively. Remark 2.5. We have the following identities: (a): (b): α (α2 − α)bT0 + αbT1 + bT−1 (β2 − β)bT0 + βbT1 + bT−1 (γ2 − γ)bT0 + γbT1 + bT−1 β γ = bα = bβ = bγ 15Xs=0 T−1+ses = bT−1 = 15Xs=0 Tses = bT0 = 15Xs=0 T1+ses = bT1 = bα bαα bαα2 + + + bβ bββ bββ2 + + + bγ bγγ bγγ2 (α − β)(α − γ) (β − α)(β − γ) (γ − α)(γ − β) (α − β)(α − γ) (β − α)(β − γ) (γ − α)(γ − β) (α − β)(α − γ) (β − α)(β − γ) (γ − α)(γ − β) TRIBONACCI AND TRIBONACCI-LUCAS SEDENIONS 13 (c): (d): (α − γ) (α − β) ((α2 − α)bK0 + αbK1 + bK−1) ((β2 − β)bK0 + βbK1 + bK−1) ((γ 2 − γ)bK0 + γbK1 + bK−1) (β − γ) (β − α) (γ − β) (γ − α) = bα = bβ = bγ 15Xs=0 K−1+ses = bK−1 =bαα−1 +bββ−1 +bγγ−1 15Xs=0 Kses = bK0 =bα +bβ +bγ 15Xs=0 K1+ses = bK1 =bαα +bββ +bγγ. Using above Remark we can find bT2, bK2 as follows: bαα3 T2+ses = bT2 = bT1 + bT0 + bT−1 = 15Xs=0 (α − β)(α − γ) + bββ3 (β − α)(β − γ) + bγγ3 (γ − α)(γ − β) (2.13) and (2.14) 15Xs=0 K2+ses = bK2 = bK1 + bK0 + bK−1 =bαα2 +bββ2 +bγγ2. Of course, (2.13) and (2.14) can be found directly from (2.4) and (2.5). Now, we present the formulas which give the summation of the first n Tribonacci and Tribonacci-Lucas numbers. Lemma 2.6. For every integer n ≥ 0, we have (2.15) and (2.16) Ti = T0 + 1 2 (Tn+2 + Tn − 1) = 1 2 (Tn+2 + Tn − 1) nXi=0 Ki = Kn+2 + Kn 2 . nXi=0 Proof. (2.15) and (2.16) can be proved by mathematical induction easily. For a proof of (2.15) with a telescopik sum method see [16] or with a matrix diagonalisation proof, see [25] or see also [9]. For a proof of (2.16), see [19]. Since K0 = 3 and Kn+2 + Kn 2 . Ki = Kn+2 + Kn − 6 2 , it follows that nPi=1 Ki = nPi=0 14 Y UKSEL SOYKAN There is also a formula of the summation of the first n negative Tribonacci numbers: T−i = 1 2 nXi=1 (1 − T−n−1 − T−n+1). For a proof of the above formula, see Kuhapatanakul and Sukruan [28]. Next, we present the formulas which give the summation of the first n Tribonacci and Tribonacci-Lucas sedenions. Theorem 2.7. The summation formula for Tribonacci and Tribonacci-Lucas sedenions are (2.17) and (2.18) respectively, where nXi=0 bTi = nXi=0 bKi = 1 2 (bTn+2 + bTn + c1) 1 2 (bKn+2 + bKn + c2) c1 = −1 − e1 − 3e2 − 5e3 − 9e4 − 17e5 − 31e6 − 57e7 − 105e8 − 193e9 −355e10 − 653e11 − 1201e12 − 2209e13 − 4063e14 − 7473e15 and c2 = −6e1 − 8e2 − 14e3 − 28e4 − 50e5 − 92e6 − 170e7 − 312e8 − 574e9 −1056e10 − 1842e11 − 3572e12 − 6570e13 − 12084e14 − 22226e15. Proof. Using (2.1) and (2.15), we obtain nXi=0 bTi = nXi=0 nXi=0 = (T0 + ... + Tn) + e1(T1 + ... + Tn+1) Ti + e1 Ti+1 + e2 Ti+2 + ... + e15 nXi=0 Ti+15 nXi=0 and +e2(T2 + ... + Tn+2) + ... + e15(T15 + ... + Tn+15). 2 nXi=0 bTi = (Tn+2 + Tn − 1) + e1(Tn+3 + Tn+1 − 1 − 2T0) +e2(Tn+4 + Tn+3 − 1 − 2(T0 + T1)) ... +e15(Tn+17 + Tn+15 − 1 − 2(T0 + T1 + ... + T14)) = bTn+2 + bTn + c1 TRIBONACCI AND TRIBONACCI-LUCAS SEDENIONS 15 where c1 = −1 + e1(−1 − 2T0) + e2(−1 − 2(T0 + T1)) + ... + e15(−1 − 2(T0 + ... + T14)). Hence We can compute c1 as 1 2 nXi=0 bTi = (bTn+2 + bTn + c1). c1 = −1 − e1 − 3e2 − 5e3 − 9e4 − 17e5 − 31e6 − 57e7 − 105e8 − 193e9 −355e10 − 653e11 − 1201e12 − 2209e13 − 4063e14 − 7473e15. This proves (2.17). Similarly we can obtain (2.18). 3. Some Identities For The Tribonacci and Tribonacci-Lucas Sedenions In this section, we give identities about Tribonacci and Tribonacci-Lucas sedenions. Theorem 3.1. For n ≥ 1, the following identities hold: (α − β)(α − γ) (a): bKn = 3bTn+1 − 2bTn − bTn−1, (b): bTn + bTn = 2Tn, bKn + bKn = 2Kn, + bβ (β + 1) βn+1 (c): bTn+1 + bTn = bα (α + 1) αn+1 + bγ (γ + 1) γn+1 (d): bKn+1 + bKn =bα (α + 1) αn +bβ (β + 1) βn +bγ (γ + 1) γn, + bββ(1 + β)n nPi=0(cid:0)n i(cid:1)bFi = bαα(1 + α)n + bγγ(1 + γ)n nPi=0(cid:0)n i(cid:1)bKi =bα(1 + α)n +bβ(1 + β)n +bγ(1 + γ)n. (β − α)(β − γ) (γ − α)(γ − β) (α − β)(α − γ) (β − α)(β − γ) (γ − α)(γ − β) (e): (f ): , , Proof. Since Kn = 3Tn+1 − 2Tn− Tn−1 (see for example [11]), (a) follows. The others can be established easily. Theorem 3.2. For n ≥ 0, m ≥ 3, we have (a): bTm+n = Tm−1bTn+2 + (Tm−2 + Tm−3)bTn+1 + Tm−2bTn, (b): bTm+n = Tm+2bTn−1 + (Tm+1 + Tm)bTn−2 + Tm+1bTn−3, (c): bKm+n = Kn−1bTm+2 + (bTm+1 + bTm)Kn−2 + Kn−3bTm+1, (d): bKm+n = Km+2bTn−1 + (Km+1 + Km)bTn−2 + Km+1bTn−3. Proof. (a) and (d) can be proved by strong induction on m and (c) can be proved by strong induction on n. For (b), replace n with n − 3 and m with m + 3 in (a). References [1] Baez, J., The octonions, Bull. Amer. Math. Soc. 39 (2), 145-205, 2002. [2] Biss, D.K., Dugger, D., and Isaksen, D.C., Large annihilators in Cayley-Dickson algebras, Communication in Algebra, 2008. [3] Bilgici, G., Toke¸ser, U,. Unal, Z., Fibonacci and Lucas Sedenions, Journal of Integer Sequences, Article 17.1.8, 20, 1-11. 2017. 16 Y UKSEL SOYKAN [4] Bruce, I., A modified Tribonacci sequence, The Fibonacci Quarterly, 22 : 3, pp. 244 -- 246, 1984. [5] Cariow, A., Cariowa G., An Algorithm for Fast Multiplication of Sedenios, Information Proccessing Letters, Volume 113, Issue, 9, 324-331, 2013. [6] Catalani, M., Identities for Tribonacci-related sequences - arXiv preprint, https://arxiv.org/pdf/math/0209179.pdf math/0209179, 2002. [7] Catarino, P., k-Pell, k-Pell -- Lucas and modified k-Pell sedenions, Asian-European Journal of Mathematics, 2018 [8] Cawagas, E.R., On the Structure and Zero Divisors of the Cayley-Dickson Sedenion Algebra, Discussiones Mathematicae, General Algebra and Applications 24, 251-265, 2004. [9] Cerda-Morales, G., On a Generalization for Tribonacci Quaternions, Mediterranean Journal of Mathematics, 14:239, 1 -- 12, 2017. [10] Cerda-Morales, G., The Third Order Jacobsthal Octonions: Some Combinatorial Properties, arXiv:1710.00602v1, [Math.RA], 2 Oct 2017. [11] Cerda-Morales, G., A Three-By-Three Matrix Representation of a Generalized Tribonacci sequence, arXiv:1807.03340v1, [Math.CO], 9 Jul 2018. [12] Cerda-Morales, G., The Unifying Formula for All Tribonacci-Type Octonions Sequences and Their Properties, arXiv:1807.04140v1 [math.RA], [Math.CO], 10 Jul 2018. [13] Choi, E., Modular tribonacci Numbers by Matrix Method, J. Korean Soc. Math. Educ. Ser. B: Pure Appl. Math. Volume 20, Number 3 (August 2013), Pages 207 -- 221, 2013. [14] C¸ imen, C., Ipek, A., On Jacobsthal and Jacobsthal-Lucas Sedenios and Several Identities Involving These Numbers, Mathematica Aeterna, Vol. 7, No.4, 447-454, 2017. [15] C¸ imen, C., Ipek, A, On Jacobsthal and Jacobsthal-Lucas Octonions, Mediterr. J. Math., 14:37, 1-13, 2017. [16] Devbhadra, S. V., Some Tribonacci Identities, Mathematics Today Vol.27(Dec-2011) 1-9, 2011. [17] Elia, M., Derived Sequences, The Tribonacci Recurrence and Cubic Forms, The Fibonacci Quarterly, 39:2, pp. 107-115, 2001. [18] Feinberg, M., Fibonacci -- Tribonacci, The Fibonacci Quarterly, 1 : 3 (1963) pp. 71 -- 74, 1963. [19] Frontczak, R., Sums of Tribonacci and Tribonacci-Lucas Numbers, International Journal of Mathematical Analysis, Vol. 12, No. 1, 19-24, 2018. [20] Gul, K., On k-Fibonacci and k-Lucas Trigintaduonions, International Journal of Contemporary Mathematical Sciences, Vol. 13, no. 1, 1 - 10, 2018. [21] Halici, S., Karata¸s, A., On a Generalization for Fibonacci Quaternions. Chaos Solitons and Fractals 98, 178 -- 182, 2017. [22] Horadam, A. F., Complex Fibonacci Numbers and Fibonacci quaternions, Amer. Math. Monthly 70, 289 -- 291, 1963. [23] Imaeda, K., Imaeda, M., Sedenions: algebra and analysis, Applied Mathematics and Computation, 115, 77-88, 2000. [24] Ke¸cilioglu O, Akku¸s, I., The Fibonacci Octonions, Adv. Appl. Clifford Algebr. 25, 151 -- 158, 2015. [25] Kılı¸c, E., Tribonacci Sequences with Certain Indices and Their Sums, Ars. Comb., 86, 13-22, 2008. [26] Koplinger, J., Signature of gravity in conic sedenions, Applied Mathematics and Computation, 188, 942-947, 2007. [27] Koplinger, J., Gravity and eletromagnetism an conic sedenions, Applied Mathematics and Computation, 188, 948-953, 2007. [28] Kuhapatanakul, K., Sukruan, L., The Generalized Tribonacci Numbers With Negative Subscripts, Integer, 14, 1-6, 2014. [29] Moreno, G., The zero divisors of the Cayley-Dickson algebras over the real numbers, Bol. Soc. Mat. Mexicana (3) 4 , 13-28,1998. [30] Muses, C.A., Hypernumber and quantum field theory with a summary of physically applicable hypernumber arithmetics and their geometrics, Applied Mathematics and Computation, 6, 63-94, 1980. TRIBONACCI AND TRIBONACCI-LUCAS SEDENIONS 17 [31] Okubo, S., Introduction to Octonions and Other Non-Associative Algebras in Physics, Cambridge University Press, Cam- bridge 1995. [32] Polatlı, E., A Generalization of Fibonacci and Lucas Quaternions, Advances in Applied Clifford Algebras, 26 (2), 719-730, 2016. [33] Lin, P. Y., De Moivre-Type Identities For The Tribonacci Numbers, The Fibonacci Quarterly, 26, pp. 131-134, 1988. [34] Pethe, S., Some Identities, The Fibonacci Quarterly, 26, pp. 144 -- 246, 1988. [35] Scott, A., Delaney, T., Hoggatt Jr., V., The Tribonacci sequence, The Fibonacci Quarterly, 15:3, pp. 193 -- 200, 1977. [36] Shannon, A., Tribonacci numbers and Pascal's pyramid, The Fibonacci Quarterly, 15:3, pp. 268-275, 1977. [37] Spickerman, W., Binet's formula for the Tribonacci sequence, The Fibonacci Quarterly, 20, pp.118 -- 120, 1981. [38] Spickerman, W., Joyner, R. N., Binets's formula for the Recursive sequence of Order K, The Fibonacci Quarterly, 22, pp.327-331, 1984. [39] Yalavigi, C. C., Properties of Tribonacci numbers, The Fibonacci Quarterly, 10 : 3, pp. 231 -- 246, 1972. [40] Yilmaz, N, and N. Taskara, N., Tribonacci and Tribonacci-Lucas Numbers via the Determinants of Special Matrices, Applied Mathematical Sciences, 8, no. 39, 1947-1955, 2014. This figure "tribonacci1.gif" is available in "gif"(cid:10) format from: http://arxiv.org/ps/1808.09248v2
1912.06674	1	1912	2019-12-13T19:42:21	Toward an intuitive understanding of the structure of near-vector spaces	[ "math.RA" ]	In this paper, we analyze the definition Andr\'e proposed for near-vector spaces to make it more transparent. We also study the class of near-vector spaces over division rings and give a characterization of regularity that gives a new insight into the decomposition of near-vector spaces into regular subspaces. We explicitly describe span and deduce a new characterization of subspaces.	math.RA	math	Toward an intuitive understanding of the structure of near-vector spaces K-T Howell & S Marques∗ Department of Mathematical Sciences, Stellenbosch University, Stellenbosch, 7600, e-mail: [email protected] & [email protected] South Africa Abstract: In this paper we analyse the definition Andr´e proposed for near-vector spaces to make it more transparent. We also study the class of near-vector spaces over division rings and give a characterisation of regularity that gives a new insight into the decomposition of near-vector spaces into regular subspaces. We explicitly describe span and deduce a new characterisation of subspaces. Keywords: Near-vector spaces; Nearrings; Nearfields; Division rings; Span; Subspaces; Linear Algebra 2010 Mathematics Subject Classification: 16Y30; 12K05 1 Introduction The notion of a structure with less linearity than a traditional vector space has been studied by a number of authors. First Beidleman [2] used near-ring modules to construct a near-vector space; whereas Andr´e [1] used an additive group together with a set of endomorphisms of the group, satisfying certain conditions. Next Karzel [6] defined a near-vector space structure mimicking a vector space structure but without the scalars acting as endomorphisms on the underlying group. In the first part of the paper we analyse Andr´e's definition and in doing so we attempt to highlight how Andr´e's definition seems to be the most suitable and natural definition to work with since it allows a lot of flexibility in the structure. We state a structural Lemma 3.4 for general vector spaces that hopefully reveals the connection between vector spaces and near-vector spaces and how the generalisation affects the structure. We also study one of the main new features of near-vector spaces, namely the non-unique additive structures on the underlying multiplicative group. In particular, we state the Key Lemma 3.10 for this paper that give a necessary condition for two additive structure to be the same. Several examples are exhibited to highlight the special features of near-vector spaces. In the second part of the paper we focus on constructions of near-vector spaces over division rings. We first focus on regularity. The main result (Theorem 4.2) of this section fully characterises the uniqueness of the induced additive structure into regularity, division ring and quasi-kernel structure. This permitted us to give an alternate proof of the decomposition into regular subspaces when the underlying set can be endowed with a division ring structure. We then focus on the structure of Span. Our main result in that section, Theorem 4.7, describes precisely the span of any one element set in V . Surprisingly, it reveals that Span and linear combinations agree in that setting as they do in classical linear algebra. Generalising this result of Span to a general set, we also deduce a very useful characterisation of subspaces which actually corresponds to the usual characterisation in classical linear algebra while the initial definition required a difficult check for the generating condition. 1 2 Preliminary material In this section we define concepts that are analogous to those that are central to traditional linear algebra. In [1] the notion of a near-vector space was defined as: Definition 2.1. ([1], Definition 4.1, p.9) A non-trivial near-vector space is a pair (V, A) which satisfies the following conditions: 1. (V, +) is a group and A is a set of endomorphisms of V ; 2. A contains the endomorphisms 0, id and −id; 3. A∗ = A\{0} is a subgroup of the group Aut(V ); 4. If αx = βx with x ∈ V and α, β ∈ A, then α = β or x = 0, i.e. A acts fixed point free on V ; 5. The quasi-kernel Q(V ) of V , generates V as a group. Here, Q(V ) = {x ∈ V ∀α, β ∈ A, ∃γ ∈ A such that αx + βx = γx}. Note that the trivial near-vector space has to be considered separately since it results in A∗ being empty. We will write Q(V )∗ for Q(V )\{0} throughout this paper and just Q if it does not cause confusion. Also, we note that we write scalars on the left and and as a result, make use of left near-fields. We will use F to denote a near-field and Fd its distributive elements. For more on near-fields we refer the reader to [11] and [12]. From Andr´e's definition, it is natural to define the concept of a homomorphism as follows: Definition 2.2. ([9], Definition 3.2, p.57) We say that two near-vector spaces (V1, A1) and (V2, A2) are homomorphic (written (V1, A1) ∼= (V2, A2)) if there are group homomorphisms θ : (V1, +) → (V2, +) and η : (A∗1,·) → (A∗2,·) such that θ(αx) = η(α)θ(x) for all x ∈ V1 and α ∈ A∗1. We will write homomorphisms as pairs (θ, η). This permits us to compare near-vector spaces. We will also need the notion of a subspace and span as they will have an important role to play as in traditional linear algebra. Definition 2.3. ([8], Definition 2.3, p.3) If (V, A) is a near-vector space and ∅ 6= V ′ ⊆ V is such that V ′ is the subgroup of (V, +) generated additively by AX = {ax x ∈ X, a ∈ A}, where X is an independent subset of Q(V ), then we say that (V ′, A) is a subspace of (V, A), or simply V ′ is a subspace of V if A is clear from the context. Definition 2.4. ([10], Definition 3.2, p.3235) Let (V, A) be a near-vector space, then the span of a set S of vectors is defined to be the intersection W of all subspaces of V that contain S, denoted span S. It is straightforward to verify that W is a subspace, called the subspace spanned by S, or conversely, S is called a spanning set of W and we say that S spans W . Moreover, if we define span ∅ = {0}, then it is not difficult to check that span S is the set of all possible linear combinations of S if S ⊆ Q(V ). However; if S contains elements outside of the quasi-kernel then it is not clear that these two coincide (See Theorem 4.7 ). For this reason we define: 2 Definition 2.5. Let V be a near-vector space. For every v ∈ V , we define the linear combinations of v as the set L(v) = {α1(v) + ··· + αt(v)t ∈ N and αi ∈ A, f or i ∈ {1, . . . , t}}. In a near-vector space linear independence is defined in terms of the elements of the quasi- kernel Q(V ). Definition 2.6. ([1], p.302) Let (V, A) be a near-vector space. We say that a set S ⊆ Q(V ) is linearly independent if for any v1,··· , vn in S and α1,··· , αn ∈ A such that if then α1v1 + ··· + αnvn = 0 α1 = ··· = αn = 0. Otherwise we say they are linearly dependent. Definition 2.7. ([1], p.303) Let (V, A) be a near-vector space, then an indepen- dent generating set for Q(V ) is called a basis of V and its cardinality is called the dimension of V . As for a vector space, one can prove that a near-vector space has a basis by showing that from the set of elements of Q(V ) generating V , one can always extract a basis. Thus any near-vector space admits a basis in Q(V ). It is routine as in linear algebra to prove that there is a well defined notion of dimension, i.e. if a near-vector space has a finite basis all the bases have the same number of elements. See [1] for more details on this and a proof that any near-vector space admits a basis by enlarging an existing linear independent set. The dimension of an element is defined as follows: Definition 2.8. ([10], Definition 3.5, p.3236) For v ∈ V \{0} we define the dimen- sion of v to be n = min(m ∈ N v = m Xi=1 αiui, with ui ∈ Q(V ) \ {0}, αi ∈ A \ {0}, i = 1, . . . , m) , we denote it by dim(v) = n and dim(v) = 0 if v is the zero vector. The concept of regularity is a central notion in the study of near-vector spaces. Andr´e called the regular spaces the building blocks of near-vector space theory. They happen to be well-behaved, as we will see. This led to the Decomposition Theorem, where any near-vector space is decomposed into regular parts. See Theorem 4.13, p. 3.6 in [1]. Definition 2.9. ([1] Definition 4.7, p.11) A near-vector space is regular if any two vectors of Q(V )\{0} are compatible, i.e. if for any two vectors u and v of Q(V ) there exists a λ ∈ A\{0} such that u + vλ ∈ Q(V ). Q(V ) = V . Note that every near vector space (V, A) with dim(V ) ≤ 1, is regular and has The addition in V naturally gives rise to an addition in A as follows: 3 Definition 2.10. ([1], p.299) Let (V, A) be a near-vector space and let v ∈ Q(V )\{0}. Define the operation +v on A by (α +v β)v := αv + βv (α, β ∈ A). definition will become clear to the reader in Section 3 and 4. With this addition, (A, +v,·) is a near-field (see [1]). The essentiality of this This addition gives rise to: Definition 2.11. ([1], Definition 2.6, p.301) Let (V, A) be a near-vector space and let u ∈ Q(V )\{0}. Define the kernel Ru(V ) = Ru of (V, A) by the set Ru := {v ∈ V (α +u β)v = αv + βv for every α, β ∈ A}. In Theorem 4.2 in Section 4, we will see how the notion of a kernel relates to regularity. 3 Analysis of the definition of Near-vector spaces Definition 3 might be a bit disconcerting to some. As a result, we decided to revisit the definition and try to understand how essential each assumption is to a good notion of what a near-vector space should be as the natural way to widely generalise a vector space after the introduction of near-fields. As part of this we will later see how the concepts of subspace, span and linear combinations interplay. Traditionally in linear algebra, a vector space is a field together with an abelian group endowed with an endomorphism action by the field on the abelian group, called the scalar multiplication. A near-vector space structure does not just result in a weakening of one of the distributive laws, but allows more generality. In the definition we do not begin by fixing a near-field, but instead fix a multiplicative group. From Andr´e's definition (Definition 3), one can construct a near-field such that the near-vector space can be viewed in the expected way mentioned above. Nevertheless, the choice of the near-field is not unique, as we will see later on. As result of a lack of uniqueness of the underlying near-field, the direct generalisation of the traditional construction is weaker than what is proposed by Andr´e. The essence of the geometry behind linear algebra stems from the existence of a coordinate system. The geometry resides in the notion of a basis which gives rise to unit vectors and coordinate axes which are one-dimensional subspaces. In order to reproduce this idea in a more general setting, we start with an additive group V and a set A such that there exist elements v ∈ V such that Av represents a set of coordinate axes and A becomes a near-field induced by (V, +) that is in bijection with Av. Cleverly, Andr´e noted that one does not need this property to hold for every v ∈ V to guarantee the existence of a coordinate system, one only needs to ensure that such v's generate V . As we will see this is guaranteed with the property that Q(V ) = {v ∈ V L(v) = Av} = {v ∈ V Av = span(v)} generates V (see Definition , 5.). Note that, for v ∈ Q(V ), we have L(v) = span(v) = Av. 4 Thus picking v ∈ Q(V ) guarantees that Av = span v = L(v). This shows that the coordinate axes Av are now subspaces of V, which we would expect. The fact that Q(V ) generates V allows us to see that V = span({v ∈ V L(v) = Av}). The notion of a basis will give us access to a coordinate system from some Av with v ∈ Q(V ). We are still expecting an underlying near-field structure. We will now explain how this structure can be revealed without being fixed in advance. To define a well-defined operation on A induced from that of V , we need for all α, β ∈ A that αv + βv = γv ∈ Av, for a unique γ ∈ A. The existence of γ is equivalent to v belonging to Q(V ) and its uniqueness is guaranteed by requiring fixed point freeness (see Definition , 4.). As a consequence of the existence and uniqueness of γ, if we fix a nonzero v in Q(V ), then for any α, β ∈ A, we can define an operation +v on A that sets α +v β to be this γ. What is left to show now is that the addition of V naturally induces a structure of a near-field on A. This mimics what we have for traditional vector spaces, where the underlying structure would be a field. The difference is that the underlying near-field structure is not fixed beforehand and non unique. We will use 0A and 0V to denote the identities of A and V, respectively. In order for the group structure of V to induce a group structure on A, we need that for all v ∈ Q(V ), 0A · v = 0V . Therefore 0A acts as the zero endomorphism on V which is generated by elements of Q(V ). In order to have a meaningful endomorphism action of A on V that induces a near-field structure on A, we will also need a multiplicative structure on A∗ = A\{0}. We will use 1A to denote the multiplicative identity of A. In order to ensure A acts as an endomorphism on V we will need 1A to act as the identity endomorphism Id. If v ∈ Q(V ), then 1Av − 1Av = 0V , by the group structure of (V, +), thus ((1A) +v (−1A))v = 0V . Now by the fixed point free property (1A) +v (−1A) = 0A. Thus −1A is the inverse of 1A, so that for all v ∈ V, (−1A)(v) = −v and −1A act as −Id in V . Moreover, since A induces an endomorphism action on V , if we use −a to denote the additive inverse of a ∈ A we then have (−1A · a)v = (−1A)(av) = −(av). As before we can prove that −a = −1A(a) is the additive inverse of A. So that a ∈ A, implies −a ∈ A. Note that as a consequence of −Id acting as an endomorphism of (V, +) we have that −(v + w) = −w − v = −v − w, for all v, w ∈ V therefore (V, +) must be an abelian group. Let us summarise what has been identified. As in traditional linear algebra we start with a group (V, +) and a set of endomorphisms of V , (A,·). To have access to a coordinate system (basis) formed from an underlying near- field structure so that we are able to study the geometric properties of the system, we need 5 1. A set of coordinate axes Av that generate (V, +) as an additive group. This is guaranteed by the property "The quasi-kernel Q(V ) of V , generates V as a group. " in Definition 3. 2. That (V, +) induces a near-field structure on (A,·), more precisely there exists a group operation +′ on A induced by the operation + on V such that (A, +′,·) is a near-field. The properties: "A contains the endomorphisms 0, id and −id", "A∗ = A\{0} is a subgroup of the group Aut(V );" and the fixed point free property precisely ensures that (A, +v,·) is a near- field for any v ∈ Q(V )\{0}. Remark 3.1. We note that Andr´e's definition (Definition 3 2.) requires the exis- tence of elements in A that acts as IdA and −IdA on V , that will imply that A will contain an element of multiplicative order 2. By Cauchy and Lagrange's Theorem, (A∗,·) is a group with even order if and only if there is a x 6= 1A ∈ A∗ such that x2 = 1A. For any v ∈ V, we have that x(v + xv) = v + xv since V is abelian and by the fixed point free property, since x 6= 1A, we have that v + xv = 0 and xv = −v. When the characteristic of V is not 2, this element x acts as −Id in A. However, when the characteristic of V is 2, x and 1A have the same action, contradicting the fixed point freeness. To conclude, when the characteristic of V is not 2, A∗ will have exactly one element of order 2 while in characteristic 2, A∗ cannot have any element of order 2. Note that if A∗ is finite will imply that if the characteristic of V is 2, A∗ will have even order, while if the characteristic is not 2, the order of A∗ has to be odd. From linear algebra the most basic example of a vector space is a field over itself, hence it would be essential to have that the additive group of a near-field be a near-vector space over its multiplicative group. Example 3.2. ([1]) Let F be a near-field that is not MC (Z2) (see [12], Proposition 8.1, p.249), then we have that: 1. (F, +) is a group and (F,·) is a set of endomorphisms of F ; 2. 0F acts as an endomorphism by assumption since F is zero-symmetric. Clearly 1F is an endomorphism while −1F ∈ F and is an endormorphism by Proposition 8.10, p. 151 in [12]; 3. It is clear that F ∗ = F\{0} is a subgroup of the group Aut(F ); 4. F acts fixed point free on itself since F has multiplicative inverses; 5. The quasi-kernel is the set Q(F ) = {(ki)λ λ ∈, ki ∈ Fd} = F (Theorem 4.4 p.304 in [1]). For any nonzero x ∈ F, {x} is a basis of (F, F ) of dimension 1. Thus (F, F ) is a near-vector space. In order to state the next lemma, we need some definitions. Definition 3.3. For I an index set, possibility infinite, we define A(I) = {a : I → Aa is zero for all but finitely many i ∈ I}, and the standard basis elements ei ∈ A(I) for any i ∈ I defined on I as ei(j) = δi,j =(cid:26) 0 1 if i 6= j otherwise. 6 The following lemma gives a way to visually compare near-field theory to classical linear algebra, emphasising the structure that completes the analysis above. Lemma 3.4. The following assertions are equivalent: 1. (V, A) is a near-vector space. 2. V ≃φ A(I), ei ∈ Q(A(I)) and the additive structure of the near-vector space (A(I), A) is given by +′ defined for any a, b ∈ A(I) point-wise by (a +′ b)(i) = a(i) +φ−1(ei) b(i) and the scalar multiplication induced by the action of A on V. Proof. We know that V admits a basis {v1,··· , vn}. It is not hard to prove then that 1. ⇒ 2. Then, we obtain the isomorphism φ by composing with the isomorphism V ≃ ⊕i∈I Avi. ψ : ⊕i∈IAvi → A(I) sending vi to ei and extended by linearity by sending any element in ⊕i∈I Avi, Pi∈I αivi to Pi∈I αiei. The scalar multiplication on A(I) is induced by the action of A on V component- wise via the bijections Avi ≃ A for any i ∈ I which sends αvi to α for i ∈ I. We note that this is well-defined by the fixed point free property. It is not difficult to prove that this will define a near- vector space isomorphism for the near-vector space structure identified in the statement. The converse is also clear. Remark 3.5. We suppose that the scalar multiplication on A(I) is induced by the given multiplicative structure of A as follows: ·′ is defined point-wise for any α ∈ A and a ∈ A(I) by: (α ·′ a)(i) = ηi(α)a(i), where ηi is a surjective map to ensure that any element of A acts on every compo- nent so that the entire action of A is transmitted to A(I). Then ηi is a near-field isomorphism from (A, +b,·) to (A, +,·), where b ∈ Q(A(I))\{0}. tivity follows from the action of A. Indeed, the injectivity is guaranteed by the fixed point freeness. The multiplica- Moreover, let b ∈ Q(A(I)) nonzero, for all α, β ∈ A, we have point-wise (α ·′ b + β ·′ b)(i) = ηi(α)b(i) + ηi(β)b(i) = (ηi(α) + ηi(β))b(i). In the other hand α ·′ b + β ·′ b = (α +b β) ·′ b = ηi(α +b β)b(i). So that by the point free property ηi(α) + ηi(β) = ηi(α +b β), as stated. When +a = +b = + for any a, b ∈ Q(A(I))\{0}, then we have a near-field automorphism of (A, +,·). 7 To emphasise the importance of −1 ∈ A and the quasi-kernel generating V beyond the induced structure of near-field in A, we give the following two examples: Example 3.6. Take and Then V = R A = R+ ∪ {0}. Q(V ) = R. Note that all the axioms of Definition 3 are satisfied, except that A contains no elements acting as −IdA in V . But despite the fact we cannot naturally obtain the structure of a near-field in R+ from the field structure of R as explained above.The set {−1, 1} is a generating set of R, but it is not linearly independent since for x, y ∈ R, both nonzero, implies that x = y. Thus we also do not have a basis. x + y(−1) = 0 A-groups have been studied. They meet all the requirements of Andr´e's definition, but assumption 5. of Definition 3 (see [1] for more on A-groups.) As an illustration we give the following example where only this assumption is not satisfied. It shows that the notion of an A-group would not lead to a good notion of what intuitively we could expect a near-vector space to be. Example 3.7. Take and Then V = Z 3Z ⊕ Z A = {−1, 0, 1}. Q(V ) = Z 3Z ⊕ {0}. The only missing assumption in the definition of near vector space is that Q(V ) does not generate V . The following example illustrates the flexibility that the definition of a homo- morphism allows. In the next section we will identify the important properties of near-vector spaces which will explain the unnatural phenomena in the example be- low. Example 3.8. We use +3 to denote the addition on R defined by for all x, y ∈ R by x +3 y = 3px3 + y3. Then we can show that R+3 = (R, +3,·) is a field. It is clear that R is closed under +3, 0 is the identity element, +3 is commutative and −x is an inverse for x. As for distributivity, we have α(x +3 y) = α 3px3 + y3 = 3p(αx)3 + (αy)3 = (αx) +3 (αy). 8 Note that any odd power could be used to define the addition, giving infinite field structures on R. Next we prove that the mapping φ : (R, +3,·) → (R, +,·) x 7→ x3 is a field isomorphism. It is well-known that φ is a bijection and for all x, y ∈ R, φ(x +3 y) = x3 + y3 = φ(x) + φ(y). We can construct a near-vector space isomorphism from this. Since φ induces both a multiplicative and additive isomorphism if we define ·3 on for all x, α ∈ R by α ·3 x = α3x = φ(α)x, This homomorphism induces a commutative diagram: (R,·) × (R, +3) m (R, +3) φ×φ φ / (R,·) × (R, +) m / (R, +) where m sends (x, y) to xy, the usual multiplication of x and y in R. Note that R+3 is a vector space over R+3, but only a near-vector space over R. Clearly, R is a vector space over R. Even though they are not isomorphic in the traditional vector space sense, (φ, φ) is a near-vector space isomorphism that gives more flexibility than traditional linear algebra would have allowed by only fixing the multiplicative structure of the underlying set. Example 3.9. We consider Q(√2) and Q(√3). These are non- isomorphic fields with isomorphic multiplicative groups with Q(√2)∗ = Q(√3)∗ = Z × {±1} since the ring of integers of Q(√2) (respectively Q(√3)) is Z(√2) (resp. Z(√3)) are unique factorization domains. Fixing A = Z×{±1}∪{0} we have that (Q(√2), A) and Q(√3), A) are near-vector This also permits us to construct the near-vector space (Q(√3)⊕Q(√2), A) which spaces with (A, +√2,·) and (A, +√3,·) non-isomorphic. has no equivalent in traditional linear algebra. The different stuctures of the near-field (A, +v,·) for a near vector space (V, A) are known to be isomorphic when w ∈ Av(= Span(V )), for any v ∈ Q(V )\{0} (Theorem 2.5, p.300 in [1]). The main results of the section rely upon the following result that holds for any near-vector space and reveals an interesting necessary condition for +v = +w for v, w ∈ Q(V )\{0}. Proposition 3.10 (Key Lemma). Let (V, A) be a near-vector space with dim(V ) > 1 and Q(V ) = V and S = {vi ∈ Q(V )i ∈ I} a linearly independent set (possibly infinite) of V. If v =Pi∈I θivi and v′ =Pi∈I θ′ivi are both in Q(V ) where all but finitely many of the θi and θ′i are zero and we have that there exist i0, j0 ∈ I with i0 6= j0, such that +θi0 vi0 . Then +v = +v′ = +θivi = +θ′ = +θ′ j vj , for all i, j ∈ I. vj0 j0 9 / / Proof. By assumption v, v′ ∈ Q(V ), therefore for all α, β ∈ A we get that (α +v β)v = α(v) + β(v) = Pi∈I αθivi +Pi∈I βθivi =Pi∈I (αθivi + βθivi) = Pi∈I (α +θivi β)θivi. (α +v β)v =Xi∈I (α +v β)θivi. Moreover, Therefore, 0 = Pi∈I (α +v β)θivi −Pi∈I (α +θivi β)θivi = Pi∈I ((α +v β) +θivi (−(α +θivi β)))θivi. Since S is linearly independent and for any i ∈ I, so we have that (α +v β) +θivi (−(α +θivi β)) = 0, α +v β = α +θivi β. For all θivi (This is deduced from the fixed point freeness of the scalar multiplication which proves that the inverse of (α +v β) for + is −(α +v β).). Thus +v = +θivi for all i ∈ I. The same can be done to show that +v′ = +θ′ j vj . Proving that +v′ = +θ′ j0 = +θi0 vi0 = +v. vj0 Remark 3.11. Let (V, A) be a near-vector space with dim(V ) > 1 and Q(V ) = V and S = {vi ∈ Q(V )i ∈ I} a linearly independent set (possibly infinite) of V. If v =Pi∈I θivi ∈ Q(V ), then +v = +θivi, for all i ∈ I. 4 Structural results of Near-vector spaces constructed from division rings 4.1 Regular spaces and their decomposition In the next lemma which is not difficult to prove, we discover how the addition +v that turns A into a near-field for v ∈ Q(V )\{0} results in a special structure on (A, +v,·). Lemma 4.1. Let V be a near-vector space and v ∈ Q(V ){0}. Then the following are equivalent: 1. +v = +θv for all θ ∈ A; 2. (A, +v,·) is a division ring. 10 Proof. It is not difficult to see that: +v = +θv if and only if (α +v β)θ = (αθ +v βθ), ∀α, β ∈ A. The following theorem, attempts to understand the relationship between proper- ties of Q(V ), +v and regularity. Theorem 4.2. Let (V, A) be a near-vector space. The following assumptions are equivalent: 1. For any v ∈ Q(V )\{0}, V is a vector space over the near-field (A, +v,·) ; 2. There is a v ∈ Q(V )\{0}, such that V is a vector space over the near-field (A, +v,·); 1.' For any v ∈ Q(V )\{0}, V is a vector space over the near field (A, +v,·) and (A, +v,·) is a division ring; 2.' There is a v ∈ Q(V )\{0}, such that V is a vector space over the near-field (A, +v,·) and (A, +v,·) is a division ring; 3. Q(V ) = V and (A, +v,·) is a division ring, for all v ∈ Q(V )\{0}. 4. +v = +w for all v, w ∈ Q(V )\{0}. 5. Rw(V ) = V for all w ∈ Q(V )\{0}. 6. V is regular and for any v ∈ V , +v = +θv for all v ∈ Q(V )\{0} and θ ∈ A. 7. V is regular and for any v ∈ Q(V )\{0}, (A, +v,·) is a division ring. Proof. 1. ⇔ 2. ⇔ 1.′ ⇔ 2.′ is not hard to prove using Lemma 4.1. 2. ⇒ 3 If there is v ∈ V such that V is a vector space over the near-field (A, +v,·) then for any w ∈ V we have αw + βw = (α +v β)w. Therefore, w ∈ Q(V ). 3. ⇒ 4. Let (V, A) be a near-vector space such that Q(V ) = V . We denote B = {vi ∈ Q(V )i ∈ I} a basis for V . For any two nonzero v, w ∈ V , either v = θw and +v = +w, according to Lemma 4.1 since (A, +v,·) is a division ring. Otherwise we have v =Pi∈I θivi ∈ V and v′ =Pi∈I θ′ivi ∈ V , and there exist i 6= j ∈ I such that θi and θ′j are nonzero, therefore we can take w = θivi + θ′jvj, and since by assumption w ∈ Q(V ) we can apply Proposition 3.10 to obtain that +v = +w = +v′. 4. ⇔ 5 ⇒ 6. ⇔ 7. is trivial the last equivalence being a consequence again of Lemma 4.1. 11 θ ∈ A. 6. ⇒ 4. Suppose that V is regular and for any v ∈ V , +v = +θv for all v ∈ V and Let {vi ∈ Q(V ), i ∈ I} be a basis for V . For any i 6= j ∈ I, Then, from the regularity of V , there is a λi,j ∈ A such that Again applying Proposition 3.10, we get ω = vi + λi,jvj ∈ Q(V ). +ω = +vi = +λi,j vj = +vj =: + where the last equality holds since by assumption (A, +vj ,·) is a division ring. Let v ∈ V be nonzero, we have v =Pi∈I θivi. Indeed, for any α, β ∈ A, (αθivi + βθivi) =Xi∈I αv + βv =Xi∈I Therefore v ∈ Q(V ) and Q(V ) = V . Classical linear algebra is a particular case of a near-vector space corresponding precisely to the regular case, when (A, +v,·) is a field for every nonzero v ∈ Q(V ). As an application of the previous theorem, we reprove the Decomposition Theo- rem for the division ring case. (αθi + βθi)vi =Xi∈I (α + β)θivji = (α + β)v. Corollary 4.3. Let (V, A) be a near-vector space such that for all v ∈ Q(V )∗, (A, +v,·) is a division ring. Then V is the direct sum of regular near-vector spaces Vj for j ∈ K for some index set K such that each u ∈ Q(V )∗ lies in precisely one summand Vj. The subspaces Vj are maximal regular near-vector spaces. Proof. Let B = {vi i ∈ I} be a canonical basis of V for some index set J. We put Vi = {v ∈ V +v = +vi}. We claim that V = ⊕i∈KVi, and that the Vi are maximal regular subspaces of V where K is a index set such that for each i ∈ J there is a unique k ∈ K such that vi ∈ Vk. For i ∈ K, the Vi are subspaces of V since they are closed under addition by Proposition 3.10 and closed under scalar multiplication by Lemma 4.1. We claim that they are maximal regular subspaces of V . They are regular by Theorem 4.2, since (A, +v,·) is a division ring and we have that +v = +w for all v, w ∈ Vi. Moreover, Q(Vi) = Vi for i ∈ K. We claim that V = ⊕j∈KVj. The sum is clearly direct since we have a basis. It is also clear that ⊕j∈KVj ⊆ V . Let v ∈ V, then since B is a basis of V , it is clear that v = Pi∈I θivi for some θi ∈ A and vi ∈ B. We set Sj = {i ∈ Ivi ∈ Vj} for all j ∈ K, then v = Pj∈K wj with wj =Pi∈Sj To show that each u ∈ Q(V )∗ lies in precisely one summand Vj we prove that Q(V ) = ⊎k i=1Q(Vi) ⊆ Q(V ) is clear. Now let u ∈ Q(V ), then +u = +ui for some unique i ∈ K, by Proposition 3.10, since we are supposing that (A, +v,·) is a division ring for every v ∈ Q(V )\{0}. Finally we need to show the maximality of each Vj for j ∈ K. Suppose that there exists a regular subspace M such that Vj ⊂ M. Then since M is regular, by Theorem 4.2, +v = +w for all v, w ∈ M. Take a m ∈ M\Vj and v ∈ Vj. Then +m = +v, which contradicts our initial partition. For the uniqueness of the decomposition, suppose that there are two decompositions V = ⊕j∈KVj and V = ⊕s∈RWs. If v ∈ Vj for some j ∈ K, then by Theorem 4.2, +v = +vj . But since V = ⊕s∈RWs, vi ∈ Ws for θivi. In particular, wj ∈ Vj. Thus v ∈ ⊕j∈KVj and so V = ⊕j∈KVj. i=1Q(Vi). ∪k 12 some s ∈ R and again by Theorem 4.2, +vj = +w for all w ∈ Ws. Now we have that +w = +vi = +v, so v ∈ Ws. Similarly, Ws ⊆ Vj. Corollary 4.4. If (V, A) is as before and B = {vi i ∈ I} a canonical basis of V, K a index set such that for each i ∈ I there is a unique k ∈ K such that vi ∈ Vk and we choose j ∈ K. • {vi vi ∈ Vj} is a basis for Vj. • Referring to Lemma 3.4, ⊕s i=1Avr is a subspace of Vj where vr ∈ Vj for r ∈ {1, . . . , s}. Example 4.5. Consider the near-vector space (R3, R), where scalar multiplication is defined for all (x, y, z) ∈ R3 and α ∈ R by α(x, y, z) = (α3x, α5y, α3z). Then by applying Corollary 4.3, we have that V1 = {(a, b, c) ∈ R3∀α, β ∈ R, α +(a,b,c) β = (α3 + β3) 1 3 = {(a, 0, c)a, c ∈ R}, while V2 = {(a, b, c) ∈ R3∀α, β ∈ R, α +(a,b,c) β = (α5 + β5) 1 5} = {(0, b, 0)b ∈ R} and R3 = V1 ⊕ V2. 4.2 The structure of Span It is immediate to prove the following lemmas: Lemma 4.6. Let (V, A) be a near-vector space. v0,··· , vn ∈ Q(V ) are linearly dependent if there is i0 ∈ {1,··· , n} such that vi0 ∈ span(v1,··· , vn). We have explicitly seen in Lemma 3.4 how V ≃ A(I) allows us to retranslate the following lemma to any near-vector space. Theorem 4.7. Let V = A(I) as in Lemma 3.4, 2. and (A, +v,·) be a division ring for any nonzero v ∈ V . Let v ∈ V. According to Corollary 4.3, we have the regular decomposition V = ⊕j∈KVj, so that we can write v uniquely as v =Pj∈K vi where vi ∈ Vi. If N is the subset of K, N = {i ∈ Kvi 6= 0}, we have that and span(v) = L(v) = ⊕Avi dim(v) = N. Proof. Let v be as in the statement, i.e. vi. v =Xi∈N For convenience we order the elements in N as {i1, . . . , in} where n = N. 13 Since the vi are in different regular components and we are assuming the division ring condition, using Theorem 4.2, we know that +vi 6= +vj for any i 6= j ∈ N . Therefore there exist α, β ∈ A such that Then, taking α +vi1 β 6= α +vi2 β. w = (αv + βv) − (α +vi2 β)v ∈ L(v) ⊆ span(v), we have that w =Pi∈N1 wi where N1 ⊆ N\{i2}, i1 ∈ N1 and wi ∈ Vi nonzero. Repeating this process successively, we can construct an element ω1 = θ1vi1 ∈ L(v) = span(v) and more generally restarting from v we can construct ωj = θjvij ∈ L(v) = span(v). Since Aθjvij = Avij , this proves that ⊕i∈N Avi ⊆ span(v). But since ⊕i∈N Avi ⊕i∈N Avi = L(v) and since it is generated by elements of Q(V ), we get the result is a near-vector space contained in L(v), we therefore have ⊕i∈N Avi = L(v) = span(v). We now give an example to illustrate the previous result. Example 4.8. Consider the near-vector space ((Z11)3, Z11), where scalar multipli- cation is defined for all (x, y, z) ∈ R3 and α ∈ R by α(x, y, z) = (α3x, α5y, α3z). Note that Q(V ) = {(a, 0, c)a, c,∈ Z11} ∪ {(0, b, 0)b ∈ Z11}. Take for example, v = (2, 5, 6) ∈ (Z11)3, then v /∈ Q(Z11), and span((2, 5, 6)) = Z11(2, 0, 6) + Z11(0, 5, 0). If we take, for example w = (3, 0, 4), then w ∈ Q(Z11), and span((3, 0, 4)) = Z11(3, 0, 4). An easy but useful consequence of the previous theorem is the following. Corollary 4.9. Let (V, A) be a near-vector space. For every S = {vi ∈ Q(V )i ∈ I}, where I is an index set, span(S) = L(S) = ⊕t∈T Avit where {vit , t ∈ T} is the biggest linearly independent subset of S. More generally we have, 14 Corollary 4.10. Let (V, A) be a non-regular near-vector space and suppose (A, +v,·) is a division ring for all v ∈ Q(V )\{0}. Let S = {vi ∈ V i ∈ I} where I is an index set, possibly infinite. Then we have that span(S) = L(S) = ⊕i∈T Aωi, where ωi ∈ Q(V ), for all i ∈ T . Proof. For any j ∈ I, vj = Xi∈Nj vj i according to the regular decomposition where vj i ∈ Vi nonzero. It is not hard to see that span(vj) = ⊕i∈Nj Avj i ⊆ span(S), where Therefore, Nj = {i ∈ Ivj i 6= 0}. Xi∈I span(vj) ⊆ span(S). From the previous corollary we know that span({vj ij ∈ I}) = ⊕t∈Ti,j Awi,j t, where {wi,j tt ∈ Ti,j} is the biggest linearly independent subset of {vj ij ∈ I}. We leave the details to the reader to conclude that span(S) = L(S) = ⊕j∈I,i∈Nj,t∈Ti,j Awi,j t, proving the result. From the previous corollary, we can rectify an error made in Lemma 2.4 in [8], p.2426, where it was shown that a subset of a near-vector space is a subspace if and only if it is closed under addition and scalar multiplication. Of course, the one direction is obvious. The problem with the converse was that the proposed generating set was not necessarily contained in the subspace. We rectify it here for the case where we assume +v = +w for any v, w ∈ Q(V )\{0}. Corollary 4.11. Let (V, A) be a non-regular near-vector space and suppose (A, +v,·) is a division ring for all nonzero v ∈ Q(V ). W is a subspace of V if and only it is non-empty, closed under addition and scalar multiplication. Corollary 4.12. Let (V, A) be a non-regular near-vector space and suppose (A, +v,·) is a division ring for all v ∈ Q(V )\{0}. There exist v and w ∈ V \Q(V ) v 6= w such that span(v) = span(w). Proof. Take v1, v2 ∈ Q(V ) linearly independent and not in the same regular com- ponent and v = v1 + v2 and w = θv1 + v2 ∈ V \Q(V ), where θ 6= 1, v 6= w, but span(v) = span(w). 15 The following two corollaries shed more light on why taking elements outside of Q(V ) as basis elements would be counter-intuitive to our general intuition with respect to a basis. Corollary 4.13. Let (V, A) be a non-regular near-vector space with dim(V ) > 2 and suppose (A, +v,·) is a division ring for all v ∈ Q(V )\{0}. Then there exisst v and w ∈ V \Q(V ) such that v /∈ span(w) and w /∈ span(v) and span(v)∩ span(w) 6= {0}. Proof. Take v1, v2, v3 ∈ Q(V ) linearly independent not in the same regular subspace in the decomposition of V and v = v1 + v2 and w = v2 + v3 ∈ V \Q(V ), then we have v /∈ span(w) and w /∈ span(v), but span(v) ∩ span(w) = Av2 6= {0}. Acknowledgment The authors would like to express their gratitude for funding by the National Re- search Foundation (Grant number: 93050) and Stellenbosch University. The authors would like to thank Jacques Rabie for Example 3.6 and Charlotte Kestner for Ex- ample 3.7. References [1] Andr´e J. Lineare Algebra uber Fastkorpern. Math Z 1974: 136; 295-313. [2] Beidleman, J.C. On near-rings and near-ring modules, PhD, Pennsylvanian State University, USA, 1966. [3] Chistyakov D, Howell K-T, Sanon SP. On representation theory and near-vector spaces. Linear Multilinear Algebra 2019: 67:7; 4951510. [4] Dorfling S, Howell K-T, Sanon SP. The decomposition of finite dimensional Near-vector spaces. Communications in Algebra 2018: 46:7; 3033-3046. [5] Howell K-T. Contributions to the Theory of Near-Vector Spaces, PhD, Univer- sity of the Free State, Bloemfontein, South Africa, 2008. [6] H. Karzel, Fastvektorraume, unvollstandige Fastkorper und ihre abgeleiteten Strukturen. Erscheint in Mitt. Sem. Univ. Giessen, 1984. [7] K-T Howell, Near-vector spaces determined by finite fields and their fibra- tions, Turkish Journal of Mathematics 2019: 43; 25492560. [8] Howell K-T. On subspaces and mappings of Near-vector spaces. Communica- tions in Algebra 2015: 43:6; 2524-2540. [9] Howell K-T, Meyer JH. Near-vector spaces determined by finite fields. Journal of Algebra 2010: 398; 55-62. [10] Howell K-T, Sanon S. On spanning sets and generators of Near-vector spaces. Turkish Journal of Mathematics 2018: 42; 32323241. [11] Meldrum JDP. Near-rings and Their Links with Groups. New York, NY, USA: Advanced Publishing Program, 1985. 16 [12] Pilz G. Near-rings: The Theory and its Applications, Revised Edition. North Holland, New York, 1983. [13] van der Walt, APJ. Matrix near-rings contained in 2-primitive near-rings with minimal subgroups. Journal of Algebra 1992: 148; 296-304. 17
1509.05916	1	1509	2015-09-19T17:01:38	Aspects on weak, s-CS and almost injective rings	[ "math.RA" ]	It is not known whether right CF-rings (FGF-rings) are right artinian (quasi-Frobenius). This paper gives a positive answer of this question in the case of weak CS (s-CS) and GC2 rings. Also we get some new results on almost injective rings.	math.RA	math	ASPECTS ON WEAK, s-CS AND ALMOST INJECTIVE RINGS NASR. A. ZEYADA AND AMR K. AMIN Abstract. It is not known whether right CF -rings (F GF -rings) are right artinian (quasi-Frobenius). This paper gives a positive answer of this question in the case of weak CS (s-CS) and GC2 rings. Also we get some new results on almost injective rings. 1. Introduction A module M is said to satisfy C1-condition or called CS-module if every submod- ule of M is essential in a direct summand of M . Patrick F. Smith [20] introduced weak CS modules. A right R-module M is called weak CS if every semisimple sub- module of M is essential in a summand of M . I. Amin, M. Yousif and N. Zeyada [1] introduced soc-injective and strongly soc-injective modules, for any two modules M and N , M is soc-N -injective if any R-homomorphism f : soc(N ) −→ M extends to N . R is called right (self-) soc-injective, if the right R-module RR is soc-injective. M is strongly soc-injective if M is soc-N -injective for any module N . They proved that every strongly soc-injective module is weak CS. N. Zeyada [24] introduced the notion of s-CS, for any right R-module M , M is called s-CS if every semisimple submodule of M is essential in a summand of M . A ring R is called a right CF ring if every cyclic right R-module can be embedded in a free module. A ring R is called a right F GF ring if every finitely generated right R-module can be embedded in a free right R-module. In section 2, we show that the right CF , weak CS (s-CS) and GC2 rings are artinian. Zeyada, Hussein and Amin introduced the notions of almost and rad-injectivity [23]. In the third section we make a correction to the result [23, Theorem 2.12] and we get a new results using these notions. Throughout this paper R is an associative ring with identity and all modules are unitary R-modules. For a right R-module M , we denote the socle of M by soc(M ). Sr and Sl are used to indicate the right socle and the left socle of R, respectively. For a submodule N of M , the notations N ⊆ess M and N ⊆⊕ M mean that N is essential and direct summand, respectively. We refer to [2], [5], [7], [12] and [15] for all undefined notions in this paper. 2. Generalizations of CS-modules and rings Lemma 1. For a right R-module M , the following statements are equivalent: (1) M is weak CS. (2) M = E ⊕T where E is CS with soc(M ) ⊆ess E. Key words and phrases. CS, weak-CS, S-CS, rad-injective rings, almost injective rings, Kasch rings, Quasi-Frobenius rings. 1 2 NASR. A. ZEYADA AND AMR K. AMIN (3) For every semisimple submodule A of M , there is a decomposition M = M1 ⊕ M2 such that A ⊆ M1 and M2 is a complement of A in M . Proof. (1) =⇒ (2). Let M be a weak CS. Then soc (M ) is essential in a sum- mand, so M = E ⊕T with soc(M ) ⊆ess E. Now if K is a submodule of E, then soc(K) ⊆ess L where L is a summand of M and L ⊆ess (K + L). But L is closed, so K ⊆ L. Since E ⊆ess (L + E) and E is closed in M , so L ⊆ E and E is CS. (2) =⇒ (1). If E is CS and a summand of M with soc(M ) ⊆ess E, then every submodule of soc (M ) is a summand of E and a summand of M . (1 =⇒ 3). Let A be a submodule of soc (M ). By (1), there exists M1 ⊆⊕ M such that A ⊆ess M1. Write M = M1 ⊕ M2 for some M2 ⊆ M . Since M2 is a complement of M1 in M and A is essential in M1, then M2 is a complement of A in M . (3 =⇒ 1). Let A be a submodule of soc (M ). By (2), there exists a decomposition M = M1 ⊕ M2 such that A ⊆ M1 and M2 is a complement of A in M . Then (A ⊕ M2) ⊆ess M = M1 ⊕ M2 and A ⊆ M1 then A ⊆ess M1. Hence M is weak CS module. (cid:3) Recall that, a right R-module M is s-CS if every singular submodule of M is essential in a summand [24]. Proposition 1. If M is a right R-module, then the following statements are equiv- alent: (1) M is s-CS. (2) The second singular submodule Z2 (M ) is CS and a summand of M. (3) For every singular submodule A of M , there is a decomposition M = M1 ⊕ M2 such that A ⊆ M1 and M2 is a complement of A in M . Proof. (1) ⇐⇒ (2). [24, Proposition 14]. (1) ⇐⇒ (3). Similar argument of the proof of the above Lemma. (cid:3) Given a right R-module M we will denote by Ω(M ) [respectively C(M )] a set of representatives of the isomorphism classes of the simple quotient modules (respec- tively simple submodules) of M . In particular, when M = RR, then Ω(R) is a set of representatives of the isomorphism classes of simple right R-modules. Lemma 2. Let R be a ring, and let PR be a finitely generated quasi-projective CS-module, such that Ω(P ) ≤ C(P ). Then Ω(P ) = C(P ), and PR has finitely generated essential socle. Proof. See [17, Lemma 7.28]. (cid:3) Proposition 2. Let R be a ring. Then R is a right P F -ring if and only if RR is a cogenerator and R is weak CS. Proof. Every right P F -ring is right self-injective and is a right cogenerator by [17, Theorem 1.56]. Conversely, if R is weak CS and R is cogenerator then R = E ⊕T where E is CS with Sr ⊆ess E. By the above Lemma, E has a finitely generated, essential right socle. Since E is right finite dimensional and RR is a cogenerator, let Sr = S1 ⊕ S2 ⊕ ..... ⊕ Sm and Ii = I(Si) be the injective hull of Si, then there exists an embedding σ : Ii −→ RI for some set I. Then π ◦ σ 6= 0 for some projection π : RI −→ R, so (π ◦ σ)Si 6= 0 and hence is monic. Thus π ◦ σ : Ii −→ R is monic, and so R = E1⊕ .... ⊕ Em ⊕ T where soc (T ) = 0. So R is a right P F -ring. (cid:3) Proposition 3. [24, Proposition 16] Let R be a ring. Then R is a right P F -ring 3 if and only if RR is a cogenerator and (cid:0)Z 2 r(cid:1)R is CS. Proposition 4. The following statements are equivalent: (1) Every right R-module is weak CS. (2) Every right R-module with essential socle is CS. (3) For every right R-module M , M = E ⊕K where E is CS with soc (M ) ⊆ess E. Proposition 5. The following statements are equivalent: (1) Every right R-module is s-CS. (2) Every Goldie torsion right R-module is CS. (3) For every right R-module M , M = Z2 (M ) ⊕ K where Z2 (M ) is CS. Dinh Van huynh, S. K. Jain and S. R. L´opez-Permouth [?] proved that if R is simple such that every cyclic singular right R-module is CS, then R is right noetherian. Corollary 1. If R is simple such that every cyclic right R-module is s-CS, then R is right noetherian. Proposition 6. If R is a weak CS and GC2, and right Kasch, then R is semiper- fect. Proof. Since R is a weak CS, so E is CS by Lemma 1 and R = E ⊕ K for some right ideal K of R and so E is a finitely generated projective module. By Lemma 2, E has a finitely generated essential socle. Then, by hypothesis, there exist simple submodules S1, · · · , Sn of E such that {S1, · · · , Sn} is a complete set of represen- tatives of the isomorphism classes of simple right R-modules. Since E is CS, there exist submodules Q1, · · · , Qn of E such that Q1, · · · Qn is an direct summands of E and (Si)R ⊆ess (Qi)R for i = 1, · · · , n. Since Qi is an indecomposable projective and GC2 R-module, it has a local endomorphism ring; and since Qi is projective, J(Qi) is maximal and small in Qi. Then Qi is a projective cover of the simple module Qi/J(Qi). Note that Qi ∼= Qj clearly implies Qi/J(Qi) ∼= Qj/J(Qj); and the converse also holds because every module has at most one projective cover up to isomorphism. It is clear that Qi ∼= Qj if and only if Si ∼= Sj if and only if i = j. Thus, {Q1/J(Q1), · · · , Qn/J(Qn)} is a complete set of representatives of the iso- morphism classes of simple right R-modules. Hence every simple right R-module has a projective cover. Therefore R is semiperfect. (cid:3) The following example show that the proof of [24, Proposition 13] is not true, since the endomorphism ring of an indecomposable projective module which is an essential extension of a simple module may be not a local ring. So we add an extra condition that R is right GC2 to prove the Proposition. Example 1. Let R be the ring of triangular matrices, R = (cid:26)(cid:18) a 0 Take P1 = (cid:26)(cid:18) a 0 0 (cid:19) : a ∈ Z, b ∈ Q(cid:27) and P2 = (cid:26)(cid:18) 0 0 0 b c (cid:19) : a ∈ Z, b, c ∈ Q(cid:27). b c (cid:19) : c ∈ Q(cid:27), we see that P1 is indecomposable projective module with simple essential socle and P2 is pro- jective simple module. The socle of P1 is isomorphic to P2 and its endomorphism ring is isomorphic to Z which is not local. 4 NASR. A. ZEYADA AND AMR K. AMIN Proposition 7. If R is right s-CS and GC2, and right Kasch, then R is semiper- fect. Proof. Since R is a weak CS, so E is CS by Lemma 1 and R = E ⊕ K for some right ideal K of R and so E is a finitely generated projective module. By Lemma 2, E has a finitely generated essential socle. Then, by hypothesis, there exist simple submodules S1, · · · , Sn of E such that {S1, · · · , Sn} is a complete set of represen- tatives of the isomorphism classes of simple right R-modules. Since E is CS, there exist submodules Q1, · · · , Qn of E such that Q1, · · · Qn is an direct summands of E and (Si)R ⊆ess (Qi)R for i = 1, · · · , n. Since Qi is an indecomposable projective and GC2 R-module, it has a local endomorphism ring; and since Qi is projective, J(Qi) is maximal and small in Qi. Then Qi is a projective cover of the simple module Qi/J(Qi). Note that Qi ∼= Qj clearly implies Qi/J(Qi) ∼= Qj/J(Qj); and the converse also holds because every module has at most one projective cover up to isomorphism. It is clear that Qi ∼= Qj if and only if Si ∼= Sj if and only if i = j. Thus, {Q1/J(Q1), · · · , Qn/J(Qn)} is a complete set of representatives of the iso- morphism classes of simple right R-modules. Hence every simple right R-module has a projective cover. Therefore R is semiperfect. (cid:3) Lemma 3. Let R be a semiperfect, left Kasch, left min-CS ring. Then the following hold:: (1) Sl ⊆ess e ∈ R. R R and soc(Re) is simple and essential in Re for all local idempotents (2) R is right Kasch if and only if Sl ⊆ Sr. (3) If {e1, ..., en} are basic local idempotents in R then {soc(Re1), ..., soc(Ren)} is a complete set of distinct representatives of the simple left R-modules. Proof. See [17, Lemma 4.5]. (cid:3) Recall that a ring R is right minfull if it is semiperfect, right mininjective, and soc(eR) 6= 0 for each local idempotent e ∈ R. Corollary 2. If R is commutative s-CS (weak CS) and Kasch, then R is minfull. Proof. Since every Kasch ring is C2, so R is semiperfect by Proposition 6 (Propo- sition 7). Thus using the above Lemma and [17, Proposition 4.3] R is minfull. (cid:3) Theorem 1. If R is right weak CS (s-CS), GC2 and every cyclic right R-module can be embedded in a free module (right CF ring ) then R is right artinian Proof. If R is right weak CS (s-CS) right CF , then by Lemma 1 (Lemma 1) R = E ⊕ K where E is CS and soc (K) = 0 (Z (K) = 0). Thus by Proposition 6 (Proposition 7), R is semiperfect. The above Lemma gives Sr ⊆ess RR, so K = 0. Hence R is CS and R is right artinain by [9, Corollary 2.9]. (cid:3) Proposition 8. Let R be a right F GF , right weak CS (s-CS) and right GC2 ring. Then R is QF . Proof. It clear by Proposition 6 (Proposition 7) , and [8, Theorem 3.7]. (cid:3) 5 3. Almost injective Modules Definition 1. A right R-module M is called almost injective, if M = E ⊕ K where E is injective and K has zero radical. A ring R is called right almost injective, if RR is almost injective. In [23], the statement of Theorem 2.12 is not true, so the proof of (3) =⇒ (1). The following Proposition is the true version of [23, Theorem 2.12]. Then we rewrite the related results. Proposition 9. For a ring R the following are true: (1) R is semisimple if and only if every almost-injective right R-module is in- jective. (2) If R is semilocal, then every rad-injective right R-module is injective. Proof. (1). Assume that every almost-injective right R-module is injective, then every right R-module with zero radical is injective. Thus every semisimple right R-module is injective and R is right V -ring. Hence, every right R-module has a zero radical. Therefore, every right R-module is injective and R is semisimple. The converse is clear. (2). Let R be a semilocal ring and M be a rad-injective right R-module. Consider a homomorphism f : K −→ M where K is a right ideal of R. Since R is semilocal, there exists a right ideal L of R such that K + L = R and K ∩ L ⊆ J [13]. Then there exists a R-homomorphism g : R −→ M such that g (x) = f (x) for every x ∈ K ∩ L. Define F : R −→ M by F (x) = f (k) + g (l) for any x = k + l where k ∈ K and l ∈ L. It is clear that F is a well-defined R-homomorphism such that F K= f . i.e. F extends f . Therefore M is injective. (cid:3) A ring R is called quasi-Frobenius (QF ) if R is right (or left) artinian and right (or left) self-injective. Also, R is QF if and only if every injective right R-module is projective. Theorem 2. R is a quasi-Frobenius ring if and only if every rad-injective right R-module is projective. Proof. If R is quasi-Frobenius, then R is right artinian, and by Proposition 9 (2), every rad-injective right R-module is injective. Hence, every rad-injective right R- module is projective. Conversely, if every rad-injective right R-module is projective, then every injective right R-module is projective. Thus, R is quasi-Frobenius. (cid:3) Recall that a ring R is called a right pseudo-Frobenius ring (right P F -ring) if the right R-module RR is an injective cogenerator. Proposition 10. The following are equivalent: (1) R is a right P F -ring. (2) R is a semiperfect right self-injective ring with essential right socle. (3) R is a right finitely cogenerated right self-injective ring. (4) R is a right Kasch right self-injective ring. Theorem 3. If R is right Kasch right almost-injective, then R is semiperfect. Proof. Let R be right Kasch and RR = E ⊕ T , where E is injective and T has zero radical. If J = 0, then every simple right ideal of R is projective and R is semiperfect (for R is right Kasch). Now suppose that J 6= 0. Clearly, every 6 NASR. A. ZEYADA AND AMR K. AMIN simple singular right R-module embeds in E. In particular, every simple quotient of E is isomorphic to a simple submodule of E, and so E is a finitely generated injective and projective module containing a copy of every simple quotient of E. By [8, Lemma 18], E has a finitely generated essential socle. Then by hypothesis, there exist simple submodules S1, ..., Sn of E such that {S1, ..., Sn} is a complete set of representatives of the isomorphism classes of simple singular right R-modules. Since E is injective, there exist submodules Q1, ..., Qn of E such that Q1⊕···⊕Qn is a direct summand of E and (Si)R ⊆ess (Qi)R for i = 1, 2, ..., n. Since Qi is an indecomposable injective R-module, it has a local endomorphism ring. The projectivity of Qi implies that J(Qi) is maximal and small in Qi. Then Qi is the projective cover of the simple module Qi/J(Qi). Note that Qi ∼= Qj clearly implies Qi/J(Qi) ∼= Qj/J(Qj) and the converse also holds because every module has at ∼= Qj if and most one projective cover up to isomorphism. But it is clear that Qi only if Si ∼= Sj, if and only if i = j. Moreover, every Qi/J(Qi) is singular. Thus, {Q1/J(Q1), ..., Qn/J(Qn)} is a complete set of representatives of the isomorphism classes of the simple singular right R-modules. Hence, every simple singular right R-module has a projective cover. Since every non-singular simple right R-module is projective, we conclude that R is semiperfect. (cid:3) Proposition 11. The following are equivalent: (1) R is a right P F -ring. (2) R is a semiperfect right rad-injective ring with soc(eR) 6= 0 for each local idempotent e of R. (3) R is a right finitely cogenerated right rad-injective ring. (4) R is a right Kasch right rad-injective ring. (5) R is a right rad-injective ring and the dual of every simple left R-module is simple. Proof. (1) ⇔ (2) By Proposition 9 (2). (1) ⇒ (3) Clear. (3) ⇒ (1) Since R is a right rad-injective ring, it follows from [23, Proposition 2.5] that R = E ⊕ K, where E is injective and K has zero radical. Since R is a right finitely cogenerated ring, K is a finitely cogenerated right R-module with zero radical. Hence, K is semisimple. Therefore, by [22, Corollary 8], R is a right P F -ring. (1) ⇒ (4) Clear. (4) ⇒ (1) If R is right Kasch right rad-injective, then R is right almost-injective ([23, Proposition 2.5]). Thus R is semiperfect (3). Hence R is injective by Propo- sition 9 (2). Therefore, R is right P F . (1) ⇒ (5) Since every right P F -ring is left Kasch and left mininjective, the dual of every simple left R-module is simple by [16, Proposition 2.2]. (5) ⇒ (1) By [23, Proposition 2.10], R is a right min − CS ring (i.e., every minimal right ideal of R is essential in a summand). Thus, by [10, Theorem 2.1], R is semiperfect with essential right socle. Proposition 9 (2) entails that R is right self-injective, and hence right P F by Proposition 10. (1) ⇐⇒ (4) and (1) ⇐⇒ (5) are direct consequences of [23, Proposition 2.5]. (cid:3) A result of Osofsky [18, Proposition 2.2] asserts that a ring R is QF if and only if R is a left perfect, left and right self-injective ring. This result remains true for rad-injective rings. 7 Proposition 12. The following are equivalent: (1) R is a quasi-Frobenius ring. (2) R is a left perfect, left and right rad-injective ring. Proof. (1) ⇒ (2) It is well known. (2) ⇒ (1) By hypothesis, R is a semiperfect right and left rad-injective ring. By (cid:3) Proposition 9 (2), R is right and left injective, hence R is quasi-Frobenius. Note that the ring of integers Z is an example of a commutative noetherian almost-injective ring which is not quasi-Frobenius. Definition 2. A ring R is called right CF -ring (F GF -ring) if every cyclic (finitely generated) right R-module embeds in a free module. It is not known whether right CF -rings (F GF -rings) are right artinian (quasi-Frobenius rings). In the next re- sult, a positive answer is given if we assume in addition that the ring R is right rad-injective. Proposition 13. The following are equivalent: (1) R is quasi-Frobenius. (2) R is right CF and right rad-injective. Proof. (1) ⇒ (2) It is well known. (2) ⇒ (1) Since every simple right R-module embeds in R, R is a right Kasch ring. By Proposition 11, R is right self-injective with finitely generated essential right socle. Thus, every cyclic right R-module has a finitely generated essential socle, and by [21, Proposition 2.2], R is right artinian, hence quasi-Frobenius. (cid:3) References [1] I. Amin, M. Yousif, N. Zeyada, Soc-injective rings and modules, Comm. Algebra 33 (2005) 4229 -- 4250. [2] F. W. Anderson, K. R. Fuller, Rings and Categories of Modules, Springer-Verlag, Berlin-New York, 1974. [3] J. E. Bjork, Rings satisfying certain chain conditions, J. Reine Angew. Math.245 (1970) 63 -- 73. [4] V. Camillo, W.K. Nicholson, M.F. Yousif, Ikeda-Nakayama rings, J. Algebra 226 (2000) 1001 -- 1010. [5] N. V. Dung, V. D. Huynh, P. F. Smith, and R. Wisbauer, Extending Modules, Pitman Reasearch Notes in Math. Longman, 1994. [6] C. Faith, Rings with ascending chain conditions on annihilators, Nagoya Math. J.27 (1966) 179 -- 191. [7] C. Faith, Algebra II, Ring Theory, Springer-Verlag, Berlin-New York, 1976. [8] J. L. G´omez Pardo, P.A. Guil Asensio, Essential embedding of cyclic modules in projectives, Trans. Amer. Math. Soc.349 (1997) 4343 -- 4353. [9] J. L. G´omez Pardo, P. A. Guil Asensio, Rings with finite essential socle, Proc. Amer. Math. Soc. 125 (1997) 971-977. [10] J. L.G´omez Pardo, M. F. Yousif, Semiperfect min-CS rings, Glasg. Math.J.41(1999) 231 -- 238. [11] Dinh Van Huynh, S. K. Jain, and S. R. Lopez-Permouth, When cyclic singular modules over a simple ring are injective, J. Algebra 263 (2003) 188-192. [12] F. Kasch, Modules and Rings, Academic Press, New York, 1982. [13] C. Lomp, On semilocal modules and rings, Comm. Algebra 27 (4) (1999) 1921 -- 1935. [14] G. O. Michler, O.E. Villamayor, On rings whose simple modules are injective, J. Algebra 25 (1973) 185 -- 201. [15] S. H. Mohamed, B. J. Muller, Continuous and Discrete Modules, Cambridge University Press, Cambridge, 1990. 8 NASR. A. ZEYADA AND AMR K. AMIN [16] W. K. Nicholson, M.F. Yousif, Mininjective rings, J. Algebra187 (1997) 548 -- 578. [17] W. K. Nicholson, M.F. Yousif, Quasi-Frobenius Rings, Cambridge Tracts in Math., 158, Cambridge University Press, Cambridge, 2003. [18] B. L. Osofsky, A generalization of quasi-Frobenius rings, J. Algebra4 (1966) 373 -- 387. [19] Liang Shen, Jianlong Chen, New characterizations of quasi-Frobenius rings, Comm. Algebra 34 (2006) 2157 -- 2165. [20] P. F. Smith, CS -modules and weak CS -modules, Noncommutative Ring Theory,, Springer LNM 1448 (1990) 99-115. [21] P. Vamos, The dual of the notion of 'finitely generated', J. London Math. Soc. 43 (1968) 186 -- 209. [22] M. F. Yousif, Y. Zhou, N. Zeyada, On pseudo-Frobenius rings, Canad. Math. Bull.48 (2) (2005) 317 -- 320. [23] N. Zeyada, S. Hussein, A. Amin, Rad-injective and Almost-injective Modules and Rings, algebra colleq. Volume: 18, 3(2011), 411-418. [24] N. Zeyada, N. Jarboui, s-CS Modules and Rings, Int. J. of Algebra Vol. 7, 2013, no. 2, 49 - 62.. Cairo University, Faculty of Science, Department of Mathematics, Egypt Current address: University of Jeddah, Faculty of Science, Department of Mathematics, Saudi Arabia E-mail address: [email protected] Current address: Umm Al-Qura University, University college, Departement of Mathematics, Saudi Arabia Beni-Suef University, Faculty of Science, Departement of Mathematics, Egypt E-mail address: [email protected]
1503.06668	2	1503	2015-05-19T14:15:38	Nil-clean companion matrices	[ "math.RA" ]	We describe nil-clean companion matrices over fields.	math.RA	math	NIL-CLEAN COMPANION MATRICES SIMION BREAZ AND GEORGE CIPRIAN MODOI Abstract. We describe nil-clean companion matrices over fields. 1. Introduction Nicholson proved in [16] that in the study of some important properties of rings it is important to know when we can decompose some (or all) elements of a ring as a sum of an idempotent and an element with some special properties. For instance, an element r of a ring R is • clean if r = u + e, where e is idempotent and u is a unit of R, [16]; • weakly-clean if r = u ± e, where e is idempotent and u is a unit of R, [1]; • nil-clean if r = n + e, where e is idempotent and n is nilpotent, [11]; • weakly nil-clean if r = n ± e, where e is idempotent and n is nilpotent, [10]. The classes of clean and nil-clean rings are closed with respect standard construc- tions, e.g. direct products of (nil-)clean rings are (nil-)clean, and matrix rings over (commutative nil-)clean rings are (nil-)clean, [12] (resp. [4]). However, the classes of weakly (nil-)clean rings are not closed under these constructions (see [1] and [5]). Moreover, while all matrix rings over fields are clean, when we consider nil- clean rings there are strongly restrictions: if a matrix ring over a division ring F is nil-clean then F has to be isomorphic to F2, [13]. It can be useful to know the (nil-)clean elements in some rings which are not (nil-)clean. For instance, strongly clean matrices (i.e. they have a decomposition r = u + e such that eu = ue) are characterized in [14, Theorem 3] for general rings and in [3, Theorem 37] for matrices over commutative local rings are studied (we reffer to [8] for a particular case). In particular it would be also nice to characterize nil-clean elements in matrix rings over division rings. For the case of strongly nil- clean elements (i.e. they have a decomposition r = e + n such that en = ne) we refer to [11, Theorem 4.4]. From this result we conclude that an n × n matrix over a division ring D is strongly nil-clean if and only if its characteristic polynomial has the form X (n−m)(X − 1)m. For other studies of (nil-)clean elements in various rings we refer to [2] and [7]. Since in the proofs of the fact that the matrix ring Mn(F) over the field F (resp. F = F2) is (nil-)clean the Frobenius (rational) normal form can be used, cf. [15], [19] and [4], it is useful to know when a companion matrix is nil-clean. In the main result of the present paper (Theorem 5) we characterize companion matrices over fields which are nil-clean. As a corollary of the main result we obtain a characterization Date: August 3, 2021. 2000 Mathematics Subject Classification. 15A24, 15A83, 16U99. This research is supported by the grant UEFISCDI grant PN-II-ID-PCE-2012-3-0100. 1 2 SIMION BREAZ AND GEORGE CIPRIAN MODOI for finite prime fields by the fact that every big enough companion matrix is nil- clean (Corollary 8). Moreover, it is proved that all these matrices have nil-clean like decompositions: for every polynomial χ of degree n such that the coefficient of X n−1 is 0, all big enough companion matrices can be decomposed as A = E + B where E is an idempotent and B is a matrix whose characteristic polynomial is χ. We can view this result as a partial answer to the general (open) problem, which is connected to the main result of [6], which asks to characterize the matrices A over F which can be decomposed A = B + C such that χB = f and g(C) = 0, where f and g are two fixed polynomials over a field F. In fact the problem of decompositions of a companion matrix in this way can be reduced to a matrix completion problem (for nil-clean decomposition we used Lemma 1 and Proposition 3). We refer to [9] for a nice survey on this subject. In this paper, F will denote a field, Mn(F) is the ring of all n × n matrices over F, and Fp is the prime field of characteristic p (p is a prime). If A ∈ Mn(F) we will denote by χA = det(XI − A) the characteristic polynomial of A. 2. Matrix completion results which involve idempotents Let F be a field and n a positive integer. For every tuple (c0, c1, . . . , cn−1) ∈ Fn, we denote by (C) C = Cc0,c1,...,cn−1 = 0 1 0 · · · 0 0   0 0 1 · · · 0 0 · · · · · · · · · · · · · · · · · · −c0 −c1 −c2 · · · 0 0 0 · · · 0 −cn−2 1 −cn−1   the n × n companion matrix determined by (c0, c1, . . . , cn−1). The characteristic polynomial associated to C is denoted by χc0,c1,...,cn−1 = X n + cn−1X n−1 + · · · + c1X + c0. We are grateful to an anonymous referee for pointing out that the matrix in the following lemma not only have the same characteristic polynomial, but is even similar to the respective companion matrix: Lemma 1. Let f = X n + fn−1X n−1 + . . . + f1X + f0 ∈ F[X] be a monic polynomial. For every tuple (c1, . . . , cn−1) ∈ Fn−1 there exists a unique tuple (α0, α1, . . . , αn−1) ∈ Fn such that the matrix M = −αn−1 −αn−2 1 0 · · · 0 0 0 1 · · · 0 0   · · · −α1 −α0 −c1 · · · · · · −c2 · · · · · · · · · · · · 0 0 · · · 0 1 −cn−2 −cn−1 ∈ Mn(F)   is similar to the companion matrix Cf0,f1,...,fn−1 of f . NIL-CLEAN COMPANION MATRICES 3 Proof. First we claim that the n-tuple (α0, α1, . . . , αn−1) ∈ Fn is uniquely deter- mined by the equality χM = f . The characteristic polynomial of M is χM = det(XIn − M ) = X + αn−1 αn−2 −1 · · · 0 0 X · · · 0 0 · · · α1 0 · · · · · · · · · · · · X · · · −1 X + cn−1 α0 c1 · · · cn−2 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) . (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Expanding it after the first line we obtain after some immediate computations: χM = (X + αn−1) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) c1 · · · cn−2 0 X · · · · · · · · · · · · · · · X 0 0 · · · −1 X + cn−1 0 X · · · · · · · · · · · · · · · X 0 0 · · · −1 X + cn−1 c2 · · · cn−2 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) + . . . + α3(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) −1 X + cn−1 (cid:12)(cid:12)(cid:12)(cid:12) cn−2 X + α1(X + cn−1) + α0 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) + αn−2 + α2(cid:12)(cid:12)(cid:12)(cid:12) X 0 −1 X 0 −1 X + cn−1 cn−3 cn−2 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = (X + αn−1)χc1,...,cn−1 + αn−2χc2,...,cn−1 + . . . + α1χcn−1 + α0. Therefore χM can be written as: χM = X n + (αn−1 + cn−1)X n−1 + (αn−2 + αn−1cn−1 + cn−2)X n−2 + . . . + (αn−i + αn−i+1cn−1 + . . . + αn−1cn−i+1 + cn−i)X n−i + . . . + (α1 + α2cn−1 + . . . + αn−1c2 + c1)X + (α0 + α1cn−1 + . . . + αn−1c1 + c0), where c0 = 0 is added only for uniformity. The system (αn−i + αn−i+1cn−1 + . . . + αn−1cn−i+1 + cn−i) = fn−i, 1 ≤ i ≤ n with the unknowns α0, α1, . . . , αn−1 has obviously a unique solution, proving the claim made at the very beginning of this proof. Finally we observe that the minimal polynomial of the matrix M is equal to its characteristic polynomial. Indeed by direct computation we can see that for every 1 ≤ k < n the first column of the matrix M k has 1 on the position (k + 1, 1) and only 0 bellow, that is on all positions (i, 1) with k + 1 < i ≤ n. This implies the matrix obtained by evaluating at M any monic polynomial of degree k < n has an entry 1 on the position (k + 1, 1), hence it is not zero. Therefore the minimal polynomial of M is of degree at least n, hence it is forced to coincide with χM . In these conditions we only have to note the equality χM = f is enough in order to conclude that M is similar to the companion matrix of f . (cid:3) Remark 2. In Lemma 1 above we may want αn−1, . . . , αn−i to be fixed, 1 ≤ i ≤ n, and we can determine uniquely fn−1, . . . , fn−i and αn−i−1, . . . , α0 such that M is similar to the companion matrix of the polynomial f . In the following we will make use of a particular case of this remark, namely when i = 1 and αn−1 is fixed. Noting that fn−1 = αn−1 + cn−1 we will determine α0, α1, . . . , αn−2 such that χM = X n + (αn−1 + cn−1)X n + fn−2X n−2 + . . . + f1X + f0, where f0, f1, . . . , fn−2 are arbitrary in F. 4 SIMION BREAZ AND GEORGE CIPRIAN MODOI Proposition 3. Let n = m+k be a positive integer, where m, k ∈ N∗. Fix constants c0, c1, . . . , cn−1 ∈ F and denote as above C = Cc0,c1,...,cn−1. For every polynomial g ∈ F[X] of degree at most n − 2 there exist two matrices E, M ∈ Mn(F) such that (1) C = E + M ; (2) E is idempotent, and (3) χM = X n + (k1 + cn−1)X n−1 + g. Proof. Let g = fn−2X n−2 + . . . + f1X + f0 ∈ F[X]. Consider the matrix E = (cid:20) E11 E12 E21 E22 (cid:21) ∈ Mn(F), where E11 = Ik ∈ Mk(F), E21 and E22 are both 0 (in Mm×k(F), respectively Mm(F)) and E12 = 0 0 · · · 0 0 0 · · · 0 αn−2 αn−3   0 0 · · · 0 α0 − c0 α1 − c1 · · · · · · · · · · · · αk−2 − ck−2 · · · αk αk−1 − ck−1 · · ·   where α0, α1, . . . , αn−2 ∈ F. Then it may be immediately verified that E = E2 is an idempotent. We will prove by induction on k ≥ 1 that there are uniquely determined α0, α1, . . . , αn−2 such that M = C −E has the characteristic polynomial equal to f = X n + (k1 + cn−1)X n−1 + g. The step k = 1 is just Lemma 1 (with fixed αn−1 = 1 and fn−1 = 1 + cn−1, as in Remark 2). Now suppose the claim is true for k − 1 ≥ 1, and let M = C − E as before. Expanding after the first column we compute: χM = det(XIn − M ) X + 1 −1 · · · 0 0 · · · 0 0 = (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) = (X + 1) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) Clearly the coefficient of X n−1 in χM is k1 + cn−1. 0 0 0 · · · 0 · · · 0 X + 1 · · · · · · · · · · · · X + 1 αn−2 · · · · · · · · · · · · · · · 0 0 · · · 0 0 −1 · · · 0 0 X · · · 0 0 X + 1 · · · 0 0 · · · 0 0 0 · · · 0 · · · · · · · · · · · · X + 1 αn−2 · · · · · · · · · · · · −1 · · · 0 0 X · · · 0 0 ck · · · cn−2 α0 α1 · · · αk−1 0 · · · 0 · · · · · · · · · · · · αk−2 0 · · · · · · · · · · · · X · · · −1 X + cn−1 0 · · · · · · · · · · · · αk−2 0 · · · · · · · · · · · · X · · · −1 X + cn−1 α1 · · · αk−1 ck · · · cn−2 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) + α0. NIL-CLEAN COMPANION MATRICES 5 By division with remainder theorem we obtain f = (X + 1)q + r with deg r ≤ 0, and χM = f if and only if q = X + 1 · · · 0 0 · · · 0 (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) 0 · · · 0 · · · · · · · · · · · · X + 1 αn−2 · · · · · · · · · −1 · · · 0 X · · · 0 α1 · · · αk−1 · · · · · · · · · · · · · · · · · · X + cn−1 ck · · · (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) and r = α0. Then the determinant giving q is the same form as det(XIn − M ) but of dimension (n − 1) × (n − 1) where n − 1 = (k − 1) + m and the coefficient of X n−2 is (k − 1)1 + cn−1. We apply induction hypothesis in order to determine (uniquely) α1, · · · , αn−2. (cid:3) 3. Nil-clean companion matrices We will use the results proved in the previous section to give a complete descrip- tion of nil-clean companion matrices over fields. We start with a result which is valid for all matrices. Recall that an element of a ring is unipotent if it is a sum 1 + b such that b is nilpotent. Lemma 4. Let A ∈ Mn(F) be a (not necessarily companion) matrix which is nil- clean. The following are true: (1) There exists a non-negative integer k such that trace (A) = k · 1. (2) If char(F) = 0 and trace (A) = k · 1 then (a) k ≤ n; (b) k = 0 if and only if A is nilpotent; (c) k = n if and only if A is unipotent. (3) If char(F) = p > 0 then (a) there exists k ∈ {1, . . . , p} such that trace (A) = k · 1, and k ≤ n or k = p; (b) if n < k = p then A is nilpotent; (c) if k = n < p then A is unipotent; (d) if k = n = p then A is unipotent or nilpotent. Proof. If A = E + N with E idempotent and N nilpotent then trace (A) = trace (E) + trace (N ) = trace (E). Moreover it is known that if E is an idem- potent matrix then trace (E) = rank(E) · 1 (see [5, Lemma 19]), so there is k ∈ N, k ≤ n such that trace (A) = k · 1. This k is unique if char(F) = 0 and is unique only modulo p if char(F) = p, so the statements (2)(a) and (3)(a) are obvious. (2)(b) If k = 0 then rank E = 0, hence E = 0, and this implies that A is nilpotent. The converse is obvious. (2)(c) If k = n then rank E = n, hence E = In, and this implies that A is unipotent. The converse is obvious. (3)(b) Since k = p we have trace (A) = 0, hence trace (E) = 0 and it follows that p is a divisor for rank E. But rank E < p, and it follows that rank E = 0, hence A is nilpotent. (3)(c) If k = n then rank E ≡ n(mod p), and it is easy to see that in our hypothesis we have rank E = n, hence E = In, and this implies that A is unipotent. (3)(d) If A = E + N where E = E2 is an idempotent and N is nilpotent then rank E·1 = 0, hence rank E ∈ {0, p}. This implies E ∈ {0n, In}. Now the conclusion is obvious. (cid:3) 6 SIMION BREAZ AND GEORGE CIPRIAN MODOI We are ready to enunciate the promised characterization for nil-clean companion matrices. Before this, let us remark that it is easy to verify if a (companion) matrix C is nilpotent or unipotent since these matrices are characterized by the conditions χC = X n, respectively χC = (X − In)n. Theorem 5. Let C = Cc0,c1,...,cn−1 ∈ Mn(F) be a companion matrix. The following are equivalent: (1) C is nil-clean. (2) One of the following conditions is true: (i) C is nilpotent (i.e. c0 = · · · = cn−1 = 0); (ii) C is unipotent (i.e. ci = (−1)i(cid:0) n n−i(cid:1) for all i ∈ {0, . . . , n − 1}); (iii) char(F) = 0 and there exists a positive integer k such that −cn−1 = k · 1 and n > k > 0; (iv) char(F) = p 6= 0 and there exists an integer k ∈ {1, . . . , p} such that −cn−1 = k · 1 and n > k. Proof. (1)⇒(2) Suppose that C is nil-clean, but it is not nilpotent nor unipotent. If char(F) = 0 then we can use Lemma 4 to observe that there exists a unique non-negative integer k ≤ n such that −cn−1 = k · 1. Since C is not nilpotent nor unipotent we obtain 0 < k < n. If char(F) = p > 0 then we can use again Lemma 4 to observe that there exists a unique integer k ∈ {1, . . . , p} such that −cn−1 = k · 1. If 0 6= E 6= In is an idempotent such that C − E is a nilpotent matrix then rank E ≡ k (mod p). Since rank E 6= 0 we have n ≥ rank E ≥ k. But E 6= In, hence rank E 6= n, and we obtain n > k. (2)⇒(1) If C is nilpotent or unipotent, then it is obviously nil-clean. Moreover, if we are in one of the casses (iii) or (iv), we can apply Proposition 3 for g = 0 to obtain a nil-clean decomposition for C. (cid:3) Remark 6. (a) The conditions (i) -- (iv) stated in Theorem 5 are not independent: if n > p then the nilpotent n × n companion matrix verify (iv). (b) In general the nil-clean decompositions obtained using Proposition 3 are not the unique nil-clean decompositions for companion matrices. For instance, if cn−1 = 1 then Cc0,...,cn−1 − C0,...,0 is idempotent. Example 7. (a) If n = 2 and p is a prime then the companion matrix C = Cc0,c1 over Fp is nil-clean, if and only if c0 = c1 = 0 (C is nilpotent) or c0 = 1 and c1 = −2 (C is unipotent) or c1 = −1 (in this case C = (cid:20) 0 −c0 0 1 (cid:21) +(cid:20) 0 0 1 0 (cid:21)). (b) If n = 3 and C = Cc0,c1,c2 is a companion matrix over Fp we have the following cases: • if c2 = 0, C is nil-clean exactly in one of the following situations: (i) p = 2, (ii) p = 3 and C is unipotent or nilpotent, (iii) p ≥ 5 and C is nilpotent; • if c2 = −1 then C is nil-clean since N =   nilpotent and C − N is idempotent; −1 c1 − 1 2c1 − 1 1 0 −c1 0 1 1   is NIL-CLEAN COMPANION MATRICES 7 • if c2 = −2 then C is nil-clean since N =   and C − N is idempotent; 1 0 −1 1 −1 p − 3 0 2 1 is nilpotent   • if p ≥ 5 and c2 = −3 then C is nil-clean if and only if it is unipotent; • if p ≥ 5 and c2 = −k with k ∈ {4, . . . , p − 1} then C is not nil-clean. (c) Finally, we illustrate the decompositions obtained by using Proposition 3 for the case n = 4 and p = 3: • for c3 = 0 we obtain that N =   −1 0 1 −1 0 0 0 −1 1 0 0 1 −1 0 1 0   is nilpotent and Cc0,c1,c2,0 − N is idempotent; • for c3 = 1 we obtain that N =   −1 1 0 0 c2 − 1 0 1 0 2 − 1 c1 − c2 − 1 −c1 − c2 −c1 −c2 1 0 0 1   is nilpotent and Cc0,c1,c2,1 − N is idempotent; • for c3 = −1 we obtain that N =   0 −1 1 −1 c2 0 0 1 0 0 −1 1 0 −c2 1 −1   is nilpotent and Cc0,c1,c2,0 − N is idempotent. As a corollary of this theorem we obtain a generalization for [4, Theorem 1]: Corollary 8. Let n ≥ 3 be a positive integer. The following are equivalent for a field F: (1) F ∼= Fp for a prime p < n; (2) every companion matrix C ∈ Mn(F) is nil-clean; (3) if C ∈ Mn(F) is a companion matrix then for every polynomial g ∈ F[X] of degree at most n − 2 there exist two matrices E, M ∈ Mn×n(F) such that C = E + M , E is idempotent, and χM = X n + g. Proof. (1)⇒(3) For every companion matrix C we can write trace (C) = k · 1 with k ∈ {1, . . . , p}, and we can apply Proposition 3. (3)⇒(2) This is obvious. (2)⇒(1) Since every element of the field F can be the trace of a companion matrix, it follows from Lemma 4 that every element from F has the form k · 1, k ∈ N. This implies that there exists a prime p such that F ∼= Fp. Moreover, if we supose p ≥ n, then use can use Theorem 5 to observe that the companion matrix C = C(−1)n+1,0,...,0 is not nil-clean (it is not nilpotent nor unipotent, and it does not verify the condition (iv) from Theorem 5). Therefore p < n, and the proof is complete. (cid:3) 8 SIMION BREAZ AND GEORGE CIPRIAN MODOI Remark 9. It is not hard to see that if all companion matrices which appear in the Frobenius normal form of a matrix A are nil-clean then A is also nil-clean. It would be nice to know if the converse of this remark is also true. For a proof of a very particular case we refer to [5], where it is proved that for p = 3 and every n ≥ 2 the matrix An = diag(1, −1, 0, . . . , 0) ∈ Mn(F3) is not nil-clean. Note that the Frobenius normal form for the matrix A4 is diag(0; C0,−1,0), where 0 ∈ M1(F3) is obviously nil-clean and C0,−1,0 ∈ M3(F3) is not nil-clean. On the other side, let us remark that if we consider only diagonals of companion matrices (non necessarily Frobenius normal forms) it is possible to obtain nil-clean matrices even the companion matrices are not nil-clean. For instance it is not hard to see that the minimal polynomial of the matrix A = diag(C0,0; C−1,0) ∈ M4(F3) is of degree 4. Therefore A is nil-clean since it is similar to a 4 × 4 companion matrix over F3. On the other side C−1,0 ∈ M2(F3) is not nil-clean. Acknowledgements. This study has the starting point in the work of S¸tefana Sorea, [18]. She proved by some direct computations that all 4 × 4 companion matrices over F3 are nil-clean. We would like to thank to the referee for his/her critical remarks, which helped us to improve the paper. References [1] M.-S Ahn, D.D. Anderson, Weakly clean rings and almost clean rings, Rocky Mountain J. Math. 36 (2006), 783 -- 798. [2] N. Arora, S. Kundu, Semiclean rings and rings of continuous functions, Journal of Commu- tative Algebra 6 (2014), 1 -- 16. [3] G. Borooah, A. J. Diesl, T. J. Dorsey, Strongly clean matrix rings over commutative local rings, J. Pure Appl. Algebra, 212 (2008), 281 -- 296. [4] S. Breaz, G. Calugareanu, P. Danchev and T. Micu, Nil-clean matrix rings, Linear Algebra Appl. 439 (2013), 3115 -- 3119. [5] S. Breaz, P. Danchev, Y. Zhou, Rings in which every element is either a sum or a difference of a nilpotent and an idempotent, preprint arXiv:1412.5544 [math.RA]. [6] V. Camillo, J.J. Sim´on, The Nicholson-Varadarajan theorem on clean linear transformations, Glasg. Math. J. 44 (2002), 365 -- 369. [7] H. Chen, Matrix nil-clean factorizations over abelian rings, preprint arXiv:1407.7509 [math.RA] [8] J. Chen, X. Yang, Y. Zhou, When is the 2 × 2 matrix ring over a commutative local ring strongly clean? J. Algebra 301 (2006), 280-293. [9] G. Cravo, Matrix completion problems, Linear Algebra Appl. 430 (2009), 2511 -- 2540. [10] P.V. Danchev and W.Wm. McGovern, Commutative weakly nil clean unital rings, preprint. [11] A.J. Diesl, Nil clean rings, J. Algebra 383 (2013), 197 -- 211. [12] J. Han, W.K. Nicholson, Extensions of clean rings, Commun. Algebra 29 (2001), 2589 -- 2595. [13] M.T. Kosan, T-K. Lee, Y. Zhou, When is every matrix over a division ring a sum of an idempotent and a nilpotent?, Linear Algebra Appl. 450 (2014), 7 -- 12. [14] W.K. Nicholson, Strongly clean rings and Fittings lemma, Comm. Algebra 27 (1999), 3583 -- 3592. [15] W.K.Nicholson, K. Varadarajan, Countable linear transformations are clean, Proc. Am. Math. Soc. 126 (1998), 61 -- 64. [16] W.K. Nicholson, Lifting idempotents and exchange rings, Trans. Amer. Math. Soc. 229 (1977), 269 -- 278. [17] W.K. Nicholson and Y. Zhou, Rings in which elements are uniquely the sum of an idempotent and a unit, Glasgow Math. J. 46 (2004), 227 -- 236. [18] S. Sorea, Identities in rings, Bachelor Thesis, Babe¸s-Bolyai University, 2014. [19] X. Yang, Y. Zhou, Strong cleanness of the 2 × 2 matrix ring over a general local ring, J. Algebra 320 (2008), 2280 -- 2290. NIL-CLEAN COMPANION MATRICES 9 Babes¸-Bolyai University, Faculty of Mathematics and Computer Science, Str. Mihail Kogalniceanu 1, 400084 Cluj-Napoca, Romania E-mail address: [email protected] Babes¸-Bolyai University, Faculty of Mathematics and Computer Science, Str. Mihail Kogalniceanu 1, 400084 Cluj-Napoca, Romania E-mail address: [email protected]
1810.04566	1	1810	2018-10-09T16:37:16	The structure of idempotent translatable quasigroups	[ "math.RA" ]	We prove the main result that a groupoid of order n is an idempotent k-translatable quasigroup if and only if its multiplication is given by x.y = (ax+by)(mod n), where a+b = 1(mod n), a+bk = 0(mod n) and (k,n)= 1. We describe the structure of various types of idempotent, k-translatable quasigroups, some of which are connected with affine geometry and combinatorial algebra, and their parastrophes. We prove that such parastrophes are also idempotent, translatable quasigroups and determine when they are of the same type as the original quasigroup. In addition, we find several different necessary and sufficient conditions making a k-translatable quasigroup quadratical.	math.RA	math	The structure of idempotent translatable quasigroups Wieslaw A. Dudek and Robert A. R. Monzo Abstract. We prove the main result that a groupoid of order n is an idempotent and k- translatable quasigroup if and only if its multiplication is given by x · y = (ax + by)(mod n), where a + b = 1(mod n), a + bk = 0(mod n) and (k, n) = 1. We describe the structure of various types of idempotent k-translatable quasigroups, some of which are connected with affine geometry and combinatorial algebra, and their parastrophes. We prove that such parastrophes are also idempotent k-translatable quasigroups and determine when they are of the same type as the original quasigroup. In addition, we find several different necessary and sufficient conditions making a k-translatable quasigroup quadratical. 1 Introduction Many collections of algebraic objects are equivalent to others. That is, there is a bijection between the collections V and V′, which lets us move back and forth freely between them. When this is the case, we write V ≡ V′. This equivalence may hold only up to isomorphism. In this case we write V ≈ V′. For example, if G, I, H, WQ, GHI, GH, EBM and NWQ are the collec- tions of groups, inverse semigroups, heaps, Ward quasigroups, generalized heaps that appear as the standard ternary operation of an inverse semigroup, generalized heaps, equivalence bimodules and the natural ternary operations of Ward quasi- groups, then G ≡ WQ, G ≈ H, G ≡ NWQ, NWQ ≈ H, H ≈ WQ, GHI ≡ I and GH ≈ EBM (cf. [13, 15, 20]). One hope might be that one could use results in, or properties of, one algebraic system to assist in proving important results that are difficult to prove if we remain only within an equivalent (≡ or ≈) algebraic system. It would be useful also if results in one system implied interesting results in the equivalent system. In this paper, we consider the collection QQ of all quadruples (Q, ·, Ls, Rs), where (Q, ·) is a quadratical quasigroup with a pair (Ls, Rs) of left and right translations respectively for some s ∈ Q, and the collection A of all quadruples (Q, +, λ, ρ), where (Q, +) is a commutative group with a pair of (commuting) automorphisms (λ, ρ) such that λx + ρx = x and ρλx + ρλx = x for all x ∈ Q. Note that from λx + ρx = x it follows that λ is an automorphism of (Q, +) if 2010 Mathematics Subject Classification: 20M15, 20N02 Keywords: Quasigroup, quadratical quasigroup, k-translability. 2 W. A. Dudek and R. A. R. Monzo and only if ρ is an automorphism of (Q, +). In this case λρ = ρλ. We prove that QQ ≡ A. One major result of this paper is that a quadratical quasigroup of order n is k-translatable if and only if it is induced by the additive group of integers modulo n = 4t+1, where t is some positive integer (Theorem 3.7). This is proved using the equivalence between QQ and A; in fact, the proof jumps back and forth between these equivalent collections. Moreover, in Section 3 we find necessary and sufficient conditions on a k-translatable quasigroup for it to be quadratical (Theorems 3.3 and 3.6). In Section 4 we prove the main result that a groupoid of order n is an idempo- tent k-translatable quasigroup if and only if its multiplication is given by x · y = (ax + by)(mod n), a + b = 1(mod n), a + bk = 0(mod n) and (k, n) = 1 (Lemma 4.1). Theorem 4.26 determines the structure of k-translatable quadratical, hexag- onal, GS, right modular, left modular, Stein, ARO and C3 quasigroups. The structure of the parastrophes of such quasigroups is determined in Theorem 5.1. In Theorem 5.3 we prove that these parastrophes are all idempotent translatable quasigroups and find the value of translatability. In Table 3 we determine when a parastrophe of an idempotent k-translatable quadratical (hexagonal, GS, ARO, C3, right modular or Stein) quasigroup is quadratical (hexagonal, GS, ARO, C3, right modular or Stein). In Table 4 we find necessary and sufficient conditions for a parastrophe to be quadratical (hexagonal, GS, ARO, C3, right modular or Stein). 2 Preliminary definitions and results Recall that a groupoid (Q, ·) has property A if it satisfies the identity xy ·x = zx·yz [8, 19]. It is called right solvable (left solvable) if for any {a, b} ⊆ Q there exists a unique x ∈ G such that ax = b (xa = b). It is left (right) cancellative if xy = xz implies y = z (yx = zx implies y = z). It is a quasigroup if it is left and right solvable. Note that right solvable groupoids are left cancellative, left solvable groupoids are right cancellative and quasigroups are cancellative. Volenec [19] defined quadratical groupoids as right solvable groupoids satisfying property A and proved some basic properties of these groupoids. Below, we list several such properties. Throughout the remainder of this paper we will use, without mention, the fact that quadratical groupoids are quasigroups that satisfy these properties. Theorem 2.1. A quadratical groupoid is left solvable and satisfies the following identities: The structure of idempotent translatable quasigroups 3 yx · xy = x, (1) x = x2, (2) x · yx = xy · x, (3) x · yx = xy · x = yx · y, (4) (5) x · yz = xy · xz, (6) xy · z = xz · yz, (7) xy · zw = xz · yw, (8) x(y · yx) = (xy · x)y, (9) (xy · y)x = y(x · yx), (10) xy = zw ⇐⇒ yz = wx, (idempotency) (elasticity, f lexibility) (strong elasticity) (bookend) (lef t distributivity) (right distributivity) (mediality) (alterability). Definition 2.2. QQ is defined as the collection of quadruples (Q, ·, λ, ρ), where (Q, ·) is a quadratical quasigroup with commuting automorphisms λ and ρ satis- fying for all x, y, z ∈ Q and some w ∈ Q the following conditions: (11) xy · λz = ρx · yz, (12) λx · ρx = x, (13) ρ−1x · λ−1y = ρ−1y · λ−1x, (14) ρ−1x · λ−1w = y. Proposition 2.3. If (Q, ·, λ, ρ) ∈ QQ, then λ = Lρx·λx, ρ = Rρx·λx and ρx · λx = ρy · λy for all x, y ∈ Q. Proof. For all x, y ∈ Q, by (13), ρ−1x · λ−1y = ρ−1y · λ−1x. Hence, ρλ(ρ−1x · λ−1y) = ρλ(ρ−1y · λ−1x). Since ρλ = λρ, λρ−1 = ρ−1λ, so λx · ρy = λy · ρx. By alterability, ρy · λy = ρx · λx. By (12), λx · ρx = x. But (Q, ·) is bookend and so λx = (ρx · λx) · (λx · ρx) = (ρx · λx) · x and ρx = (λx · ρx) · (ρx · λx) = x · (ρx · λx). Therefore, ρ = Rρx·λx and λ = Lρx·λx. Proposition 2.4. If (Q, ·) is a quadratical quasigroup, then (Q, ·, Ls, Rs) ∈ QQ for all s ∈ Q. Proof. Right and left distributivity imply that Rs and Ls are endomorphisms for each s ∈ Q. Thus, they are automorphisms of (Q, ·). Elasticity implies that Rs and Ls commute. By mediality, xy · Lsz = Rsx · yz, so (11) is valid. Bookend implies x = Lsx· Rsx , so (12) is valid. Mediality also implies Lsx· Rsy = Lsy · Rsx. Then, since RsLs = LsRs implies RsL−1 s y) = Lsx · Rsy = Lsy · Rsx = LsRs(R−1 s x and (13) is valid. Finally, right solvability implies that for all x, y ∈ Q there exists u ∈ Q such that R−1 is an automorphism, there exists w ∈ Q such that R−1 s w = y and (14) is valid. By definition then, (Q, ·, Ls, Rs) ∈ QQ. s x · u = y. Since L−1 s x). Hence R−1 s x · L−1 s y = R−1 s y · L−1 s y · L−1 s = L−1 s Rs, LsRs(R−1 s x · L−1 s x · L−1 s Definition 2.5. A is defined as the collection of quadruples (A, +, λ, ρ), where (A, +) is a 2-divisible commutative group with automorphisms λ and ρ such that: (15) ρx + λx = x, (16) λρx + λρx = x 4 W. A. Dudek and R. A. R. Monzo for all x ∈ A. Note that (15) and λ, ρ automorphisms imply ρλx = λx − λ2x = λρx, so ρλ = λρ. Lemma 2.6. If (A, +, λ, ρ) ∈ A, then for each x ∈ A there exists only one x ∈ A such that x = x + x. Proof. Indeed, by (16), for x = y+y we have λρx = λρ(y+y) = λρy+λρy = y. Definition 2.7. For (A, +, λ, ρ) ∈ A we define new product on A by putting: (17) x ⊕ y = ρx + λy for all x, y ∈ A. Similarly, for (Q, ·, λ, ρ) ∈ QQ we define: (18) x ⊙ y = ρ−1x · λ−1y for all x, y ∈ Q. Then (A, ⊕) and (Q, ⊙) are quasigroups. Proposition 2.8. Suppose that (A, +, λ, ρ) ∈ A and (Q, ·, λ, ρ) ∈ QQ. Then (19) (20) (x ⊕ y) + (z ⊕ w) = (x + z) ⊕ (y + w) f or x, y, z, w ∈ A, and (x · y) ⊙ (z · w) = (x ⊙ z) · (y ⊙ w) f or x, y, z, w ∈ Q. Proof. We have (x ⊕ y) + (z ⊕ w) = (ρx + λy) + (ρz + λw) = ρ(x + z) + λ(y + w) = (x + z) ⊕ (y + w) and (x · y) ⊙ (z · w) = ρ−1(x · y) · λ−1(z · w) = (ρ−1x · λ−1z) · (ρ−1y · λ−1w) = (x ⊙ z) · (y ⊙ w). The following Lemma follows from the proof of the main Theorem in [7]. Lemma 2.9. If (A, +, λ, ρ) ∈ A, then λρ = ρλ, (A, ⊕) is a quadratical quasigroup and (x ⊕ y) ⊕ z = (z ⊕ x) ⊕ (y ⊕ z) for all x, y, z ∈ A. Theorem 2.10. If (A, +, λ, ρ) ∈ A, then the following statements are equivalent: (21) x ⊕ y = z ⊕ w, (22) x + (w ⊕ y) = z + (y ⊕ w) and y + (x ⊕ z) = w + (z ⊕ x). Proof. (21) ⇒ (23). Since, by Lemma 2.9, (A, ⊕) is quadratical, it is alterable. Hence, y ⊕ z = w ⊕ x. By definition, x ⊕ y = z ⊕ w and y ⊕ z = w ⊕ x imply that ρx + λy = ρz + λw and ρy + λz = ρw + λx. So, (ρx + λy) + (ρw + λx) = (ρz + λw) + (ρy + λz) and (ρx + λy) + (ρy + λz) = (ρz + λw) + (ρw + λx). This proves (22). (22) ⇒ (21). By definition, (23) (ρx + λx) + (ρw + λy) = (ρz + λz) + (ρy + λw), (24) (ρy + λy) + (ρx + λz) = (ρw + λw) + (ρz + λx). Then, ρy + λz = (ρz + λw) + (ρw + λx) − (ρx + λy) = (ρx + λy) + (ρw + λx) − (ρz + λw). Therefore, 2(ρx + λy) = 2(ρz + λw), which, by Lemma 2.6, means that ρx + λy = ρz + λw. This proves (21). The structure of idempotent translatable quasigroups 5 Theorem 2.11. A ≡ QQ. Proof. We prove that the mappings Ψ : A → QQ; (A, +, λ, ρ) 7→ (A, ⊕, λ, ρ) and Φ : QQ → A; (Q, ·, λ, ρ) → (Q, ⊙, λ, ρ) are mutually inverse mappings. Let (A, +, λ, ρ) ∈ A. Then, by Lemma 2.9, (A, ⊕) is a quadratical quasigroup. Moreover, (x ⊕ y) ⊕ λz = (ρ2x + ρλy) + λ2z = ρ2x + λ(ρy + λz) = ρx ⊕ (y ⊕ z). So, (11) holds (for ⊕). Also, λx ⊕ ρx = ρλx + λρx = λρx + λρx = x, which proves (12). Then, ρ−1x ⊕ λ−1y = x + y = y + x = ρ−1y ⊕ λ−1x and so (13) holds. Since y = x + (y − x), y = ρ−1x ⊕ λ−1w for w = y − x. This proves (14). Finally, λ(x ⊕ y) = λ(ρx + λy) = ρλx + λ2y = λx ⊕ λy and so λ is an automorphism of (A, ⊕) and, similarly, so is ρ. Hence, (A, ⊕, λ, ρ) ∈ QQ and Ψ is well-defined. Conversely, let (Q, ·, λ, ρ) ∈ QQ. Then λρ = ρλ and ρ−1λ−1 = λ−1ρ−1. Now (11), for x = ρ−2a, y = λ−1ρ−1b and z = λ−2c, gives (ρ−2a · ρ−1λ−1b) · λ−1c = ρ−1a · (λ−1ρ−1b · λ−2c). Hence, (a ⊙ b) ⊙ c = ρ−1(ρ−1a · λ−1b) · λ−1c = (ρ−2a · ρ−1λ−1b)·λ−1c = ρ−1a·(λ−1ρ−1b·λ−2c) = a⊙(b⊙c) and so (Q, ⊙) is a semigroup. By, (13), it is commutative. From (14) it follows that for all x, y ∈ Q there exists w ∈ Q such that y = x ⊙ w. Hence, (Q, ⊙) is a commutative group. Now since ρx ⊙ λx = x · x = x, (15) holds. Also, ρ(x ⊙ y) = ρ(ρ−1x · λ−1y) = x · λ−1ρy = ρx ⊙ ρy. So ρ is an automorphism of (Q, ⊙). Similarly, λ is an automorphism of (Q, ⊙). Finally, by (12), λρx ⊙ λρx = ρλx ⊙ λρx = ρ−1(ρλx) · λ−1(λρx) = λx · ρx = x and so (16) holds too. Consequently, (Q, ⊙, λ, ρ) ∈ A and Φ is well-defined. Clearly, Ψ and Φ are mutually inverse mappings. 3 k-translatable quasigroups All groupoids considered in this section are finite. For simplicity we assume that they have form Q = {1, 2, . . . , n} with the natural ordering 1, 2, . . . , n, which is always possible by renumeration of elements. Moreover, instead of i ≡ j(mod n) we will write [i]n = [j]n. Additionally, in calculations of modulo n, we assume that 0 = n. Recall that a groupoid (Q, ·) is k-translatable if and only if for all i, j ∈ Q we have i · j = [i + 1]n · [j + k]n, or equivalently, i · j = a[k−ki+j]n , where a1, a2, . . . , an is the first row of the multiplication table of this groupoid. Then the sequence a1, a2, . . . , an is called a k-translatable sequence. For the original definition of k-translatability see [9] or [10]. Note that a groupoid may be k-translatable for one ordering but not for another, i.e., a1, a2, . . . , an may be a k-translatable sequence for the ordering 1, 2, . . . , n but in this groupoid for the ordering 2, 1, 3, . . . , n may not be any k- translatable sequence. However the following result is true [10, Lemma 2.7]. Lemma 3.1. The sequence a1, a2, . . . , an is a k-translatable sequence of (Q, ·) with respect to the ordering 1, 2, . . . , n if and only if ak, ak+1, . . . , an, a1, . . . , ak−1 is a k-translatable sequence of (Q, ·) with respect to the ordering n, 1, 2, . . . , n − 1. 6 W. A. Dudek and R. A. R. Monzo If (Q, ·) is a groupoid, then (Q, ∗), where x ∗ y = y · x is called the dual groupoid. It is clear that (Q, ∗) is idempotent (medial) if and only if (Q, ·) is idempotent (medial). Moreover, as a consequence of Theorem 4.1 in [10] we obtain the following result. Proposition 3.2. The dual groupoid of a left cancellative k-translatable groupoid of order n is k∗-translatable if and only if [kk∗]n = 1. Below we present several characterizations of finite k-translatable quadratical quasigroups. Theorem 3.3. Let (Q, ·) be a k-translatable quasigroup of order n. Then the following statements are equivalent: (Q, ·) is quadratical, [i + k(z · i)]n = [(j · z) + k(i · j)]n f or all i, j, z ∈ Q, [(z · i) + k(j · z)]n = [ki + (i · j)]n and [k2]n = [−1]n, (a) (b) (c) (d) Q is idempotent, medial and [k2]n = [−1]n. Proof. (a) ⇔ (b). In a k-translatable quasigroup (Q, ·) we have i · j = a[k−ki+j]n and ai = aj if and only if i = j. So, the condition A: (i · j) · i = (z · i) · (j · z) defining a quadratical quasigroup is equivalent to a[k−k(i·j)+i]n = a[k−k(z·i)+(j·z)]n , i.e., to [i + k(z · i)]n = [(j · z) + k(i · j)]n. (a) ⇔ (c). A quadratical quasigroup is alterable (Theorem 2.1). By Corollary 2.18 in [10],a k-translatable left cancellative groupoid (and so a quasigroup too) is alterable if and only if [k2]n = [−1]n. The condition A defining a quadratical quasigroup is, as in the previous part of this proof, equivalent to [i + k(z · i)]n = [(j · z) + k(i · j)]n. Multiplying both sides of this equation by k we obtain (c), and conversely, multiplying (c) by k we obtain (b). This proves the equivalence (a) and (c). (a) ⇔ (d). By Theorem 2.1, a quadratical quasigroup is idempotent, medial and alterable. If it is k-translatable, then [k2]n = [−1]n, by Corollary 2.18 in [10]. Conversely, by Corollary 2.18 in [10], a k-translatable quasigroup with the property [k2]n = [−1]n is alterable. Since it is idempotent and medial, it is elastic. Idempotency, elasticity and alterability imply bookend. By Theorem 2.30 in [8], a bookend, idempotent and medial quasigroup is quadratical. Corollary 3.4. A k-translatable, left cancellative, right distributive groupoid of order n is a quadratical quasigroup if and only if [k2]n = [−1]n. Proof. A quadratical groupoid is an alterable quasigroup. If it is a k-translatable, then [k2]n = [−1]n, by Corollary 2.18 in [10]. Conversely, a right distributive and left cancellative groupoid is idempotent and elastic. If it is k-translatable, then, by Proposition 2.13 from [10], it is left distributive too, and [k2]n = [−1]n shows, by Corollary 2.18 in [10], that it is alterable. An alterable, idempotent, elastic groupoid satisfies j = j ·j = (i·j)·(j ·i), The structure of idempotent translatable quasigroups 7 so it is bookend. An alterable, left distributive, bookend groupoid is a quadratical quasigroup [8, Theorem 2.30]. Corollary 3.5. A k-translatable, right solvable and right distributive, alterable groupoid of order n is a quadratical quasigroup and [k2]n = [−1]n. Proof. By Proposition 2.9 from [10] such groupoid is an idempotent quasigroup. Since it is alterable, [k2]n = [−1]n. By Corollary 3.4 it is quadratical. Theorem 3.6. Let (Q, ·) be a k-translatable, quasigroup of order n. Then the following statements are equivalent: (a) (Q, ·) is quadratical, (b) (Q, ·) is right distributive and [k2]n = [−1]n, (c) (Q, ·) is right distributive and alterable, (d) (Q, ·) is left distributive and alterable. Proof. A k-translatable quadratical quasigroup of order n satisfies (b), (c) and (d). Hence (a) implies (b), (c) and (d). By Corollary 3.4, (b) implies (a). By Corollary 3.5, (c) implies (a). Thus (a), (b) and (c) are equivalent. A k-translatable quasigroup satisfying (d) is idempotent. Hence, it is right distributive [10, Proposition 2.13]. Thus (d) implies (c), and in the consequence (a). This completes the proof. Finally, we will describe k-translatable quasigroups induced by the additive group Zn of positive integers modulo n. As is known (cf. [7]), all quadratical quasigroups are uniquely determined by some commutative groups and their two commuting automorphisms. In particular, a quadratical quasigroup induced by the additive group Zn of positive integers modulo n has the form x·y = [ax+(1−a)y]n, where a ∈ Zn and [2a2 −2a+1]n = 0 [8, Theorem 4.8]. For a fixed a ∈ Zn such a quasigroup may be k-translatable for only one value of k [9, Theorem 9.3]. This is possible only for [k2]n = [−1]n [10, Corollary 2.18]. Moreover, a quadratical quasigroup induced by Zn is k- translatable if and only if its dual quasigroup (n − k)-translatable [9, Theorem 9.4]. Theorem 3.7. A quadratical quasigroup of order n is k-translatable if and only if it is induced by the additive group Zn. Proof. Necessity. Let (Q, ·) be a k-translatable quadratical quasigroup of order n. Then x · y = ρx + λy for some commutative group (Q, +) of order n and two of its commuting automorphisms ρ and λ. Let e be the neutral element of (Q, +). Then e = ρe + λe = e · e = [e + 1]n · [e + k]n = ρ[e + 1]n + λ[e + k]n 8 W. A. Dudek and R. A. R. Monzo because (Q, ·) is k-translatable. Using Lemma 3.1 repeatedly, we can choose an ordering of Q such that e = n. Hence, e = n = n · n = 1 · k = ρ1 + λk, which means that in the group (Q, +) we have λk = −ρ1. Now, we prove by induction on i that ρi = [i(ρ1)]n for every i ∈ Q. Clearly, ρ1 = 1(ρ1). Assume ρj = j(ρ1) for all j ≤ i − 1. Then, [(i − 1)(ρ1)]n = ρ(i − 1) = ρ(i − 1) + λe = (i − 1) · n = i · k = ρi + λk = ρi − ρ1, which implies ρi = [i(ρ1)]n for all i ∈ Q. In particular, e = n = ρn = n(ρ1). Define ϕ : (Q, +) → (Zn, +) as ϕ(ρi) = i. Then, ϕ is one-to-one and onto. Also, ϕ(ρi + ρj) = ϕ([i(ρ1)]n + [j(ρ1)]n) = ϕ([i + j]n(ρ1)) = ϕ(ρ[i + j]n) = [i + j]n = [ϕ(ρi) + ϕ(ρj)]n. So, (Q, +) and Zn are isomorphic. Hence, (Q, ·) is induced by the additive group Zn. Sufficiency. A quadratical quasigroup induced by the additive group Zn has the form x · y = [ax + (1 − a)y]n, where a ∈ Zn and [2a2 − 2a + 1]n = 0 [8, Theorem 4.8]. Since [a + (1 − 2a)(1 − a)]n = [1 − 2a + 2a2]n = 0, this quasigroup is [1 − 2a]n-translatable [9, Lemma 9.1]. As a simple consequence of the above theorem and Proposition 3.4 in [8] we obtain the following two corollaries. Corollary 3.8. A quadratical quasigroup of order n is k-translatable if and only if it is induced by the group Z4t+1, for some positive integer t. Corollary 3.9. The set of all orders of k-translatable quadratical quasigroups is 2 · · · pαt {n : n = pα1 t }, where the pi are different primes such that pi ≡ 1(mod 4) for all i = 1, 2, . . . , t. 1 pα2 4 Idempotent k-translatable groupoids Let's recall that an idempotent k-translatable groupoid of order n is left cancella- tive [10, Lemma 2.10]. Thus in a k-translatable sequence a1, a2, . . . , an of this groupoid all elements are different, i.e., ai = aj if and only for i = j. Such a groupoid may not be right cancellative, but if (k, n) = 1, then it is a quasigroup [10, Theorem 2.14]. For every odd n and every k > 1 such that (k, n) = 1 there is only one (up to isomorphism) idempotent k-translatable quasigroup [10, Theorem 2.12]. For even n there are no such quasigroups [9, Theorem 8.9]. Lemma 4.1. In an idempotent k-translatable groupoid (Q, ·) with a k-translatable sequence a1, a2, . . . , an, for c = 2 · 1 and x, y ∈ Q (i) [kc − c − 2k + 1]n = 0, (ii) [(1 − k)x]n = [(1 − k)y]n implies x = y, (iii) a[i+1]n = [ai + (2 − c)]n = [1 + i(2 − c)]n, (iv) x · y = [(c − 1)x + (2 − c)y]n, The structure of idempotent translatable quasigroups 9 (v) (Q, ·) is a quasigroup if and only if (k, n) = 1, (vi) if (Q, ·) is a quasigroup, then n is odd. Proof. (i). Since (Q, ·) is idempotent and k-translatable, c · c = c = 2 · 1 = 1 · [1 − k]n = [1 + (c − 1)]n · [(1 − k) + k(x − 1)]n = c · [1 − k + kc − k]n. This, by left cancellativity, implies c = [1 − 2k + kc]n. (ii). [(1 − k)x]n = [(1 − k)y]n gives [x − y]n = [k(x − y)]n, i.e., t = [tk]n for t = [x − y]n. Then, by (i), we have 0 = [t(kc − c − 2k + 1)]n = [−t]n = [y − x]n. Hence x = y. (iii). Since ai = ai · ai = a[k−kai+ai]n , by left cancellativity i = [k − kai + ai]n. Thus, [i − k]n = [(1 − k)ai]n. Also [(1 − k)(2 − c)]n = [2 − c − 2k + kc]n = 1, by (i). Now, using these two facts, we obtain [(1−i)(ai +(2−c))]n = [(1−k)ai +1]n = [i− k + 1]n = [(1 − k)ai+1]n. From this, by (ii), we deduce that a[i+1]n = [ai + (2 − c)]n. But a1 = 1 ·1 = 1, so a2 = [1 + (2 − c)]n, and, by induction, a[i+1]n = [1 + i(2 − c)]n. (iv). We have x · y = aq, where q = [k − kx + y]n. This, by (iii), gives x·y = aq = [1+(k−kx+y−1)(2−c)]n = [1+(2−c)(k−1)−k(2−c)x+(2−c)y]n (i) = [(kc − 2k)x + (2 − c)y]n (i) = [(c − 1)x + (2 − c)y]n. (v). This follows from Theorem 2.14 [10] and Lemma 2.15 [10]. (vi). Observe that (i) can be rewritten in the form [(k − 1)(c − 1) − k]n = 0. This, for odd k, means that n is odd. From (k, n) = 1 it follows that n must be odd also for even k. So, in both cases n is odd. Theorem 4.2. A naturally ordered groupoid (Q, ·) of order n is idempotent and k-translatable if and only if x · y = [ax + by]n for all x, y ∈ Q and some a, b ∈ Zn and (1) [a + b]n = 1, (2) 1 · 1 = 1, 1 · k = n. [a + bk]n = 0, or equivalently, Proof. By Lemma 4.1(iv), an idempotent k-translatable groupoid (Q, ·) of order n has the form x · y = [ax + by]n, where a = [c − 1]n, b = [2 − c]n and c = 2 · 1. Then obviously, [a + b]n = 1 and x · y = [x + 1]n · [y + k]n, which gives [a + bk]n = 0. The converse statement is obvious. Note that in the above theorem (b, n) = 1. Indeed, since (Q, ·) is left can- cellative, x · y = x · w implies y = w. But this is possible only in the case when (b, n) = 1. So, ψ(x) = [bx]n is an automorphism of the additive group Zn, but ϕ(x) = [ax]n may not be one-to-one. For example, the groupoid with the operation x · y = [4x + 5y]8 is idempotent and 4-translatable, but it is not a quasigroup. Corollary 4.3. If a naturally ordered idempotent groupoid of order n has the form x · y = [ax + by]n, then it is k-translatable if and only if [a + bk]n = 0 and (k, n) divides (a, n). Proof. Indeed, (k, n) = d together with [a + bk]n = 0 gives da. 10 W. A. Dudek and R. A. R. Monzo Corollary 4.4. If a naturally ordered idempotent quasigroup (Q, ·) of order n has the form x · y = [ax + by]n, then it is k-translatable if and only if [a + bk]n = 0 and (k, n) = 1. Corollary 4.5. An idempotent k-translatable groupoid is medial. Corollary 4.6. A naturally ordered groupoid (Q, ·) of order n is a k-translatable quadratical quasigroup if and only if x · y = [ax + by]n for all x, y ∈ Q and some a, b ∈ Zn such that (1) [a + b]n = 1, (2) 1 · 1 = 1, 1 · k = n, [a2 + b2]n = 0. [a + bk]n = 0, [2ab]n = 1, or equivalently, Proof. Since (Q, ·) is idempotent and k-translatable, by our Theorem 4.2 and Theorem 4.8 in [8], the multiplication in such a quasigroup is given by x · y = [ax + by]n, where a, b ∈ Zn are such that [a + b]n = 1 and [2a2 − 2a + 1]n = 0. The last implies [2ab]n = 1. By translatability we also have [a + bk]n = 0. Conversely, if a groupoid of order n satisfies the above conditions, then b = [1 − a]n, which, by [2ab]n = 1, implies [2a2 − 2a + 1]n = 0 and (a, n) = (b, n) = 1. So, this groupoid is quadratical [8, Theorem 4.8]. Since (a, n) = (b, n) = 1, it is a quasigroup. Note that by [8, Proposition 3.4], the order of a finite quadratical quasigroup is n = 4t + 1 for some t = 0, 1, 2, . . . Corollary 4.7. A quadratical quasigroup of the form x · y = [ax + by]n is k- translatable only for k = [1 − 2a]n. Its dual quasigroup is [2a − 1]n-translatable. Proof. In such a quasigroup [2a2 − 2a + 1]n = 0, [2ab]n = 1 and [a + bk]n = 0. Multiplying the last equation by 2a we obtain [2a2+k]n = 0, whence k = [−2a2]n = [1 − 2a]n. Since, [(1 − 2a)(2a − 1)]n = 1, by Proposition 3.2, the dual quasigroup is [2a − 1]n-translatable. Using Theorem 4.2 we can determine the structure of various k-translatable quasigroups strongly connected with the affine geometry and combinatorial de- signs. We will start with hexagonal quasigroups, which are connected with affine ge- ometry (see for example [18]), i.e., with idempotent medial quasigroups satisfying the identity x · yx = y. By Lemma 4.1, k-translatable hexagonal quasigroups have odd order n and (k, n) = 1. Proposition 4.8. A naturally ordered groupoid (Q, ·) of order n is a k-translatable hexagonal quasigroup if and only if [k2 − k + 1]n = 0 and x · y = [(1 − k)x + ky]n. Proof. Necessity. Let (Q, ·) be a naturally ordered k-translatable hexagonal quasigroup of order n. Since, by definition, (Q, ·) is idempotent, by Lemma 4.1, we have x · y = [(c − 1)x + (2 − c)y]n, where c = 2 · 1. From hexagonality we obtain 1 · c = 1 · (2 · 1) = 2, which, by k-translatability, implies 1 · (c − 1) = c. Therefore, The structure of idempotent translatable quasigroups 11 (Q, ·) is n − (c − 2)-translatable and so k = [2 − c]n. Consequently, c = [2 − k]n and x · y = [(1 − k)x + ky]n. Moreover, 0 = [a + bk]n = [1 − k + k2]n. Sufficiency. It is easy to see that a naturally ordered groupoid (Q, ·) of odd order n with x · y = [(1 − k)x + ky]n, where [k2 − k + 1]n = 0, is idempotent, medial, k-translatable and satisfies the identity x·yx = y. Since 0 = [k2 −k +1]n = [k(k−1)+1]n, we also have (k, n) = (k−1, n) = 1. Thus, (Q, ·) is a quasigroup. Corollary 4.9. A naturally ordered groupoid (Q, ·) of order n is a k-translatable hexagonal quasigroup if and only if x · y = [(c − 1)x + (2 − c)y]n, where c = 2 · 1, k = [2 − c]n and [c2 − 3c + 3]n = 0. Corollary 4.10. A naturally ordered hexagonal quasigroup (Q, ·) of order n may be k-translatable only for k = [2 − c]n, where c = 2 · 1. Corollary 4.11. A hexagonal quasigroup of order n is k-translatable if and only if its dual quasigroup is [1 − k]n-translatable. Proof. This is a consequence of Proposition 3.2 and fact that 1 = [k − k2]n = [k(1 − k)]n. Corollary 4.12. There is (up to isomorphism) only one k-translatable commu- tative hexagonal quasigroup. It is induced by the group Z3 and has the form x · y = [2x + 2y]3. Proof. Indeed, in this case [1 − k]n = k and k = n − 1. So, [3]n = 0. Thus, n = 3 and consequently, x · y = [2x + 2y]3. GS-quasigroups (golden section quasigroups) defined in [17] are used to describe various objects in affine geometry (see for example [12] and [17]). They are defined as idempotent quasigroups satisfying the (mutually equivalent) identities x(xy · z) · z = y and x · (x · yz)z = y. It is not difficult to see that these quasigroups are medial [17]. Hence, by Lemma 4.1, k-translatable GS-quasigroups have odd order n and (k, n) = 1. Proposition 4.13. A naturally ordered groupoid (Q, ·) of order n is a k-translata- ble GS-quasigroup if and only if x· y = [(k − 1)x+ (2 − k)y]n and [k2 − 3k + 1]n = 0. Proof. By Theorem 4.2, the multiplication in a k-translatable GS-quasigroup of order n is defined by x · y = [ax + by]n, where [a + b]n = 1, [a + bk]n = 0 and (a, n) = 1. From x(xy · z) · z = y we obtain [a2 − a − 1]n = 0 and [ab + 1]n = 0. Multiplying [a + bk]n = 0 by a we get [a2 − k]n = 0. So, k = [a2]n = [a + 1]n. Thus, a = k − 1. Consequently x · y = [(k − 1)x + (2 − k)y]n and [k2 − 3k + 1]n = 0. Conversely, a groupoid satisfying these conditions is idempotent and k-transla- table. It also satisfies the identity x(xy · z) · z = y. From [(k − 1)(2 − k) + 1]n = [−(k2 − 3k + 1)]n = 0 it follows that (k − 1, n) = (2 − k, n) = 1. Thus, this groupoid is a GS-quasigroup. 12 W. A. Dudek and R. A. R. Monzo Corollary 4.14. A naturally ordered groupoid (Q, ·) of order n is a k-translata- ble GS-quasigroup if and only if its dual groupoid is a [3 − k]n-translatable GS- quasigroup. Corollary 4.15. A commutative GS-quasigroup is k-translatable only for k = 4. It has the form x · y = [3x + 3y]5. Proof. Indeed, if it is commutative, then [k − 1]n = [2 − k]n, i.e., [2k]n = 3. From 2 · 1 = 1 · 2 it follows k = n − 1. Thus 3 = [2k]n = [−2]n. So, n = 5 and k = 4. Other quasigroups associated with the affine geometry are ARO-quasigroups defined as idempotent medial quasigroups satisfying the identity xy · y = yx · x (cf. [21]). Thus, k-translatable ARO-quasigroups have odd order n and (k, n) = 1. Proposition 4.16. A naturally ordered groupoid (Q, ·) of order n is a k-translatable ARO-quasigroup if and only if x · y = [ax + by]n for some a, b ∈ Zn such that [a + b]n = 1, [a + bk]n = 0 and [2a2]n = 1. Proof. If (Q, ·) is a k-translatable ARO-quasigroup, then, by Theorem 4.2, x · y = [ax + by]n, [a + b]n = 1 and [a + bk]n = 0 for some a, b ∈ Zn. The identity xy · y = yx · x implies [2a2]n = 1. Conversely, a groupoid (Q, ·) satisfying these conditions is a k-tanslatable ARO-groupoid. From [2a2]n = 1 it follows that (a, n) = 1. Now, if db and dn, then, by [a + bk]n = 0, da, so (b, n) = 1. Thus (Q, ·) is a quasigroup. Now we describe several types of idempotent k-translatable quasigroups as- sociated with combinatorial designs [5]. First, we will describe idempotent k- translatable quasigroups satisfying the identity (xy · y)y = x. Idempotent quasi- groups satisfying this identity are called C3 quasigroups and correspond to a class of resolvable Mendelsohn triple systems (see, for example, [3]). It was shown in [1] that finite C3 quasigroups exist only for orders n = 3t + 1. By Lemma 4.1, k-translatable C3 quasigroups have odd order n and (k, n) = 1. Proposition 4.17. A naturally ordered groupoid (Q, ·) of order n is a k-translatable C3 quasigroup if and only if x · y = [ax + by]n for some a, b ∈ Zn such that [a + b]n = 1, [a + bk]n = 0 and [a3]n = 1. Proof. Proof. By Theorem 4.2, the multiplication in a k-translatable C3 quasi- group is defined by x·y = [ax+by]n, where a, b ∈ Zn, [a+b]n = 1 and [a+bk]n = 0. From (xy · y)y = x it follows that [a3]n = 1. Conversely, a groupoid satisfying these conditions is an idempotent, k-trans- latable C3 groupoid. From [a3]n = 1 it follows that it is right cancellative and, hence, it is a quasigroup. Corollary 4.18. A commutative C3 quasigroup is k-translatable only for k = 6 and has the form x · y = [4x + 4y]7. The structure of idempotent translatable quasigroups 13 Proof. In this case a = b, [2a]n = 1 and [a2 + a + 1]n = 0. Thus, 0 = [2(a2 + a + 1)]n = [a + 3]n. So, n = 7 and k = 6. Proposition 4.19. A naturally ordered groupoid (Q, ·) of order n is a k-translatable ARO-quasigroup if and only if x · y = [ax + by]n for some a, b ∈ Zn such that [a + b]n = 1, [a + bk]n = 0 and [2a2]n = 1. Proof. If (Q, ·) is a k-translatable ARO-quasigroup, then, by Theorem 4.2, x · y = [ax + by]n, [a + b]n = 1 and [a + bk]n = 0 for some a, b ∈ Zn. The identity xy · y = yx · x implies [2a2]n = 1. Conversely, a groupoid (Q, ·) satisfying these conditions is a k-tanslatable ARO-groupoid. From [2a2]n = 1 it follows that (a, n) = 1. Now, if db and dn, then, by [a + bk]n = 0, da, so (b, n) = 1. Thus (Q, ·) is a quasigroup. A groupoid (Q, ·) satisfying the identity x · xy = yx is called a Stein groupoid. Left cancellative Stein groupoids are idempotent and right cancellative. So, idem- potent k-translatable Stein groupoids are idempotent k-translatable quasigroups. Stein quasigroups have applications in the theory of Latin squares (cf. [11] and [16]) and combinatorial designs (cf. [2]). Below we show that k-translatable Stein quasigroups are associated with left modular groupoids, i.e., groupoids satisfying the identity x · yz = z · yx. Their dual groupoids are associated with groupoids satisfying the identity xy · z = zy · x and are called right modular. By Lemma 4.1, k-translatable Stein quasigroups are of odd order n and (k, n) = 1. As a simple consequence of Theorem 4.2 we obtain the following two lemmas. Lemma 4.20. A naturally ordered idempotent k-translatable groupoid (Q, ·) of order n is left modular if and only if x · y = [ax + by]n for some a, b ∈ Zn, [a + b]n = 1, [a + bk]n = 0 and [a2 − 3a + 1]n = 0. Lemma 4.21. A naturally ordered idempotent k-translatable groupoid (Q, ·) of order n is right modular if and only if x · y = [ax + by]n for some a, b ∈ Zn, [a + b]n = 1, [a + bk]n = 0 and [a2 + a − 1]n = 0. Proposition 4.22. A naturally ordered groupoid (Q, ·) of order n is a k-translatable Stein quasigroup if and only if [k2 − k − 1]n = 0 and x · y = [(k + 1)x − ky]n. Proof. Let (Q, ·) be a k-translatable Stein quasigroup of order n. Then, by Theo- rem 4.2, x · y = [ax + by]n for some a, b ∈ Zn, where [a + b]n = 1, [a + kb]n = 0, [b2]n = a and [a2 − 3a + 1]n = 0. From 0 = [a + bk]n we obtain 0 = [ab + b2k]n = [a(b + k)]n = [(a − 3)a(b + k)]n = [−(b + k)]n. Thus, b = [−k]n. Therefore x · y = [(k + 1)x − ky]n. Obviously [k2 − k − 1]n = 0. Conversely, a groupoid satisfying these conditions is an idempotent k-trans- latable Stein groupoid. Since [(k + 1)(k − 2)]n = −1, (Q, ·) is right cancellative and, so, it is a quasigroup. 14 W. A. Dudek and R. A. R. Monzo Corollary 4.23. k-translatable Stein quasigroups cannot be commutative. Corollary 4.24. A groupoid (Q, ·) is a k-translatable Stein quasigroup if and only if it is an idempotent k-translatable left modular quasigroup. Corollary 4.25. A groupoid (Q, ·) of order n is a k-translatable Stein quasigroup if and only if its dual groupoid is an idempotent [k − 1]n-translatable right modular quasigroup. As a consequence of the above results we obtain the following two theorems. Theorem 4.26. A naturally ordered groupoid (Q, ·) of order n with the multipli- cation defined by x · y = [ax + by]n, where a, b ∈ Zn, [a + b]n = 1 and [a + bk]n = 0 is a k-translatable quasigroup that is (1) quadratical if and only if [2a2 − 2a + 1]n = 0, (2) hexagonal if and only if [a2 − a + 1]n = 0, (3) GS-quasigroup if and only if [a2 − a − 1]n = 0, (4) right modular quasigroup if and only if [a2 + a − 1]n = 0, (5) left modular quasigroup if and only if [a2 − 3a + 1]n = 0, (6) Stein quasigroup if and only if [a2 − 3a + 1]n = 0, (7) ARO-quasigroup if and only if [2a2]n = 1, (8) C3 quasigroup if and only if [a3]n = 1. Theorem 4.27. A naturally ordered groupoid (Q, ·) of order n with the multipli- cation defined by x · y = [ax + by]n, where a, b ∈ Zn, [a + b]n = 1 and [a + bk]n = 0, is a k-transalatable (1) quadratical quasigroup if and only if k = [1 − 2a]n, (2) hexagonal quasigroup if and only if k = [1 − a]n, (3) GS-quasigroup if and only if k = [a + 1]n, (4) right modular quasigroup if and only if k = [−1 − a]n, (5) left modular quasigroup if and only if k = [a − 1]n, (6) ARO-quasigroup if and only if k = [−1 − 2a]n, (7) C3 quasigroup if and only if [(1 − a2)k]n = 1. Quasigroups satisfying the identity x(xy · z) = (y · zx)x are called Cheban quasigroups. The structure of idempotent translatable quasigroups 15 Proposition 4.28. There are no idempotent k-translatable Cheban quasigroups. Proof. The Cheban identity implies (1) [2a3−3a2−a+1]n = 0, (2) [a3−3a2+a]n = 0 and (3) [a3 − 2a + 1]n = 0. From (2) and (3), multiplying each identity by a, we obtain (4) [3a3]n = [3a2 − a]n. By (2) and (4), [3a3]n = [a3]n and so [2a3]n = 0. Multiplying (2) by 2a we obtain [2a2]n = 0. Multiplying (2) by 2 we get [2a]n = 0. Now, multiplying (1) by 2 we obtain [2]n = 0. So, n = 2, a contradiction because n must be odd. A Schröder quasigroup is a quasigroup satisfying the identity xy · yx = x. Proposition 4.29. There are no idempotent k-translatable Schröder quasigroups. Proof. By Theorem 4.2, in an idempotent k-translatable Schröder quasigroup of order n must be [2a2]n = [2a]n. Hence, multiplying [a + bk]n = 0 by 2a we obtain [2a]n = 0. Thus 2 · n = [2a + bn]n = n = n · n, a contradiction. From the above results it follows that the only idempotent k-translatable quasi- group that is C3 and quadratical is x · y = [3x + 11y]13. The only idempotent k-translatable quasigroup that is C3 and ARO is x · y = [2x + 6y]7. The only idempotent k-translatable quasigroup that is right modular and quadratical is x · y = [2x + 4y]5 . The only idempotent k-translatable quasigroup that is hexago- nal and ARO is x · y = [5x + 3y]7. The only idempotent k-translatable quasigroup that is the dual of a C3 quasigroup and ARO is x · y = [27x + 5y]31. There are no idempotent k-translatable quasigroups that are ARO and right modular, Stein and C3, Stein and right modular, right modular and GS, ARO and Stein, GS and ARO, Stein and GS, GS and C3, Hexagonal and Stein, Hexagonal and C3, Hexagonal and right modular, hexagonal and GS, quadratical and hexagonal, quadratical and ARO, quadratical and GS, C3 and right modular, or quadratical and Stein. 5 Parastrophes of k-translatable quasigroups Each quasigroup (Q, ·) determines five new quasigroups Qi = (Q, ◦i) with the operations ◦i defined as follows: x ◦1 y = z ⇔ x · z = y, x ◦2 y = z ⇔ z · y = x, x ◦3 y = z ⇔ z · x = y, x ◦4 y = z ⇔ y · z = x, x ◦5 y = z ⇔ y · x = z. Such defined quasigroups are called conjugates or parastrophes of (Q, ·). Since x ◦1 y = y ◦4 x, x ◦2 y = y ◦3 x and x ◦5 y = y · x, parastrophes of a given quasigroups are pairwise dual. Moreover, parastrophes of some quasigroups are pairwise equal 16 W. A. Dudek and R. A. R. Monzo or isotopic (cf. [11], [14] and [6]). Parastrophes of idempotent quasigroups are idempotent, but parastrophes of translatable quasigroups are not translatable in general. Indeed, a 3-translatable quasigroup Q defined by the multiplication table with the first row 1, 4, 3, 2, 8, 7, 6, 5 has only one translatable parastrophe, namely its dual quasigroup, i.e., (Q, ◦5). In Theorem 5.3 we will prove that parastrophes of an idempotent k-translatable quasigroup also are idempotent and k-translatable, but for other value of k. Let Qa = (Q, ·) be a naturally orderd idempotent and k-translatable quasigroup of order n with the multiplication x · y = [ax + by]n, a, b ∈ Zn. Then, [a + b]n = 1 and (a, n) = (b, n) = 1. The last means that aa′ + ns = 1 = bb′ + nt for some a′, b′, s, t, i.e., [aa′]n = [bb′]n = 1. It is clear that elements a′, b′ ∈ Zn are uniquely determined. Moreover, a′ = b′ is equivalent to a = b. Theorem 5.1. Multiplications of parastrophes of a quasigroup Qa have the form: (1) x ◦1 y = [(1 − b′)x + b′y]n, (2) x ◦2 y = [a′x + (1 − a′)y]n, (3) x ◦3 y = [(1 − a′)x + a′y]n, (4) x ◦4 y = [b′x + (1 − b′)y]n, (5) x ◦5 y = [bx + ay]n, where [aa′]n = [bb′]n = 1. Proof. (1). By definition, x◦1y = z ⇔ x·z = y ⇔ [ax+bz]n = y, and consequently, [bz]n = [y − ax]n. From this, multiplying by b′, we obtain z = [b′y − ab′x]n = [−ab′x + b′y]n. So, x ◦1 y = [−ab′x + b′y]n = [(1 − b′)x + b′y]n. (2). Similarly, x ◦2 y = z ⇔ z · y = x ⇔ [az + by]n = x. Whence, x ◦2 y = [a′x − a′by]n = [a′x + a′(a − 1)y]n = [a′x + (1 − a′)y]n. Analogously we can prove (3) and (4). (5) is obvious. Corollary 5.2. All parastrophes of a quasigroup Qa are isotopic to the group Zn. Theorem 5.3. A quasigroup Qa is k-translatable for k = [1−b′]n. Its parastrophes are k∗-translatable, where (1) k∗ = a for (Q, ◦1), (2) k∗ = [1 − k]n for (Q, ◦2), (3) k∗ = [1 − a]n for (Q, ◦3), The structure of idempotent translatable quasigroups 17 (4) k∗ = a′ for (Q, ◦4), (5) k∗ = [1 − a′] for (Q, ◦5). Proof. From [a + (1 − b′)b]n = [a + b − 1]n = 0 it follows that a quasigroup Qa is k-translatable for k = [1 − b′]n. Similartl, from x ◦1 y = [(1 − b′)x + b′y]n and [(1 − b′) + b′a]n = [(1 − a′) + b′(1 − b)]n = 0 it follows that (Q, ◦1) is k∗-translatable for k∗ = a. Analogously we can verify other cases. Lemma 5.4. For a quasigroup Qa and its parastrophes the following relationships are possible: (1) Qa = Q1 ⇔ x · y = [2x − y]n, (2) Qa = Q2 ⇔ x · y = [−x + 2y]n, (3) Qa = Q3 ⇔ [ab]n = 1, (4) Qa = Q4 ⇔ [ab]n = 1, (5) Qa = Q5 ⇔ a = b. Proof. Indeed, Qa = Q1 implies b = b′. Thus 1 = [bb′]n = [1 − 2a + a2]n, i.e., [a2]n = [2a]n, whence, multiplying by a′, we obtain a = 2, and consequently b = n − 1. So, x · y = [2x − y]n. The converse statement is obvious. Other statements can be proved analogously. Theorem 5.5. For parastrophes of a quasigroup Qa the following cases are pos- sible: (a) Qa = Q1 = Q2 = Q3 = Q4 = Q5 ⇔ x · y = [2x + 2y]3, (b) Qa = Q3 = Q4 6= Q1 = Q2 = Q5 ⇔ [ab]n = 1, a 6= b, (c) Qa = Q1 6= Q2 = Q3 6= Q4 = Q5 ⇔ x · y = [2x − y]n and n > 3, (d) Qa = Q2 6= Q1 = Q4 6= Q3 = Q5 ⇔ x · y = [−x + 2y]n and n > 3, (e) Qa = Q5 6= Q1 = Q3 6= Q2 = Q4 ⇔ a = b and n > 3, (f ) Qa 6= Q1 6= Q2 6= Q3 6= Q4 6= Q5 ⇔ a 6= b 6= 2 and n > 3. Proof. Let Qa be an idempotent k-translatable quasigroup of order n. Then n is odd. For n = 3 we have only one possibility: x · y = [2x + 2y]3. Then, as it is not difficult to see, all parastrophes of Qa are equal to Qa. Conversely, by Lemma 5.4, Qa = Q1 = Q2 implies [2x − y]n = [−x + 2y]n. Thus n = 3 and x · y = [2x + 2y]3. This proves (a). Now let n > 3. Then, by Lemma 5.4, Qa = Q3 ⇔ [ab]n = 1 ⇔ a′ = b, b′ = a ⇔ Qa = Q3 = Q4 6= Q1 = Q2 = Q5 and a 6= b, since a = b implies n = 3. This 18 W. A. Dudek and R. A. R. Monzo proves (b). If Qa = Q1, then, by Lemma 5.4, x · y = [2x − y]n. Then obviously, Qa = Q1 6= Q2 = Q3 6= Q4 = Q5. This proves (c). The proof of (d) is analogous. The case (e) is obvious. So, Qa = Qi implies one of statements (a) − (d). To prove (f ) suppose that Qa 6= Qi for all i = 1, . . . , 5. Then also Qs 6= Qt for all 1 6 s < t 6 5. Indeed, Q1 = Q2 means that 1 = [a′ + b′]n, whence, multiplying by ab, we obtain [ab]n = [b + a]n = 1. So, by Lemma 5.4, Qa = Q3. Q1 = Q3 implies Qa = Q5. Q1 = Q4 gives 1 = [2b′]n, and consequently, b = 2, a = n − 1. Thus Qa = Q2, by Lemma 5.4. From Q1 = Q5 we obtain a = b and a′ = b′. This implies Q1 = Q3, and in the consequence, Qa = Q5. In the case Q2 = Q3 we have 1 = [2a′]n, whence a = 2, b = n − 1 and Qa = Q1 by Lemma 5.4. Q2 = Q4 implies Qa = Q5. If Q2 = Q5, then a′ = b, b′ = a. Consequently Qa = Q4. Further, Q3 = Q4 implies [a′ + b′]n = 1, whence, as in the case Q1 = Q2, we obtain Qa = Q3. Q3 = Q5 and Q4 = Q5 imply Qa = Q2 and Qa = Q1, respectively. Thus in any case Qs = Qt implies Qa = Qi for some i = 1, 2, . . . , 5. Hence, if Qa 6= Qi for all i = 1, . . . , 5, then also Qs 6= Qt for all 1 6 s < t 6 5. This proves (f ). The example x · y = [3x + 9y]11 shows that the case (f ) is possible. Using Theorems 4.26, 4.27 and 5.3 we can calculate the k∗-translatability of the parastrophes of several types of idempotent k-translatable quasigroups as a function of a. Results of calculations are presented below. quadratical hexagonal GS ARO [1 − 2a]n [1 − a]n [a + 1]n [−1 − 2a]n a [2a]n [1 − a]n [2 − 2a]n [2a − 1]n a a [1 − a]n [1 − a]n a a [−a]n [1 − a]n [a − 1]n [2 − a]n a [2 + 2a]n [1 − a]n [2a]n [1 − 2a]n Stein [a − 1]n a [2 − a]n [1 − a]n [3 − a]n [a − 2]n Qa Q1 Q2 Q3 Q4 Q5 right modular [−1 − a]n a [a + 2]n [1 − a]n [a + 1]n [−a]n Table 1. Translatability as a function of a. If Qa is k-translatable, then its parastrophes are k∗-translatable for the follow- ing values of k∗. [1−k−a]n [1 − k]n [1 − k]n Qa quadratical hexagonal Q1 Q2 Q3 Q4 Q5 [1 − k]n [k + a]n [k + 1]n [−k]n [1 − k]n k k GS ARO Stein [k − 1]n [−1−k−a]n [k + 1]n [1 − k]n [1 − k]n [−k]n [k−2a]n [2 − k]n [k − 2]n [3 − k]n [k − 1]n [k+a+2]n [−1 − k]n [k + 2]n [1 − k]n right modular [−1 − k]n [1 − k]n [k + 2]n [−k]n [k + 1]n Table 2. Tranlatability of parastrophes as a function of k. Using Theorem 4.26 and Theorem 5.1, we can readily see that all parastro- phes of a hexagonal quasigroup Qa are hexagonal. Moreover, a quasigroup Qa is The structure of idempotent translatable quasigroups 19 hexagonal if and only if one of its parastrophes is hexagonal. In other types of idempotent k-translatable quasigroups the situation is more complicated. For ex- ample, parastrophes of a quadratical k-translatable quasigroup Qa are quadratical only for some values of a and n. Qa quadratical GS Q1 a = 2, n = 5 never Q2 a = 4, n = 5 never Q3 a = 4, n = 5 never Q4 a = 2, n = 5 never Q5 always always ARO Stein r. modular C3 never a = 2, n = 7 never always a = 5, n = 7 never a = 5, n = 7 never never never always never never never never a = 2, n = 7 always a = 2, n = 7 a = 4, n = 7 a = 4, n = 7 Table 3. Parastrophe types References [1] F.E. Bennett,On a class of n2 × 4 orthogonal arrays and associated quasigroups, Congressus Numerantium 39 (1983), 117 − 122. [2] F.E. Bennett, The spectra of a variety of quasigroups and related combinatorial designs, Discrete Math. 77 (1989), 29 − 50. [3] F.E. Bennet, E. Mendelsohn and N.S. Mendelsohn, Resolvable perfect cyclic designs, J. Combin. Theory (A), 29 (1980), 142 − 150. [4] A.H. Clifford and G.B. Preston, The algebraic theory of semigroups, vol. 1, Math. Surveys 7, AMS, 1961. [5] Ch.J. Colbourn and J. Dinitz, (editors), The CRC handbook of combinatorial designs, 2nd ed., Discrete Math. Appl., Chapman and Hall/CRC (2007). [6] W.A. Dudek, Quadratical quasigroups, Quasigroups and Related Systems 23 (2015), 221 − 230. [7] W.A. Dudek, Parastrophes of quasigroups, Quasigroups and Related Systems 4 (1997), 9 − 13. [8] W.A. Dudek and R.A.R. Monzo, On the fine structure of quadratical quasi- groups, Quasigroups and Related Systems 24 (2016), 205 − 218. [9] W.A. Dudek and R.A.R. Monzo, Translatable quadratical quasigroups, (sub- mitted), arXiv:1708.08830v1 [math.RA] 27 Aug 2017. [10] W.A. Dudek and R.A.R. Monzo, Translatability and translatable semigroups, (submitted). [11] D. Keedwell and J. Dénes, Latin squares and their applications, Noth-Holland, 2015. [12] Z. Kolar-Begović and V. Volenec, Affine regular dodecahedron in GS-quasi- groups, Quasigroups and Related Systems 13 (2005), 229 − 236. [13] M.V. Lawson, Generalized heaps, inverse semigroups and Morita equivalence, Al- gebra Universalis 66 (2011), 317 − 330. 20 W. A. Dudek and R. A. R. Monzo [14] C.C. Lindner and D. Steedly, On the number of conjugates of a quasigroups, Algebra Universalis 5 (1975), 191 − 196. [15] R.A.R. Monzo, The ternary operations of groupoids, arXiv:1510.07955v1 [math.GM] [16] V. Shcherbacov, Elements of quasigroup theory and applications, Champan and Hall Book, 2017. [17] V. Volenec, GS quasigroups, Čas. pest. mat. 115 (1990), 307 − 318. [18] V. Volenec, Hexagonal quasigroups, Arch. Math. (Brno), 27 (1991), 113 − 122. [19] V. Volenec, Quadratical groupoids, Note di Matematica 13 (1993), 107 − 115. [20] V. Volenec, Heaps and right solvable Ward groupoids, J. Algebra 156 (1993), 1 − 4. [21] V. Volenec, Z. Kolar-Begović and R. Kolar-Šuper, ARO-quasigroups, Quasi- groups and Related Systems 18 (2010), 213 − 228. W.A. Dudek Faculty of Pure and Applied Mathematics, Wroclaw University of Science and Technology, 50-370 Wroclaw, Poland Email: [email protected] R.A.R. Monzo Flat 10, Albert Mansions, Crouch Hill, London N8 9RE, United Kingdom E-mail: [email protected]
1810.12386	2	1810	2019-06-08T18:37:32	Non-singular derivations of solvable Lie algebras in prime characteristic	[ "math.RA" ]	We study solvable Lie algebras in prime characteristic $p$ that admit non-singular derivations. We show that Jacobson's Theorem remains true if the quotients of the derived series have dimension less than~$p$. We also study the structure of Lie algebras with non-singular derivations in which the derived subalgebra is abelian and has codimension~one. The paper presents some new examples of solvable, but not nilpotent, Lie algebras of derived length~3 with non-singular derivations.	math.RA	math	NON-SINGULAR DERIVATIONS OF SOLVABLE LIE ALGEBRAS IN PRIME CHARACTERISTIC MARCOS GOULART LIMA AND CSABA SCHNEIDER Abstract. We study solvable Lie algebras in prime characteristic p that admit non- singular derivations. We show that Jacobson's Theorem remains true if the quotients of the derived series have dimension less than p. We also study the structure of Lie algebras with non-singular derivations in which the derived subalgebra is abelian and has codimension one. The paper presents some new examples of solvable, but not nilpotent, Lie algebras of derived length 3 with non-singular derivations. 1. Introduction By Jacobson's famous theorem [Jac55, Theorem 3], a finite-dimensional Lie algebra over a field of characteristic zero with a non-singular derivation is nilpotent. More re- cently [KK95, BM12] showed that a finite-dimensional Lie algebra in characteristic zero that admits a periodic derivation (that is, a derivation of finite multiplicative order) has nilpotency class at most two; moreover, if the order of the derivation is not a multiple of 6, then the Lie algebra is abelian. As shown by examples in [Sha99, Mat02], Jacobson's Theorem fails in positive characteristic. A classification of finite-dimensional simple Lie algebras with non-singular derivations was given in [BKK95]. In this paper, we explore the structure of finite-dimensional solvable, but not nilpotent, Lie algebras that admit a non-singular derivation. We present some auxiliary results in Section 2 before exhibiting, in Section 3, new exam- ples of non-nilpotent Lie algebras with non-singular derivations. Among our examples, the reader can find Lie algebras of derived length three with nilpotent quotients of arbitrary high nilpotency class, and we believe that such examples appear in the research literature for the first time. Then in Section 4 we present an approach to Jacobson's Theorem for solvable Lie algebras using compatible pairs of derivations. Compatible pairs of automor- phisms were used by Eick [Eic04] to describe and compute the automorphism group of a solvable Lie algebra. This method leads to the following theorem which also covers the case 2010 Mathematics Subject Classification. 17B05,17B30,17B40,15A21. Key words and phrases. Lie algebras, non-singular derivations, periodic derivations, cyclic spaces. Most of the work presented in this paper was carried out as the first author's PhD project; he thanks the Universidade Federal de Ouro Preto for financially supporting his studies. The second author was a recepient of the CNPq grants 308773/2016-0 (Bolsa de Produtividade em Pesquisa) and 421624/2018-3 (Universal ). We thank the referee for his or her useful suggestions, particularly for those that simplified the arguments in Section 3 and in the proof of Lemma 6.1. 1 2 MARCOS GOULART LIMA AND CSABA SCHNEIDER of positive characteristic. The terms of the derived series of a Lie algebra L are denoted by Lpiq; see Section 2 for the precise definition. Theorem 1.1. Let L be a finite-dimensional solvable Lie algebra over a field F of charac- teristic p ě 0 and suppose that L admits a non-singular derivation of finite order. If either p " 0 or dim Lpiq{Lpi`1q ă p for all i then L is nilpotent. The aforementioned examples in [Sha99, Mat02] (see also Example 2.6) show that, in prime characteristic, the condition on dim Lpiq{Lpi`1q cannot be weakened in Theorem 1.1. In the second part of the paper, we study the class of non-nilpotent Lie algebras that have an abelian ideal of codimension one and also admit a non-singular derivation. Note that the examples of [Sha99, Mat02] belong to this class. Using the concept of px, pq-cyclic modules introduced in Section 5, we prove our second main result in Section 6 that shows that the overall structure manifested by the known examples holds in the wider class of such algebras. Theorem 1.2. Let L be a finite-dimensional Lie algebra of derived length 2 over an al- gebraically closed field F of prime characteristic p such that dimpL{L1q " 1. Let x P LzL1 and assume that x induces a non-singular endomorphism of finite order on L1. Then L has a non-singular derivation of finite order if, and only if, L1 can be written as a direct sum of px, pq-cyclic modules. Theorem 1.2 can be interpreted as saying that finite-dimensional non-nilpotent Lie al- gebras with non-singular derivations are not as uncommon as the authors believed they were before starting their research on this topic. The reason why one does not normally see them in small dimensions is explained by Theorem 1.1 that implies that the dimension of such a solvable algebra is at least char F ` 1. Based on Theorem 1.2, we close Section 6 with an application to Lie algebras that arise from representations of the Heisenberg Lie algebra and admit non-singular derivations. The main results of this paper are obtained as applications of several theorems in linear algebra. In particular, the matrix theoretical Lemma 2.7 lies at the heart of the proof of Theorem 1.1, while Theorem 1.2 relies on the concept of px, pq-cyclic spaces which can be viewed as a restriction of the widely used concept of x-cyclic spaces. 2. Basic concepts The symbol ''' will be used to denote the direct sum of algebras, while the direct sum of vector spaces will be denoted by '`'. If V is a vector space, then EndpV q denotes the associative algebra of endomorphisms of V with the product given by composition, while glpV q denotes the Lie algebra of these endomorphisms with the product given by the Lie bracket. The Lie algebra of all derivations of an algebra K is denoted by DerpKq and is a Lie subalgebra of glpKq. NON-SINGULAR DERIVATIONS IN PRIME CHARACTERISTIC 3 We denote by Lpiq the i-th term of the derived series of a Lie algebra L; that is, Lp0q " L and, for i ě 1, Lpiq " rLpi´1q, Lpi´1qs. The terms Lp1q and Lp2q are also denoted by L1 and L2, respectively. The smallest integer n such that Lpnq " 0 is called the derived length of L. Suppose that K and I are Lie algebras. An action or a representation of K on I is a Lie algebra homomorphism ψ : K Ñ DerpIq. Additionally, if I is an abelian Lie algebra, then I is called a K-module. For k P K and v P I, the element ψpkqpvq will usually be denoted by rk, vs. If I is an ideal of a Lie algebra K, then K acts on I and the endomorphism of I induced by an element k P K is denoted by adI kpvq " rk, vs. The corresponding representation of K into glpIq is denoted by adI. We write ad for the adjoint representation adK : K Ñ DerpKq. To avoid an excess of brackets, we use the following notation for k, v P K: k. Hence, for k P K and v P I, adI (1) rk, . . . , rk, rk, vsss " rk, . . . , k, k , vs " rkn, vs. loooomoooon n times Thus, for v P I and for k P K, padI kqnpvq " rkn, vs for all n ě 1. The following lemma is well known; see [SF88, Proposition 1.3]. Lemma 2.1. Let V be a vector space over a field F. If x, y P EndpV q, then rxn, ys " n p´1qin ÿi"0 ixn´iyxi for all n ě 1. Let K and I be Lie algebras such that K acts on I via the homomorphism ψ : K Ñ DerpIq. We define the semidirect sum K iψ I as the vector space K ` I endowed with the product operation given by rpk1, v1q, pk2, v2qs " prk1, k2s, rk1, v2s ´ rk2, v1s ` rv1, v2sq. When the K-action on I is clear from the context, then we usually suppress the homomor- phism 'ψ' from the notation and write simply K i I. If L is a Lie algebra, such that L has an ideal I, and a subalgebra K in such a way that L " K ` I, then L -- K iψ I where ψ is the restriction of adI to K. In a semidirect sum K i I, an element pk, vq P K ` I will often be written as k ` v. Suppose that K and I are as in the previous paragraph. An element pα, βq P DerpKq ' DerpIq is said to be a compatible pair if the linear transformation pα, βq : K i I Ñ K i I that maps pk, vq ÞÑ pαpkq, βpvqq is a derivation of K i I. This is equivalent to the condition that (2) βprk, vsq " rαpkq, vs ` rk, βpvqs for all k P K, v P I. The set of compatible pairs of DerpKq ' DerpIq is denoted by ComppK, Iq. It is easy to check that ComppK, Iq is a Lie subalgebra of DerpKq ' DerpIq. 4 MARCOS GOULART LIMA AND CSABA SCHNEIDER If L is a Lie algebra, x, y P L, δ P DerpLq, then Leibniz's Formula [dG00, equation (1.11)] gives that (3) δnprx, ysq " n ÿk"0n k"δkpxq, δn´kpyq‰ for all n ě 1. This implies the following lemma which will be used often in this paper. Lemma 2.2. Let F be a field of prime characteristic p. (1) If L is a Lie algebra over F and δ P DerpLq, then δp P DerpLq. (2) Let I and K be Lie algebras such that K acts on I and let pα, βq P ComppK, Iq. Then pαp, βpq P ComppK, Iq. Proof. Equation (3) implies statement (1). As explained above, pα, βq can be viewed as a derivation of K i I by setting pα, βqpk, vq " pαpkq, βpvqq for k P K and v P I. Then pα, βqp " pαp, βpq. By statement (1), pα, βqp P DerpK i Iq, and this implies (2). (cid:3) An endomorphism α of a finite-dimensional vector space V is said to be diagonalizable if V admits a basis consisting entirely of eigenvectors of α. In such a basis, the matrix of α is diagonal. For a non-singular linear transformation α of finite order, let α denote the order of α. The following lemma is well known. Lemma 2.3. Suppose that L is a finite-dimensional Lie algebra over a field F of charac- teristic p ě 0 and let δ be a non-singular derivation of L with finite order. Then there exists an algebraic extension K of F such that one of the following is valid: (1) p " 0 and δ is diagonalizable over K; (2) p is a prime, δ " npt with p ∤ n, and δpt diagonalizable over K. is a non-singular derivation that is Let L be a Lie algebra over a field F and let A be an abelian group. Suppose that for all a P A, La is a linear subspace of L such that L " ` La aPA and rLa, Lbs ď La`b holds for all a, b P A. Then L is said to be an A-graded Lie algebra. The elements in the subspaces La are said to be homogeneous. There is a strong connection between gradings and diagonalizable derivations on a Lie algebra L. Indeed, if δ P DerpLq is a diagonalizable derivation of L, and, for α P F, Lα is the α-eigenspace of δ (taking Lα " 0 when α is not an eigenvalue of δ), then L " ` Lα αPF is a grading on L with respect to the additive group of F. Conversely, if L is F-graded, then we define a derivation δ on L by setting δpxq " αx for all x that lie in the homogeneous component Lα. If L0 " 0, then the derivation δ is non-singular. NON-SINGULAR DERIVATIONS IN PRIME CHARACTERISTIC 5 The following result is a bit more general than [LGM02, Corollary 5.2.7], but can be proved using the same argument; see also [Sha99, Lemma 2.1]. Lemma 2.4. Let L be a finite-dimensional Lie algebra which is graded by some abelian group. Suppose that L satisfies the graded n-Engel condition for some n ě 1; that is rxn, ys " 0 for all homogeneous elements x, y P L. Then L is nilpotent. For the following classical theorem of Jacobson; see [Jac55, Theorem 3]. Theorem 2.5. Let L be a finite-dimensional Lie algebra over a field of characteristic 0 and suppose that there exists a subalgebra D of the algebra of derivations of L such that (1) D is nilpotent; (2) if x P L such that δpxq " 0 for all δ P D then x " 0. Then L is nilpotent. In particular, if L admits a non-singular derivation, then L is nilpo- tent. In Theorem 2.5, the hypothesis of zero characteristic is essential to prove that every element in a homogeneous component is nilpotent. Theorem 2.5 fails to hold in prime characteristic as shown by the following example. Example 2.6 ([Mat02, Theorem 2.1]). Let V be a p-dimensional vector space over a field F of characteristic p ą 0 with basis v0, . . . , vp´1. Define the linear map x P glpV q by xpviq " vi`1 for 0 ď i ď p ´ 2 and xpvp´1q " v0. Let K be the abelian Lie algebra generated by x and consider V as a K-module. Set L " K i V . Then L is a solvable non-nilpotent Lie algebra of derived length 2. Suppose that α, β P Fzt0u such that α ‰ γβ for all γ P Fp (such elements exist if F ě p2). Define the linear map δ : L Ñ L as δ : " x ÞÑ αx; vi ÞÑ pβ ` pi ` 1qαqvi for 0 ď i ď p ´ 1. Then δ is a non-singular derivation of L. It is known that the converse of Jacobson's Theorem is not valid. Dixmier and Lis- ter [DL57] presented finite-dimensional nilpotent Lie algebras admitting only nilpotent derivations. We end this section with a result concerning matrices. For a field F and for a natural number n, let Mpn, Fq denote the set of n n matrices over F. Lemma 2.7. Let F be a field of characteristic p ě 0, let n be a natural number and let A, B, C be n n matrices over F. Assume also that either p " 0 or n ă p. If rA, Bs " C ` λB for some λ P F and rB, Cs " 0, then rA, Brs " rBr´1C ` λrBr for all r ě 1. In particular, if λ ‰ 0 and C is nilpotent, then B is nilpotent. Proof. The lemma can be proved following the proof of [Ber09, Fact 3.17.13] and making the necessary adaptations in the case of prime characteristic. (cid:3) 6 MARCOS GOULART LIMA AND CSABA SCHNEIDER 3. Examples of non-nilpotent Lie algebras with non-singular derivations In this section we present some examples of non-nilpotent and solvable Lie algebras that admit non-singular derivations, including examples of derived length three and examples with nilpotent quotients of arbitrary high nilpotency class. Our first example is a Lie algebra of the form L " KiI where K is a nilpotent Lie algebra of maximal class and I is a faithful K-module. Let F be a field of prime characteristic p and let I be an F-vector space of dimension 2p. Suppose that tv1, . . . , v2pu is a basis of I and define the the elements x, y P EndpIq with the following rules x : v1 ÞÑ v2, v2 ÞÑ v3, . . . vp´1 ÞÑ vp, vp ÞÑ v1, vp`1 ÞÑ vp`2, vp`2 ÞÑ vp`3, . . . v2p´1 ÞÑ v2p, v2p ÞÑ vp`1; y : vp`1 ÞÑ v1 and vi ÞÑ 0 if i ‰ p ` 1. Proposition 3.1. Let K be the Lie subalgebra of glpIq generated by x and y as defined above and set L " K i I. Then the following hold. (1) K is a nilpotent Lie algebra of dimension p ` 1 and nilpotency class p. (2) L " K i I is not nilpotent and has derived length 3. (3) If F ě p2, then L admits a non-singular derivation of finite order. Proof. (1) Let rw1, w2, . . . , wrs be a right-normed product in K such that wj P tx, yu for j P t1, . . . , ru. As is well known, such products generate K. We claim that rw1, w2, . . . , wrs " 0 unless rw1, w2, . . . , wrs is either of the form rx, . . . , x, ys or of the form rx, . . . , x, y, xs. Let I1 be the vector subspace generated by tv1, . . . , vpu. By the definition of x, y P EndpIq, (4) xpI1q " I1, ypIq ď I1 and ypI1q " 0. Denoting the symmetric group of degree r by Sr, there exist coefficients cσ P F for all σ P Sr such that (5) rw1, w2, . . . , wrs " s ÿσPSr cσw1σw2σ wrσ. If y appears more than once in the sequence w1, w2, . . . , wr, then w1σw2σ . . . wrσ " 0 holds for all σ, by (4). Suppose now that y appears exactly once in the sequence w1, w2, . . . , wr and suppose that wj " y. If j ď r ´ 2, then rw1, w2, . . . , wrs " rx, x, . . . , y, . . . , x, xs " 0. Thus the only possibilities for a non-zero right-normed product are rx, . . . , x, ys and rx, . . . , x, y, xs as claimed. By Lemma 2.1, rxp´1, ysvp`1 " " p´1 p´1qip ´ 1 p´1qip ´ 1 i xp´1´iyxivp`1 i xp´1´iyvp`1`i " xp´1yvp`1 " xp´1v1 " vp. ÿi"0 ÿi"0 p´1 NON-SINGULAR DERIVATIONS IN PRIME CHARACTERISTIC 7 Therefore rxp´1, ys ‰ 0. On the other hand, xp acts as identity on I. Further, by Lemma 2.1, rxp, ys " xpy ´ yxp " 0. Thus, noting that rx, . . . , x, y, xs " ´rx, . . . , x, x, ys and interpreting rx0, ys as y, the set txu Y trxn, ys 0 ď n ď p ´ 1u is a linear generating set for K. This implies that K has nilpotency class p and dimension at most p ` 1. On the other hand, a nilpotent Lie algebra of nilpotency class p has dimension at least p ` 1. Thus dim K " p ` 1. (2) The derived series of L is L ą (cid:10)rx, ys, rx, x, ys, . . . , rxp´1, ys, I(cid:11) ą hv1, . . . , vpi ą 0. Hence L is solvable of derived length 3. On the other hand, as x induces a non-singular transformation on I, we have that rxn, Is " I for all n ě 1, and so L is not nilpotent. (3) We claim that L admits a grading with respect to the additive group of F. First, I can be graded by assigning vkp`i degree b ´ ka ` i ´ 1. Now the linear transformations x and y preserve this grading in the sense that xpIαq Ď Iα`1 and ypIαq Ď Iα`a for all α P F. Consequently, the Lie algebra K, generated by x and y, is F-graded; the degrees of x and y are 1 and a, respectively, while the degree of zj is a ` j for j P t1, . . . , p ´ 1u. Thus the semidirect sum L is graded over the additive group of the field F, and so the linear map δ, which multiplies each homogeneous element by its degree, is a derivation (see the discussion after Lemma 2.3). By the choice of a, b P F, the homogeneous component L0 is trivial, and so δ is non-singular. (cid:3) Our next example involves the Heisenberg Lie algebra, which is the 3-dimensional Lie algebra H " hx, y, zi over a field F with the multiplication rx, ys " z and rx, zs " ry, zs " 0. The Lie algebra H is nilpotent with nilpotency class 2. Suppose that char F is prime and let I be a vector space of dimension 2p over F with basis tv1, . . . , v2pu. Define the endomorphisms x, y, and z of I as follows: x : v1 ÞÑ v2, v2 ÞÑ v3, . . . , vp´1 ÞÑ vp, vp ÞÑ v1, vp`1 ÞÑ vp`2, vp`2 ÞÑ vp`3, . . . , v2p´1 ÞÑ v2p, v2p ÞÑ vp`1; y : vi ÞÑ 0 for all i P t1, . . . , pu, vp`1 ÞÑ 0, vp`2 ÞÑ v2, vp`3 ÞÑ 2v3, . . . , v2p ÞÑ pp ´ 1qvp; z : vi ÞÑ 0 for all i P t1, . . . , pu, vp`1 ÞÑ v2, vp`2 ÞÑ v3, . . . , v2p´1 ÞÑ vp, v2p ÞÑ v1. Easy computation shows that rx, yspviq " zvi for all i and that rx, zs " ry, zs " 0. Hence the Lie algebra H " hx, y, zi is a Heisenberg Lie algebra over F. Set L " H i I. Proposition 3.2. The Lie algebra L is solvable of derived length 3, but not nilpotent. If F ě p2, then L admits a non-singular derivation. Proof. The derived series of L is L ą hz, Ii ą hv1, . . . , vpi ą 0, 8 MARCOS GOULART LIMA AND CSABA SCHNEIDER and so the derived length of L is 3. Further, rxn, Is " I for all n ě 1, and so L is not nilpotent. Suppose that F ě p2. There are a, b P pFzFpq such that b ´ a R Fp. Following the steps seen in the proof of Proposition 3.1, setting the degree b ´ ka ` i ´ 2 for each vkp`i, I is a graded Lie algebra. The linear maps x, y and z are graded on I, with degrees 1, a and 1 ` a, respectively. Thus, L is an F-graded Lie algebra with L0 " 0 and hence there is a corresponding non-singular derivation δ as explained after Lemma 2.3. (cid:3) 4. Jacobson's Theorem through compatible pairs The aim of this section is to prove Theorem 1.1. For a Lie algebra K and for a K-module I, we let, as in Section 2, ComppK, Iq denote the set of compatible pairs in DerpKq'DerpIq. Using the representation ψ : K Ñ DerpIq, we can rewrite equation (2) as (6) rβ, ψpkqs " ψpαpkqq for all k P K. Thus pα, βq P ComppK, Iq if and only if (6) is valid. Letting ad : DerpIq Ñ DerpIq denote the adjoint representation of I, equation (6) can be rewritten as (7) adβψpkq " ψpαpkqq for all k P K. Therefore, pα, βq P ComppK, Iq if, and only if, the following diagram commutes: K α K ψ oe ψ DerpIq ad β / DerpIq. Theorem 4.1. Let K be a finite-dimensional solvable Lie algebra over an algebraically closed field F of characteristic p ě 0 and let I be a K-module under the representation ψ : K Ñ DerpIq. Let pα, βq P ComppK, Iq such that α is non-singular with finite order. If either p " 0 or dim I ă p, then ψpkq is nilpotent for all k P K. Proof. If char F " 0, then α is diagonalizable by Lemma 2.3. If char F is prime, then the same lemma implies, for a suitable t, that αpt is a non-singular and diagonalizable , βptq P ComppK, Iq. Hence, by Lemma 2.2, we may assume derivation of K such that pαpt without loss of generality that pα, βq P ComppK, Iq with α non-singular, diagonalizable, and of finite order. Let x1, . . . , xs be a basis of K such that αpxiq " λixi. Let B be a basis for I and, for all a P glpIq, denote by JaK the matrix of a in B. Then, by equation (6), rJβK, JψpxiqKs " JψpαpxiqqK " λiJψpxiqK. Applying Lemma 2.7 with A " JβK, B " JψpxiqK, C " 0 and λ " λi (which is non-zero) we conclude that JψpxiqK is nilpotent for 1 ď i ď s. Now, Lie's Theorem is valid for the solvable Lie algebra K also in the case when char F ą 0 and dim I ă p (see [Hum78, Section 4.1 and Exercise 2 on page 20]). Thus there is a basis of I such that ψpKq is upper triangular. Working in this basis, since JψpxiqK is nilpotent and upper triangular, it must / / / NON-SINGULAR DERIVATIONS IN PRIME CHARACTERISTIC 9 be strictly upper triangular (that is, with zeros in the diagonal) for all i. Therefore JψpkqK is strictly upper triangular, and hence nilpotent, for all k P K. (cid:3) From the previous theorem, we obtain Theorem 1.1 which can be viewed as a version of Jacobson's Theorem on non-singular derivations for solvable Lie algebras which is valid also in prime characteristic. Proof of Theorem 1.1. Since the solvability of L and dim Lpiq{Lpi`1q are invariant under extensions of the ground field, we may assume that F is algebraically closed. We prove the statement by induction on the derived length k of L. When k " 1, then L is clearly nilpotent. Suppose that the result holds for Lie algebras of derived length k and assume that L has derived length k`1. Then I " Lpkq is an abelian ideal of L. Setting K " L{I, we have that K acts on I by the adjoint representation adI : K Ñ DerpIq. Further, since the terms of the derived series are invariant under derivations, a non-singular derivation δ P DerpLq gives rise to a compatible pair pα, βq P ComppK, Iq where α is the derivation induced on K by δ and β is the restriction of δ to I. Since δ is non-singular, so are α P DerpKq and β P DerpIq. Note that K is solvable of derived length k and K piq{K pi`1q -- Lpiq{Lpi`1q for all i ď k ´ 1. Hence the induction hypothesis is valid for K and we obtain that K is nilpotent. In addition, Theorem 4.1 implies that adI k is nilpotent for all k P K. Therefore, L{I is nilpotent and adI x : I Ñ I is nilpotent for all x P L. It follows from [Hum78, Exercise 10, page 14] that L is nilpotent. (cid:3) 5. Primary decomposition and p-cyclic spaces In this section we review a primary decomposition concept and define px, pq-cyclic spaces that appear in Theorem 1.2. Let V be a finite-dimensional vector space over a field F and let x P EndpV q. Let q P Frts be a univariate polynomial and define (8) V0pqpxqq " tv P V there is m ą 0 such that qpxqmv " 0u. The set V0pqpxqq is a vector subspace of V which is invariant under x. Let qx be the minimal polynomial of x and suppose that qx " qk1 is the factorization of qx into monic irreducible factors, such that qi ‰ qj for 1 ď i ă j ď r. Then V decomposes as a direct sum of subspaces 1 qkr r V " V0pq1pxqq ` ` V0pqrpxqq, with each space V0pqipxqq being invariant under x. Furthermore, the minimal polynomial of the restriction of x to V0pqipxqq is qki i . A proof of this result can be found in [dG00, Lemma A.2.2]. We can generalize this decomposition to subalgebras of glpV q generated by more than one element. The following definition was stated in [dG00, Definition 3.1.1]. Definition 5.1. Let V be a finite-dimensional vector space over a field F and let K ď glpV q be a subalgebra. A decomposition V " V1 ` ` Vs of V into K-submodules Vi is said to be primary if the minimal polynomial of the restriction of x to Vi is a power of an 10 MARCOS GOULART LIMA AND CSABA SCHNEIDER irreducible polynomial for all x P K and 1 ď i ď s. The subspaces Vi are called primary components. If for any two components Vi and Vj pi ‰ jq, there is an x P K such that the minimal polynomials of the restrictions of x to Vi and Vj are powers of different irreducible polynomials, then the decomposition is called collected. In general, a K-module V may not have a primary (or collected primary) decomposition into K-submodules, but such a decomposition is guaranteed to exist if K, as a subalgebra of glpV q, is nilpotent; see [dG00, Theorem 3.1.10]. Lemma 5.2 ([dG00, Proposition 3.1.7]). Suppose that V is a finite-dimensional vector space over a field F, let x, y P glpV q and let q P Frts be a polynomial. Suppose that rxn, ys " 0 for some n ě 1. Then V0pqpxqq is invariant under y. Lemma 5.3. Let K be a nilpotent Lie algebra over a field F of characteristic p ą 0 and let I be a finite-dimensional K-module. Let L " K i I, x P K and qptq " t ´ a with some a P F. Let δ P DerpLq such that δpIq ď I and δpxq " bx for some b P F. Then I0pqpxqq is δ-invariant. Proof. Suppose that w P I0pqpxqq; that is, there is m ą 0 such that px ´ a idqpm w " 0. As char F " p, we have (9) 0 " px ´ a idqpm w " pxpm ´ apm idq w " xpm w ´ apm w. As δ P DerpLq, using the right-normed convention introduced in equation (2), δpxpm wq " δprxpm , wsq " rδpxq, . . . , x, ws ` rx, δpxq, . . . , x, ws ` ` rx, . . . , x, δpwqs " rax, . . . , x, ws ` rx, ax, . . . , x, ws ` ` rx, . . . , x, δpwqs " pm a rxpm " xpm δpwq. , ws ` rxpm , δpwqs (10) Combining (10) and (9) we obtain 0 " δp0q " δpxpm w ´ apm wq " δpxpm wq ´ apm δpwq " xpm δpwq ´ apm δpwq. Hence, and δpwq P I0pqpxqq. px ´ a idqpm δpwq " xpm δpwq ´ apm δpwq " 0 (cid:3) Let V be a finite-dimensional vector space over a field F and x P EndpV q. The vector space V is said to be x-cyclic if there is v P V such that txkpvq k ě 0u is a generating set for V . As is well known, V is x-cyclic if and only if the degree of the minimal polynomial of the endomorphism x coincides with dim V , which amounts to saying that the minimal polynomial of x is equal to its characteristic polynomial. It is well known that if V is a vector space and x P EndpV q, then V can be decomposed as a direct sum of x-cyclic subspaces; see [Rom08, Theorem 7.6] and [KR16, Theorem 1.5.8 and Corollary 1.5.14] for more details. NON-SINGULAR DERIVATIONS IN PRIME CHARACTERISTIC 11 In the non-nilpotent Lie algebra with non-singular derivation given by Example 2.6, the vector space V is x-cyclic, but it also satisfies some stronger conditions. This motivates the following concept. Definition 5.4. Let V be a vector space over a field F of prime characteristic p and let x P glpV q. The vector space V is px, pq-cyclic if the following hold: (1) V is x-cyclic; (2) qxptq P Frtps and qxp0q ‰ 0. Thus if V is px, pq-cyclic, as in Definition 5.4, then the minimal polynomial qxptq is of the form c0 ` c1tp ` ` cn´1tpn´1qp ` tnp with c0 ‰ 0. Hence, over a perfect field F, qxptq can be written as the p-th power of a polynomial q0ptq. 6. Lie algebras with an abelian ideal of codimension one In this section we consider non-singular derivations of a Lie algebra L such that the derived subalgebra L1 is abelian and has codimension one. If x P LzL1, then L can be written as L " K i L1 where K " hxi and the action of K on L1 is given by the restriction of adL1 to K. The Lie algebra L presented in Example 2.6 satisfies this condition and we observed at the end of Section 5 that L1 is an px, pq-cyclic module. The objective of this section is to prove Theorem 1.2 stated in the Introduction which shows that this phenomenon can be observed in a more general case. We start with the easier direction of Theorem 1.2. Lemma 6.1. Let K be a one-dimensional Lie algebra over an algebraically closed field F of prime characteristic p with K " xxy and let I1, I2, . . . , Is be px, pq-cyclic K-modules. Then the Lie algebra L given by the semidirect sum has a non-singular derivation with sp ` 1 distinct eigenvalues. L " K i pI1 ` I2 ` ` Isq Proof. As F is algebraically closed, we can choose b, a1, . . . , as P F such that ajb´1 R Fp, for all 1 ď j ď s, and (11) taj ` ib 1 ď j ď s and 0 ď i ď p ´ 1u " ps. By assumption, Ij is px, pq-cyclic, for 1 ď j ď s, and so there is a basis Bj " tvj 0, vj 1, . . . , vj rj p´1u of Ij such that the matrix of x in Bj is the companion matrix of the minimal polynomial ´c0,j ´ cp,jtp ´ ´ cprj´1qp,jtprj ´1qp ` trj p with ci,j P F. Let 1 ď j ď s. Then i`1 rj´1 "x, vj rj p´1ı " i‰ " vj ÿi"0 "x, vj for 0 ď i ă rjp ´ 1 and cip,jvj ip. 12 MARCOS GOULART LIMA AND CSABA SCHNEIDER Define an F-grading on L by setting the degrees of x and the vj i , for all possible i and j, to be b and aj ` ib, respectively. The fact that this assignment defines a grading on L follows easily from the last two displayed equations. By the discussion after Lemma 2.3, there is a corresponding derivation δ on L. By the choice of a, b P F, the homogeneous component L0 is trivial, and so δ is non-singular. The eigenvalues of δ are b and the elements of the set in (11), and hence δ has sp ` 1 distinct eigenvalues, as claimed. (cid:3) The following lemma will be used as a first step in the proof of Theorem 1.2. Lemma 6.2. Let K " xxy be a one-dimensional Lie algebra over F of prime characteristic p. Let I be a finite-dimensional K-module such that x induces an invertible endomorphism of I and set L " K iI. Assume that δ is a non-singular derivation of L such that δpxq " x and δI is diagonalizable. If v P I is an eigenvector of δ, then the K-submodule xvyK is px, pq-cyclic. Proof. Define the sequence v0 " v and vi`1 " rx, vis for i ě 0. Then the set tv0, v1, . . . , u generates xvyK and xvyK is x-cyclic. As I has finite dimension, there is a k ą 0 such that tv0, v1, . . . , vk´1u is linearly independent and tv0, v1, . . . , vku is linearly dependent. Since x and v0 are δ-eigenvectors, easy induction shows that each vi is an eigenvector of δ associated to the eigenvalue a ` i, where a is the eigenvalue associated with v. Note that a, a ` 1, . . . , a ` pp ´ 1q are distinct eigenvalues, and hence the set tv0, v1, . . . , vp´1u is linearly independent. In particular, k ě p. If the eigenvectors vi and vj are associated with eigenvalues a ` i and a ` j, then vi and vj are associated with the same eigenvalue if and only if i " j pmod pq. Suppose that k " rp ` t with 0 ď t ď p ´ 1. Since a set of eigenvectors associated to pairwise distinct eigenvalues is linearly independent, the eigenvector vk must be a linear combination of the eigenvectors vi with i ď k ´ 1 that have the same eigenvalue as vk, which is a ` t. Hence, (12) vk " c0vt ` c1vp`t ` c2v2p`t ` ` cr´1vpr´1qp`t. If t ‰ 0, then we can replace every vi by rx, vi´1s in equation (12) and obtain (13) rx, vk´1s " c0rx, vt´1s ` c1rx, vp`t´1s ` ` cr´1rx, vpr´1qp`t´1s. Since x induces an injective endomorphism on I, vk´1 " c0vt´1 ` c1vp`t´1 ` c2v2p`t´1 ` ` cr´1vpr´1qp`t´1. This contradicts to the assumption that tv0, v1, . . . , vk´1u is linearly independent. Thus, t " 0 and k " rp. Equation (12) implies also that vk " c0v0 ` c1vp ` ` cr´1vpr´1qp and so, the characteristic polynomial of x is qxptq " trp ´ cpr´1qtpr´1qp ´ ´ c2t2p ´ c1tp ´ c0. If c0 " 0, then replacing each vi with rx, vi´1s as above implies that the set tv0, v1, . . . , vk´1u is linearly dependent. Therefore c0 ‰ 0. As xvyK is x-cyclic, the minimal polynomial of the restriction of x to xvyK is qx and xvyK is px, pq-cyclic. (cid:3) NON-SINGULAR DERIVATIONS IN PRIME CHARACTERISTIC 13 For an endomorphism x of a vector space I, we denoted the minimal polynomial of x by qx. When we want to emphasize the domain of x, we use the notation qx,I. If v P I, then qx,v denotes the minimal polynomial of x with respect to v. That is, qx,v is the smallest degree, non-zero, monic polynomial such that qx,vpxqpvq " 0. It is well known that qx,v qx,I for all v P I. The proof of the following theorem was inspired by the proof of Theorem 6.6 in [BR02]. Lemma 6.3. Let K " xxy be a Lie algebra of dimension 1 over an algebraically closed field F of characteristic p ą 0. Let I be a finite-dimensional K-module such that x induces an invertible endomorphism of finite order on I and set L " K i I. Assume that δ is a non-singular derivation of L such that δpxq " x and δI is diagonalizable. Assume, further, that qx,Iptq " pt ´ λqm with some λ P F and m ě 1 and that the δ-eigenvalues on I are a, a ` 1, . . . , a ` p ´ 1 with some a P F. Then I is the direct sum of px, pq-cyclic subspaces, each of which is generated by a δ-eigenvector. Proof. We prove this lemma by induction on dim I. By Lemma 6.2, dim I ě p, and so the base case of the induction is when dim I " p. In this case, if v P I is a δ-eigenvector, then hviK is px, pq-cyclic of dimension greater than or equal to p, and hence I " hviK. Thus the lemma is valid when dim I " p. Suppose now that dim I ě p ` 1 and that the lemma is valid for spaces of dimension less than dim I. By our conditions, I " Ea ` ` Ea`p´1 where Eb denotes the b-eigenspace of δ in I. Since Ťb Eb generates I as a vector space, there is some eigenvector v0 P I such that qx,v0ptq " qx,Iptq " pt ´ λqm. We may assume without loss of generality that v0 P Ea. Let I0 be the K-module generated by v0, and let J " I{I0. Since v0 is a δ-eigenvector, I0 is px, pq-cyclic by Lemma 6.2, and hence p m. In particular, qx,v0ptq " pt ´ λqm " pt ´ λqm0p, where m0 ě 1. Note that I0 is an ideal of L that is invariant under δ. Considering J as a K-module, we can consider the Lie algebra K i J -- L{I0 that satisfies the conditions of the lemma. Since dim J ă dim I, the induction hypothesis applies to J, and we may write J " J1 ` ` Jk where the Ji are px, pq-cyclic subspaces of J and each Ji is generated by a δ-eigenvector, wi ` I0, say. Since I0 has a basis consisting of δ-eigenvectors, the δ- eigenvalues in J " I{I0 are a, a ` 1, . . . , a ` p ´ 1. We claim that wi can be chosen such that wi P Ea. Since x induces an invertible endomorphism of finite order on I (and hence on I{I0), we have that hw ` I0iK " hxpw ` I0qiK holds for all w P I. Hence, by possibly swapping wi with xℓwi for some suitable ℓ ě 0, we may assume without loss of generality that wi ` I0 is a δ-eigenvector corresponding to eigenvalue a. As δpwiq ` I0 " awi ` I0, δpwiq ´ awi " u P I0. Since I is the sum of the δ-eigenspaces Eb, we may write wi " za ` za`1 ` ` za`p´1 with zb P Eb. Further, since I0 is spanned by δ-eigenvectors, we have that I0 " pI0 X Eaq ` ` pI0 X Ea`p´1q. Thus we have u " ua ` ua`1 ` ` ua`p´1 with ub P Eb X I0. Hence, δpwiq ´ awi " aza ` pa ` 1qza`1 ` ` pa ` p ´ 1qza`p´1 ´ apza ` za`1 ` ` za`p´1q " ua ` ua`1 ` ` ua`p´1. 14 MARCOS GOULART LIMA AND CSABA SCHNEIDER Since eigenvectors with different eigenvalues are linearly independent, we obtain ua`j " j za`j for all j ě 0. This implies that za`j " j´1 ua`j P I0 holds for all j ě 1. Therefore wi " za ` ua`1 ` 2´1ua`2 ` ` pp ´ 1q´1ua`p´1 P za ` I0. Therefore we may replace wi by za and so, we may assume without loss of generality that wi P Ea. Since Ji is px, pq-cyclic, qx,Jiptq " pt ´ λqmip with some mi ě 1. We claim, for all i " 1, . . . , k, that there is some vi P Ea X pwi ` I0q such that qx,viptq " qx,Jiptq " pt ´ λqmip. We prove this claim for i " 1. Since qx,J1ptq " pt ´ λqm1p, we have px ´ λqm1ppw1 ` I0q " 0, and so px ´ λqm1ppw1q P I0. Thus, there is some polynomial h P Frts with deg h ă m and px ´ λqm1ppw1q " hpxqpv0q. On the other hand, w1 P Ea, and hence px ´ λqm1ppw1q " rpx ´ λqpsm1pw1q " rpxp ´ λpqsm1pw1q P Ea, which gives hpxqpv0q P Ea, since xp fixes each eigenspace Eb. Write hptq " h0ptq ` h1ptq ` ` hp´1ptq such that hjptq " ajtj ` ap`jtp`j ` a2p`jt2p`j ` , for all 0 ď j ď p´1. Suppose that hj ‰ 0 for some j ą 0. Observe that hjpxqpv0q P Ea`j. As hpxqpv0q P Ea and eigenvectors associated to different eigenvalues are linearly independent, we have hjpxqpv0q " 0. Thus, qx,v0 hj. On the other hand, deg hj ă m " deg qx,v0, which implies that hj " 0 for all j ą 0. Hence, we can assume that h " h0 " h with some h P Frts. Now observe that p 0 " px ´ λqmpw1q " px ´ λqm´m1ppx ´ λqm1ppw1q " px ´ λqm´m1phpxqpv0q. Since qx,v0ptq " pt ´ λqm, we have that pt ´ λqm pt ´ λqm´m1phptq, and so pt ´ λqm1p hptq. Therefore there is some q P Frts such that qptqpt ´ λqm1p " hptq " hptqp. This also implies that q " qp with some q. Now set v1 " w1 ´ qpxqpv0q. Since qpxqpv0q P I0, we have v1 P w1 ` I0. Further, qpxqpv0q " qpxqppv0q P Ea, and hence v1 P Ea. This implies also that qx,J1 qx,v1. On the other hand, px ´ λqm1ppv1q " px ´ λqm1ppw1 ´ qpxqpv0qq " px ´ λqm1ppw1q ´ px ´ λqm1pqpxqpv0q " 0. Thus qx,v1ptq " pt ´ λqm1p " qx,J1ptq, as claimed. For i " 1, . . . , k, let Ii " hviiK. We claim that I " I0 ` ` Ik. First, and so Ji " hwi ` I0iK " hvi ` I0iK " pIi ` I0q{I0 I{I0 " pI1 ` I0q{I0 ` ` pIk ` I0q{I0. This implies that I " I0 ` I1 ` ` Ik. Further, the direct decomposition of I{I0 also implies that dim I0 `ři dimpIi ` I0q{I0 " dim I. On the other hand, since qx,vi " qx,Ji, we also obtain that dim Ii " dim Ji " dimpIi ` I0q{I0. Therefore dim I0 ` dim I1 ` ` dim Ik " dim I. Hence the direct decomposition I " I0 ` I1 ` ` Ik is valid. (cid:3) NON-SINGULAR DERIVATIONS IN PRIME CHARACTERISTIC 15 Now we can prove Theorem 1.2. The proof of Theorem 1.2. One direction of the theorem follows directly from Lemma 6.1, and so we are only required to show the other direction. Let δ P DerpLq be a non-singular derivation of finite order. Set I " L1. Note that I is δ-invariant and δpxq " αx ` a with some α P Fzt0u and a P I. Defining δ1 P EndpLq as δ1paq " δpaq for a P I and δ1pxq " αx we obtain that δ1 P DerpLq and also that I and K " hxi are δ1-invariant. Further, if the order of δ1I is npr where p ∤ n, then the derivation δ2 " pδ1qpr is non-singular and the order of the restriction of δ2 to I is coprime to p. Hence the restriction of δ2 to I is diagonalizable. In addition, multiplying δ2 by a scalar we may also assume that δ2pxq " x. By the argument in the previous paragraph, we may assume without loss of generality that L has a non-singular derivation δ of finite order such that δpxq " x, δpIq " I and that the restriction of δ to I is diagonalizable. Let qx,Iptq " pt ´ λ1qm1 pt ´ λkqmk be the minimal polynomial of x considered as an element of glpIq. As K is one-dimensional, the collected primary decomposition of I into K-modules is I " I0px ´ λ1q ` ` I0px ´ λkq, and Lemma 5.3 implies that the I0px ´ λiq are δ-invariant. Hence we may assume without loss of generality that k " 1 and qx,Iptq " pt ´ λqm. Further, I can be decomposed as I " Ea1 ` ` E as where, for ai P F, Eai is the sum of the eigenspaces Eai, . . . , Eai`p´1 of δ that correspond to the eigenvalues ai, ai ` 1, . . . , ai ` p ´ 1, respectively. Since x is a δ- eigenvector corresponding to eigenvalue 1, xpEai`jq ď Eai`j`1 holds for all j P t0, . . . , p´2u and xpEai`p´1q ď Eai. Thus each Eai is x-invariant. Therefore we may assume that I " Ea with some a P F. Now the theorem follows from Lemma 6.3. (cid:3) We present an application of Theorem 1.2 concerning the representations of the Heisen- berg Lie algebra with the property that the semidirect sum admits a non-singular deriva- tion. Theorem 6.4. Let F be the algebraic closure of Fp where p is an odd prime. Let H be the Heisenberg Lie algebra over F. Let ψ : H Ñ glpIq be a faithful representation, and suppose that L " H i I is non-nilpotent. If L has a non-singular derivation of finite order such that δpIq ď I, then dim I ě p ` 3. Proof. Let δ P DerpLq be a non-singular derivation of finite order such that δpIq ď I. Using the argument at the beginning of the proof of Theorem 1.2, we can suppose that δ is diagonalizable, δpIq " I and δpHq " H. Hence L can be viewed as a graded Lie algebra over the additive group of F as explained after Lemma 2.3. As L is non-nilpotent, by Lemma 2.4, there are homogeneous δ-eigenvectors k, v P L such that rkm, vs ‰ 0 for all m ą 0. Write k " kH ` kI and v " vH ` vI, such that kH, vH P H and kI , vI P I. Notice that kH, vH, kI , vI are δ-eigenvectors. In fact, if δpkq " bk, for some b P F, then δpkHq ` δpkIq " δpkH ` kIq " bpkH ` kIq. 16 MARCOS GOULART LIMA AND CSABA SCHNEIDER Hence, as H and I are invariant under δ and L " H i I, δpkHq " bkH and δpkIq " bkI. Analogously, if δpvq " cv for some c P F, then δpvHq " cvH and δpvIq " cvI. We claim that there is h P H and v P I such that rhm, vs ‰ 0 for all m ě 1. We have that, rkm, vH ` vIs ‰ 0 for all m ě 1, and so, rkm, vHs ‰ 0 or rkm, vIs ‰ 0 for all m ě 1. Suppose first that rkm, vHs ‰ 0 for all m ě 1. Then, since rkH, vHs P ZpHq and since I is abelian, rk2, vH s " rkH ` kI, rkH ` kI, vH ss " rkH ` kI, rkH, vHs ` rkI , vHss " rkH, rkH, vHss ` rkH, rkI, vHss ` rkI, rkH, vHss ` rkI, rkI, vHss " rkH, rkI, vHss ` rkI, rkH, vH ss. This means that rk2, vHs P I. Therefore, as I is abelian, rk3, vHs " rk, rk2, vHss " rkH ` kI , rk2, vH ss " rkH, rk2, vHss and easy induction shows that rpkHqm, rk2, vHss " rkm`2, vHs ‰ 0 holds for all m ě 1. Hence the choice of h " kH and v " rk2, vHs is as claimed. If rkm, vIs ‰ 0, for all m ě 1, then let v " vI and let h " kH. Hence, an argument similar to the one in the previous case shows that rhm, vs ‰ 0 for all m ě 1. Let h P H and v P I be δ-eigenvectors such that rhm, vs ‰ 0, for all m ě 1, as in the previous paragraph. Let q be the minimal polynomial of ψphq as element of EndpIq and suppose that q " qs1 is the factorization of q into irreducible factors. Then, by the argument presented at the beginning of Section 5, I can be written as the direct sum I " I0pq1phqq ` . . . ` I0pqrphqq. By Lemma 5.2, each I0pqiphqq is an H-module. Let I1 be the sum of I0pqiphqq such that qiptq ‰ t, and set I0 " I0phq. Thus, 1 . . . qsr r L " H i pI0 ` I1q . Also, I1 ‰ 0, since v R I0. By Lemma 5.3, I0 and I1 are H-modules and δ-invariant. It follows that the Lie algebras L0 " H i I0 and L1 " H i I1 have a non-singular derivation. Observe that, by the construction of L1, h acts non-singularly on I1. Hence, L1 is non- nilpotent. Let δ1 be the restriction of δ to L1. The derivation δ1 P DerpL1q is non-singular and has finite order. As h is an eigenvector of δ1 and I is δ-invariant, the Lie algebra xhy i I is δ-invariant, and so the restriction of δ to xhy i I is a non-singular derivation of finite order. Note that h induces a non-singular linear transformation on I1 and this transformation has finite order, since F is the algebraic closure of Fp, and so all the entries of the matrix representing h in a basis of I1 are elements of a suitable finite field. Thus, by Theorem 1.2, I1 can be written as a direct sum of ph, pq-cyclic modules, and so dim I1 " np for some n ě 1. In particular, dim I1 ě p. The action of H must be faithful either on I1 or on I0. For, if H were not faithful on I0 and on I1, then ZpHq would act trivially on both I1 and I0, and hence ZpHq would act trivially on I. This contradicts to the assumption that I is a faithful H-module. As δpHq " H, δpZpHqq " ZpHq. Hence, if z P ZpHqzt0u, then δpzq " dz, since dimpZpHqq " 1. If I1 is NON-SINGULAR DERIVATIONS IN PRIME CHARACTERISTIC 17 a faithful module, then there is u P I1 a δ-eigenvector associated to the eigenvalue e P F, such that rz, us ‰ 0. It follows that δprz, usq " pd ` eqrz, us. Then, since d ‰ 0, u and rz, us are linearity independent. If dim I1 " p, then by [Sze14, Corollary 2.4] the representation is irreducible and there exists f P F such that rz, ws " f w for all w P I. This, however, contradicts the fact that u and rz, us are linearity independent, and so dim I1 ě 2p. As p ě 3, dim I1 ě p ` 3. If I1 is not faithful, then I0 is, and, by [Sze14, Theorem 3.1], dim I0 ě 3. In both cases, dim I " dim I0 ` dim I1 ě p ` 3. (cid:3) The following example shows that the bound p ` 3 in Theorem 6.4 is sharp. Example 6.5. Let F be a field of characteristic p, and let I be a vector space of dimension p ` 3 over F with basis B " tv1, v2, v3, u0, u1, . . . , up´1u. We define a representation of the Heisenberg Lie algebra H " hx, y, zi on I as follows: x : v2 ÞÑ v1, u0 ÞÑ u1, u1 ÞÑ u2, . . . , up´1 ÞÑ u0; y : v3 ÞÑ v2; z : v3 ÞÑ v1. The Lie algebra L " H i I is not nilpotent, but solvable with derived length 3. Suppose that a, b, c, d P F and define δ : L Ñ L by δpxq " ax, δpyq " by, δpzq " pa ` bqz, δpuiq " pc ` iaq, δpv1q " pa ` b ` dqv1, δpv2q " pb ` dqv2 and δpv3q " dv3. If a, b, c, d are chosen so that the eigenvalues of δ are non-zero, then δ is a non-singular derivation of H i I. The details can be verified as in Proposition 3.1 and are left to the reader. References [Ber09] Dennis S. Bernstein. Matrix mathematics. Princeton University Press, Princeton, NJ, second edition, 2009. Theory, facts, and formulas. [BKK95] Georgia Benkart, Alexei I. Kostrikin, and Michael I. Kuznetsov. Finite-dimensional simple Lie algebras with a nonsingular derivation. J. Algebra, 171(3):894 -- 916, 1995. [BM12] Dietrich Burde and Wolfgang Alexander Moens. Periodic derivations and prederivations of Lie algebras. J. Algebra, 357:208 -- 221, 2012. [BR02] T. S. Blyth and E. F. Robertson. Further linear algebra. Springer Undergraduate Mathematics Series. Springer-Verlag London, Ltd., London, 2002. [dG00] Willem A. de Graaf. Lie algebras: theory and algorithms, volume 56 of North-Holland Mathe- [DL57] matical Library. North-Holland Publishing Co., Amsterdam, 2000. J. Dixmier and W. G. Lister. Derivations of nilpotent Lie algebras. Proc. Amer. Math. Soc., 8:155 -- 158, 1957. [Eic04] Bettina Eick. Computing the automorphism group of a solvable Lie algebra. Linear Algebra Appl., 382:195 -- 209, 2004. [Hum78] James E. Humphreys. Introduction to Lie algebras and representation theory, volume 9 of Grad- uate Texts in Mathematics. Springer-Verlag, New York-Berlin, 1978. Second printing, revised. [Jac55] N. Jacobson. A note on automorphisms and derivations of Lie algebras. Proc. Amer. Math. Soc., 6:281 -- 283, 1955. [KK95] A. I. Kostrikin and M. I. Kuznetsov. Two remarks on Lie algebras with nondegenerate derivation. Trudy Mat. Inst. Steklov., 208(Teor. Chisel, Algebra i Algebr. Geom.):186 -- 192, 1995. Dedicated 18 MARCOS GOULART LIMA AND CSABA SCHNEIDER to Academician Igor Rostislavovich Shafarevich on the occasion of his seventieth birthday (Rus- sian). [KR16] Martin Kreuzer and Lorenzo Robbiano. Computational linear and commutative algebra. Springer, Cham, 2016. [LGM02] C. R. Leedham-Green and S. McKay. The structure of groups of prime power order, volume 27 of London Mathematical Society Monographs. New Series. Oxford University Press, Oxford, 2002. Oxford Science Publications. [Mat02] S. Mattarei. The orders of nonsingular derivations of modular Lie algebras. Israel J. Math., 132:265 -- 275, 2002. [Rom08] Steven Roman. Advanced linear algebra, volume 135 of Graduate Texts in Mathematics. Springer, [SF88] New York, third edition, 2008. Helmut Strade and Rolf Farnsteiner. Modular Lie algebras and their representations, volume 116 of Monographs and Textbooks in Pure and Applied Mathematics. Marcel Dekker, Inc., New York, 1988. [Sha99] Aner Shalev. The orders of nonsingular derivations. J. Austral. Math. Soc. Ser. A, 67(2):254 -- 260, [Sze14] 1999. Group theory. Fernando Szechtman. Modular representations of Heisenberg algebras. Linear Algebra Appl., 457:49 -- 60, 2014. (Schneider) Departamento de Matem´atica, Instituto de Ciencias Exatas, Universidade Federal de Minas Gerais, Av. Antonio Carlos 6627, Belo Horizonte, MG, Brazil. Email: [email protected], URL: http://www.mat.ufmg.br/„csaba (Lima) Departamento de Ciencias Exatas e Aplicadas, Universidade Federal de Ouro Preto (Campus Joao Monlevade), Rua 36, no 115, Loanda, Joao Monlevade, MG, CEP 35931-008, Brazil Email: [email protected]
1711.00368	1	1711	2017-11-01T14:37:36	Hom-Lie-Yamaguti Superalgebras	[ "math.RA" ]	(Multiplicative) Hom-Lie-Yamaguti superalgebras which generalize Hom-Lie supertriple systems (and subsequently ternary multiplicative Hom-Nambu superalgebras) and Hom-Lie superalgebras in the same way as Lie-Yamaguti superalgebras [Frac] generalize Lie supertriple systems and Lie superalgebras are defined. We show that the category of (multiplicative) Hom-Lie-Yamaguti superalgebras is closed under twisting by self-morphisms. Construction of some examples of Hom-Lie-Yamaguti superalgebras is given. The notion of an $n^{th}-$derived (binary) Hom-superalgebras is extended to the one of an $n^{th}-$derived binary-ternary Hom-superalgebras and it is shown that the category of Hom-Lie-Yamaguti superalgebras is closed under the process of taking $n^{th}-$ derived Hom-superalgebras.	math.RA	math	HOM-LIE-YAMAGUTI SUPERALGEBRAS DONATIEN GAPARAYI 1 AND SYLVAIN ATTAN 2 1 ECOLE NORMALE SUPÉRIEURE (E.N.S), BP 6983 BUJUMBURA, BURUNDI 2 DÉPARTEMENT DE MATHÉMATIQUES UNIVERSITÉ D'ABOMEY-CALAVI 01 BP 4521, COTONOU 01, BÉNIN [email protected], 2 [email protected] Abstract. (Multiplicative) Hom-Lie-Yamaguti superalgebras which generalize Hom-Lie super- triple systems (and subsequently ternary multiplicative Hom-Nambu superalgebras) and Hom-Lie superalgebras in the same way as Lie-Yamaguti superalgebras [12] generalize Lie supertriple systems and Lie superalgebras are defined. We show that the category of (multiplicative) Hom-Lie-Yamaguti superalgebras is closed under twisting by self-morphisms. Construction of some examples of Hom- Lie-Yamaguti superalgebras is given. The notion of an n −derived (binary) Hom-superalgebras −derived binary-ternary Hom-superalgebras and it is shown that is extended to the one of an n the category of Hom-Lie-Yamaguti superalgebras is closed under the process of taking n −derived Hom-superalgebras. th th th Mathematics Subject Classification: 17A30, 17A32, 17D99 Keywords: Lie-Yamaguti superalgebra [12] (i.e.generalized Lie supertriple system, Lie superal- gebra), Hom-Lie-Yamaguti superalgebra (i.e.generalized Hom-Lie supertriple system, Hom-Lie su- peralgebra). 1. Introduction [12] A Lie-Yamaguti superalgebra is a triple (L, ∗, {, , }) where L = L0⊕L1 is K−vector superspace i.e Z2−graded vector space, " ∗ " a bilinear map (the binary superoperation on L) and "{, , }" a trilinear map (the ternary superoperation on L) such that (SLY 1) x ∗ y = −(−1)xyy ∗ x, (SLY 2) {x, y, z} = −(−1)xy{y, x, z}, (SLY 3) {x, y, z} + (−1)x(y+z){y, z, x} + (−1)z(x+y){z, x, y} + J(x, y, z) = 0, (SLY 4) {x ∗ y, z, u} + (−1)x(y+z){y ∗ z, x, u} + (−1)z(x+y){z ∗ x, y, u} = 0, (SLY 5) {x, y, u ∗ v} = {x, y, u} ∗ v + (−1)u(x+y)u ∗ {x, y, v}, (SLY 6) {x, y, {u, v, w}} = {{x, y, u}, v, w} + (−1)u(x+y){u, {x, y, v}, w} +(−1)(x+y)(u+v){u, v, {x, y, w}}, for all x, y, z, u, v, w in L and (1) J(x, y, z) = (x ∗ y) ∗ z + (−1)x(y+z)(y ∗ z) ∗ x + (−1)z(x+y)(z ∗ x) ∗ y. The relation (1) is called super-Jacobian. Observe that if x ∗ y = 0, for all x, y in L, then a Lie- Yamaguti superalgebra (L, ∗, {, , }) reduces to a Lie supertriple system (L, {, , }) as defined in [17] and if {x, y, z} = 0 for all x, y, z in L, then (L, ∗, {, , }) is a Lie superalgebra (L, ∗) [17]. Recall that a Lie supertriple system [17] is a pair (T, {, , }) where T = T0 ⊕ T1 is a K−vector superspace and 1 2 SYLVAIN ATTAN AND DONATIEN GAPARAYI. "{, , }" a trilinear map (the ternary superoperation on T ) such that (i) {x, y, z} = (−1)xy{y, x, z} (ii) {x, y, z} + (−1)x(y+z){y, z, x} + (−1)z(x+y){z, x, y} = 0 (iii) {x, y, {u, v, w}} = {{x, y, u}, v, v} + (−1)(x+y)u{u, {x, y, v}, w} +(−1)(x+y)(u+v){u, v, {x, y, w}} for all x, y, z in T and a Lie superalgebra [17] is a pair (A, ∗) where A = A0 ⊕ A1 is a K−vector superspace with " ∗ " a bilinear map (the binary superoperation on A) such that for all x, y, z in A. (i) x ∗ y = (−1)xyy ∗ x (ii) J(x, y, z) = 0 A Hom-type generalization of a kind of algebra is obtained by certain twisting of the defining identies by a linear self-map, called the twisting map, in such way that when the twisting map is identity map, then one recovers the original kind of algebra. In this scheme, e.g., associative algebras and Leibniz algebras are twisted into Hom-associative algebras and Hom-Leibniz algebras respectively [16] and, likewise, Hom-type analogues of Novikov algebras, alternative algebras, Jordan algebras or Malcev algebras are defined and discussed in [15],[19], [20]. Also, Hom-Lie triple systems [21], n−ary Hom-Nambu and Hom-Nambu-Lie algebras [5], were introduced in. In the some way, this generalization of binary or ternary algebras has been extended on the binary-ternary algebras. Indeed, Hom-Akivis algebras [10], Hom-Lie-Yamaguti algebras [8] and Hom-Bol algebras [6] which generalize Akivis algebras, Lie-Yamaguti algebras and Bol algebras respectively are introduced. One could say that the theory of Hom-algebras originated in [9] (see also [13],[14]) in the study of defor- mations of Witt and Virasoro algebras (in fact, some q−deformations of Witt and Virasoro algebras have a structures of a Hom-Lie algebra [9]). Some algebraic abstractions of this study are given in [16], [18]. For further more information on other Hom-type algebras, one may refer to, e.g., [5], [7], [15], [19], [20], [21]. In [1], the authors introduce Hom-associative superalgebras and Hom-Lie superalgebras which generalize associative superalgebras [17] and Lie superalgebras [17] respectively. They provide a way for constructing Hom-Lie superalgebras from Hom-associative superalgebras which extend the fundamental construction of Lie superalgebras from associative superalgebras via supercommuta- tor bracket (see the Proposition 1.1. in [17]). Indeed, they show also that the supercommutator bracket defined using the multiplication in a Hom-associative superalgebra leads naturally to Hom- Lie superalgebras. In [2] the authors introduce the Hom-alternative, Hom-Malcev and Hom-Jordan superalgebras wich are the generalization of Alternative, Malcev and Jordan superalgebras respec- tively. Our present study extends the Hom-type generalization of binary superalgebras to the one of ternary superalgebras or binary-ternary superalgebras. The purpose of this paper is to introduced Hom-type generalization of Lie-Yamaguti superalgebras [12], called Hom-Lie-Yamaguti superalge- It is also to extend the notion of an nth−derived (binary) Hom-superalgebras [2] to the bras. one of an nth−derived ternary or binary-ternary Hom-superalgebras and we shown that ternary or binary-ternary superalgebras are closed under the process of taking nth−derived Hom-superalgebras. The rest of this paper is organized as follows. In Section two, we recall basic definitions in Hom- superalgebras theory and useful results about Hom-associative superalgebras and Hom-Lie superalge- bras. In [1], the autors show that the supercommutator bracket defined using the multiplication in a HOM-LIE-YAMAGUTI SUPERALGEBRAS 3 Hom-associative superalgebra leads naturally to Hom-Lie superalgebra. Also, we recall the notion of nth−derived (binary) Hom-superalgebra introduced in [1] and as example, we show that Hom-Lie su- peralgebras are closed under the process of taking nth−derived (binary) Hom-superalgebras (see the Proposition 2.5). In the third Section, we introduce ternary and binary-ternary Hom-superalgebras. In particulary, Hom-Lie-Yamaguti superalgebras which are binary-ternary Hom-superalgebras are defined. Hom-Lie-Yamaguti superalgebras generalize Hom-Lie triple supersystems (and subsequently ternary multiplicative Hom-Nambu superalgebras) and Hom-Lie superalgebras. We provide that any non-Hom-superassociative Hom-superalgebra is a Hom-supertriple system. We also extend a Yau's theorem [18] on Hom-Lie algebras to Hom-Lie-Yamaguti superalgebras and we conclude that these category of Hom-superalgebras are closed under twisting by self-morphisms (see Theorem 3.10). Some examples of Hom-Lie-Yamaguti superalgebras are constructed using Theorem 3.10 via Corol- lary 3.11. In the last Section, we extend the notion of nth− derived (binary) Hom-superalgebra to the case of ternary and binary-ternary Hom-superalgebras. We show that the category of Hom-Lie- Yamaguti superalgebras are closed under the process of taking nth−derived Hom-superalgebras. 2. Some basics on superalgebras We recall some basic facts about Hom-superalgebras, including Hom-associative and Hom-Lie superalgebras [1]. Also, we recall the notion of (binary) nth−derived Hom-superalgebras. As ex- ample, we show that the category of Hom-Lie superalgebras is closed under the process of taking nth−derived Hom-superalgebras [1]. Definition 2.1. 0 ⊕ A′ A′ f (Ai) ⊂ A′ (i) Let f : (A, ∗, α) → (A′, ∗′, α′) be a map, where A = A0 ⊕ A1 and A′ = 1 are Z2−graded vector spaces. The map f is called an even (resp. odd ) map if i (resp. f (Ai) ⊂ A′ i+1), for i = 0, 1. (ii) A Hom-superalgebra is a triple (A, ∗, α) in which A = A0 ⊕ A1 is a K−super-module, ∗ : A × A → A is an even bilinear map, and α : A → A is an even linear map such that α(x ∗ y) = α(x) ∗ α(y) (multiplicativity). Remark 2.2. For convenience, we assume throughout this paper that all Hom-superalgebras are multiplicative. Definition 2.3. Let (A, ∗, α) be a Hom-superalgebra, that is a K−vector superspace A together with a multiplication " ∗ " and an even linear self-map α. (i) [2] The Hom-superassociator of A is the trilinear map asα : A × A × A → A defined as (2) asα = ∗ ◦ (∗ ⊗ α − α ⊗ ∗). In terms of elements, the map asα is given by (ii) [1] The Hom-superalgebra (A, ∗, α) is called a Hom-associative superalgebra if asα(x, y, z) = (x ∗ y) ∗ α(z) − α(x) ∗ (y ∗ z). asα(x, y, z) = 0, ∀ x, y, z ∈ A (iii) [2] The Hom-super-Jacobian of A is the trilinear map Jα : A × A × A → A defined as Jα(x, y, z) := (x ∗ y) ∗ α(z) + (−1)x(y+z)(y ∗ z) ∗ α(x) + (−1)z(x+y)(z ∗ x) ∗ α(y) for all x, y, z ∈ A. If α = Id, then a Hom-super-Jacobian reduces to (1). (iv) [1] The Hom-superalgebra (A, ∗, α) is called a Hom-Lie superalgebra if (3) x ∗ y = −(−1)xyy ∗ x Jα(x, y, z) = 0 (the Hom-super-Jacobi identity) for all x, y, z ∈ A. If α = Id, a Hom-Lie superalgebra reduces to a usual Lie superalgebra. 4 SYLVAIN ATTAN AND DONATIEN GAPARAYI. Recall that the supercommutator bracket defined using the multiplication in any Hom-associative superalgebra leads naturally to a Hom-Lie superalgebra (see the Proposition 2.6 in [1]). Others Hom-Lie superalgebras can be constructed from Lie superalgebras using the Theorem 2.7 in [1]. In the following, we recall the notion of (binary) nth−derived Hom-superalgebra. Definition 2.4. [2] Let (A, ∗, α) be a Hom-superalgebra and n ≥ 0 an integer ( " ∗ " is the binary operation on A). The Hom-superalgebra An defined by An := (A, ∗(n), α2n ), where (x ∗(n) y) := α2n−1(x ∗ y), ∀ x, y ∈ A is called the nth−derived Hom-superalgebra of A. For simplicity of exposition, ∗(n) is written as ∗(n) = α2n−1◦∗. Then notes that A0 = (A, ∗, α), A1 = (A, ∗(1) = α ◦ ∗, α2), and An+1 = (An)1. Proposition 2.5. Let (A, [, ], α) be a multiplicative Hom-Lie superalgebra. Then the nth−derived Hom-superalgebra An = (A, [, ](n) = α2n−1 ◦ [, ], α2n ) is also a multiplicative Hom-Lie superalgebra for each n ≥ 0. Proof: Observe that [x, y](n) = −(−1)xy[y, x](n). Indeed, [x, y](n) = α2n−1([x, y]) = α2n−1((−1)xy[y, x]) = (−1)xyα2n−1([y, x]) = (−1)xy[y, x](n). Then, the superantisymmetry of [x, y](n) holds in An. Next, we have [[x, y](n), α2n (z)](n) + (−1)x(y+z)[[y, z](n), α2n (x)](n) + (−1)x(y+z)[[z, x](n), α2n (y)](n) (z)]) + (−1)x(y+z)α2n−1([α2n−1([y, z]), α2(x)]) = α2n−1([α2n−1([x, y]), α2n + (−1)z(x+y)α2n−1([α2n−1([z, x]), α2n = α2(2n−1)([[x, y], α2(z)] + (−1)x(y+z)[[y, z], α2(x)] + (−1)z(x+y)[[z, x], α2(y)]) = α2(2n−1)(0) = 0 (y)]) (by (3)) and so the Hom- superJacobi identity (3) holds in A(n). Thus, we conclude that A(n) is a (multi- plicative) Hom-Lie superalgebra. In section 4, we extend this notion of nth−one derived (binary) Hom-superalgebra to the case of ternary and binary-ternary Hom-superalgebras. 3. Ternary and binary-ternary Hom-superalgebras In this section, we introduce Hom-Lie supertriple systems and consequently ternary multiplica- tive Hom-Nambu superalgebras. Hom-Lie-Yamaguti superalgebras which are binary-ternary Hom- superalgebras are defined. Hom-Lie-Yamaguti superalgebras reduce to Lie-Yamaguti superalgebras [12] when the twisting map is identity map. Note also that Hom-Lie-Yamaguti superalgebras gen- eralize Hom-Lie supertriple systems (and subsequently ternary multiplicative Hom-Nambu superal- gebras) and Hom-Lie superalgebras. We finish this section by giving some examples of Hom-Lie- Yamaguti superalgebras constructed by using Theorem 3.10 via Corollary 3.11. 3.1. Definitions and Theorem of construction. In this subsection, we give some definitions of ternary Hom-superalgebras and we show that any non-Hom-superassociative Hom-superalgebra has a natural structure of Hom-supertriple system. In the Theorem 3.10, we show that Hom-Lie- Yamaguti superalgebras are closed under twisting by self-morphisms respectively. Definition 3.1. A ternary Hom-superalgebra is triple (T, {, , }, α = (α1, α2)) constitued by a vector K−superspace "T = T0 ⊕ T1", a trilinear map {, , } : T × T × T −→ T, and even linear maps αi : T → T, i = 1, 2, called the twisting maps. The algebra (T, {, , }, α = (α1, α2)) is said multiplicative if α1 = α2 := α and α({x, y, z}) = {α(x), α(y), α(z)} for all x, y, z ∈ T . HOM-LIE-YAMAGUTI SUPERALGEBRAS 5 Definition 3.2. A (multiplicative) ternary Hom-Nambu superalgebra is (a multiplicative) ternary Hom-superalgebra (T, {, , }, α) satisfying {α(x), α(y), {u, v, w}} = {{x, y, u}, α(v), α(w)} (4) + (−1)u(x+y){α(u), {x, y, v}, α(w)} + (−1)(x+y)(u+v){α(u), α(v), {x, y, w}} for all u, v, w, x, y, z ∈ T. The condition (4) is called the ternary Hom-Nambu superidentity. general, the ternary Hom-Nambu superidentity reads: In {α1(x), α2(y), {u, v, w} = {{x, y, u}, α1(v), α2(w)} + (−1)(x+y)u{α1(u), {x, y, v}, α2(w)} + (−1)(x+y)(u+v){α1(u), α2(v), {x, y, w}} for all u, v, w, x, y, z ∈ T, where α1 and α2 are linear self-maps of T. Definition 3.3. A (multiplicative) Hom-supertriple system is a (multiplicative) ternary Hom-superalgebra (T, {., ., .}, α) such that (i) {x, y, z} = −(−1)xy{y, x, z}, (ii) {x, y, z} + (−1)x(y+z){y, z, x} + (−1)z(x+y){z, x, y} = 0 ∀ x, y, z ∈ T. Remark 3.4. Our definition here is motivated by the concern of giving a Hom-type analogue of the relationships between nonassociative superalgebras and supertriple systems (see Proposition 3.5). Proposition 3.5. Any non-Hom-associative superalgebra is a Hom-supertriple system. Proof : Let (A, ∗, α) be a non-Hom-associative superalgebra. Define the supercommutator by := x ∗ y − (−1)xyy ∗ x and the ternary superoperation by {x, y, z} := [[x, y], α(z)] − [x, y] asα(x, y, z) + (−1)xyasα(y, x, z) for all homogeneous elements x, y, z ∈ A and where the su- perassociator as(x, y, z) is given by (2). Then we have {x, y, z} = −(−1)xy{y, x, z} and (cid:9)(x,y,z) (−1)xz{x, y, z} = 0. Thus (A, {, , }, α) is a Hom-supertriple system. (cid:3) Remark 3.6. For α = Id (the identity map), we recover the supertriple system with ternary superop- eration {x, y, z} = [[x, y], z]−as(x, y, z)+(−1)xyas(y, x, z) that is associated to each nonassociative superalgebra and where [x, y] and as(x, y, z) are supercommutator and superassociator respectively for all homogeneous elements x, y, z. Definition 3.7. A Hom-Lie supertriple system is a Hom-supertriple system (A, {, , }, α) satisfy- ing the ternary Hom-Nambu superidentity (4). When α = Id, a Hom-Lie supertriple system reduces to a Lie supertriple system. One can note that Hom-Bol superalgebras introduced in [11] may be viewed as some generalization of Hom-supertriple systems. In the following, we now give the definition of the basic object of this paper and we consider construction methods for Hom-LY superalgebras. These methods allow to find examples of Hom-LY superalgebras starting from ordinary LY superalgebras or even from Malcev superalgebras. Recall that from a Malcev superalgebra (A, ∗) and consider on A the ternary superoperation (5) {x, y, z} := x ∗ (y ∗ z) − (−1)xyy ∗ (x ∗ z) + (x ∗ y) ∗ z, ∀ x, y, z ∈ A, 6 SYLVAIN ATTAN AND DONATIEN GAPARAYI. then (A, ∗, {, , }) is a Lie-Yamaguti superalgebra [12]. First, as the main tool, we show that the category of (multiplicative) Hom-LY superalgebras is closed under self-morphisms (see the Theorem 3.10). Definition 3.8. A Hom-Lie Yamaguti superalgebra (Hom-LY superalgebra for short) is a quadruple (L, ∗, {, , }, α) in which L is K-vector superspace, " ∗ " a binary superoperation and "{, , }" a ternary superoperation on L, and α : L → L an even linear map such that (SHLY 1) α(x ∗ y) = α(x) ∗ α(y), (SHLY 2) α({x, y, z}) = {α(x, )α(y), α(z)}, (SHLY 3) x ∗ y = −(−1)xyy ∗ x, (SHLY 4) {x, y, z} = −(−1)xy{y, x, z}, (SHLY 5) (cid:9)(x,y,z) (−1)xz[(x ∗ y) ∗ α(z) + {x, y, z}] = 0, (SHLY 6) (cid:9)(x,y,z) (−1)xz[{x ∗ y, α(z), α(u)}] = 0, (SHLY 7) {α(x), α(y), u ∗ v} = {x, y, u} ∗ α2(v) + (−1)u(x+y)α2(u) ∗ {x, y, v}, (SHLY 8) {α2(x), α2(y), {u, v, w}} = {{x, y, u}, α2(v), α2(w)} + (−1)u(x+y){α2(u), {x, y, v}, α2(w)} + (−1)(x+y)(u+v){α2(u), α2(v), {x, y, w}}, for all u, v, w, x, y, z ∈ L and where (cid:9)(x,y,z) denotes the sum over cyclic permutation of x, y, z. Note that the conditions (SHLY 1) and (SHLY 2) mean the multiplicativity of (L, ∗, {, , }, α). Remark 3.9. (1) If α = Id, then the Hom-LY superalgebra (L, ∗, {, , }, α) reduces to a LY su- peralgebra (L, ∗, {, , }) (see (SLY 1) − (SLY 6)). (2) If x ∗ y = 0, for all x, y ∈ L, then (L, ∗, {, , }, α) becomes a Hom-Lie supertriple system (L, {, , }, α2) and, subsequently, a ternary Hom-Nambu superalgebra (since, by Definition 3.7, any Hom-Lie supertriple system is automatically a ternary Hom-Nambu superalgebra). (3) If {x, y, z} = 0 for all x, y, z ∈ L, then the Hom-LY superalgebra (L, ∗, {, , }, α) becomes a Hom-Lie superalgebra (L, ∗, α). Theorem 3.10. Let Aα := (A, ∗, {, , }, α) be a Hom-LY superalgebra and let "β" be an even en- domorphism of the superalgebra (A, ∗, {, , }) such that βα = αβ. Let β0 = Id and, for any n ≥ 1 βn := β ◦ βn−1. Define on A the superoperations x ∗β y := βn(x ∗ y), {x, y, z}β := β2n({x, y, z}), for all x, y, z ∈ A. Then, Aβn := (A, ∗β , {, , }β , βnα) are Hom-LY superalgebras, with n ≥ 1. Proof. First, we observe that the condition βα = αβ implies βnα = αβn, n ≥ 1. Next we have (βnα)(x ∗β y) = (βnα)(βn(x) ∗ βn(y)) = βn((αβn)(x) ∗ (αβn)(y)) = (αβn)(x) ∗β (αβn)(y) = (βnα)(x) ∗β (βnα)(y) and we get (SHLY 1) for Aβn. Likewise, the condition βα = αβ implies (SHLY 2). The identities (SHLY 3) and (SHLY 4) for Aβn follow from the skew-supersymmetry of " ∗ " and "{, , }" respectively. HOM-LIE-YAMAGUTI SUPERALGEBRAS 7 Consider now (cid:9)x,y,z [(x ∗β y) ∗β (βnα)(z) + {x, y, z}β ]. Then (x ∗β y) ∗β (βnα)(z) + (−1)x(y+z)(y ∗β z) ∗β (βnα)(x) + (−1)z(x+y)(z ∗β x) ∗β (βnα)(y) +{x, y, z}β + (−1)x(y+z){y, z, x}β + (−1)z(x+y){z, x, y}β = βn(βn(x ∗ y) ∗ (βn(α(z))) + (−1)x(y+z)βn(βn(y ∗ z) ∗ (βn(α(x))) +(−1)z(x+y)βn(βn(z ∗ x) ∗ (βn(α(y))) + β2n({x, y, z}) + (−1)x(y+z)β2n({y, z, x} +(−1)z(x+y)β2n({z, x, y} = β2n((x ∗ y) ∗ α(z)) + (−1)x(y+z)β2n((y ∗ z) ∗ α(x)) +(−1)z(x+y)β2n((z ∗ x) ∗ α(y)) + β2n({x, y, z}) + (−1)x(y+z)β2n({y, z, x} +(−1)z(x+y)β2n({z, x, y} = β2n[(x ∗ y) ∗ α(z) + (−1)x(y+z)(y ∗ z) ∗ α(x) +(−1)z(x+y)(z ∗ x) ∗ α(y) + {x, y, z} + (−1)x(y+z){y, z, x} +(−1)z(x+y){z, x, y}] = β2n(0) (by (SHLY 5) for Aα) = 0 and thus we get (SHLY 5) for Aβn. Next, {x ∗β y, (βnα)(z), (βnα)(u)}β = {β3n(x ∗ y), β3n(α(z)), β3n(α(u))} = β3n({x ∗ y, α(z), α(u)}). Therefore (cid:9)(x,y,z) (−1)xz{x ∗β y, (βnα)(z), (βnα)(u)}β =(cid:9)(x,y,z) (−1)xz[β3n({x ∗ y, α(z), α(u)})] = β3n((cid:9)(x,y,z) (−1)xz({x ∗ y, α(z), α(u)})) = β3n(0) (by (HLY 6) for Aα) = 0 So that we get (SHLY 6) for Aβn . Further, using (SHLY 7) for Aα and condition αβ = βα, we compute {(βnα)(x), (βnα)(y), u ∗β v}β = β3n({α(x), α(y), u ∗ v}) = β3n({x, y, u} ∗ α2(v) + (−1)u(x+y)α2(u) ∗ {x, y, v}) = βn(β2n({x, y, u}) ∗ (β2nα2)(v)) + (−1)u(x+y)βn((β2nα2)(u) ∗ β2n({x, y, v})) = {x, y, u}β ∗β (β2nα2)(v) + (−1)u(x+y)(β2nα2)(u) ∗β {x, y, v}β = {x, y, u}β ∗β (βnα)2(v) + (−1)u(x+y)(βnα)2(u) ∗β {x, y, v}β . Thus (SHLY 7) holds for Aβn . Using repeatedly the condition αβ = βα and the identity (SHLY 8) for Aα, the verification for (SHLY 8)for Aβn is as follows. 8 SYLVAIN ATTAN AND DONATIEN GAPARAYI. {(βnα)2(x), (βnα)2(y), {u, v, w}β }β = β2n({(β2nα2)(x), (β2nα2)(y), β2n({u, v, w})}) = β4n({α2(x), α2(y), {u, v, w}}) = β4n({{x, y, u}, α2(v), α2(w)}) + β4n((−1)u(xy){α2(u), {x, y, v}, α2(w)}) + β4n((−1)(x+y)(u+v){α2(u), α2(v), {x, y, w}}) = β2n({β2n({x, y, u}), (β2nα2)(v), (β2nα2)(w)}) + (−1)u(xy)β2n({(β2nα2)(u), β2n({x, y, v}), (β2nα2)(w)}) + (−1)(x+y)(u+v)β2n({(β2nα2)(u), (β2nα2)(v), β2n({x, y, w})}) = {{x, y, u}β , (βnα)2(v), (βnα)2(w)}β + (−1)u(xy){(βnα)2(u), {x, y, v}β , (βnα)2(w)}β + (−1)(x+y)(u+v){(βnα)2(u), (βnα)2(v), {x, y, w}β }β Thus (SHLY 8) holds for Aβn . Therefore, we get that Aβ is a Hom-LY superalgebra. This finishes (cid:3) the proof. In [18], D. Yau etablished a general method of construction of Hom-algebras from their corre- sponding untwisted algebras. From Theorem 3.1 we have the following method construction of Hom-LY superalgebras from LY superalgebras(this yields examples of Hom-Ly superalgebras). This method is an extension to binary-ternary superalgebras of D. Yau's result ([18], Theorem 2.3). Such an extension to binary-ternary algebras is first mentioned in [10], Corollary 4.5. Corollary 3.11. Let (A, ∗, [, , ]) be a LY algebra and β an endomorphism of (A, ∗, [, , ]). If define on A a binary operation "∗" and a ternary operation {, , } by x∗y := β(x ∗ y), {x, y, z} := β2([x, y, z]), then (A, ∗, {, , }, β) is a Hom-LY algebra. Proof. The proof follows if observe that Corollary 3.11 is Theorem 3.10 when α = Id and n = 1. (cid:3) Proposition 3.12. Let (L, [, ]α, α) be a (multiplicative) Hom-Lie superalgebra. Define on (L, [, ]α, α) a ternary superoperation by (6) {x, y, z}α := [[x, y], α(z)]. Then, (L, [, ]α, {, , }α, α) is a Hom-Lie-Yamaguti Superalgebras. Proof: Straightforward calculations by verification of (SHLY 1−SHLY 8) identities of Definition (cid:3) 3.8 3.2. Examples of Hom-Lie-Yamaguti superalgebras. In the following, we give some examples of Hom-Lie-Yamaguti superalgebras which are constructed firstly from the example 2.8 in [1] and using Proposition 3.12. In the second from the example 3.2 given in [3] (see also in [2]) and using the Theorem 3.10 via Corollary 3.11 . Then, we obtain some family of Hom-Lie-Yamaguti algebras of dimension 5 and dimension 3 respectively. Indeed, HOM-LIE-YAMAGUTI SUPERALGEBRAS 9 Example 3.13. Consider the family of Hom-Lie superalgebra osp(1, 2)λ = (osp(1, 2), [, ]αλ , αλ) given in the example 2.8 in [1]. The Hom-Lie superalgebra bracket [, ]αλ on the basis elements is given, for λ 6= 0, by: [H, X]αλ = 2λ2X, [H, Y ]αλ = −2 λ2 Y, [X, Y ]αλ = H, [Y, G]αλ = 1 λ F, [X, F ]αλ = λG, [H, F ]αλ = − 1 λ F, [H, G]αλ = λG, [G, F ]αλ = H, [G, G]αλ = −2λ2X, [F, F ]αλ = 2 λ2 Y, where αλ : osp(1, 2) → osp(1, 2) is a linear map defined by 1 λ and osp(1, 2) = V0 ⊕ V1 is a Lie superalgebra where V0 is generate by: αλ(X) = λ2X, αλ(Y ) = Y, αλ(H) = H, αλ(F ) = 1 λ2 F, αλ(G) = λG, H =   0 1 0 0 0 0 0 0 −1 and V1 is generated by: F =   such that 0 0 1 0 0 0 0 0 0  , X =      , G =   0 0 0 1 0 0 0 1 0   0 0 0 0 0 0 1 0 0  , Y =      . 0 1 0 0 0 −1 0 0 0 [H, X] = 2X, [H, F ] = −F, [H, Y ] = −2Y, [H, G] = G, [X, Y ] = H, [G, F ] = H, [Y, G] = F, [G, G] = −2X, [X, F ] = G, [F, F ] = 2Y, Now, we define on osp(1, 2)λ = (osp(1, 2), [, ]αλ , αλ) a ternary superoperation by (6). Then, sLY (1, 2)λ = (osp(1, 2), [, ]αλ , {, , }αλ , αλ) is a family of Hom-Lie-Yamaguti superalgebras where the supercommutator [, ]αλ is defined as above in this example 3.13 and the ternary superoperation {, , }αλ (we give only the ones with non zero values in the left hand side and using the identity (SHLY 4) of Definition 3.8, one can deduce the others values wich are non zero) is defined by 2H = {H, X, Y }αλ = {H, Y, X}αλ = 2{H, F, G}αλ = 2{H, G, F }αλ = 2{X, F, F }αλ = −2{Y, G, G}αλ = −{F, F, X}αλ −2λ4X = 1 2 {H, X, H}αλ = −{X, Y, X}αλ = {F, G, X}αλ = {H, G, G}αλ = {X, F, G}αλ 4 λ4 − 2 λ2 Y = {H, Y, H}αλ = 2{Y, X, Y }αλ = 2{F, G, Y }αλ = −2{H, F, F }αλ = 2{Y, G, F }αλ = −{F, F, H}αλ F = 2{H, F, H}αλ = {H, Y, G}αλ = 2{H, G, Y }αλ = 2{X, Y, F }αλ = −{F, F, G}αλ = 2{X, F, Y }αλ = −2{Y, G, H}αλ = −2{F, G, F }αλ −λ2G = {H, G, H}αλ = − 1 2 {H, X, F }αλ = −{H, F, X}αλ = −{X, Y, G}αλ = {X, F, H}αλ = {Y, G, X}αλ = {F, G, G}αλ These Hom-Lie-Yamaguti superalgebras are not Lie-Yamaguti superalgebras for λ 6= ±1. Indeed, the left hand side of the identity (SLY 3), for α = Id, leads to (cid:9)(H,Y,X) (−1)HX ([H, Y ], X] + {H, Y, X}) = 2(cid:18) 1 − λ4 λ2 (cid:19) H 10 SYLVAIN ATTAN AND DONATIEN GAPARAYI. and also the left hand side of the identity (SLY 4), for α = Id, leads to Then, they not vanish for λ 6= ±1. (cid:9)(H,Y,X) (−1)HX{[H, Y ], X, G} = 2(cid:0)1 − λ2(cid:1) G. Example 3.14. From the example M 3(3, 1) of a non-Lie Malcev superalgebra given in [3] (see also in [2]) and defining on it a ternary superoperation by (5), we obtain a Lie-Yamaguti super- algebra (sLY (3, 1), [, ], {, , }) of dimension 4. defined with respect to a basis (e1, e2, e3, e4), where (sLY (3, 1))0 = span(e1, e2, e3) and (sLY (3, 1))1 = span(e4), by the following multiplication table (= −[e3, e1]), [e2, e3] = 2e2 (= −[e3, e2]), [e3, e4] = −e4 (= −[e4, e3]), [e1, e3] = −e1 [e4, e4] = e1 + e2, {e1, e3, e3} = 2e1 {e3, e4, e3} = 2e4 {e4, e4, e3} = −e1 − 4e2. (= −{e3, e1, e3}), {e2, e3, e3} = 8e2 (= −{e4, e3, e3}), {e3, e4, e4} = e1 − 2e2 (= −{e3, e2, e3}), (= −{e4, e3, e4}) Consider the even superalgebra endomorphism α1 : sLY (3, 1) → sLY (3, 1) with respect to the same basis define by α1(e1) = a2e1, α1(e2) = a2e2, α1(e3) = be1 + ce2 + e3, α1(e4) = a2e4 ∀ a, b, c ∈ K For each such even superalgebra endomorphism α1 and using the Theorem 3.10 via Corollary 3.11, there is a Hom-Lie-Yamaguti superalgebra sLY (3, 1)α1 = (sLY (3, 1), [, ]α1 , {, , }α1 ) such that (= −[e3, e1]α1), (= −[e4, e3]α1 ), [e1, e3]α1 = −a2e1 [e3, e4]α1 = −ae4 {e1, e3, e3}α1 = 2a4e1 {e3, e4, e3}α1 = −2a5e4 {e4, e4, e3}α1 = a4((2a − 1)e1 + 2(a + 1)e2). [e2, e3]α1 = 2a2e2 [e4, e4]α1 = a2(e1 + e2), (= −[e3, e2]α1), (= −{e3, e1, e3}α1 ), {e2, e3, e3}α1 = 8a4e2 (= −{e3, e2, e3}α1), (= −{e4, e3, e3}α1 ), {e3, e4, e4}α1 = a4(e1 − 2e2) (= −{e4, e3, e4}α1 ) While for example (cid:9)(e3,e4,e4) (−1)e3e4([[e3, e4], e4] + {e3, e4, e4}) = a4(2(a − 1)e1 + (2a + 3)e2), then for a 6= 0, these Hom-Lie-Yamaguti superalgebras are note Lie-Yamaguti superalgebras. In the same a way, when we consider an even superalgebra endomorphism α2 : sLY (3, 1) → sLY (3, 1) with respect to the same basis define by α2(e3) = be1+ce2+ 1 K and using the Theorem 3.10 via Corollary 3.11, we obtain a twisting of sLY (3, 1) into a family of Hom-Lie superalgebras define by [e3, e4] = −de2 (= −[e4, e3]) which are also Hom-Lie-Yamaguti superalgebras in particulary case where the ternary superoperation is zero. Observe that they are also Lie superalgebras and consequently they are Lie-Yamaguti superalgebras where the ternary superop- eration is zero. 2 e3, α2(e4) = de4, ∀ a, b, c, d ∈ 4. nth−derived binary-ternary Hom-superalgebras In this section, we extend the notion of nth−derived (binary) Hom-superalgebra to the case of ternary and binary-ternary Hom-superalgebras. In particulary, we introduce nth−derived of Hom- Lie-Yamaguti superalgebra and we shown that the category of Hom-Lie-Yamaguti superalgebras is closed under the process of taking nth−derived Hom-superalgebras. HOM-LIE-YAMAGUTI SUPERALGEBRAS 11 Definition 4.1. Let A := (A, {, , }, α) be a ternary Hom-superalgebra and n ≥ 0 an integer. Define on A the nth−derived ternary operation {, , }(n) by (7) {x, y, z}(n) := α2n+1−2({x, y, z}), ∀ x, y, z ∈ A. Then A(n) := (A, {, , }(n), α2n ) will be called the nth−one derived ternary Hom-superalgebra of A. Now denote {x, y, z}(n) = α2n+1−2({x, y, z}). Then we note that A0 = (A, {, , }, α), A1 = (A, {x, y, z}(1) = α2 ◦ {x, y, z}, α2), and An+1 = (An)1. Definition 4.2. Let A := (A, ∗, {, , }, α) be a binary-ternary Hom-superalgebra and n ≥ 0 an integer. Define on A the nth−derived binary operation and the nth−derived ternary operation [, , ](n) by (8) (9) x ∗(n) y := α2n−1(x ∗ y) {x, y, z}(n) := α2n+1−2({x, y, z}), ∀ x, y, z ∈ A. Then A(n) := (A, ∗(n), {, , }(n), α2n ) will be called the nth−one derived (binary-ternary) Hom- superalgebra of A. Denote ∗(n) = α2n−1 ◦ ∗ and {x, y, z}(n) = α2n+1−2({x, y, z}). Then we note that A0 = (A, ∗, {, , }, α), A1 = (A, ∗(1) = α ◦ ∗, {x, y, z}(1) = α2 ◦ {x, y, z}, α2), and An+1 = (An)1. One observes that, from Definition 4.2, if set {x, y, z} = 0, ∀ x, y, z ∈ A, we recover the nth−derived (binary) Hom-superalgebra of Definition 2.4 and if x∗y = 0 ∀ x, y ∈ A, then we have an nth−derived (ternary) Hom-superalgebra (see Definition 4.1) Proposition 4.3. Let A := (A, {, }, α) be a multiplicative Hom-Lie supertriple system. Then for each n ≥ 0, the nth−derived Hom-superalgebra A(n) = (A, {, , }(n) = α2n+1−2 ◦ {, , }, α2n ) of A is a multiplicative Hom-Lie supertriple system. Proof: The proof of this Proposition 4.3 can be constitued throughout the proof of Theorem 4.4 (cid:3) below if the binary superoperation is zero. In the following result, we shown that the category of Hom-Lie-Yamaguti superalgebras is closed under the process of taking nth−derived Hom-superalgebras. Theorem 4.4. Let Aα := (A, [, ], {, , }, α) be a Hom-Lie-Yamaguti. Then, for each n ≥ 0 nth derived Hom-algebra A(n) := (A, [, ](n) = α2n−1 ◦ [, ], {, , }(n)) = α2n+1−2 ◦ {, , }, α2n ) is a Hom-Lie-Yamaguti algebra. Proof: The identies (SHLY 1) − (SHLY 4) for A(n) are obvious. The checking of SHLY 5 for )(z)](n) + (−1)xy{x, y, z}(n)). A(n) is as follows. Indeed, consider (cid:9)(x,y,z) ((−1)xy[[x, y](n), (α2n Then )(z)](n) + (−1)xz{x, y, z}(n)) (cid:9)(x,y,z) ((−1)xz[[x, y](n), (α2n =(cid:9)(x,y,z) ((−1)xz[α2n−1(α2n−1([x, y])), (α2n −1(α(z)))])+ (cid:9)(x,y,z) (α2n+1−2((−1)xz{x, y, z})) = α2n+1−2((cid:9)(x,y,z) ((−1)xz[[x, y], α(z)])+ (cid:9)(x,y,z) ((−1)xz{y, z, x})) = α2n+1−2((cid:9)(x,y,z) ((−1)xz[[x, y], α(z)] + (−1)xz{x, y, z})) = α2n+1−2(0) (by (SHLY 5) for Aα) = 0 and thus we get (SHLY 5) for A(n). 12 SYLVAIN ATTAN AND DONATIEN GAPARAYI. Next, {[x, y](n), (α2n )(z), (α2n Therefore )(u)}(n) = {α2n−1([x, y]), (α2n )(z), (α2n = α2n+1−2({α2n−1([x, y]), (α2n = α2n+1−2α2n−1({[x, y], α(z), α(u)}) = α3(2n−1)({[x, y], α(z), α(u)}). )(u)}(n) )(z), (α2n )(u)}) (cid:9)(x,y,z) ((−1)xz{[x, y](n), (α2n )(z), (α2n )(u)}(n)) = (cid:9)(x,y,z) [α3(2n −1)((−1)xz{[x, y], α(z), α(u)})] = α3(2n−1)((cid:9)(x,y,z) ((−1)xz{[x, y], α(z), α(u)})) = α3(2n−1)(0) (by (SHLY 6) for Aα) = 0 So that we get (SHLY 6) for A(n). Further, using (SHLY 7) for Aα and condition of multiplicativity linearity of α, we compute {(α2n )(y), α2n−1([u, v])}) )(y), [u, v](n)}(n) )(x), (α2n )(x), (α2n = α2n+1−2({(α2n = α2n+1−2α2n−1({α(x), α(y), [u, v]}) = α2n+1−2α2n−1([{x, y, u}, α2(v)] + (−1)u(x+y)[α2(u), {x, y, v}]) = α2n−1([{x, y, u}(n), (α2n = [{x, y, u}(n), (α2n )2(v)](n) + (−1)u(x+y)[(α2n )2(v)] + (−1)u(x+y)[(α2n )2(u), {x, y, v}(n)](n). )2(u), {x, y, v}(n)]) Thus (SHLY 7) holds for A(n). Using repeatedly the condition of multiplicativity and the identity (SHLY 8) for Aα, the verification for (SHLY 8) for A(n) is as follows. {(α2n )2(x), (α2n )2(y), {u, v, w}(n)}(n) )2)(x), (α2n )(x), (α2n+1 )2)(y), α2n+1−2({u, v, w})}) )(y), α2n+1−2({u, v, w})}) = α2n+1−2({(α2n = α2n+1−2({(α2n+1 = (α2n+1−2)2({α2(x), α2(y), {u, v, w}}) = (α2n+1−2)2({{x, y, u}, α2(v), α2(w)}) + (α2n+1−2)2((−1)u(xy){α2(u), {x, y, v}, α2(w)}) + (α2n+1−2)2((−1)(x+y)(u+v){α2(u), α2(v), {x, y, w}}) = (α2n+1−2)({(α2n+1−2)({x, y, u}), (α2n+1 −2)(α2(v)), (α2n+1 −2)(α2(w))}) + (α2n+1−2)((−1)u(xy){(α2n+1−2)(α2(u)), (α2n+1−2)({x, y, v}), (α2n+1 −2)(α2(w))}) + (α2n+1−2)((−1)(x+y)(u+v){(α2n+1−2)(α2(u)), (α2n+1−2)(α2(v)), (α2n+1 −2)({x, y, w})}) = {{x, y, u}(n), (α2n + (−1)u(xy){(α2n + (−1)(x+y)(u+v){(α2n )2(v), (α2n )2(u), {x, y, v}(n), (α2n )2(v), {x, y, w}(n)}(n) )2(u), (α2n )2(w)}(n) )2(w)}(n) HOM-LIE-YAMAGUTI SUPERALGEBRAS 13 Thus (SHLY 8) holds for A(n). Therefore, we get that A(n) is a Hom-LY superalgebra. This (cid:3) finishes the proof. Remark 4.5. (1) If [x, y](n) = 0, for all x, y ∈ L, then nth−derived of Hom-Lie-Yamaguti super- algebras becomes an nth−derived of Hom-Lie supertriple system and, subsequently, nth−derived of ternary Hom-Nambu superalgebra (since, by Definition 3.7, any Hom-Lie supertriple sys- tem is automatically a ternary Hom-Nambu superalgebra). (2) If {x, y, z}(n) = 0, for all x, y, z ∈ L, then the nth−derived of Hom-LY superalgebra becomes a nth−derived of Hom-Lie superalgebra. References [1] , K. Abdaoui, F. Ammar and A. Makhlouf, Hom-Lie Superalgebras and Hom-Lie admissible Superalgebras, Journal of algebra, Vol. 324, Issue 7, 1513-1528 (2010). [2] , K. Abdaoui, F. Ammar and A. Makhlouf, Hom-alternative, Hom-Malcev and Hom-Jordan Superalgebras, arXiv: 1304.1579v1 [math.RA] Apr 2013. [3] H. Albuquerque and A. Elduque, Classification of Mal'tsev Superalgebras of small dimensions, Algebra and Logic, Vol. 35 (6) (1996) 512-554. [4] H. Albuquerque A. P. Santana Akivis superalgebras and speciality, arXiv: 0710.4535v1 [math.RA] 24 oct 2007. [5] H. Ataguema, A. Makhlouf and S.D. Silvestrov, Generalization of n-ary Nambu algebras and beyond, J. Math. Phys., 50 (2009), 083501. [6] S. Attan, A. Nourou Issa, Hom-Bol algebras, Quasigroups and related Systems 21 (2013), 131-146. [7] Y. Fregier, A. Gohr, Unital algebras of Hom-associative type and surjective or injective twistings, J. of Gen. Lie Theo. and Appl. Vol. 3, (2009), No. 4, 285-295. [8] D. Gaparayi and A. N. Issa, A twisted generalization of Lie-Yamaguti algebras, Int. J. Algebra 6 (2012), no. 7, 339-352. [9] J. T. Hartwig, D. Larsson and S. D. Silvestrov, Deformations of Lie algebras using σ−derivations, J. Algebras, 292 (2006), 314-361. [10] A. N. Issa, Hom-Akivis algebras, Comment. Math. Univ. Carolin. 52 (4) (2011), 485-500. 292 (2006), 314-361. [11] A. N. Issa, Supercommutator algebras of right (Hom)-alternative superalgebras, arXiv:1710.02706v1[math.RA] [12] A. Koulibaly, M.F. Ouedraogo Supersystèmes Triples de Lie associés aux superalgèbres de Malcev Africa Matem- atika, série 3, Volume 14 (2002). [13] D. Larsson and S. D. Silvestrov, Quasi-Hom-Lie algebras, central extensions and 2-cycle-like identies, J. Algebra 288 (2005), 321-344. [14] D. Larsson and S. D. Silvestrov, Quasi-Lie algebras, Comptemp.Math. 391 (2005). [15] A. Makhlouf, Hom-Alternative algebras and Hom-Jordan algebras, Int. Elect. J. Alg., 8 (2010), 177-190. Meeting, 143-177, Editional Circulo Rojo, Almeria, 2010. [16] A. Makhlouf, Silvestrov S.D., Hom-algebra structures, J. Gen. Lie Theory Appl. 2 (2008), 51-64. [17] S. Okubo, Jordan-Lie Super Algebra and Jordan-Lie Triple System, Journal of algebra 198, 388-411 (1997). [18] D. Yau, Hom-algebras and homology, J. Lie Theory, 19 (2009), 409-421. [19] D. Yau, Hom-Novikov algebras, J. Phys. A 44 (2011), 085202. [20] D. Yau, Hom-Maltsev, Hom-alternative and Hom-Jordan algebras, International Electronic Journal of algebras, 11 (2012), 177-217. [21] D. Yau , On n-ary Hom-Nambu and Hom-Nambu-Lie algebras, J. Geom. Phys. 62 (2012), 506-522.
1205.2489	1	1205	2012-05-11T11:33:45	Left unital Kantor triple systems and structurable algebras	[ "math.RA" ]	Left unital Kantor triple systems will be shown to correspond to structurable algebras endowed with an involutive automorphism. A related result is proved for (-1,-1) Freudenthal-Kantor triple systems. Some consequences for the associated 5-graded Lie algebras and superalgebras are deduced too. In particular, left unital (-1,-1) Freudenthal-Kantor triple systems are shown to be intimately related to Lie superalgebras graded over the root system of type B(0,1).	math.RA	math	LEFT UNITAL KANTOR TRIPLE SYSTEMS AND STRUCTURABLE ALGEBRAS ALBERTO ELDUQUE⋆, NORIAKI KAMIYA⋆⋆, AND SUSUMU OKUBO⋆⋆⋆ Abstract. Left unital Kantor triple systems will be shown to coincide with the triple systems attached to structurable algebras endowed with an involutive automorphism σ, with triple product given by the formula xyz = Vx,σ(y)(z). A related result is proved for (−1, −1) Freudenthal-Kantor triple systems. Some consequences for the associated 5-graded Lie algebras and superalgebras are deduced too. In particular, left unital (−1, −1) Freudenthal-Kantor triple systems are shown to be intimately related to Lie superalgebras graded over the root system of type B(0, 1). A structurable algebra over a field is a unital (binary) algebra with involution 1. Introduction (A,·,¯) that satisfies (x − ¯x, y, z) = (y, ¯x − x, z), (1.1) (1.2) for any u, v, x, y, z ∈ A, where (x, y, z) = (x · y) · z − x · (y · z) is the associator of x, y, z, and (1.3) [Vu,v, Vx,y] = VVu,v x,y − Vx,Vv,uy, Vx,y(z) = (x · ¯y) · z + (z · ¯y) · x − (z · ¯x) · y. (See [AF93].) For fields of characteristic not two or three, it is known that (1.1) follows from (1.2) [A78]. On the other hand, a Kantor triple system (or generalized Jordan triple system of second order [K72, K73]) is a vector space U endowed with a trilinear map U × U × U → U satisfying: [L(u, v), L(x, y)] = L(uvx, y) − L(x, vuy), K(K(u, v)x, y) = K(u, v)L(x, y) + L(y, x)K(u, v), (1.4) (1.5) for any u, v, x, y ∈ U , where the maps L(x, y) and K(x, y) are given by L(x, y) : z 7→ xyz, K(x, y) : z 7→ xzy − yzx, for x, y, z ∈ U . Given a structurable algebra (A,·,¯), we may consider the triple product {xyz} = Vx,y(z). Then (A,{xyz}) is a Kantor triple system [F94]. (1.6) Date: May 7, 2012. ⋆ Supported by the Spanish Ministerio de Econom´ıa y Competitividad and FEDER (MTM MTM2010-18370-C04-02) and by the Diputaci´on General de Arag´on (Grupo de Investigaci´on de ´Algebra). ⋆⋆ Supported by a Grant-in-aid for Scientific Research no. 19540042(C),(2), of the Japan Society for the Promotion of Science. ⋆⋆⋆ Supported in part by U.S. Department of Energy Grant No. DE-FG02-91 ER40685. 1 2 ALBERTO ELDUQUE, NORIAKI KAMIYA, AND SUSUMU OKUBO Moreover, if e = 1 is the unity element of the structurable algebra (A,·,¯), then we have {eex} = x, 2{xee} + {exe} = 3x, (1.7) (1.8) for any x ∈ A. Conversely, over a field of characteristic not two or three, if (U,{xyz}) is a Kantor triple system, and if e is an element in U satisfying (1.7) and (1.8), then Faulkner proved in [F94, Lemma 1.7] that (U,{xyz}) is the Kantor triple system obtained from a structurable algebra (U,·,¯) defined on U , with unity e, so that {xyz} = (x · ¯y) · ¯z + (z · ¯y) · x − (z · ¯x) · y as in (1.3) and (1.6). (A different proof is given in [KO10, Theorem 5.1].) In [F94], Faulkner studied a class of symmetric spaces called rotational, and he proved that they are intimately connected to real structurable algebras. To do so, he needed to prove first that if a Kantor triple system contains a distinguished element e satisfying only (1.7) (we say then that e is a left unit of the Kantor triple system) then, assuming the characteristic is not two, three or five, still there exists a structurable algebra attached to the triple system, but in a indirect way. The definition of this structurable algebra is involved, and the triple product of the Kantor triple system does not have a clear description in terms of the binary product in the structurable algebra. In this paper it will be proved that, over fields of characteristic not two or three, we may attach to any Kantor triple system with a left unit a structurable algebra (A,·,¯) endowed with an involutive automorphism σ, defined on the vector space of the Kantor triple system, such that now the triple product is given by the simple expression xyz = Vx,σ(y)(z), = (x · σ(¯y)) · z + (z · σ(¯y)) · x − (z · ¯x) · σ(y), (1.9) and conversely. That is, the Kantor triple systems with a left unit are precisely the Kantor triple systems obtained from structurable algebras (A,·,¯) endowed with an involutive automorphism σ, so that the triple product is recovered by (1.9). Moreover, the arguments used for left unital Kantor triple systems can be used to study left unital (−1,−1) Freudenthal-Kantor triple systems. Besides, the 5-graded Lie algebra naturally attached to any left unital Kantor triple system is shown to be graded over the nonreduced root system BC1, and of type B1, while the Lie superalgebra attached to any left unital (−1,−1) Freudenthal-Kantor triple system is graded over the root system of the simple Lie superalgebra B(0, 1). Conversely, any B(0, 1)-graded Lie superalgebra gives rise to a left unital (−1,−1) Freudenthal- Kantor triple system. It should be mentioned here that the case of the generalized Jordan triple systems containing an element e satisfying eex = xee = x instead of (1.7) and (1.8) has been studied to be intimately related with a variety of nonconmutative Jordan algebras in [EKO05]. Some balanced (−1,−1) Freudenthal-Kantor triple systems are then related to the algebras in this variety that have degree two. Throughout the paper we fix a ground field F of characteristic not two or three. 2. Left unital Kantor triple systems and structurable algebras with involutive automorphisms Let us first recall some definitions. Given a triple system (U, xyz) we will denote by L(x, y) the linear operator on U given by L(x, y)z = xyz, for x, y, z ∈ U . LEFT UNITAL KANTOR TRIPLE SYSTEMS AND STRUCTURABLE ALGEBRAS 3 Definition 2.1 ([K72]). (i) A triple system (U, xyz) is said to be a generalized Jordan triple system if it satisfies (2.1) (2.2) (2.3) (2.4) uv(xyz) = (uvx)yz − x(vuy)z + xy(uvz) for any u, v, x, y, z ∈ U or, equivalently, [L(u, v), L(x, y)] = L(L(u, v)x, y) − L(x, L(v, u)y), for u, v, x, y ∈ U . (ii) A generalized Jordan triple system (U, xyz) is said to be left unital if there is an element e ∈ U such that L(e, e) = id (i.e., eex = x for any x ∈ U ). The element e is said to be a left unit. (iii) A generalized Jordan triple system (U, xyz) is called a Kantor triple system if it satisfies K(K(u, v)x, y) = K(u, v)L(x, y) + L(y, x)K(u, v), for any u, v, x, y ∈ U , where K(x, y)z = xzy − yzx. Remark 2.2. The left unit is not necessarily unique, as shown by the generalized Jordan triple system (U, xyz), with xyz = (xy)z for a symmetric bilinear form. Any element u with (uu) = 1 is a left unit here. Given a left unital generalized Jordan triple system (U, xyz), fix a left unit e and (cid:3) consider the linear maps: (ρ : U → U, x 7→ xee, µ : U → U, x 7→ exe. (2.5) Lemma 2.3. Let (U, xyz) be a left unital generalized Jordan triple system with left unit e. Then, for any x, y ∈ U , the following conditions hold L(xye, e) = L(e, yxe), (2.6) L(ρ(x), e) = L(e, µ(x)), L(µ(x), e) = L(e, ρ(x)), ρ2 = µ2, ρµ = µρ. (2.7) (2.8) Proof. Equation (2.2) with x = y = e, so that L(x, y) = id, gives (2.6). Now (2.7) is obtained by imposing x = e or y = e in (2.6). Also, (2.7) implies ρ(x)ee = eµ(x)e and µ(x)ee = eρ(x)e, or ρ2 = µ2 and ρµ = µρ. (cid:3) Lemma 2.4. Let (U, xyz) be a left unital Kantor triple system with left unit e. Then, for any u, v ∈ U , the following conditions hold: K(u, e)e = (ρ − id)(u), (2.9) K(u, v) = K(K(u, v)e, e), 1 2 (ρ − id)(ρ − 3id) = 0. In particular, ρ and µ are invertible linear operators. (2.10) (2.11) Proof. Equation (2.9) follows from the definition of the operators K(x, y) in (2.4), while (2.10) is a straightforward consequence of (2.3). Now, from (2.9) and (2.10) we get (ρ − id)(u) = K(u, e)e = whence (2.11). 1 2 K(K(u, e)e, e) = 1 2 (ρ − id)K(u, e)e = 1 2 (ρ − id)2(u), (cid:3) 4 ALBERTO ELDUQUE, NORIAKI KAMIYA, AND SUSUMU OKUBO Therefore, if (U, xyz) is a left unital Kantor triple system with left unit e, then U = U1 ⊕ U3, with Ui = {x ∈ U : ρ(x) = ix}, i = 1, 3. Define the binary product ⋆ on U by means of (2.12) x ⋆ y = eµ−1(x)y for x, y ∈ U . Lemma 2.5. With these conventions, we have: (i) e is the unity of the algebra (U, ⋆): e ⋆ x = x = x ⋆ e for any x ∈ U . (ii) For any x, y ∈ U exy = µ(x) ⋆ y, xey = ρ(x) ⋆ y. (iii) For any x, y, z ∈ U xyz = x ⋆ (µ(y) ⋆ z) − µ(y) ⋆ (x ⋆ z) + µ(cid:0)µρ−1(x) ⋆ y(cid:1) ⋆ z. (iv) For any x, y ∈ U we have xye = Λx,y(eyx), with (2.13) (2.14) (2.15) (2.16) Λx,y =  for x ∈ U1, id − ρ for x ∈ U3, y ∈ U1, 1 3 ρ for x, y ∈ U3, 2x ⋆ y − y ⋆ x, 3y ⋆ x, 4x ⋆ y − y ⋆ x, if x, y ∈ U1 or x, y ∈ U3, if x ∈ U1 and y ∈ U3, if x ∈ U3 and y ∈ U1. where ρ = 3id − 2ρ. (v) For any x, y ∈ U , ρ(x ⋆ y) =  and Proof. The assertion in (i) is clear. For (ii) note that µ(x)⋆y = e(cid:0)µ−1µ(x)(cid:1)y = exy, ρ(x) ⋆ y = e(µ−1ρ(x))y = e(ρµ−1(x))y = L(e, ρµ−1(x))(y) = L(x, e)y = xey, because of (2.7). For (iii) we start with which we rewrite, using (ii), as ex(eyz) = (exe)yz − e(xey)z + ey(exz), µ(x) ⋆(cid:0)µ(y) ⋆ z(cid:1) = µ(x)yz − µ(ρ(x) ⋆ y) ⋆ z + µ(y) ⋆ (µ(x) ⋆ z), which is equivalent to the assertion in (iii). For (iv), if x ∈ U1, (2.9) gives K(x, e)e = 0, and hence K(x, e) = 0 by (2.10) and xye = K(x, e)y + eyx = eyx. However, if x ∈ U3, (2.9) gives K(x, e)e = 2x and K(K(x, e)y, e)e equals ((ρ − id)K(x, e)y, by (2.10), K(x, e)L(y, e)e + L(e, y)K(x, e)e = K(x, e)ρ(y) + 2eyx, by (2.3). Thus (ρ − id)K(x, e)y = 2eyx + K(x, e)ρ(y) for any y ∈ U . Now, if y ∈ U1 this gives (ρ − 2id)K(x, e)y = 2eyx, but (ρ − 2id)2 = id by (2.11), so we obtain xye − eyx = K(x, e)y = 2(ρ − 2id)(eyx), or xye = (2ρ − 3id)(eyx) = − ρ(eyx). If y ∈ U3 the same argument gives (ρ− 4id)K(x, e)y = 2eyx, and ρ(ρ − 4id) = −3id, so we get xye − eyx = K(x, e)y = − 1 3 2ρ(eyx) or xye = 3 (3id − 2ρ)(eyx) = 1 3 ρ(eyx). 1 LEFT UNITAL KANTOR TRIPLE SYSTEMS AND STRUCTURABLE ALGEBRAS 5 For (v) we start with which we rewrite as ex(yee) = (exy)ee − y(xee)e + ye(exe), exρ(y) = ρ(exy) − yρ(x)e + yeµ(x). With x 7→ µ−1(x) and using (ii), this is equivalent to x ⋆ ρ(y) = ρ(x ⋆ y) + ρ(y) ⋆ x − yρµ−1(x)e. Using (ii) and (iii) this gives: ρ(x ⋆ y) = x ⋆ ρ(y) − ρ(y) ⋆ x +  ρ(x) ⋆ y − ρ(ρ(x) ⋆ y), 1 3 ρ(ρ(x) ⋆ y), if y ∈ U1, if x ∈ U1 and y ∈ U3, if x, y ∈ U3. If x, y ∈ U1 we get ρ(x ⋆ y) = x ⋆ y − y ⋆ x + x ⋆ y = 2x ⋆ y − y ⋆ x. If x, y ∈ U3 we get ρ(x⋆y) = 3x⋆y−3y ⋆x+ ρ(x⋆y), that is, ρ(x⋆y) = 2x⋆y−y ⋆x again. If x ∈ U1 and y ∈ U3 we get ρ(x ⋆ y) = 3x ⋆ y − 3y ⋆ x− ρ(x ⋆ y), that is, ρ(x ⋆ y) = 3y ⋆ x. Finally, if x ∈ U3 and y ∈ U1, we obtain ρ(x ⋆ y) = x ⋆ y − y ⋆ x − 3x ⋆ y = 4x ⋆ y − y ⋆ x. (cid:3) Proposition 2.6. Let (U, xyz) be a left unital Kantor triple system with left unit e, and consider the linear map σ = µ−1 ρ (ρ = 3id − 2ρ as above). Then σ is an involutive automorphism of (U, xyz). Proof. σ commutes with µ and ρ and σ2 = µ−2(3id− 2ρ)2 = ρ−2(9id− 12ρ + 4ρ2) = ρ−2(3(ρ − id)(ρ − 3id) + ρ2) = id by (2.8) and (2.11), and hence (2.14) shows that it is enough to prove that σ is an automorphism of the algebra (U, ⋆) defined by (2.12). Therefore we must prove σ(µ(x) ⋆ µ(y)) = σ(µ(x)) ⋆ σ(µ(y)) or ρ(cid:0)µ(x) ⋆ µ(y)(cid:1) = µ(cid:0)ρ(x) ⋆ ρ(y)(cid:1) for x, y ∈ U . Equation (2.6) gives (xye)ee = e(yxe)e, or ρ(xye) = µ(yxe), which is equivalent (Lemma 2.5) to ρΛx,y(µ(y) ⋆ x) = µΛy,x(µ(x) ⋆ y). Change x to µ(x) and use (2.8) to obtain (2.18) Set α(x) = i if x ∈ Ui, i = 1, 3, and β(x) = 1 for x ∈ U1 and β(x) = −3 for x ∈ U3. Hence ρ(x) = α(x)x and ρ(x) = β(x)x for any x ∈ U1 ∪ U3. Then (2.18) can be rewritten as ρΛx,y(cid:0)µ(y) ⋆ µ(x)(cid:1) = µΛy,x(cid:0)ρ2(x) ⋆ y(cid:1). (2.17) ρΛx,y(cid:0)µ(y) ⋆ µ(x)(cid:1) = = α2(x) β(x)β(y) β(x) β(y) Λy,xµ(cid:0)ρ(x) ⋆ ρ(y)(cid:1) Λy,xµ(cid:0)ρ(x) ⋆ ρ(y)(cid:1), ρ(x ⋆ y) = ρΛ−1 y,xΛx,y(y ⋆ x), β(y) β(x) for x, y ∈ U1 ∪ U3, because α2(x) = β2(x). Then, (2.17) is satisfied if and only if or for x, y ∈ U1 ∪ U3. y ⋆ x = β(x) β(y) Λ−1 x,yΛy,xρ−1 ρ(x ⋆ y), (2.19) for x, y ∈ U1 or x, y ∈ U3, for x ∈ U1, y ∈ U3, for x ∈ U3, y ∈ U1. Λ−1 id, − ρ, − ρ−1, x,yΛy,x =  x,yΛy,xρ−1 ρ = 2id − ρ, 3 ρ(2id − ρ) = 1 3 ρ, 3 ρ−1ρ−1 ρ = 3ρ−1,  Hence, β(x) β(y) Λ−1 1 for x, y ∈ U1 or x, y ∈ U3, for x ∈ U1, y ∈ U3, for x ∈ U3, y ∈ U1, (cid:3) 6 ALBERTO ELDUQUE, NORIAKI KAMIYA, AND SUSUMU OKUBO Equation (2.11) gives ρ−1 ρ = ρ−1(3id − 2ρ) = 2id − ρ. Also (2.15) gives and (2.19) follows at once from Lemma 2.5(v). Lemma 2.7. Let (U, xyz) be a Kantor triple system with an involutive automor- phism σ. Define a new triple product by {xyz} = xσ(y)z. Then (U,{xyz}) is a Kantor triple system too, and σ is an automorphism of (U,{xyz}). Proof. Denote by L(x, y) : z 7→ {xyz}, and K(x, y) : z 7→ {xzy}−{yzx}, the L and K operators of the new triple system. Then L(x, y) = L(x, σ(y)) and K(x, y) = K(x, y)σ. For any u, v, x, y ∈ U we have: [ L(u, v), L(x, y)] = [L(u, σ(v)), L(x, σ(y))] = L(uσ(v)x, σ(y)) − L(x, σ(v)uσ(y)) = L({uvx}, y) − L(x,{vuy}), and K( K(u, v)x, y) = K(K(u, v)σ(x), y)σ = K(u, v)L(σ(x), y)σ + L(y, σ(x))K(u, v)σ = K(u, v)σL(x, σ(y)) + L(y, σ(x))K(u, v)σ = K(u, v) L(x, y) + L(y, x) K(u, v), because σL(x, y) = L(σ(x), σ(y))σ for any x, y and σ2 = id. Hence (U,{xyz}) is a Kantor triple system. The assertion on σ being an automorphism of (U,{xyz}) is clear. (cid:3) Lemma 2.8. Let (U, xyz) be a left unital Kantor triple system with left unit e, and let σ be the involutive automorphism σ = µ−1(3id − ρ) as in Proposition 2.6. Con- sider the new Kantor triple system (U,{xyz}) with {xyz} = xσ(y)z as in Lemma 2.7. Then {eex} = x and 2{xee} + {exe} = 3x for any x ∈ U . Conversely, let (U,{xyz}) be a Kantor triple system containing an element e such that {eex} = x and 2{xee} + {exe} = 3x for any x ∈ U , and endowed with an involutive automorphism σ such that σ(e) = e. Define a new triple product on U by xyz = {xσ(y)z}. Then (U, xyz) is a left unital Kantor triple system with left unit e, σ is an automorphism of (U, xyz), and σ = µ−1(3id− 2ρ), for ρ and µ given in (2.5). Proof. For the first part, note that {eex} = eex = x as σ(e) = e, and 2{xee} + {exe} = 2xee + eσ(x)e = (2ρ − µσ)(x) = (2ρ − (3id − 2ρ))(x) = 3x for any x ∈ U . For the converse, Lemma 2.7 shows that (U, xyz) is a Kantor triple system with involutive automorphism σ. Besides eex = {eex} = x, because σ(e) = e. Moreover, 3x = {xee} + {exe} = 2xee + eσ(x)e = (2ρ + µσ)(x) for any x ∈ U . Hence 2ρ + µσ = 3id and σ = µ−1(3id − 2ρ). (cid:3) We arrive to the main result of the paper: LEFT UNITAL KANTOR TRIPLE SYSTEMS AND STRUCTURABLE ALGEBRAS 7 Theorem 2.9. Let (U, xyz) be a left unital Kantor triple system with left unit e. Define a linear map and a (binary) multiplication on U by means of (¯x = 2x − xee, x · y = ¯xey − ¯xσ(¯y)e + yex, (2.20) for x, y ∈ U . Then (U,·,¯)) is a structurable algebra with unity e, σ = µ−1(3id− 2ρ) is an involutive automorphism of (U,·,¯) and the triple product on U is recovered as in equation (1.9): xyz = (x · σ(¯y)) · z + (z · σ(¯y)) · x − (z · ¯x) · σ(y), (2.21) for any x, y, z ∈ U . Conversely, if (A,·,¯) is a structurable algebra endowed with an involutive auto- morphism σ, and e = 1 is the unity of A, then (A, xyz) is a left unital Kantor triple system with left unit e, where xyz is defined by the formula in (2.21). Moreover, σ = µ−1(3id − 2ρ), and the involution and the multiplication in the structurable algebra are related to this triple product by equation (2.20). Proof. If (U, xyz) is a left unital Kantor triple system with left unit e, Lemma 2.8 shows that with σ = µ−1(3id − 2ρ) and {xyz} = xσ(y)z, (U,{xyz}) is a Kantor triple system with {eex} = x, 2{xee} + {exe} = 3x, and where σ is an involutive automorphism such that σ(e) = e. Now [F94, Lemma 1.7] (see also [KO10, Theorem 5.1]) proves that (U,·,¯) is a structurable algebra with unity e, involution ¯x = 2x−{xee} = 2x− xee and multiplication x· y = 1 2 ({xey} +{exy} +{xye}−{eyx}). Then [F94, (24 -- 28)] show that in this case the triple product {xyz} is recovered as {xyz} = (x · ¯y) · z + (z · ¯y) · x − (z · ¯x) · y. In particular so that {¯xey} = ¯x · y + y · ¯x − y · x, {¯x¯ye} = ¯x · y + y · ¯x − x · ¯y, {yex} = y · x + x · y − x · ¯y, ¯xey − ¯xσ(¯y)e + yex = {¯xey} − {¯x¯ye} + {yex} = x · y, for any x, y ∈ U , and xyz = {xσ(y)z} = (x · σ(¯y)) · z + (z · σ(¯y)) · x − (z · ¯x) · σ(y), for any x, y, z ∈ U . Conversely, if (A,·,¯) is a structurable algebra, [F94, Lemma 1.7] or [KO10, Theorem 5.1] show that with the triple product {xyz} = Vx,y(z) = (x · ¯y) · z + (z · ¯y) · x − (z · ¯x) · y for x, y, z ∈ A, (A,{xyz}) is a Kantor triple system where the unity e = 1 satisfies {eex} = x, 2{xee} + {exe} = 3x for any x ∈ A, and the involution and binary multiplication on A are recovered as follows: ¯x = 2x− {xee} and x·y = {¯xey}−{¯x¯ye}+{yex} as above. Now, if σ is an involutive automorphism of (A,·,¯), then Lemma 2.8 shows that (A, xyz) is a left unital Kantor triple system with left unit e, where xyz = {xσ(y)z}, and where σ = µ−1(3id − 2ρ). The result follows. (cid:3) Remark 2.10. Let (U, xyz) be a left unital Kantor triple system with left unit e, and let (U,·,¯) be the structurable algebra with involution and multiplication given by (2.20). Then the eigenspaces for ρ : x 7→ xee are precisely the subspaces of symmetric and skew-symmetric elements for the involution: U1 = {x ∈ U : xee = x} = {x ∈ U : ¯x = x}, U3 = {x ∈ U : xee = 3x} = {x ∈ U : ¯x = −x}. (cid:3) 8 ALBERTO ELDUQUE, NORIAKI KAMIYA, AND SUSUMU OKUBO 3. Left unital (−1,−1)-Freudenthal-Kantor triple systems In [YO84], Yamaguti and Ono considered a wide class of triple systems: the (ǫ, δ) Freudenthal-Kantor triple systems, which are useful tools in the construction of Lie algebras and superalgebras. An (ǫ, δ) Freudental-Kantor triple system (ǫ, δ are either 1 or −1) is a triple system (U, xyz) such that, if L(x, y), and K(x, y) are given by then L(x, y) : z 7→ xyz, K(x, y) : z 7→ xzy − δyzx, [L(u, v), L(x, y)] = L(uvx, y) + ǫL(x, vuy), (3.1a) (3.1b) K(cid:0)K(u, v)x, y(cid:1) = L(y, x)K(u, v) − ǫK(u, v)L(x, y), hold for any x, y, u, v ∈ U . Kantor triple systems are exactly the (−1, 1) Freudenthal-Kantor triple systems. Actually, for ǫ = −1, (3.1a) and (3.1b) coincide with (1.4) and (1.5) (or (2.1) and (2.3)), but K(x, y) is symmetric on x and y for δ = −1, and alternating for δ = 1. In particular, (−1,−1) Freudenthal-Kantor triple systems are generalized Jordan triple systems. Also, Freudenthal triple systems, symplectic triple systems and Faulkner ternary algebras are intimately related to (1, 1) Freudenthal-Kantor triple systems. (See, for instance, [E07, Theorem 4.7] and [E06, Theorem 2.18], and references therein, for the relationship between these triple systems.) With the same arguments as in Lemma 2.7 we have: Lemma 3.1. Let (U, xyz) be an (ǫ, δ) Freudenthal-Kantor triple system endowed with an automorphism σ such that σ2 = ±id. Define a new triple product by {xyz} = xσ(y)z. Then (U,{xyz}) is a (±ǫ, δ) Freudenthal-Kantor triple system, and σ is an automorphism of (U,{xyz}) too. Corollary 3.2. Let (U, xyz) be an (ǫ, δ) Freudenthal-Kantor triple system endowed with a bijective linear map σ : U → U satisfying: (cid:3) σ2 = ±µid, σ(x)σ(y)σ(z) = µσ(xyz), for any x, y, z ∈ U , where 0 6= µ ∈ F. Define a new triple product by {xyz} = xσ(y)z. Then (U,{xyz}) is a (±ǫ, δ) Freudenthal-Kantor triple system. Proof. By extending scalars if necessary, the map σ : x 7→ √µ−1σ(x) is an auto- morphism of (U, xyz) with σ2 = ±id and Lemma 3.1 applies. Example 3.3. Let (U, xyz) be an (ǫ, δ) Freudenthal-Kantor triple system. Con- sider the (ǫ, δ) Freudenthal-Kantor triple system defined on (cid:3) with componentwise multiplication. Then the map M2,1(U ) =(cid:26)(cid:18)x y(cid:19) : x, y ∈ U(cid:27) , σ :(cid:18)x y(cid:19) 7→(cid:18) y −x(cid:19) , is an automorphism of (M2,1(U ), XY Z) with σ2 = −id. Hence with {XY Z} = Xσ(Y )Z, for X, Y, Z ∈ M2,1(U ), M2,1(U ) becomes a (−ǫ, δ) Freudenthal-Kantor triple system. (cid:3) Proposition 3.4. Let (U, xyz) be an (ǫ, δ) Freudenthal-Kantor triple system, and let σ ∈ K(U, U ) (the linear span of the operators K(x, y) for x, y ∈ U ) satisfying σ2 = ǫδid. Then σ is an automorphism of (U, xyz). Therefore, with the new triple product defined by {xyz} = xσ(y)z, (U,{xyz}) is a (δ, δ) Freudenthal-Kantor triple system. LEFT UNITAL KANTOR TRIPLE SYSTEMS AND STRUCTURABLE ALGEBRAS 9 Proof. Equation (3.1b) proves (3.2) (3.3) (3.4) K(σ(x), y) = L(y, x)σ − ǫσL(x, y) for any x, y ∈ U . Therefore, we have In other words, the equation σ(x)zy − δyzσ(x) = yxσ(z) − ǫσ(xyz). σ(xyz) = −ǫσ(x)zy + ǫyxσ(z) + ǫδyzσ(x) holds for any x, y, z ∈ U . Apply σ to both sides of (3.3) to get which, using (3.3) on each summand on the right hand side, becomes: ǫδxyz = −ǫσ(σ(x)zy) + ǫσ(yxσ(z)) + ǫδσ(yzσ(x)), ǫδxyz = −ǫ(cid:16)−δxyz + ǫzσ(x)σ(y) + zyx(cid:17) + ǫ(cid:16)−ǫσ(y)σ(z)x + δxyz + ǫδxσ(z)σ(y)(cid:17) + ǫδ(cid:16)−ǫσ(y)σ(x)z + δzyx + ǫδzσ(x)σ(y)(cid:17) = 2ǫδxyz − σ(y)σ(z)x + δxσ(z)σ(y) − δσ(y)σ(x)z, and hence ǫδxyz = −δxσ(z)σ(y) + δσ(y)σ(x)z + σ(y)σ(z)x. With the substitutions x 7→ σ(x), y 7→ σ(y) and z 7→ σ(z), we obtain, σ(x)σ(y)σ(z) = ǫδ(cid:16)−δσ(x)zy + δyxσ(z) + yzσ(x)(cid:17) = −ǫσ(x)zy + ǫyxσ(z) + ǫδyzσ(x) = σ(xyz) (using (3.3)). This shows that σ is an automorphism of (U, xyz). The last assertion follows at once from Lemma 3.1. (cid:3) Corollary 3.5. Let (U, xyz) be an (ǫ, δ) Freudenthal-Kantor triple system, and let σ ∈ K(U, U ) (the linear span of the operators K(x, y) for x, y ∈ U ) satisfying σ2 = µid for a nonzero scalar µ ∈ F. Then σ(xyz) = ǫδµ−1σ(x)σ(y)σ(z) for any x, y, z ∈ U . Moreover, with the new triple product defined by {xyz} = xσ(y)z, (U,{xyz}) is a (δ, δ) Freudenthal-Kantor triple system. Proof. As in the proof of Corollary 3.2, extend scalars if necessary and consider the map σ : x 7→pǫδµ−1σ(x), which belongs to K(U, U ) and satisfies σ2 = ǫδid. The result follows now from Proposition 3.4. (cid:3) (cid:3) This Corollary allows us to give examples of (1, 1) Freudenthal-Kantor triple systems starting from structurable algebras: Example 3.6. Let (A,·,¯) be a structurable algebra, and assume there is an el- ement f ∈ A with ¯f = −f and 0 6= f ·2 ∈ F1. Write f ·2 = µ1. Note that this is always the case for the simple structurable algebras of skew-dimension one [AF84, Lemma 2.1(b)]. Consider the associated Kantor triple system (that is, (−1, 1) Freudenthal-Kantor triple system), with triple product as in (1.6). Then K(f, 1)x = {f x1} − {1xf} = Vf,x(1) − V1,x(f ) = (f − ¯f ) · x = 2f · x (see (1.3)). But (1.1) gives f · (f · x) = f ·2 · x = µx. Hence, with σ(x) = f · x, we are in the situation of Corollary 3.5, and therefore, with the new triple product given by {xyz}∼ = {x(f · y)z} = Vx,f ·y(z), A becomes a (1, 1) Freudenthal-Kantor triple system. (cid:3) 10 ALBERTO ELDUQUE, NORIAKI KAMIYA, AND SUSUMU OKUBO If (U, xyz) is a nontrivial (U 6= 0) left unital (ǫ, δ) Freudenthal-Kantor triple system, that is, there is an element e such that L(e, e) = id, then (3.1a) with u = v = e gives (1 + ǫ)L(x, y) = 0 for any x, y ∈ U . Therefore ǫ = −1. Hence only (−1, δ) Freudenthal-Kantor triple systems may be left unital. Most of the arguments in the previous sections work for (−1,−1) Freudenthal- Kantor triple systems, so Lemma 2.3 is valid for them. Lemma 2.4 has to be changed to the next result, whose proof is obtained following the same arguments step by step. Lemma 3.7. Let (U, xyz) be a left unital (−1,−1) Freudenthal-Kantor triple sys- tem with left unit e. Then, for any u, v ∈ U , the following conditions hold: K(u, e)e = (ρ + id)(u), (3.5) K(u, v) = K(K(u, v)e, e), 1 2 ρ2 = id. In particular, ρ and µ are invertible linear operators, and µ2 = id. (3.6) (3.7) (cid:3) We want to prove a result analogous to Lemma 2.8, which shows that we can modify slightly the triple product of a left unital Kantor triple system with the help of a suitable involutive automorphism, and get a new left unital Kantor triple system with stronger restrictions on the left unit. We could follow a path parallel to the one for left unital Kantor triple systems, but Proposition 3.4 allows a more direct approach. Theorem 3.8. Let (U, xyz) be a left unital (−1,−1) Freudenthal-Kantor triple system with left unit e. Then µ : x 7→ exe, is an involutive automorphism. Be- sides, consider the new (−1,−1) Freudenthal-Kantor triple system (U,{xyz}) with {xyz} = xµ(y)z. Then {eex} = x = {exe} for any x ∈ U . Conversely, let (U,{xyz}) be a (−1,−1) Freudenthal-Kantor triple system con- taining an element e such that {eex} = x = {exe} for any x ∈ U , and endowed with an involutive automorphism σ such that σ(e) = e. Define a new triple product on U by xyz = {xσ(y)z}. Then (U, xyz) is a left unital (−1,−1) Freudenthal- Kantor triple system with left unit e, σ is an automorphism of (U, xyz), and σ = µ : x 7→ exe. Proof. Since K(e, e)x = exe + exe = 2µ(x) for any x ∈ U , it follows that µ belongs to K(U, U ), and Lemma 3.7 shows that µ2 = id. Hence Proposition 3.4 shows that µ is an automorphism. Now the first part of the Theorem follows since {exe} = eµ(x)e = µ2(x) = x (Lemma 3.7). For the converse, just note that σ(x) = {eσ(x)e} = exe for any x ∈ U . (cid:3) Definition 3.9 ([EO11, Definition 3.3]). An (ǫ, δ) Freudenthal-Kantor triple sys- tem (U, xyz) is said to be special in case K(x, y) = ǫδL(y, x) − ǫL(x, y) (3.8) holds for any x, y ∈ U . K(U, U ) (the linear span of the endomorphisms K(x, y)). Moreover, (U, xyz) is said to be unitary in case the identity map belongs to If an (ǫ, δ) Freudenthal-Kantor triple system is unitary, then necessarily ǫ = δ, and the system is special (see [EO11, Proposition 3.4].) Corollary 3.10. Let (U, xyz) be a left unital (−1,−1) Freudenthal-Kantor triple system with left unit e. Consider the new (−1,−1) Freudenthal-Kantor triple system (U,{xyz}) with {xyz} = xµ(y)z. Then (U,{xyz}) is a unitary, and hence special, (−1,−1) Freudenthal-Kantor triple system. LEFT UNITAL KANTOR TRIPLE SYSTEMS AND STRUCTURABLE ALGEBRAS 11 Proof. Denote by L(x, y) and K(x, y) the L and K operators in (U,{x, yz}). Then K(e, e) : x 7→ 2{exe} = 2x, and hence K(e, e) = 2id and (U,{xyz}) is unitary. (cid:3) 4. Associated Lie algebras and superalgebras Let (U, xyz) be an (ǫ, δ) Freudenthal-Kantor triple system, then the space of 2 × 1 matrices over U : T = T(U, xyz) =(cid:26)(cid:18)x y(cid:19) : x, y ∈ U(cid:27) becomes a Lie triple system for δ = 1 and an anti-Lie triple system for δ = −1 (see [YO84, Section 3]) by means of the triple product: (4.1) (4.2) (4.3) b3(cid:19)(cid:21) b2(cid:19)(cid:18)a3 −ǫK(b1, b2) b1(cid:19)(cid:18)a2 (cid:20)(cid:18)a1 =(cid:18)L(a1, b2) − δL(a2, b1) L = span(cid:26)(cid:18) L(a, b) K(c, d) K(e, f ) and, therefore, the vector space δK(a1, a2) ǫL(b2, a1) − ǫδL(b1, a2)(cid:19)(cid:18)a3 b3(cid:19) ǫL(b, a)(cid:19) : a, b, c, d, e, f ∈ U(cid:27) is a Lie subalgebra of Mat2(cid:0)EndF(U )(cid:1)− (Given an associative algebra A, A− denotes the Lie algebra defined on A with product given by the usual Lie bracket [x, y] = xy − yx.) (for δ = −1) Hence we get either a Z2-graded Lie algebra (for δ = 1) or a Lie superalgebra . (4.4) g(U ) = g(U, xyz) = L ⊕ T where L is the even part and T the odd part. The bracket in g(U ) is given by: EndF(T)), • the given bracket in L as a subalgebra of Mat2(cid:0)EndF(U )(cid:1)− • [M, X] = M (X) for any M ∈ L and X ∈ T (note that Mat2(cid:0)EndF(U )(cid:1) ≃ • for any a1, a2, b1, b2 ∈ U : b1(cid:19) ,(cid:18)a2 (cid:20)(cid:18)a1 b2(cid:19)(cid:21) =(cid:18)L(a1, b2) − δL(a2, b1) ǫL(b2, a1) − ǫδL(b1, a2)(cid:19) . −ǫK(b1, b2) δK(a1, a2) , To simplify things, we will talk about the (anti-)Lie triple system T and the Lie (super)algebra g(U ), meaning that they are a Lie triple system and a Lie algebra for δ = 1 and an anti-Lie triple system and a Lie superalgebra for δ = −1. This (super)algebra g(U ) is Z-graded as follows: 0 ǫL(b, a)(cid:19) : a, b ∈ U(cid:27) , 0 g(U )(0) = span(cid:26)(cid:18)L(a, b) g(U )(1) =(cid:18)U 0(cid:19) , g(U )(−1) =(cid:18) 0 U(cid:19) , g(U )(2) = span(cid:26)(cid:18)0 K(a, b) g(U )(−2) = span(cid:26)(cid:18) 0 0 0 (cid:19) : a, b ∈ U(cid:27) , K(a, b) 0(cid:19) : a, b ∈ U(cid:27) , 0 so that g(U ) is 5-graded and L = g(U )(−2) ⊕ g(U )(0) ⊕ g(U )(2), T = g(U )(−1) ⊕ g(U )(1). 12 ALBERTO ELDUQUE, NORIAKI KAMIYA, AND SUSUMU OKUBO This Lie (super)algebra g(U ) is completely determined by the (anti-)Lie triple system T. In case (U, xyz) is the Kantor triple system defined on a structurable algebra (A,·,¯) by means of (1.6), then g(U ) coincides with the 5-graded Lie algebra K(A,¯) defined in [A79]. Proposition 4.1. Let σ be an involutive automorphism of an (ǫ, δ) Freudenthal- Kantor triple system. Then, with the new triple product defined by {xyz} = xσ(y)z for any x, y, z ∈ U , (U,{xyz}) is again an (ǫ, δ) Freudenthal-Kantor triple system and the associated Lie (super)algebras g(U, xyz) and g(U,{xyz}) are isomorphic (as 5-graded Lie (super)algebras). Proof. The proof of Lemma 2.7 applies here and shows that (U,{xyz}) is an (ǫ, δ) Freudenthal-Kantor triple system. Consider now the (anti-)Lie triple systems T = T(U, xyz) and T = T(U,{xyz}). Denote by [...] the triple product in T and by [...]∼ the one in T. Let Φ : T → T the linear isomorphism given by Φ(cid:18)a b(cid:19) =(cid:18) a σ(b)(cid:19) , b2(cid:19)(cid:18)a3 b3(cid:19)(cid:21)(cid:19) for any a, b ∈ U . Then, for any ai, bi ∈ U , i = 1, 2, 3, we have: b1(cid:19)(cid:18)a2 Φ(cid:18)(cid:20)(cid:18)a1 −ǫK(b1, b2)a3 + ǫL(b2, a1)b3 − ǫδL(b1, a2)b3(cid:19) = Φ(cid:18) L(a1, b2)a3 − δL(a2, b1)a3 + δK(a1, a2)b3 −ǫK(σ(b1), σ(b2))σ(a3) + ǫL(σ(b2), σ(a1))σ(b3) − ǫδL(σ(b1), σ(a2))σ(b3)(cid:19) =(cid:18) −ǫ K(σ(b1), σ(b2))a3 + ǫ L(σ(b2), a1)σ(b3) − ǫδ L(σ(b1), a2)σ(b3)(cid:19) =(cid:18) σ(b1)(cid:19)(cid:18) a2 =(cid:20)(cid:18) a1 Since g(U, xyz) and g(U,{xyz}) are determined by T and T, the result follows. (cid:3) Theorem 2.9 and Proposition 4.1 have the next direct consequence (see also L(a1, b2)a3 − δL(a2, b1)a3 + δK(a1, a2)b3 L(a1, σ(b2))a3 − δ L(a2, σ(b1))a3 + δ K(a1, a2)σ(b3) b3(cid:19)(cid:21)∼ b2(cid:19) Φ(cid:18)a3 b1(cid:19) Φ(cid:18)a2 σ(b2)(cid:19)(cid:18) a3 σ(b3)(cid:19)(cid:21)∼ =(cid:20)Φ(cid:18)a1 . [BS03]): Corollary 4.2. Let (U, xyz) be a left unital Kantor triple system. Then its Lie algebra g(U, xyz) is isomorphic, as a 5-graded Lie algebra, to the Lie algebra K(A,¯) defined in [A79] for a structurable algebra (A,·,¯). In particular, g(U, xyz) is a Lie algebra graded over the nonreduced root system BC1 and of type B1. Special (−1,−1) Freudenthal-Kantor triple systems give rise to strictly BC1- graded Lie superalgebras of type C1 (see [EO11, Corollary 4.6]). For left unital (−1,−1) Freudenthal-Kantor triple systems, we can strengthen this result, as we obtain superalgebras graded by a (Lie superalgebra) root system. The reader is referred to [BE03] for the results needed on Lie superalgebras graded over the root systems of the simple classical Lie superalgebras of type B(m, n). The B(0, 1)-graded Lie superalgebras are the Lie superalgebras that contain a subalgebra isomorphic to the orthosymplectic Lie superalgebra B(0, 1) = osp(1, 2), and such that, as a module for this subalgebra, they are a sum of copies of the adjoint module or the module obtained from it by interchanging the even and odd parts (the adjoint module with the parity changed), plus copies of the natural module or this module with the parity changed, plus a submodule with trivial action [BE03, Section 6]. In case the parity of any of the copies of the adjoint and LEFT UNITAL KANTOR TRIPLE SYSTEMS AND STRUCTURABLE ALGEBRAS 13 natural modules that appear in this decomposition is the natural one, we will say that the Lie superalgebra is a strictly B(0, 1)-graded Lie superalgebra. In this case, the coordinate superalgebra considered in [BE03] is actually an algebra (the odd part is zero). Corollary 4.3. Let (U, xyz) be a left unital (−1,−1) Freudenthal-Kantor triple system. Then its Lie superalgebra g(U, xyz) is a strictly B(0, 1)-graded Lie super- algebra. Proof. By Propositions 4.1 and 3.8, we may assume that there is an element e ∈ U such that eex = x = exe for any x ∈ U . Thus L(e, e) = id and K(e, e) = 2id. Then the subalgebra of the Lie superalgebra g(U, xyz) generated by the odd elements ( e 0 ) and ( 0 e ) is h = span(cid:26)(cid:18) 0 id 0(cid:19) ,(cid:18)0 e(cid:19) ,(cid:18)id 0 −id(cid:19) ,(cid:18)e 0 0(cid:19) ,(cid:18)0 0 0 id 0(cid:19)(cid:27) , which is isomorphic to the simple orthosymplectic Lie superalgebra B(0, 1) = osp(1, 2), whose even part is isomorphic to sl2(F) and its odd part is the two- dimensional natural irreducible module for its even part. Given any x ∈ U1 ∪ U−1, the h-submodule mx of g(U, xyz) generated by ( 0 x ) is the linear span of the elements 0 (cid:18) 0 K(e, x) 0(cid:19) ,(cid:18)0 x(cid:19) ,(cid:18)L(e, x) 0 0 −L(x, e)(cid:19) ,(cid:18)x 0(cid:19) ,(cid:18)0 K(e, x) 0 (cid:19) , 0 because L(e, x)e = x, L(x, e)e = ρ(x) = ±x and K(e, x)e = eex + xee = x + ρ(x), which is 0 for x ∈ U−1 and 2x for x ∈ U1. Equation (3.6) shows then than K(e, x) = 0 for x ∈ U−1. Therefore, if x ∈ U−1, then mx is the three-dimensional natural module for h ∼= osp(1, 2) (with the natural parity), while for x ∈ U1, this module is isomorphic to the adjoint module for h (again with the natural parity). Equation (3.6) gives K(U, U ) = K(U, e) = K(U1, e). Therefore, as a module for h, g(U, xyz) is the sum of the adjoint modules mx for x ∈ U1, the natural modules mx for x ∈ U−1, and the submodule with trivial action of h given by: 0 (cid:26)(cid:18)ϕ 0 −ϕ(cid:19) : ϕ ∈ L(U, U ), ϕ(e) = 0(cid:27) . This shows that g(U, xyz) is B(0, 1)-graded [BE03]. (cid:3) We finish the paper with a converse to Corollary 4.3. But first we need some preliminaries. Let (U, xyz) be an (ǫ, δ) Freudenthal-Kantor triple system, and consider the (anti-)Lie triple system T in (4.1) as well as the 5-graded Lie (super)algebra g(U ) in (4.4). The linear map Φ : T → T given by y(cid:19) 7→(cid:18) y −ǫδx(cid:19) Φ :(cid:18)x is easily checked to be an automorphism of the (anti-)Lie triple system T, such that Φ2 = −ǫδid, and hence it induces an automorphism of g(U ), also denoted by Φ, which satisfies Φ(cid:0)(g(U )(i)(cid:1) = g(U )(−i), for any i = 0,±1,±2, and Φ2 = id if ǫδ = −1, while Φ2 is the automorphism whose restriction to g(U )(i) is (−1)iid (For δ = −1, this is the grading automorphism of the Lie for i = 0,±1,±2. superalgebra g(U ).) 0 ) ∈ g(U )(1) with x ∈ U . Then We may identify g(U )(1) with U by identifying ( x the triple product on U is recovered as: xyz = [[x, Φ(y)], z] (4.5) 14 ALBERTO ELDUQUE, NORIAKI KAMIYA, AND SUSUMU OKUBO for any x, y, z ∈ U , where on the right hand side we use the Lie bracket in g(U ). Conversely, take ǫ, δ equal to 1 or −1 and let g be a 5-graded Lie algebra for δ = 1, or a consistently 5-graded Lie superalgebra for δ = −1 (this means that g¯0 = g−2 ⊕ g0 ⊕ g2 and g¯1 = g−1 ⊕ g1). Assume, moreover, that g is endowed with an automorphism Φ such that Φ(gi) = g−i for i = 0,±1,±2, and Φ2 = id if ǫδ = −1, while Φ2 is the automorphism whose restriction to gi is (−1)iid for i = 0,±1,±2. On U = g1 define the triple product by formula (4.5). Then the operators L(x, y) : z 7→ xyz and K(x, y) : z 7→ xzy − δyzx are given by: L(x, y)z = [[x, Φ(y)], z] = ad[x,Φ(y)](z), K(x, y)z = [[x, Φ(z)], y] − δ[[y, Φ(z)], x] = δ[x, [y, Φ(z)]] − [y, [x, Φ(z)]] = δ[[x, y], Φ(z)] = δ ad[x,y](Φ(z)). (4.6) (Note that ad[x,y] = adx ady −δ ady adx, because for δ = −1 the elements in g±1 are odd, while for δ = 1 this is clear.) Therefore, for u, v, x, y, z ∈ U we compute: [L(u, v),L(x, y)](z) = [ad[u,Φ(v)], ad[x,Φ(y)]](z) =(cid:16)ad[ad[u,Φ(v)](x),Φ(y)] + ad[x,[[u,Φ(v)],Φ(y)]](cid:17)(z) =(cid:16)ad[ad[u,Φ(v)](x),Φ(y)] −δ ad[x,Φ([[v,Φ−1(u)],y])](cid:17)(z) =(cid:16)ad[ad[u,Φ(v)](x),Φ(y)] +ǫ ad[x,Φ([[v,Φ(u)],y])](cid:17)(z) (as Φ2(u) = −ǫδu) = L(uvx, y)z + ǫL(x, vuy)z, and K(K(u, v)x, y)z = δK([[u, v], Φ(x)], y)z = [[[[u, v], Φ(x)], y], Φ(z)] = [[[u, v], [Φ(x), y]], Φ(z)], because [[u, v], y] ∈ [[g1, g1], g1] ⊆ [g2, g1] = 0, while L(y, x)K(u, v)z − ǫK(u, v)L(x, y)z = δ[[y, Φ(x)], [[u, v], Φ(z)]] − ǫδ[[u, v], Φ([[x, Φ(y)], z]) = −[[Φ(x), y], [[u, v], Φ(z)]] + [[u, v], [[Φ(x), y], Φ(z)]] = [[[u, v], [Φ(x), y]], Φ(z)]. Hence (3.1a) and (3.1b) are satisfied, and (U, xyz) is an (ǫ, δ) Freudenthal Kantor triple system. Example 4.4. Let g be a finite dimensional simple Lie algebra of rank l over an algebraically closed field F of characteristic zero. Fix a Cartan subalgebra h and denote by Σ the set of roots. Thus g = h ⊕(cid:0)⊕α∈Σgα(cid:1). Let ∆ be a system of simple roots, which splits Σ into positive and negative roots Σ = Σ+ ∪ Σ−, where Σ+ is the set of roots that are sums of simple roots and Σ− = −Σ+. Take a Chevalley basis {xα, α ∈ Σ; hi, 1 ≤ i ≤ l} (see [H72, §25]). In particular, [xα, x−α] = hα with α(hα) = 2 for any α ∈ Σ. Consider the order two automorphism Φ determined by Φ(xα) = −x−α for any α ∈ Σ (so Φh = −id). Let LEFT UNITAL KANTOR TRIPLE SYSTEMS AND STRUCTURABLE ALGEBRAS 15 ρ be the highest root. Then g is 5-graded: g = g−2 ⊕ g−1 ⊕ g0 ⊕ g1 ⊕ g2, with g±2 = Fx±ρ, g±1 = ⊕{gα : α ∈ Σ±, α 6= ±ρ, (ρα) 6= 0}, g0 = h ⊕(cid:0)⊕{gα : α ∈ Σ, (ρα) = 0}(cid:1), where (..) is the bilinear form induced by the Killing form. ((−1, 1) Freudenthal-Kantor triple system), by means of Equation (4.5) endows U = g1 with the structure of a Kantor triple system uvw = [[u, Φ(v)], w]. Moreover, since dim g2 = 1, [u, v] = huvixρ, for a skew-symmetric bilinear form h..i. Now (4.6) gives: K(u, v)w = [[u, v], Φ(w)] = huvi[xρ, Φ(w)]. Define σ ∈ EndF(U ) by σ(z) = [xρ, Φ(z)], so we get K(u, v) = huviσ for any u, v ∈ U = g1. Especially σ ∈ K(U, U ). Besides, for any z ∈ U : σ2(z) = [xρ, Φ(σ(z))] = [xρ, Φ([xρ, Φ(z)])] = [xρ, [Φ(xρ), Φ2(z)]] = [xρ, [−x−ρ, z]] = −[[xρ, x−ρ], z] = −[hρ, z] = −z, (because [xρ, z] ∈ [g2, g1] = 0) where we have used that α(hρ) = 1 for any α ∈ Σ+ with (ρα) 6= 0. Therefore, we have σ2 = −id, and Corollary 3.5 shows that σ is an automorphism of our Kantor triple system (U, uvw). Moreover, with the new triple product {uvw} = uσ(v)w, (U,{uvw}) is a (1, 1) Freudenthal-Kantor triple system. Besides, if we denote by L∗ and K ∗ the L and K operators for this new triple system, K ∗(u, v) = K(u, v)σ = huviσ2 = −huviid, so this system is balanced (its K operators are given by a bilinear form). Finally, with the triple product (uvw) = {uvw} − 1 2huviw + 1 2huwiv + 1 2hvwiu, U becomes a Freudenthal triple system (see [M68]). (cid:3) The orthosymplectic Lie superalgebra B(0, 1) = osp(1, 2) is the subalgebra of the general linear Lie superalgebra gl(1, 2) given by b =    0 µ −ν α µ ν β γ −α H =  A natural basis of b consists of the elements 0 0 0 0 0 0 0 0 0 0 1 0 0 0 −1 X =    , E =    , 0 −1 0 0 0 0 0 1 0 .  : α, β, γ, µ, ν ∈ F     , F =   Y =  ,  0 1 0 0 1 0 0 0 0 0 1 0 0 0 0 0 0 1 0 0 0   , where {H, E, F} form a basis of the even part, which is isomorphic to sl2(F), and {X, Y } of the odd part. 16 ALBERTO ELDUQUE, NORIAKI KAMIYA, AND SUSUMU OKUBO Its natural three-dimensional module is isomorphic to the following subspace of gl(1, 2): A natural basis of this module consists of the following elements:   s =    , 0 0 1 0 0 1 . 2α µ ν α 0 ν −µ 0 α X =  0 0 1 0 0 0  : α, µ, ν ∈ F    Y =  ,  1 0 0 2 0 0 H =  0 −1 0 0 0 1 0 0 0   , where H is even and X and Y are odd. The automorphism Φ of gl(1, 2) given by conjugation by the matrix   1 0 0 0 0 −1 0 0 1   leaves invariant both b and h. Its square is the grading automorphism of gl(1, 2). That is, it equals id on the even part and −id on the odd part. A simple computation shows: (4.7) Φ( X) = Y . Φ(X) = Y, If g is a B(0, 1)-graded Lie superalgebra, then it contains a subalgebra isomorphic to b, and [BE03, Theorem 6.20] shows that, up to isomorphism, g decomposes, as a module for b, as g =(cid:0)b ⊗ A(cid:1) ⊕(cid:0)s ⊗ B(cid:1) ⊕ D, for suitable vector superspaces A and B. Besides, the subalgebra isomorphic to b is identified with b ⊗ 1, for a distinguished even element 1 ∈ A, and D is the centralizer in g of this subalgebra. The action of H provides a 5-grading of g, with gi = {Z ∈ g : [H, Z] = iZ}. Hence we have: g−2 = F ⊗ A, g−1 = (Y ⊗ A) ⊕ ( Y ⊗ B), g0 = (H ⊗ A) ⊕ ( H ⊗ B) ⊕ D, g1 = (X ⊗ A) ⊕ ( X ⊗ B), g2 = E ⊗ A. (4.8) The Lie superalgebra g is strictly B(0, 1)-graded if and only if the superspaces A and B have trivial odd part, so they are standard vector spaces. The automorphism Φ of gl(1, 2) above extends to an automorphism of g, which acts just on b and h, and which will be denoted by Φ too. Therefore, if g is strictly B(0, 1)-graded, the vector space U = g1 is a (−1,−1) Freudenthal-Kantor triple system with the triple product given by xyz = [[x, Φ(y)], z] Now the promised converse of Corollary 4.3 is easy: Theorem 4.5. Let g be a strictly B(0, 1)-graded Lie superalgebra. Consider the 5-grading given by the action of the element H above. Then the element e = X ⊗ 1 is a left unit of the (−1,−1) Freudenthal-Kantor triple system (U = g1, xyz) which satisfies exe = x for any x ∈ U . Proof. We must prove that eex = exe = x for any x ∈ g1 = (X ⊗ A) ⊕ ( X ⊗ B). This amounts to prove that L(e, e) = id and K(e, e) = 2id. But (4.6) shows that L(e, e)x = ad[e,Φ(e)](x) and K(e, e)x = − ad[e,e](Φ(x)). Note that Φ(X) = Y by (4.7), so [e, Φ(e)] = [X, Y ] ⊗ 1 = H ⊗ 1, which acts as the LEFT UNITAL KANTOR TRIPLE SYSTEMS AND STRUCTURABLE ALGEBRAS 17 identity on g1, as the 5-grading is given precisely by the eigenspace decomposition relative to H ⊗ 1. Hence L(e, e) = id. On the other hand, [X, X] = 2X 2 = −2E, so [e, e] = −2E ⊗ 1, and hence K(e, e)(X ⊗ a) = 2[E ⊗ 1, Φ(X) ⊗ a] = 2[E ⊗ 1, Y ⊗ a] = 2[E, Y ] ⊗ a = 2X ⊗ a, while K(e, e)( X ⊗ b) = 2[E ⊗ 1, Φ( X)⊗ b] = 2[E ⊗ 1, Y ⊗ b] = 2[E, Y ]⊗ b = 2 X ⊗ b. Hence K(e, e) = 2id. (cid:3) References [A78] [A79] B.N. Allison, A class of nonassociative algebras with involution containing the class of Jordan algebras, Math. Ann. 237 (1978), no. 2, 133 -- 156. B.N. Allison, Models of isotropic simple Lie algebras, Commun. Algebra 7 (1979), no. 17, 1835 -- 1875. [AF84] B.N. Allison and J.R. Faulkner, A Cayley-Dickson process for a class of structurable algebras, Trans. Amer. Math. Soc. 283 (1984), no. 1, 185 -- 210. [AF93] B.N. Allison and J.R. Faulkner, Nonassociative coefficient algebras for Steinberg unitary Lie algebras, J. Algebra 161 (1993), no. 1, 1 -- 19. [BE03] G. Benkart and A. Elduque, Lie superalgebras graded by the root system B(m, n), Selecta Math. (N.S.) 9 (2003), no. 3, 313 -- 360. [BS03] G. Benkart and O. Smirnov, Lie Algebras Graded by the Root System BC1, J. Lie Theory [E06] [E07] 13 (2003), no. 1, 91 -- 132. A. Elduque, New simple Lie superalgebras in characteristic 3, J. Algebra 296 (2006), no. 1, 196 -- 233. A. Elduque, The magic square and symmetric compositions. II, Rev. Mat. Iberoam. 23 (2007), no. 1, 57 -- 84. [EKO05] A. Elduque, N. Kamiya and S. Okubo, (−1, −1)-balanced Freudenthal Kantor triple systems and noncommutative Jordan algebras, J. Algebra 294 (2005), no. 1, 19 -- 40. [EO11] A. Elduque and S. Okubo, Special Freudenthal-Kantor triple systems and Lie algebras [F94] [H72] with dicyclic symmetry, Proc. Royal Soc. Edinburgh 141A (2011), 1225 -- 1262. J.R. Faulkner, Structurable triples, Lie triples, and symmetric spaces, Forum Math. 6 (1994), no. 5, 637 -- 650. J.E. Humphreys, Introduction to Lie algebras and representation theory, Graduate Texts in Mathematics, Vol. 9, Springer-Verlag, New York, 1972. [KO10] N. Kamiya and S. Okubo, Representations of (α, β, γ) triple system, Linear Multilinear [K72] [K73] [M68] Algebra 58 (2010), no. 5-6, 617 -- 643. I.L. Kantor, Certain generalizations of Jordan algebras (Russian), Trudy Sem. Vektor. Tenzor. Anal. 16 (1972), 407 -- 499. I.L. Kantor, Models of the exceptional Lie algebras, Soviet Math. Dokl. 14 (1973), 254 -- 258. K. Meyberg, Eine Theorie der Freudenthalschen Tripelsysteme. I, II, Nederl. Akad. Wetensch. Proc. Ser. A 71=Indag. Math. 30 (1968), 162 -- 174, 175 -- 190. [YO84] K. Yamaguti and A. Ono, On representations of Freudenthal-Kantor triple systems U (ǫ, δ), Bull. Fac. School Ed. Hiroshima Univ., Part II, 7 (1984), 43 -- 51. Departamento de Matem´aticas e Instituto Universitario de Matem´aticas y Aplica- ciones, Universidad de Zaragoza, 50009 Zaragoza, Spain E-mail address: [email protected] Department of Mathematics, University of Aizu, 965-8580, Aizuwakamatsu, Japan E-mail address: [email protected] Department of Physics and Astronomy, University of Rochester, Rochester, NY 14627, USA E-mail address: [email protected]
0910.5181	2	0910	2011-03-01T17:59:27	Homomorphisms on infinite direct product algebras, especially Lie algebras	[ "math.RA" ]	We study surjective homomorphisms f:\prod_I A_i\to B of not-necessarily-associative algebras over a commutative ring k, for I a generally infinite set; especially when k is a field and B is countable-dimensional over k. Our results have the following consequences when k is an infinite field, the algebras are Lie algebras, and B is finite-dimensional: If all the Lie algebras A_i are solvable, then so is B. If all the Lie algebras A_i are nilpotent, then so is B. If k is not of characteristic 2 or 3, and all the Lie algebras A_i are finite-dimensional and are direct products of simple algebras, then (i) so is B, (ii) f splits, and (iii) under a weak cardinality bound on I, f is continuous in the pro-discrete topology. A key fact used in getting (i)-(iii) is that over any such field, every finite-dimensional simple Lie algebra L can be written L=[x_1,L]+[x_2,L] for some x_1, x_2\in L, which we prove from a recent result of J.M.Bois. The general technique of the paper involves studying conditions under which a homomorphism on \prod_I A_i must factor through the direct product of finitely many ultraproducts of the A_i. Several examples are given, and open questions noted.	math.RA	math	Comments, corrections, and related references welcomed! TEXed October 25, 2018 HOMOMORPHISMS ON INFINITE DIRECT PRODUCT ALGEBRAS, ESPECIALLY LIE ALGEBRAS GEORGE M. BERGMAN AND NAZIH NAHLUS Abstract. We study surjective homomorphisms f : QI Ai → B of not-necessarily-associative algebras over a commutative ring k, for I a generally infinite set; especially when k is a field and B is countable- dimensional over k. Our results have the following consequences when k is an infinite field, the algebras are Lie algebras, and B is finite-dimensional: If all the Lie algebras Ai are solvable, then so is B. If all the Lie algebras Ai are nilpotent, then so is B. If k is not of characteristic 2 or 3, and all the Lie algebras Ai are finite-dimensional and are direct products of simple algebras, then (i) so is B, (ii) f splits, and (iii) under a weak cardinality bound on I, f is continuous in the pro-discrete topology. A key fact used in getting (i)-(iii) is that over any such field, every finite-dimensional simple Lie algebra L can be written L = [x1, L] + [x2, L] for some x1, x2 ∈ L, which we prove from a recent result of J. M. Bois. The general technique of the paper involves studying conditions under which a homomorphism on QI Ai must factor through the direct product of finitely many ultraproducts of the Ai. Several examples are given, and open questions noted. 1. Introduction. In this note, an algebra over a commutative associative unital ring k means a k-module A given with a k-bilinear multiplication A × A → A, which we do not assume associative or unital. We shall assume k fixed, and "algebra" will mean "k-algebra" unless another base ring is specified. "Countable" will be used in the broad sense, which includes "finite". "Direct product" will be used in the sense sometimes called "complete direct product". Let us sketch our method of approach in a somewhat simpler case than we will eventually be considering. It is easy to show that if an algebra B is not a nontrivial direct product, and has no nonzero elements annihilating all of B, then any surjective homomorphism f : A1 × A2 → B from a direct product of two algebras onto B must factor through the projection onto one of A1 or A2. It follows that for such B, a homomorphism f : QI Ai → B from an arbitrary direct product onto B will factor through an ultraproduct QI Ai / U. For this to be useful, we need to know something about such ultraproducts. Assume k a field. There are three cases: factors through the projection to that algebra. First, U may be a principal ultrafilter. Then QI Ai / U can be identified with one of the Ai, and f Second, U may be a nonprincipal ultrafilter that is not card(k)+-complete. Then QI Ai / U is an algebra over the ultrapower K = kI / U, and for such U, K is an uncountable-dimensional extension field of k (Theorem 46). We shall see (Proposition 9) that if we map a K-algebra A onto a k-algebra B having nonzero multiplication, uncountable dimensionality of K forces B to be uncountable-dimensional over k as well. Hence, if we restrict attention to maps onto countable-dimensional B, this case does not occur. Finally, U may be a nonprincipal but card(k)+-complete ultrafilter. If k is finite, card(k)+-completeness is vacuous (implicit in the definition of an ultrafilter), and we cannot prove much in that case; we mainly 2000 Mathematics Subject Classification. Primary: 17A60, 17B05. Secondary: 03C20, 03E55, 08B25, 17B20, 17B30. Key words and phrases. homomorphic images of infinite direct products of nonassociative algebras; simple, solvable, and nilpotent Lie algebras; maps that factor through ultraproducts; measurable cardinals. After publication of this note, updates, errata, related references etc., if found, will be recorded at http://math.berkeley.edu/~gbergman/papers/. 1 2 GEORGE M. BERGMAN AND NAZIH NAHLUS note open questions. If k is infinite, on the other hand, the card(k)+-complete case "almost" does not occur: For such an ultrafilter to exist, the index set I must be of cardinality at least an uncountable measurable cardinal, and it is known that if such cardinals exist, they must be extremely large ("inaccessible", and more), and that the nonexistence of such cardinals is consistent with ZFC, the standard axiom system of set theory. Nevertheless, if U is such an ultrafilter, the behavior of QI Ai / U is almost as good as when U is principal. This case can be subdivided in two: If the dimensions of the Ai as k-algebras are not themselves extremely large cardinals, then that ultraproduct will again be isomorphic (though not by a projection) to one of the Ai (Theorem 47), and so will inherit all properties assumed for these. Without any restriction on the dimensions of the Ai, the ultraproduct will still satisfy many important properties that hold on the Ai, e.g., simplicity, nilpotence, or (in the Lie case) solvability (Propositions 48 and 49), again allowing us to get strong conclusions about the image B of f. Fortunately, the proofs of our main results do not require separate consideration of all these cases, but mainly the distinction between the card(k)+-complete case (which includes the principal case), and the non-card(k)+-complete case (which, as indicated, will be ruled out under appropriate hypotheses). The above sketch assumed that B was not a nontrivial direct product and had no nonzero elements annihilating all of B. We use these hypotheses in the early sections of the paper, but introduce in §6 a weaker hypothesis on B ("chain condition on almost direct factors") yielding more general statements. In §11, we give some tangential results on concepts introduced in earlier sections. Some examples showing the need for various of the hypotheses in our results are collected in §12. In §13, we note some open questions and directions for further investigation. Standard definitions and facts about ultrafilters, ultraproducts, and ultrapowers, assumed from §3 on, are reviewed in an appendix, §14. The more exotic topics of κ-complete ultrafilters and uncountable measurable cardinals are presented in another appendix, §15, and used from §5 on. The results proved here about Lie algebras were conjectured several years ago by the second author (under the assumption that the Ai were finite-dimensional, and the base field k algebraically closed of characteristic 0). The authors are indebted to Jean-Marie Bois, Siemion Fajtlowicz, Karl H. Hofmann, Otto Kegel, A. W. Knapp, Kamal Makdisi, Donald Passman, Alexander Premet, and the referree for many helpful com- ments and pointers related to this material. In [6] we obtain results similar to those of this note -- though stronger in some ways and weaker in others, and by a rather different approach. 2. Some preliminaries. Our statements in the above sketch, on when a map f : A1 × A2 → B must factor through A1 or A2 on the one hand, and on when a surjective homomorphism A → B of k-algebras, such that A admits a structure of K-algebra for a "large" extension field K of k, forces B to be large, on the other, both had hypotheses restricting the amount of "zero multiplication" in the structure of B. To see why such limitations are needed, note that if k is a field, and the Ai are k-algebras whose multiplications are the zero map, and B is, say, the one-dimensional zero-multiplication k-algebra, then homomorphisms QI Ai → B are arbitrary linear functionals on QI Ai, and these need not satisfy the conclusions of either statement. Homomorphisms based on zero multiplication are, from our point of view, very unruly, and we shall "work around" that phenomenon in various ways throughout this paper. To refer conveniently to that phenomenon, let us make Definition 1. If A is an algebra, we define its total annihilator ideal to be (1) Z(A) = {x ∈ A xA = Ax = {0}}. When A is a Lie algebra, Z(A) is thus the center of A, and our notation agrees with standard notation for the center. (But when A is an associative algebra, this notation conflicts with the common notation for the center of A.) Let us also make explicit that the definition of simple algebra excludes zero multiplication: Definition 2. An algebra A will be called simple if it is nonzero, has nonzero multiplication, and has no proper nonzero homomorphic images. HOMOMORPHISMS ON INFINITE DIRECT PRODUCT ALGEBRAS, ESPECIALLY LIE ALGEBRAS 3 A simple algebra A must be idempotent, AA = A, and have zero total annihilator, Z(A) = {0}, since AA and Z(A) are always ideals. By AA, above, we of course mean the set of sums of products of pairs of elements of A. More generally, Definition 3. For any k-submodules A′, A′′ of an algebra A, we will denote by A′A′′ the k-submodule of A consisting of all sums of products a′a′′ (a′ ∈ A′, a′′ ∈ A′′). Below, we will always write AA rather than A2, to avoid confusion with A × A. The next lemma shows that the total annihilator ideal leads to a way that algebra homomorphisms can be "perturbed", which we will have to take account of in many of our results. (In this lemma, we explic- itly write "k-algebra homomorphism" because a k-module homomorphism is also mentioned. Elsewhere, "homomorphism" will be understood to mean k-algebra homomorphism unless the contrary is stated.) Lemma 4. Let A and B be k-algebras, f : A → B a k-algebra homomorphism, and h : A → Z(B) a k-module homomorphism. Then the following conditions are equivalent: (i) f + h is a k-algebra homomorphism. (ii) AA ⊆ ker(h). (iii) h is a k-algebra homomorphism. Proof. We will show (i) ⇐⇒ (ii). The equivalence (iii) ⇐⇒ (ii) will follow as the f = 0 case of that result. Since f + h is k-linear, (i) will hold if and only if f + h respects multiplication, i.e., if and only if for all a, a′ ∈ A we have f (aa′) + h(aa′) = (f (a) + h(a))(f (a′) + h(a′)). Subtracting from this the equation f (aa′) = f (a)f (a′), and noting that all the terms remaining on the right are two-fold products in which at least one factor is a value of h, and hence lies in Z(B), making those products 0, we see that (i) is equivalent to h(aa′) = 0, i.e., (ii). (cid:3) Here, in somewhat sharpened form, is the fact stated in the second paragraph of §1. Lemma 5. Let B be a nonzero algebra. Then the following conditions are equivalent: (i) Every surjective homomorphism f : A1 × A2 → B from a direct product of two algebras onto B factors through the projection of A1 × A2 onto A1, or through the projection onto A2. (ii) B is not a sum B1 + B2 of two nonzero mutually annihilating subalgebras, i.e., nonzero subalgebras B1, B2 such that B1B2 = B2B1 = {0}. (iii) Z(B) = {0}, and B is not a direct product of two nonzero subalgebras. In particular, these conditions hold when B is a simple algebra. Proof. We shall show that ¬(i) ⇐⇒ ¬(ii) ⇐⇒ ¬(iii). If a surjective homomorphism f : A1 × A2 → B does not factor through either projection map, then B1 = f (A1) and B2 = f (A2) are both nonzero, and so give a counterexample to (ii). Conversely, given B1 and B2 as in ¬(ii), the map B1 × B2 → B given by the sum of the inclusions will be an algebra homomorphism that establishes ¬(i). For B1 and B2 as in ¬(ii) we get ¬(iii) by noting that if they have nonzero intersection, that intersection is a nonzero submodule of Z(B), while if they have zero intersection, then B ∼= B1 × B2 as algebras. Finally, in the situation of ¬(iii), if Z(B) 6= {0} then the equation B = B + Z(B) yields ¬(ii), while if B is a direct product of nonzero subalgebras, then those subalgebras give the required B1 and B2. The final assertion holds because a simple algebra satisfies (iii). (cid:3) The next lemma strengthens the above result a bit, so as to give interesting information in the case where Z(B) 6= {0} as well. We shall not use the result in this form, but it will eventually motivate the transition to the approach of §6. We remark that the implication (iii′) =⇒ (ii′) below is not a trivial consequence of (iii) =⇒ (ii) above, because dividing an algebra B by its total annihilator ideal Z(B) does not in general produce an algebra with zero total annihilator ideal, to which we could apply the latter result. Lemma 6. Let B be a nonzero algebra. Then the conditions (i′) and (ii′) below are equivalent. (i′) If a homomorphism f : A1 × A2 → B, when composed with the natural map B → B/Z(B), gives a surjection, then that composite map factors through the projection of A1 × A2 onto A1, or onto A2. (ii′) B is not equal to the sum B1 + B2 of two mutually annihilating subalgebras neither of which is contained in Z(B). 4 GEORGE M. BERGMAN AND NAZIH NAHLUS Moreover, the above conditions are implied by (iii′) B/Z(B) is not a direct product of two nonzero subalgebras. Sketch of proof. The surjectivity condition in the hypothesis of (i′) says that f (A1) + f (A2) + Z(B) = B. With this in mind, we can get the equivalence of (i′) and (ii′) as in the preceding result: given a counterexample to (i′), the subalgebras B1 = f (A1) + Z(B) and B2 = f (A2) + Z(B) give a counterexample to (ii′), while given a counterexample to (ii′), the induced map B1 × B2 → B is a counterexample to (i′). We complete the proof by showing that ¬(ii′) =⇒ ¬(iii′). Given B1 and B2 contradicting (ii′), enlarge them to B1 + Z(B) and B2 + Z(B) if necessary, so that they each contain Z(B); this clearly preserves the conditions assumed. It is now easy to show that their images in B/Z(B) contradict (iii′): Those images are subalgebras which sum to the whole algebra and annihilate one another on both sides, so it suffices to show that they have zero intersection. Since B1 and B2 both contain Z(B), an element of the intersection of their images in B/Z(B) will arise from an element x ∈ B1 ∩ B2. But such an element will annihilate both B2 and B1, hence will annihilate their sum, B, i.e., it will lie in Z(B), so the element of B/Z(B) that it yields is indeed zero. (cid:3) The above implication (iii′) =⇒ (ii′) is not reversible. For example, let B be the associative or Lie algebra spanned by the matrix units e12, e13, e23 within the algebra M3(k) of 3 × 3 matrices over a field k. (As a Lie algebra, B is sometimes called the Heisenberg algebra.) We find that Z(B) = k e13, and that B/Z(B) is a 2-dimensional k-vector space with zero multiplication, hence is the direct product of any two one-dimensional subspaces; so (iii′) fails. But we claim that (ii′) holds. Indeed, if B = B1 + B2, and neither summand is contained in Z(B), then we can find b1 ∈ B1 and b2 ∈ B2 which are linearly independent modulo Z(B). It is then not hard to show that in [b1, b2] = b1b2 − b2b1, the coefficient of e13 is given by a determinant of the coefficients of e12 and e23 in those two elements, and hence is nonzero; so B1 and B2 do not annihilate one another, either as Lie or as associative algebras. By Lemma 6, condition (i′) also holds for this example. Now a homomorphism from a direct product algebra A1 × A2 onto the zero-multiplication algebra B/Z(B) can fail to factor through the projection onto A1 or A2 (e.g., when the latter are each the one-dimensional zero-multiplication algebra). Yet condition (i′) shows that if such a homomorphism arises as a composite A1 × A2 → B → B/Z(B), it must so factor. What this example shows us is that though Z(B) is "trivial", in that its elements have zero multiplication with everything, it cannot be ignored in studying the multiplicative structure of B and the properties of homomorphisms onto B, because elements outside it can have nonzero product lying in it. 3. Factoring homomorphisms through ultraproducts. Suppose an algebra B satisfies the equivalent conditions of Lemma 5, and we map an infinite direct product QI Ai onto B. Then, since for every subset J ⊆ I we have QI Ai ∼= (QJ Ai) × (QI−J Ai), Lemma 5 gives us a vast family of factorizations of our homomorphism. How these fit together is described (in a general set-theoretic setting) in the next lemma. In stating it, we assume acquaintance with the concepts of filter, ultrafilter, reduced product and ultraproduct, summarized in §14. map, whose image has more than one element. Then the following conditions are equivalent: Lemma 7. Suppose (Ai)i∈I is a family of nonempty sets, B is a set, and f : A = QI Ai → B is a set (a) For every subset J ⊆ I, the map f factors either through the projection A → Qi∈J Ai, or through (b) The map f factors through the natural map A → A / U, where U is an ultrafilter on the index set the projection A → Qi∈I−J Ai. I, and A / U = QI Ai / U denotes the ultraproduct of the Ai with respect to this ultrafilter. When this holds, the ultrafilter U is uniquely determined by f. Proof. Not yet assuming either (a) or (b), but only the initial hypothesis, let (2) F = {J ⊆ I f factors through the projection A → Qi∈J Ai}. Thus, a subset J ⊆ I belongs to F if and only if for all a = (ai) ∈ A, f (a) is unchanged on making arbitrary changes in the coordinates of a indexed by the elements of the complementary set I − J. Now if the value of f (a) is unchanged on changing coordinates lying in a given subset, it is unchanged on changing coordinates in any smaller subset; and if is unchanged on changing coordinates in each of two subsets, then it is unchanged on changing coordinates in the union of those two sets. Translating these observations into HOMOMORPHISMS ON INFINITE DIRECT PRODUCT ALGEBRAS, ESPECIALLY LIE ALGEBRAS 5 statements about the family F of complements of sets with that property, we see that F is closed under intersections and enlargement, i.e., F is a filter on I. Looking at the definition of the reduced product of a family of sets with respect to a filter on the index set, we see that F is the largest filter such that f factors through the natural map of A to the reduced product A / F . The fact that the image of f has more than one element shows that the value of f (a) is not unchanged under arbitrary modification of all coordinates of a; so F does not contain the empty set, i.e., it is a proper filter. Finally, we note that condition (a) is equivalent to saying that for each J ⊆ I, either J or its complement lies in F ; i.e., that F is an ultrafilter, which we rename U. The equivalence of (a) and (b), and the final assertion, now follow. (cid:3) Combining the above with Lemma 5, we get Proposition 8. The equivalent conditions (i)-(iii) of Lemma 5 on a nonzero algebra B are also equivalent to: (iv) Every surjective homomorphism f : QI Ai → B from an arbitrary direct product of algebras to B factors through the natural map of that product onto the ultraproduct QI Ai / U, for some ultrafilter U on I. When this holds, the ultrafilter U is uniquely determined by f. Proof. Since (i) is the I = {1, 2} case of (iv), we have (iv) =⇒ (i). Conversely, if B satisfies (i), and we have a homomorphism f : A = QI Ai → B, then for every J ⊆ I we can apply (i) to the decomposition A = (QJ Ai) × (QI−J Ai), and conclude that f factors through the projection of A to one of these subproducts. Lemma 7 now yields (iv), and the final assertion. (cid:3) We now come to the other tool referred to in §1. 4. Extending algebra structures. Proposition 9. Suppose ϕ : k → K is a homomorphism of commutative rings, A is a K-algebra, B is a k-algebra, and f : A → B is a surjective homomorphism as k-algebras (under the k-algebra structure on A induced by its K-algebra structure). Then the kernel of the composite map A → B → B/Z(B) is an ideal of A, not only as a k-algebra, but as a K-algebra. Hence B/Z(B) acquires a K-algebra structure (unique for the property of making that composite map a homomorphism of K-algebras). Proof. It will suffice to show that for a ∈ A and c ∈ K, if f (a) ∈ Z(B), then f (c a) ∈ Z(B). So we must show that f (c a) annihilates on both sides an arbitrary element of B, which by surjectivity of f we can write f (a′) (a′ ∈ A). To do this, we compute: f (c a)f (a′) = f (c a a′) = f (a)f (c a′) = 0, the last step by the assumption that f (a) ∈ Z(B). The same calculation works for the product in the opposite order, completing the proof. (cid:3) Remark: If A were unital, then any ring-theoretic ideal of A, being closed under multiplication by K · 1, would be a K-algebra ideal. In that situation, moreover, Z(B) would be trivial, since no nonzero element of B could be annihilated by f (1) on either side. The above result shows that somehow, lacking unitality of A, we can make up for it at the other end by dividing out by Z(B). 5. First results on homomorphic images of infinite direct product algebras. In this section we will use the above tools to obtain a couple of results on homomorphic images of direct products of Lie and other algebras, under the assumption that card(I) is less than any measurable cardinal > card(k). (That condition holds vacuously, of course, if no such measurable cardinals exist.) We assume from here on the material of the final appendix, §15, on measurable cardinals, and κ+-complete and non- κ+-complete ultrafilters. In the first result below, we restrict the field k so that we can make use of a theorem of G. Brown [11] or its variant in Bourbaki [10, Ch.VIII, §11, Exercise 13(b)], from either of which it follows that in a finite- dimensional simple Lie algebra over an algebraically closed field of characteristic 0, every element is a bracket (not merely a sum of brackets, as it must be in any simple Lie algebra). In §9 we will say more precisely what Brown and Bourbaki prove, then bring in a recent result of J. M. Bois which allows us to obtain, with some more work, a stronger result. 6 GEORGE M. BERGMAN AND NAZIH NAHLUS Theorem 10. Suppose that k is an algebraically closed field of characteristic 0, that (Ai)i∈I is a family of finite-dimensional simple Lie algebras over k, that the index set I has cardinality less than any measurable cardinal > card(k), and that f : A = Qi∈I Ai → B is a surjective homomorphism to a finite-dimensional ∼= B, where the arrow is the projection Lie algebra B. Then B is semisimple, and f factors as QI Ai → Ai1 × · · · × Ain onto the product of a finite subfamily of the Ai. (In particular, f splits, i.e., is right-invertible.) Proof. By the result of Brown and Bourbaki cited above, in each Ai, every element is equal to a single bracket. Hence the same is true in A = QI Ai, and hence in the homomorphic image B of A; so in particular, B is idempotent: B = [B, B]. Now if B is trivial (0-dimensional), the desired result holds vacuously with n = 0, so assume the contrary. As a nontrivial idempotent Lie algebra, B must have a homomorphism onto a simple Lie algebra C. By Proposition 8, the composite map A → B → C must factor through the projection of A = QI Ai onto an ultraproduct A / U, for some ultrafilter U on I. By our assumption on the cardinality of I, U cannot be a nonprincipal card(k)+-complete ultrafilter. If it were nonprincipal and not card(k)+-complete, then the field K = kI / U, over which A / U is an algebra, would be uncountable-dimensional over k by Theorem 46, so by Proposition 9, the algebra C/Z(C) = C would acquire a structure of algebra over K. Hence C would be uncountable-dimensional, contradicting our hypothesis that B is finite-dimensional. Hence U must be a principal ultrafilter, determined by some i1 ∈ I, and what Proposition 8 then tells us is that the composite A → B → C factors through the projection A → Ai1 . Now if we write A = A′ × Ai1 where A′ = Qi∈I−{i1} Ai, we see that B = f (A) is the sum of the two mutually annihilating subalgebras B′ = f (A′) and f (Ai1 ). The latter subalgebra, since it does not go to zero under the map B → C, and since Ai1 is simple, must be an isomorphic image of Ai1 . In particular, it has trivial center. But the intersection of two mutually annihilating subalgebras of a Lie algebra must lie in their centers; so the subalgebras B′ = f (A′) and f (Ai1 ) have trivial intersection. Hence B = B′ × f (Ai1 ), and since B = [B, B] we must have B′ = [B′, B′]. We now repeat the argument with the map A′ → B′ in place of A → B. By induction on the dimension of B, the "left-over" part (which at this first stage we have called B′) must, after finitely many iterations, become zero, and we get a description of f, up to isomorphism, as the projection of A = QI Ai onto a finite subproduct Ai1 × · · · × Ain ∼= B. Such a projection clearly splits, giving the final assertion. (cid:3) Let us next examine homomorphisms on a direct product QI Ai of finite-dimensional solvable Lie alge- bras. We cannot expect that such homomorphisms will in general factor through finite subproducts, since the solvable Lie algebras include the abelian ones, which we noted at the beginning of §2 (under the descrip- tion "zero-multiplication algebras") can have homomorphisms on their direct products showing very unruly behavior. It is nevertheless reasonable to hope that a finite-dimensional homomorphic image of a direct product of solvable Lie algebras will be solvable. Among finite-dimensional Lie algebras, the solvable ones can be characterized in several ways: When the base field has characteristic 0, they are those admitting no homomorphisms onto simple Lie algebras. In general they are those containing no nonzero idempotent subalgebras, and also those satisfying one of a certain sequence of (successively weaker) identities. Since these conditions do not remain equivalent if one deletes finite-dimensionality, or the assumption that the algebras be Lie, or in the finite-dimensional Lie case, that k have characteristic 0, the statement we hope to obtain has several possible formulations for general algebras, all of potential interest. Of these, the one in terms of nonexistence of homomorphisms onto simple algebras is ready-made for a proof using Proposition 8. We shall obtain below a result for general algebras in arbitrary characteristic based on that proposition, then as a corollary get the desired statement on finite-dimensional solvable Lie algebras in characteristic 0. Our method will again require a restriction to avoid complications involving measurable cardinals. In §7, on the other hand, choosing a different condition that translates to "solvable" in the finite-dimensional Lie case, we will get a result which for finite-dimensional Lie algebras yields the same conclusion, without the restrictions on cardinality and characteristic. For general algebras, however, that result does not subsume the result of this section; they are independent. HOMOMORPHISMS ON INFINITE DIRECT PRODUCT ALGEBRAS, ESPECIALLY LIE ALGEBRAS 7 (The technical reason why the present generalization of solvability will require a cardinality restriction, but the version in §7 will not, is that we have not been able to prove that the property of admitting no homomorphisms onto simple algebras is preserved under countably complete ultraproducts, but we do have the corresponding statement for countable disjunctions of identities, Proposition 48.) Note that the next result (on not necessarily Lie algebras) is stronger than the above motivation might lead one to expect: the codomain algebra is only required (in the final sentence) to be countable-dimensional, rather than finite-dimensional, and still less is assumed about the dimensionalities of the Ai. Theorem 11. Suppose (Ai)i∈I is a family of algebras over an infinite field k, such that no Ai admits a homomorphism onto a simple algebra (or more generally, such that no Ai admits a homomorphism onto a countable-dimensional simple algebra). Assume further that, if there are measurable cardinals greater than card(k), then either card(I) or the supremum of the dimensions of all the Ai is less than all such cardinals. Then QI Ai admits no homomorphism onto a countable-dimensional simple algebra. dimensional simple algebra. Lemma 5 and Proposition 8 tell us that for some ultrafilter U on I, this f Proof. Suppose, by way of contradiction, that f : QI Ai → B is a homomorphism onto a countable- factors through the natural map QI Ai → QI Ai / U. If U is principal (in which case it is µ-complete for every cardinal µ), or is nonprincipal but card(k)+- complete (in which case, card(I) must be greater than or equal to some measurable cardinal µ > card(k), in the second paragraph of the theorem applies), then so that the bound on the dimensions of the Ai Theorem 47 shows that QI Ai / U is isomorphic to one of the Ai; but by hypothesis, no Ai admits a homomorphism onto B, a contradiction. On the other hand, if U is not card(k)+-complete, we can argue exactly as in the proof of Theorem 10, and conclude that B is uncountable-dimensional, though we assumed the contrary. So there exists no such f, as was to be proved. (cid:3) Here is the resulting statement about solvable Lie algebras. The dimensions of the Ai are still almost unrestricted, but we must make B finite-dimensional to turn the "no simple images" condition into solv- ability. Corollary 12. Suppose (Ai)i∈I is a family of solvable Lie algebras over a field k of characteristic 0, and suppose that, if there exists a measurable cardinal greater than card(k), then either card(I) or the supremum of the dimensions of the Ai is less than every such cardinal. (E.g., this is automatic if all Ai are of dimension ≤ the continuum; in particular, if they are finite-dimensional.) Then any finite-dimensional homomorphic image B of QI Ai is solvable. Proof. This follows from the preceding theorem, since if a Lie algebra is solvable, it admits no homomorphism onto a simple Lie algebra, and the converse holds in the finite-dimensional case in characteristic 0. (cid:3) 6. Ultraproducts and almost direct factors. The arguments of the preceding section were based on reducing the results to be proved to the consideration of homomorphisms from our infinite product onto simple algebras, and applying Proposition 8. But there are situations where that method is not enough. If the Ai are simple non-Lie algebras, we do not have Brown's theorem available to tell us that QI Ai is idempotent, from which we deduced that B had to have a simple homomorphic image. And if, in our consideration of solvable Lie algebras, we either replace solvability by nilpotence, or look at characterizations of solvability applicable in arbitrary characteristic, we again can't use that argument. For such purposes, we would like to have some variant of Proposition 8 not burdened with the condition Z(B) = {0} of Lemma 5(iii); something more in the spirit of Lemma 6 than of Lemma 5. It would also be nice to replace the assumption of finite-dimensionality of B in the conclusions of both Theorem 10 and Corollary 12 by a more general condition. We develop below a refinement of Proposition 8 in line with these two ideas. We will use the following concept, motivated by condition (ii′) of Lemma 6. Definition 13. For any algebra A, an almost direct decomposition of A will mean an expression A = B + B′, where B, B′ are ideals of A, and each is the two-sided annihilator of the other. In this situation, B and B′ will be called almost direct factors of A, and each will be called the complementary factor to the other. 8 GEORGE M. BERGMAN AND NAZIH NAHLUS Remarks: Any algebra A has a smallest almost direct factor, Z(A); its complement, the largest almost direct factor, is A. If A = B1 + B2 is an almost direct decomposition, then Z(B1) = Z(B2) = Z(A). Indeed, because B1 is the annihilator of B2 we have Z(A) ⊆ B1, and hence Z(A) ⊆ Z(B1); conversely, any a ∈ Z(B1) annihilates both B1 and B2, hence annihilates A, i.e., lies in Z(A). If Z(A) = {0}, the almost direct decompositions of A are its (internal) pairwise direct product decompo- sitions as an algebra. If Z(A) 6= {0}, an almost direct decomposition is never a direct product decomposition, since the two factors in the decomposition intersect in Z(A). However, an almost direct decomposition of A does induce a direct product decomposition of A/Z(A), as shown in the proof of Lemma 6(iii′) =⇒ (ii′). On the other hand, not every direct product decomposition of A/Z(A) need arise in that way, as shown by the 3 × 3 matrix example following that lemma. An almost direct factor of an almost direct factor of an algebra A is easily seen to be, itself, an almost direct factor of A. If we perform finitely many such successive almost direct decompositions, we get a decomposition A = B1 + · · · + Bn as a sum of ideals each of which is the annihilator of the sum of the rest. We may call such an expression an almost direct decomposition into several almost direct factors. The importance for us of almost direct decompositions lies in Lemma 14. If f : A1 × A2 → B is a surjective homomorphism of algebras, or more generally, a homo- morphism satisfying f (A1 × A2) + Z(B) = B, then f (A1) + Z(B) and f (A2) + Z(B) are complementary almost direct factors of B. Proof. By assumption, f (A1) + Z(B) and f (A2) + Z(B) sum to B, so it remains to prove that they are ideals of B, and are mutual two-sided annihilators. By symmetry, it suffices to show that f (A1) + Z(B) is an ideal, and is the two-sided annihilator of f (A2) + Z(B). Since A1 is an ideal of A1 × A2, its image f (A1) is an ideal of f (A1 × A2) + Z(B) = B, hence so is f (A1) + Z(B). Since A1 and A2 annihilate one another in A1 × A2, f (A1) + Z(B) and f (A2) + Z(B) annihilate one another in B. So it remains to show that any b ∈ B which annihilates f (A2) + Z(B) lies in f (A1) + Z(B). Let us write such an element b as f (a1) + f (a2) + z, with ai ∈ Ai, z ∈ Z(B). Since f (a1) and z automatically annihilate f (A2) + Z(B), and by assumption b = f (a1) + f (a2) + z does, it follows that f (a2) does. But as a member of f (A2), it also annihilates f (A1), hence it annihilates all of f (A1)+f (A2)+Z(B) = B, i.e., lies in Z(B). Hence the expression b = f (a1)+(f (a2)+z) expresses b as a member of f (A1)+Z(B), as required. (cid:3) Here is the generalization of finite-dimensionality that we indicated would be helpful in strengthening our results. Definition 15. We shall say that an algebra B has chain condition on almost direct factors if it has no infinite strictly ascending chain B1 ( B2 ( . . . of almost direct factors; equivalently, if it has no infinite strictly descending chain B′ 2 ) . . . of almost direct factors; equivalently, if it has no infinite chain (totally ordered set ) of almost direct factors. 1 ) B′ We do not know much general theory regarding the above chain condition. Clearly, it will hold in a ring with chain condition on two-sided ideals, which is already much weaker than finite-dimensionality. After the main results of this paper, a couple of results on the condition will be proved in §§11.1-11.2. An example in §12.2 will show that not every finitely generated associative algebra over a field satisfies it. We do not know whether every finitely generated Lie algebra over a field does Here, now, is our modified version of Proposition 8: Proposition 16. Suppose f : QI Ai → B is a surjective homomorphism of algebras, or more generally, a homomorphism such that f (QI Ai) + Z(B) = B; and suppose B has chain condition on almost direct factors. Let us abbreviate QI Ai to A, and write π : B → B/Z(B) for the canonical factor map. Then there exists a finite family of distinct ultrafilters U1, . . . , Un on I such that, if we write ϕm for the natural homomorphism A → A / Um (m = 1, . . . , n), then the composite map πf : A → B/Z(B) factors through the map (ϕ1, . . . , ϕn) : A → A / U1 × · · · × A / Un. Proof. If B = Z(B) this is vacuous, so assume the contrary. HOMOMORPHISMS ON INFINITE DIRECT PRODUCT ALGEBRAS, ESPECIALLY LIE ALGEBRAS 9 For any partition I = J ∪ (I − J), we have the direct product decomposition A = QJ Ai ×QI−J Ai; so by Lemma 14, f (QJ Ai) + Z(B) is an almost direct factor of B, with complementary almost direct factor f (QI−J Ai) + Z(B). Inclusions of subsets J give inclusions of almost direct factors f (QJ Ai) + Z(B), so by our assumption of chain condition on such factors, there must exist some J1 ⊆ I which yields an almost direct factor B1 strictly larger than Z(B), but such that every subset J ⊆ J1 yields either the same almost direct factor, B1, or the trivial almost direct factor, Z(B). Now by Lemma 14, for every J ⊆ J1 the ideals f (QJ Ai) + Z(B) and f (QJ1−J Ai) + Z(B) are com- plementary almost direct factors of B1; but by our choice of J1, these can only be B1 and Z(B) in one or the other order. We claim that the set (3) U (0) 1 = {J ⊆ J1 f (QJ Ai) + Z(B) = B1} is an ultrafilter on J1. Indeed, the class of subsets of J1 with the reverse property, f (QJ Ai)+Z(B) = Z(B), is closed under pairwise unions and passage to subsets, so U (0) enlargements, i.e., is a filter. Clearly ∅ /∈ U (0) J ⊆ J1, one of J or J1 − J is in U (0) is closed under pairwise intersections and 1 , so this filter is proper. And we have noted that for every 1 , so it is an ultrafilter. Having found an ultrafilter that roughly describes the behavior of f as a map from QJ1 Ai almost direct factor B1 = f (QJ1 Ai) + Z(B) of B, we now look at the complementary factor QI−J1 Ai of A, which f maps into the complementary almost direct factor f (QI−J1 Ai) + Z(B) of B. If the latter is not Z(B), we can repeat the above process with this map. (It was to make this work that we put the "or more generally" clause into the first sentence of this proposition and the preceding lemma. If f was into the 1 assumed surjective to B, this would not guarantee that its restriction to QI−J1 Ai would be surjective to f (QI−J1 Ai)+ Z(B).) Thus we get a subset J2 ⊆ I − J1 such that f (QJ2 Ai)+ Z(B) is a minimal nontrivial almost direct factor of f (QI−J1 Ai) + Z(B), and an ultrafilter U (0) of U (0) induces that same almost direct factor. on that subset such that every member 2 2 Iterating this process, we get a strictly decreasing sequence of almost direct factors of B associated with the sets I, I − J1, I − J1 − J2, . . . ; so our chain condition insures that this iteration cannot continue indefinitely. Thus, at some stage, say the n-th, our complementary almost direct factor must be Z(B), so the factor whose complement it is must be the whole algebra we are considering at that stage. Thus, without loss of generality we may, at that stage, take Jn = I − J1 − · · · − Jn−1 (rather than some proper Ai) + Z(B) subset thereof), giving us a partition I = Sm=1,...,n Jm; and we see that the ideals f (QJm let Um be the ultrafilter on I induced by the ultrafilter U (0) (m = 1, . . . , n) constitute an almost direct decomposition of B. Now for m = 1, . . . , n, m on Jm, i.e., Um = {J ⊆ I J ∩ Jm ∈ U (0) (4) gm : A / Um → B/Z(B) m }. For each such m we define a homomorphism as follows. Any element of A / Um is the image of some a = (ai)i∈I ∈ A = QI Ai. Let us map this by restriction to QJm Ai, and then by inclusion into A; this means replacing the components ai at indices i /∈ Jm by zero, while keeping the components at indices i ∈ Jm unchanged. Map the resulting element by f into B, and then by π into B/Z(B). If we had chosen a different representative a′ = (a′ i)i∈I ∈ A of our element of A / Um, then after restriction to Jm, this would have differed from (ai)i∈I only on a subset of Jm that is not in U (0) m . But by our construction of U (0) m , elements with support in such a subset of Jm are mapped into Z(B) by f ; so the image under πf of the element obtained from a′ equals the image under πf of the element obtained from a, showing that we have described a well-defined map (4). It is now routine to verify that gm is a homomorphism A / Um → B/Z(B), with image in (f (QJm Ai) + Z(B))/Z(B), so that the images of g1ϕ1, . . . , gnϕn annihilate one another; and that πf = g1ϕ1+· · ·+gnϕn, so that this map indeed factors through (ϕ1, . . . , ϕn). (cid:3) The above result says that under the indicated hypotheses we can, in a certain sense, approximate f : A → B "modulo Z(B) " by a homomorphism that factors through A / U1 × · · · × A / Un. It is natural to ask whether we can do so in a stronger sense, namely whether we can express f as a "perturbation", of the sort described by Lemma 4, of a genuine homomorphism f1 : A → B factoring through A / U1 × · · · × A / Un. 10 GEORGE M. BERGMAN AND NAZIH NAHLUS We can do this easily if the ultrafilters Um are principal: if each Um is the principal ultrafilter determined by im ∈ I, one finds that the desired f1 : A → B can be obtained by projecting A to Ai1 × · · · × Ain regarded as a subalgebra of A, and then mapping by f into B. For nonprincipal Um, we do not know whether such a factorization is always possible; but we shall show that it is whenever k is a field. Note that what we want is to perturb the given homomorphism f to a homomorphism f1 whose kernel contains the kernel of the natural surjection (ϕ1, . . . , ϕn) : A → A / U1 × · · · × A / Un. The following is a general result on when a homomorphism of algebras over a field has a perturbation whose kernel contains a prescribed ideal. Lemma 17. Let k be a field, f : A → B any homomorphism of k-algebras, and C an ideal of A. Then the following conditions are equivalent: (i) There exists a homomorphism f1 : A → B having C in its kernel, such that f − f1 is Z(B)-valued. (ii) f (C) ⊆ Z(B), and f (AA ∩ C) = {0}. Moreover, if B = f (A) + Z(B), then the first condition of (ii) is implied by the second. Proof. To get (i) =⇒ (ii) (which does not require the assumption that k is a field), suppose we have an f1 as in (i). Then C is carried into Z(B) by f − f1, but annihilated by f1, hence it must be carried into Z(B) by f, giving the first assertion of (ii). Further, Lemma 4 (with h = f1 − f ) tells us that f1 − f annihilates AA; and by assumption, f1 annihilates C, so f = f1 − (f1 − f ) must annihilate AA ∩ C, giving the second assertion. Conversely, assuming (ii), the second assertion thereof shows that the zero map and −f agree on AA ∩ C, whence there exists a unique k-linear map h : AA + C → B that agrees with the zero map on AA and with −f on C; and by the first condition of (ii), it will be Z(B)-valued. If k is a field, we can extend this as a vector space map (in an arbitrary way) to a map h : A → Z(B). Since h annihilates AA, Lemma 4 tells us that f1 = f + h is a k-algebra homomorphism; and since h agrees with −f on C, f + h has C in its kernel. For the final statement (which again does not require that k be a field), note that to show that f (C) ⊆ Z(B) is to show that f (C) is annihilated on each side by B, which, if B = f (A) + Z(B), is equivalent to being annihilated on each side by f (A). But multiplication on either side by f (A) carries f (C) into f (AC) + f (CA) ⊆ f (AA ∩ C), which is zero assuming the second condition of (ii). (cid:3) Corollary 18. Under the hypotheses of Proposition 16, if k is a field, then f can be written f1 + f0, where f1 is a homomorphism A → B factoring through (ϕ1, . . . , ϕn) : A → A / U1 × · · · × A / Un, and f0 is a homomorphism A → Z(B) (necessarily factoring through the natural map A → A/AA). Proof. It is easy to see that (ϕ1, . . . , ϕn) maps A surjectively to A / U1×· · ·×A / Un; hence a homomorphism on A can be factored through that map if and only if its kernel contains ker(ϕ1) ∩ . . . ∩ ker(ϕn). Hence, by Lemma 17 (including the final sentence), with C taken to be that intersection of kernels, it suffices to show that any a belonging both to ker(ϕ1) ∩ · · · ∩ ker(ϕn) and to AA is in ker(f ). Given such an a, A = A1 × A2, and a has zero component in the first factor. Hence since a ∈ AA = A1A1 + A2A2, we have a ∈ A2A2. let J = {i ∈ I ai = 0}. Thus, letting A1 = QJ Ai and A2 = QI−J Ai, we have Also, since a lies in the kernels of all ϕm, its support I − J belongs to none of the Um. Hence A2, which is also supported on that set, is likewise contained in the kernels of all the maps ϕm; so by the conclusion of Proposition 16, A2 ⊆ ker(π f ). This says that f (A2) ⊆ Z(B); hence f (a) ∈ f (A2A2) ⊆ Z(B)Z(B) = {0}, as required. The final parenthetical statement is a case of the general observation that any homomorphism from an (cid:3) algebra A to a zero-multiplication algebra factors through A/AA. If we now add to Corollary 18 the assumptions that the field k is infinite and the algebra B countable- dimensional, we can again (as in the proof of Theorem 10) use Proposition 9 and Theorem 46 to exclude the case where any of the ultrafilters Um are non-card(k)+-complete, getting Theorem 19. Suppose f : A = QI Ai → B is a surjective homomorphism of algebras over an infinite field k, where B is countable-dimensional over k and has chain condition on almost direct factors. HOMOMORPHISMS ON INFINITE DIRECT PRODUCT ALGEBRAS, ESPECIALLY LIE ALGEBRAS 11 Then there exist finitely many distinct card(k)+-complete ultrafilters U1, . . . , Un on I such that, writing ϕm for the natural homomorphism A → A / Um (m = 1, . . . , n), f can be written f1 +f0, where f1 factors through the map (ϕ1, . . . , ϕn) : A → A / U1 × · · · × A / Un, and f0 is a homomorphism A → Z(B). In particular, if card(I) is not ≥ any measurable cardinal > card(k), then each of the Um is a principal ultrafilter, so f1 factors through the projection of A onto the product of finitely many of the Ai. If it is merely assumed that none of the dimensions dimk(Ai) is ≥ a measurable cardinal > card(k), (cid:3) then the algebras A / Um are, at least, each isomorphic to one of the Ai. (Remark: if k is uncountable, the proof of Theorem 46 shows that for a non-card(k)+-complete ultrafilter U, the field kI / U will have dimension at least card(k) over k. So in that case, one can get the above result not only for B countable-dimensional, but for B of any dimensionality < card(k).) 7. Solvable algebras (version 2), and nilpotent algebras. We can now, as promised, prove a result which, for the finite-dimensional Lie case, gives essentially the same conclusion about homomorphic images of direct products of solvable Lie algebras as Corollary 12, but without the restrictions on card(I) and char(k), while for general algebras, it is independent of that result. We shall also obtain the analogous result for nilpotent algebras. Our conditions on general algebras will use the following analogs of the derived series and the lower central series of a Lie algebra. (We modify slightly a common notation for the latter, to avoid confusion with the subscripts indexing the factors in our direct products.) Definition 20. In any algebra A, we define, recursively, k-submodules A(d) (d = 1, 2, . . . ) by (d = 0, 1, . . . ) and A[d] (5) A(0) = A, A(d+1) = A(d) A(d), A[1] = A, (6) We will call A solvable if A(d) = {0} for some d, and nilpotent if A[d] = {0} for some d. A[d+1] = A A[d] + A[d]A. (The concept of nilpotence of a general algebra is standard. That of solvability is less so, but it appears in [32, p.17]. It is not hard to show that the submodules A[d] are ideals, and that the A(d) are subalgebras, and in the Lie case are ideals as well; but we shall not need these facts here.) Theorem 21. Suppose k is an infinite field, and (Ai)i∈I is a family of solvable k-algebras, in the sense of Definition 20 (e.g., solvable Lie algebras in the standard sense). Then any finite-dimensional homomorphic image of QI Ai is solvable. Proof. Say f : A = QI Ai → B is a homomorphism onto a finite-dimensional algebra. Since finite- dimensionality implies chain condition on almost direct factors, Theorem 19 shows that B is a sum of finitely many mutually annihilating homomorphic images of algebras A / Um, where the Um are card(k)+- complete ultrafilters on I, together with a subspace f0(A) ⊆ Z(B). The Um are, in particular, countably complete, and solvability is equivalent to the condition that an algebra satisfy one of the countable family of identities, (7) x = 0, x0 x1 = 0, (x00 x01) (x10 x11) = 0, . . . . Hence by Proposition 48, the condition of solvability on the Ai carries over to the algebras A / Um. The subalgebra f0(A) ⊆ Z(B) clearly also satisfies the second identity of (7). Hence, as the identities of (7) are successively weaker, at least one of them will be satisfied by all of the finitely many algebras A / Um and by f0(A). A sum of finitely many mutually annihilating algebras satisfying a common identity also satisfies that (cid:3) identity, yielding the asserted solvability. Exactly the same method yields Theorem 22. Suppose (Ai)i∈I is a family of nilpotent algebras (e.g., nilpotent Lie algebras) over an infinite (cid:3) field k. Then any finite-dimensional homomorphic image of QI Ai is nilpotent. 12 GEORGE M. BERGMAN AND NAZIH NAHLUS In this case, however, a stronger result will be proved in [4], by different methods, with "direct product" generalized to "inverse limit", and no requirement that k be infinite. (From that result of [4], an analog of Theorem 21 for inverse limits of solvable algebras is also deduced, but only for finite-dimensional Lie algebras Ai over a field of characteristic 0, for which there is an easy characterization of solvability in terms of nilpotence.) 8. Simple algebras -- general results. We would now like to use Theorem 19 to get a result on homomorphic images of products of finite- dimensional simple Lie algebras stronger than our earlier Theorem 10. Simplicity is not, like solvability or nilpotence, equivalent to a disjunction of identities, but that is not a problem: like countable disjunctions of identities, it is preserved by countably complete ultraproducts (Proposition 49). A more serious difficulty is that the preceding proofs used the fact that f0(A), a zero-multiplication algebra, was automatically nilpotent and solvable; but if f0(A) 6= {0}, it will certainly not be a product of simple algebras. By the final observation of Corollary 18, the map f0 of Theorem 19 can be nontrivial only if AA 6= A, i.e., if A is not idempotent. Simple algebras Ai are idempotent. Does this property carry over to direct products? To answer this, let us define, for every idempotent algebra A, its idempotence rank, idp-rk(A), to be the supremum, over all a ∈ A, of the least number m of summands in expressions for a as a sum of products: (8) h=1 b(h)c(h)}). idp-rk(A) = supa∈A(inf {m ≥ 0 (∃ b(1), . . . , b(m), c(1), . . . , c(m) ∈ A) a = Pm This will be a nonnegative integer (positive if A 6= {0}), or ∞; it measures the difficulty in asserting the idempotence of A in a uniform way. We can now state and prove Lemma 23. For a family of algebras Ai (i ∈ I), the following conditions are equivalent. (i) QI Ai is idempotent. (ii) Every Ai is idempotent, and there is a natural number n such that for all but finitely many i ∈ I, idp-rk(Ai) ≤ n. When the above equivalent conditions hold, idp-rk(QI Ai) = supi∈I idp-rk(Ai). Proof. We shall prove (ii) =⇒ (i) and ¬(ii) =⇒ ¬(i). Given n as in (ii), consider any (ai)i∈I ∈ A. For those i such that idp-rk(Ai) ≤ n, take a representation of ai as a sum of n products, while for each of the remaining finitely many indices i, take some representation of ai as a sum of products. (Some of the Ai may have infinite idempotence rank; but they are all assumed idempotent, so each element ai can be so written.) There will be a common upper bound N for the number of summands in all these representations, yielding a representation for (ai) as a sum of N products, proving (i). Assuming ¬(ii), note that if not all Ai are idempotent, then A cannot be. If they all are idempotent, but there is no finite n bounding the idempotence ranks of all but finitely many of them, then it is easy to construct an (ai) ∈ A such that the number of products required to express the component ai is unbounded as a function of i, and to deduce that (ai) cannot be written as a finite sum of products, proving ¬(i). The verification of the final assertion (which we won't use) is straightforward; one breaks it into two cases, the case where the idempotence ranks of all the Ai are finite, so that (ii) implies that they have a common finite bound, and the case where at least one is ∞. (cid:3) Applying this to homomorphic images of direct products of simple algebras, we can now prove Theorem 24. Suppose k is an infinite field, and f : A = QI Ai → B is a surjective homomorphism from a direct product of simple algebras to a countable-dimensional algebra B having chain condition on almost direct factors. Then (a) If there is a finite upper bound on all but finitely many of the values idp-rk(Ai) (i ∈ I), then B is isomorphic to a direct product of finitely many of the Ai. (b) Without the assumption of such a bound, B will be isomorphic to a direct product of finitely many of the Ai, and one k-vector-space with zero multiplication. In the situation of (a), the homomorphism f splits (has a right inverse). In the situation of (b), the composite homomorphism A → B → B/Z(B) splits. HOMOMORPHISMS ON INFINITE DIRECT PRODUCT ALGEBRAS, ESPECIALLY LIE ALGEBRAS 13 Proof. Let f be expressed as in Theorem 19. By Proposition 49, all of the A / Um in that description are simple. A homomorphic image of a finite direct product of simple algebras is the direct product of some subset of these; so possibly dropping some of the A / Um, we may assume that the image of f1 in B is an isomorphic copy of A / U1 × · · · × A / Un. Let (9) J = {i ∈ I dimk(Ai) is uncountable} ⊆ I. If for any of m = 1, . . . , n, the above set J were Um-large, it is easy to see that A / Um would also be uncountable-dimensional. (This does not use the countable completeness of Um; just the observation that if we had an ℵ1-tuple of k-linearly-independent elements in Ai for each i ∈ J, this would give an ℵ1- tuple of elements of A whose images in A / Um would also be linearly independent.) Then A / Um could not be embedded in B; so this does not happen. Hence each A / Um can be identified with a countably complete ultraproduct of (Ai)i∈I−J , a system of countable-dimensional k-algebras. The second paragraph of Theorem 47, with µ = card(k)+, now tells us that each A / Um is isomorphic to one of the Ai. In the situation of statement (a), Lemma 23 tells us that A is idempotent, so by the final parenthetical observation of Corollary 18, the f0 of Theorem 19 is zero. Thus, B = f1(A) ∼= A / U1 × · · · × A / Un, which we have just seen is isomorphic to the direct product of finitely many of the Ai. In the situation of (b), B will be the sum of f1(A), as above, and f0(A) ⊆ Z(B). It is not hard to see that if an algebra B is the sum of a subalgebra B0 ⊆ Z(B), and a subalgebra B1 with Z(B1) = {0}, then B is the direct product of those two subalgebras, establishing the "direct product" assertion of (b). To get the final splitting assertion in the situation of (a), note that since U1, . . . , Un are finitely many Ai, these n factors have pairwise products zero, and the m-th factor maps under f onto the isomorphic image of A / Um in B. Thus, f is, up to isomorphism, the direct product of the n canonical distinct ultrafilters, we can partition I into disjoint sets J1, . . . , Jn with Jm ∈ Um. Writing A = QJ1 Ai × · · · × QJn maps QJm Ai → QJm The situation of (b) is essentially the same, with the composite map A → B → B/Z(B) in place of f. (cid:3) Ai / Um. As shown in Theorem 47, each of these maps splits; hence so does f. Note that if we are given a family of idempotent algebras Ai not satisfying the hypothesis of (a) above, there always exist homomorphisms f from A = QI Ai onto algebras B for which the zero-multiplication summand of statement (b) is nonzero. For by Lemma 23, A will not be idempotent, hence A/AA will be a nontrivial zero-multiplication homomorphic image of A, and we can take for B any countable-dimensional homomorphic image of A/AA. (A/AA will itself be uncountable-dimensional, for one can partition (Ai)i∈I into infinitely many subfamilies, for each of which the finite-bound condition of (a) fails, and A/AA maps onto the direct product of the infinitely many zero-multiplication algebras that these yield.) 9. Simple Lie algebras (version 2). What happens, in particular, if the Ai are finite-dimensional simple Lie algebras? Will we necessarily be in situation (a) of the above theorem? This comes down to the question of whether, for a fixed base field k, there is a uniform bound on the idempotence ranks of all finite-dimensional simple Lie algebras over k. At the beginning of §5 we noted that a consequence of a theorem of G. Brown answers that question affirmatively for k algebraically closed of characteristic 0. What Brown in fact proved (in his doctoral thesis, [11]) is that over any infinite field k, every classical simple Lie algebra in the sense of Steinberg [35] has (in our language) idempotence rank 1. (The classical simple Lie algebras in that sense comprise both the infinite families An, . . . , Dn and the exceptional algebras, E6, . . . , G2.) When k has characteristic 0, these are the split simple Lie algebras in the sense of [10], which if k is also algebraically closed are all the finite-dimensional simple Lie algebras; so in this case the idempotence ranks of all finite-dimensional simple Lie algebras indeed have a common bound, 1. The Bourbaki reference that we also cited, [10, Ch.VIII, §13, Ex. 13(b)], gets the same conclusion, for algebraically closed fields of characteristic 0 only, but with some additional information. What if k is, instead, the field R of real numbers? If L is a finite-dimensional simple real Lie algebra, then L ⊗R C will be semisimple over C, hence a direct product of one or more simple complex Lie algebras, hence will have idempotence rank 1 by the results quoted. We claim that this implies that L itself has idempotence rank ≤ 2. Indeed, every a ∈ L can be written within L ⊗R C as a bracket [b + ic, d + ie] with 14 GEORGE M. BERGMAN AND NAZIH NAHLUS b, c, d, e ∈ L. Thus, a = Re(a) = [b, d] − [c, e] = [b, d] + [−c, e], a sum of two brackets. (This is noted at [25, Corollary A3.5, p.653], while Theorem A3.2 on the same page shows that every compact simple real Lie algebra, i.e., every simple real Lie algebra whose Killing form is negative definite, has idempotence rank 1.) Note that the above argument uses the fact that C has degree 2 over R. Hence it cannot be extended to give finite bounds on the idempotence ranks of simple Lie algebras over most subfields k ⊆ C, e.g., Q, since C has infinite degree over these. (The fields for which it works, those over which an algebraically closed field of characteristic 0 has finite degree, which is necessarily 2, are the real-closed fields, the fields that algebraically "look like" R.) However, there is another result in the literature, less obviously related to idempotence rank, that we can use to get what we need in a much wider class of cases. J.-M. Bois [9] proves, using the recently completed classification [31] of finite-dimensional simple Lie algebras L over algebraically closed fields of characteristic not 2 or 3, that every such algebra is generated as a Lie algebra by two elements. We shall show below, first, that such a bound on the number of generators yields something slightly stronger than a bound on the idempotence rank of L, and, then, that for that strengthened version of idempotence rank, change of base field is not a problem; so that from Bois's result on Lie algebras over algebraically closed fields, we can get the result we need for Lie algebras over general infinite fields. (Notes to the reader of [9]: Though Theorem A thereof does not state the assumption that the base field is algebraically closed, this is clear from the rest of the paper, and Bois (personal communication) confirms that it is to be understood. In [9, §1.2.2], the one part of that paper where non-algebraically-closed base fields k are considered, it is shown that if L is a Lie algebra over an infinite field k such that, on extending scalars to the algebraic closure K of k, the resulting Lie algebra L ⊗k K can be generated over K by two elements, then L can be so generated over k. But we shall see from examples in §12.6 below that in positive characteristic, a simple L can yield an L ⊗k K that is not even semisimple, so that [9, §1.2.2] is not applicable to it; and indeed that such an L can fail to be generated by 2 elements. Nevertheless, the ideas of [9, §1.2.2] will be used in proving Theorem 26 below, which states that any such simple Lie algebra has idempotence rank ≤ 2.) The fact which turns statements about numbers of generators into statements relevant to idempotence rank, part (c) of the next lemma, would be trivial if we were considering associative algebras. It takes a bit more work in the Lie case, where we must use the Jacobi identity instead of associativity, and is false in general nonassociative algebras (§11.2 below, last sentence). Statements (a) and (b) are steps in the proof that seemed worth recording. These results do not require the base ring to be a field, so we give them for general k. Lemma 25. Let L be a Lie algebra over a commutative ring k. (a) If U is a k-submodule of L, then {x ∈ L [x, L] ⊆ U } is a Lie subalgebra of L. (b) If V is a k-submodule of L that generates L as a Lie algebra, then [V, L] = [L, L]. (c) If [L, L] = L, and L is generated as a Lie algebra by a set X, then L = Px∈X [x, L]. Proof. In (a), the fact that the indicated set is closed under the k-module operations follows from the fact that U is, while closure under Lie brackets comes from the Jacobi identity: if [x, L] and [y, L] are contained in U, then (10) [ [x, y], L] ⊆ [x, [y, L] ] + [y, [x, L] ] ⊆ [x, L] + [y, L] ⊆ U. To get (b), we apply (a) with U = [V, L]. The Lie subalgebra described in (a) then contains V, hence, as V generates L, it equals L. This means that [L, L] ⊆ [V, L]; the opposite inclusion is clear. To get (c), we apply (b) with V the k-submodule spanned by X, and use the assumption [L, L] = L to (cid:3) replace [L, L] in (b) by L. (One can prove, more generally, a version of the above lemma for the action of L on an L-module M. E.g., (c) then takes the form, "If L M = M and L is generated as a Lie algebra by a set X, then M = Px∈X x M. ") If we apply (c) to the case where L can be generated by n elements, the resulting conclusion, (11) (∃ x1, . . . , xn ∈ L) L = [x1, L] + . . . + [xn, L], is formally stronger than the statement that L has idempotence rank n : the idempotence rank statement allows both arguments in the brackets giving an element a ∈ L to vary as we vary a, while (11) fixes one HOMOMORPHISMS ON INFINITE DIRECT PRODUCT ALGEBRAS, ESPECIALLY LIE ALGEBRAS 15 argument in each bracket. To see that it is strictly stronger, recall that by Brown's result, many finite- dimensional Lie algebras over fields have idempotence rank 1; but no nonzero finite-dimensional Lie algebra over a field can satisfy (∃ x1 ∈ L) L = [x1, L], since any x1 has nontrivial centralizer, so that [x1, L] has smaller dimension than L. Using the above lemma, Bois's Theorem A, and a density argument, we can now prove Theorem 26. Let k be an infinite field of characteristic not 2 or 3, and L a finite-dimensional simple Lie algebra over k. Then there exist x1, x2 ∈ L such that L = [x1, L] + [x2, L]. Proof. Let K be the algebraic closure of k, and LK = L ⊗k K. This will be a finite-dimensional Lie algebra over K, and will inherit from L the property of being idempotent; hence as a K-algebra it will have a finite-dimensional simple factor algebra M. Let q : LK → M be the canonical surjection. Since L generates LK as a K-algebra, q(L) similarly generates M, hence, in particular, is nonzero; so, as L is simple, q embeds L in M. Since k is infinite, L × L is Zariski-dense in LK × LK. (I.e., if we represent elements of the finite- dimensional K-vector-space LK ×LK in terms of coordinates in some K-basis, then any polynomial function of those coordinates which vanishes on the subset L × L vanishes everywhere.) It follows that its image q(L) × q(L) ⊆ M × M is Zariski-dense in the latter space. In what follows, let us identify L with q(L). By Bois [9, Theorem A], M can be generated as a Lie algebra over K by two elements. Moreover, as noted in [9, §1.2.2], the set of generating pairs of elements of M will be a Zariski-open subset of M × M. (I.e., for every generating pair (x1, x2) ∈ M × M, there is a finite family of polynomials in the coordinates of x1 and x2 which are nonzero at that pair, and such that every pair at which these polynomials are nonzero is again a generating pair. Roughly, this is because any Lie algebra expression f (x1, x2) has coordinates given by polynomials in the coordinates of x1 and x2, and the property that a given list of dimK(M ) such expressions spans M over K is equivalent to the nonvanishing of an appropriate determinant in the resulting coordinate polynomials. Cf. [5, §1].) The nonempty Zariski-open set of generating pairs must meet the Zariski-dense set L × L, which means that there exist x1, x2 ∈ L which generate M as a Lie algebra over K. Hence by Lemma 25(c), (12) M = [x1, M ] + [x2, M ]. We claim that this implies (13) L = [x1, L] + [x2, L]. To show this, let B be a basis of K as a k-vector-space. Then the Lie algebra LK = L ⊗k K, under the adjoint action of its sub-k-algebra L, is the direct sum of the sub-L-modules L ⊗ b (b ∈ B), each of which is isomorphic to L as an L-module, via the map x 7→ x ⊗ b. Since L is simple as a Lie algebra, it is a simple module over itself under the adjoint action, hence since LK is a direct sum of copies of that simple L-module, so is its homomorphic image M. As M is a direct sum of simple L-modules, L is a direct summand therein. Applying to (12) an L-module projection of M onto L, we get (13), as required. (cid:3) For some results on particular elements x1 and x2 in split simple Lie algebras L such that L = [x1, L] + [x2, L], and related matters, see [29]. Theorem 24(a) and Theorem 26 together give the desired result on infinite products of simple Lie alge- bras in characteristics 6= 2, 3. We also record the weaker statement that follows from Theorem 24(b) for characteristics 2 and 3 (where there is as yet insufficient structure theory to say whether a result like that of [9] holds). Theorem 27. Suppose k is an infinite field, and f : QI Ai → B a surjective homomorphism from a direct product of finite-dimensional simple Lie algebras to a finite-dimensional Lie algebra B. Then (a) If char(k) 6= 2 or 3, B is isomorphic to the direct product of finitely many of the Ai, and the map f : A → B splits. (b) If char(k) = 2 or 3, one can at least say that B is isomorphic to a direct product of finitely many (cid:3) of the Ai and an abelian Lie algebra. In that case, the composite map A → B → B/Z(B) splits. A few notes on the general concept of idempotence rank: By Theorem 8.4.5 of [12], every finite-dimensional simple associative algebra has a unit, and so has idempotence rank 1. On the other hand, we will give in §12.4 16 GEORGE M. BERGMAN AND NAZIH NAHLUS examples of arbitrarily large finite idempotence rank in non-simple finite-dimensional Lie and associative algebras, and in simple finite-dimensional non-Lie nonassociative algebras; while in §12.5, we will give an example of a finite-dimensional (non-Lie) algebra whose idempotence rank changes under change of base field. (This is the phenomenon, the possibility of which prevented us from using Brown's result to get Theorem 27 over a general field of characteristic 0.) 10. Continuity in the product topology. Any infinite product A = QI Ai of sets has a natural topology, the product of the discrete topologies on the Ai. If the Ai have group structures, and f : A → B is a homomorphism into another group B, it is not hard to show that f is continuous in the product topology on A and the discrete topology on B if and only if it factors through the projection of A onto a finite subproduct Ai1 × · · · × Ain . Indeed, "if" is immediate. To see "only if", note that by the discreteness of B, ker(f ) is open. As an open set containing the identity element, it must contain the intersection of the inverse images of neighborhoods of the identity elements ei of finitely many of the Ai. So a fortiori, it contains the intersection of the inverse images of the trivial subgroups of those Ai; which is the kernel of the projection to their product, so f factors through that projection. Let us briefly note when this continuity condition holds in the results of the preceding sections. It is not hard to see that it can never hold if f involves a factorization through a nonprincipal ultrafilter. When f is the sum of a map f1 that factors through finitely many of the Ai (corresponding to finitely many principal ultrafilters), and a possibly nonzero perturbing map f0 into Z(B), then the continuity of f depends on the continuity of f0, which in general cannot be expected: as noted in §1, such maps tend to be "unruly". However, the effect of f0 disappears if we compose f with the factor map π : B → B/Z(B). Summarizing the consequences of these considerations, we have Proposition 28. In the preceding results of this note, the map f will be continuous in the product topology on A and the discrete topology on B (equivalently, will factor through the projection to a finite subprod- uct of the Ai) in the situations of the following results whenever card(I) is less than every measurable cardinal > card(k) : Theorem 10 (where that restriction on I is already assumed ), Theorem 24(a), and Theorem 27(a). Under the same assumption on card(I), the composite map π f : A → B → B/Z(B) will be continuous in the situations of Theorem 19, Theorem 21, Theorem 22, Theorem 24(b), and Theorem 27(b). 11. Some tangential notes. We record here further observations on the material introduced in the preceding pages, which were not needed for the results developed there. (The subsections of this section are independent of one another.) 11.1. Almost direct factors, and Boolean algebras. Recall that for an associative unital algebra A, an almost direct decomposition is the same as a direct product decomposition (because Z(A) = {0}), and that in this situation, such decompositions are in bijective correspondence with the central idempotent elements of A. The set of such central idempotents, and hence the partially ordered set of almost direct factors of A, forms a Boolean algebra. Will the same be true of the partially ordered set of almost direct factors in a general algebra A ? Below, we obtain a positive answer when A is idempotent (or satisfies a slight weakening of that condi- tion), then a counterexample in the absence of that assumption. Proposition 29. Let A be an idempotent algebra, or more generally, an algebra satisfying (14) A = AA + Z(A). Then the almost direct factors of A form a Boolean algebra, with zero element Z(A), unit element A, the join of B and C given by the sum B + C, and the meet given by the intersection B ∩ C, which is also equal to BC + Z(A) and to CB + Z(A). Proof. Let us write a 7→ a for the quotient map A → A/Z(A). Any almost direct decomposition A = B + B′ is determined by the induced direct product decomposition A = B × B′ (cf. remarks following Definition 13), hence by the projection operator of A onto B. We shall prove below that under the present hypotheses, the projection operators so induced by any two almost direct decompositions A = B + B′ and HOMOMORPHISMS ON INFINITE DIRECT PRODUCT ALGEBRAS, ESPECIALLY LIE ALGEBRAS 17 A = C + C′ commute, and that the image B ∩ C of their composite corresponds to an almost direct factor B ∩ C = BC + Z(A) = CB + Z(A) of A, with complement B′ + C′. Now a set of pairwise commuting projection operators (i.e., idempotent endomorphisms) on any abelian group generates a Boolean algebra of such operators, with the meet and join of operators e and f (given by ef and e + f − ef respectively) corresponding to the intersection and the sum of the image subgroups; so these results will prove our claims. Given almost direct decompositions A = B + B′ and A = C + C′, let us multiply these two equations together and add Z(A). By (14), this yields A = BC + BC′ + B′C + B′C′ + Z(A), which we can rewrite (15) A = (BC + Z(A)) + (BC′ + Z(A)) + (B′C + Z(A)) + (B′C′ + Z(A)). (Since A is not assumed associative or Lie, we do not yet know that the summands in (15) are ideals of A, only that they are k-submodules.) Let us verify first that the decomposition a = aBC + aBC ′ + aB ′C + aB ′C ′ of an element a ∈ A arising from (15) is unique modulo Z(A). For this, it will suffice to show that in any decomposition of 0, (16) 0 = zBC + zBC ′ + zB ′C + zB ′C ′ into summands in the above four k-submodules, all of these summands must lie in Z(A). Now since B and C are ideals, the term BC in the first summand of (15) is contained in both B and C, hence that summand BC + Z(A) annihilates B′ and C′, hence annihilates all the summands in (15) other than itself; so if the summand zBC of (16) does not lie in Z(A), i.e., does not annihilate all of A, this can only be because it fails to annihilate the first summand of (15). But all the other summands on the right-hand side of (16) do annihilate the first summand of (15), as does the left-hand term, 0; so zBC must also. This completes the verification that it lies in Z(A); and by the same argument, so do all the terms of (16), as claimed. Hence, passing to quotients modulo Z(A), the decomposition (17) A = BC + BC′ + B′C + B′C′. is a direct product decomposition of k-modules. Using again the same kind of reasoning, note that when we decompose an element of A by (15), the component in each summand whose expression in (15) involves B annihilates B′, and inversely. Hence given such a decomposition a = aBC + aBC ′ + aB ′C + aB ′C ′, the expression a = (aBC + aBC ′) + (aB ′C + aB ′C ′ ) decomposes a into an element annihilating B′, i.e., a member of B, and an element annihilating B, i.e., a member of B′. But the decomposition of a coming from the relation A = B + B′ is unique up to summands in B ∩ B′ = Z(A); hence the idempotent endomorphism of A given by projection on the first summand in A = B + B′ must coincide with the projection of (17) onto the sum of its first and second summands. Similarly, the idempotent endomorphism of A arising from the decomposition A = C + C′ must be the projection of (17) onto the sum of its first and third summands. These two projections commute, since their product in either order is the projection of (17) onto its first summand. Since the range of the product of two commuting idempotent endomorphisms of an abelian group is the intersection of their ranges, we have BC = B ∩ C. Taking inverse images in A, this gives (18) BC + Z(A) = B ∩ C, as claimed; and by symmetry we likewise have CB + Z(A) = B ∩ C. The equality (18) shows that BC + Z(A) is an ideal; we must still verify that it is an almost direct factor in A. We claim that it and (BC′ + Z(A)) + (B′C + Z(A)) + (B′C′ + Z(A)) are each other's two- sided annihilators. We have seen that they annihilate each other. On the other hand, by the method of reasoning used immediately after (16), if an element a annihilates BC + Z(A), the first component of a decomposition of a as in (15) lies in Z(A); and since Z(A) also lies in the other three summands of (15), a will lie in the sum of those three summands. So the two-sided annihilator of BC + Z(A) is indeed (BC′ + Z(A)) + (B′C + Z(A)) + (B′C′ + Z(A)). That BC + Z(A) is likewise the two-sided annihilator of that sum is shown in the same way. This completes our proof. (cid:3) The above result covers not only the case where A is idempotent, but the opposite extreme, where A has zero multiplication, since then A = Z(A) = AA + Z(A). (In that case, our Boolean algebra is trivial.) But let us now show that when A 6= AA + Z(A), the conclusion of Proposition 29 need not hold. Let A = R2 × R, with multiplication (v, a) ∗ (w, b) = (0, v · w), where v · w is the dot product of vectors in R2. Then Z(A) = {0} × R, and for every one-dimensional subspace V ⊆ R2, we have the almost direct decomposition A = (V ×R)+(V ⊥ ×R), where ( )⊥ denotes orthogonal complement in R2. Thus, the almost 18 GEORGE M. BERGMAN AND NAZIH NAHLUS direct factors lying strictly between Z(A) and A form an infinite set of pairwise incomparable elements V × R (though for each such element, only one of the others is its "complementary almost direct factor" as we have defined the term). Hence the partially ordered set of almost direct factors of A is not a Boolean algebra. The above algebra A is, incidentally, associative, since all three-fold products are zero. 11.2. Weakening the definition of an almost direct decomposition. In our definition of an almost direct decomposition A = B + B′ of an algebra A, the condition that B and B′ be ideals can be formally weakened to say that they are subalgebras. For the latter condition on B says that it is closed under left and right multiplication by B, and since it annihilates B′, it is trivially closed under left and right multiplication by that subalgebra; hence it closed under left and right multiplication by B + B′ = A. If we ask whether it is enough to assume that B and B′ are k-submodules summing to A, each of which is the other's two-sided annihilator, the answer is mixed. If A is associative, we can still conclude that they will be almost direct factors. For since B annihilates B′ on both sides, associativity implies that BB does the same; hence it is contained in the annihilator of B′, namely B, proving that B is a subalgebra. We get the same conclusion if A is a Lie algebra: the Jacobi identity shows that [B′, [B, B]] ⊆ [B, [B′, B]] = {0}, hence [B, B] is contained in the annihilator B of B′. But for general A, the corresponding statement is false. Indeed, for any k, let A be the k-algebra which is free as a k-module on two elements x and y, with multiplication given by (19) xy = yx = 0, xx = y, yy = x. Then clearly, kx and ky are each other's two-sided annihilators, and sum to A, but are not subalgebras. Even if one of B, B′ is a subalgebra, the other may not be, as we can see by replacing the relation xx = y in (19) by xx = x, while leaving the other relations unchanged. Incidentally, taking X = {x} in the algebra defined by (19), we find that the analog of Lemma 25(c) fails (X generates A, but XA 6= A), showing that that result does not hold for general k-algebras. 11.3. "Early" ultrafilters. Just as many calculus texts come in two versions, "early transcendentals" and "late transcendentals", so the development of §§2-6 has an alternative version, in which we obtain our ultrafilters early, before the "either/or" conditions such as Lemma 5(i) by which we summoned them in our present development. In such a development, one would associate to any map f from a product of nonempty sets A = QI Ai to a set B the family Ff of J ⊆ I such that f factors through QJ Ai. This turns out to be a filter, the largest filter F such that f factors through A/F . Any filter is an intersection of ultrafilters; let us call the set of ultrafilters containing Ff the "support" of f. One verifies that f factors through the natural map A/U. Finally, bringing in the assumption that the Ai and B are algebras and f a surjective homomorphism, one can use the argument of Proposition 8 to show that if B satisfies the conditions of Lemma 5, then the support of f is a singleton {U}, while the argument of Proposition 16 shows that if B satisfies the weaker property of chain condition on almost direct factors, then πf has support in a finite set of ultrafilters. A → QU ⊇Ff The proofs of the propositions mentioned used the fact that in a direct product of algebras, elements with disjoint support have trivial product. One might get similar results on direct products of groups (or even monoids) using the fact that in a direct product of these, elements with disjoint supports commute. (This is suggested, of course, by the way the brackets of Lie algebras arise from the noncommutativity of Lie groups.) We leave this for the interested reader to investigate. A very different "early ultrafilters" approach is taken in [6, §3]. 11.4. On idempotence rank, and related functions. In examining the properties of the idempotence rank function on idempotent algebras, it is helpful to look at a more general version of that situation. For simplicity, let k be a field. Consider any 4-tuple (A, B, C, m), where A, B and C are k-vector-spaces, and m is a surjective linear map A ⊗k B → C. (Thus, m gives the same information as a k-bilinear map A × B → C whose image spans C.) Let us define (20) max-rank(m) = supc∈C inf t∈m−1(c) rank(t), where rank(t) denotes the rank of t as a member of the tensor product A ⊗k B, i.e., the minimum number of decomposable tensors a ⊗ b that must be summed to get t. We see that when A = B = C is the HOMOMORPHISMS ON INFINITE DIRECT PRODUCT ALGEBRAS, ESPECIALLY LIE ALGEBRAS 19 underlying vector space of an idempotent k-algebra A, and m the map corresponding to the multiplication of A, then max-rank(m) is in fact idp-rk(A). The function max-rank(m) has a family resemblance to the function r(M ) introduced in [2] for a subspace M of a tensor product A ⊗k B, and defined by (21) r(M ) = inf t∈M−{0} rank(t). The contexts of the two definitions are essentially the same: what we are given in each is equivalent to a short exact sequence 0 → M → A ⊗k B → C → 0 of k-vector-spaces, with middle term a tensor product. However, neither of these invariants of that short exact sequence seems to be expressible in terms of the other. In fact, if our vector spaces are finite-dimensional, we can form the dual short exact sequence 0 → C∗ → A∗ ⊗k B∗ → M ∗ → 0, and look at the same two invariants for it, getting, altogether, four invariants from our original sequence, none of which seems to be expressible in terms of the others. As noted in [2], r(M ) can decrease, but not increase, under extension of base field; for when we make such an extension, the set of elements over which the infimum of (21) is taken is enlarged, while the rank-function on elements lying in the original tensor product remains unchanged. (If we take for M the kernel of the map A ⊗k A → A corresponding to the multiplication operation of an algebra, then a decrease in r(M ) from a value > 1 to 1 under base extension from k to K means that from a k-algebra A without zero divisors, we get a K-algebra A ⊗k K with zero divisors. The reverse cannot happen, of course.) Since the definition (20) of max-rank(m) involves both a supremum and an infimum, that function can potentially increase or decrease under base extension. The possibility of its decreasing is what made it impossible for us to go from Brown's result showing that idp-rk(L) = 1 for L a finite-dimensional simple Lie algebra over C to the corresponding statement for subfields of C. In §12.5 we will see examples of finite-dimensional, idempotent (but non-Lie, nonassociative, non-simple) algebras A whose idempotence ranks do increase and decrease under base extensions. We do not know whether either can happen when A is a simple Lie algebra. What we used in §8 (in conjunction with the results of [9]), instead of the unsuccessful approach indicated above, was an argument via what might be called the "one-variable-constant idempotence rank function", the least n such that (11) holds. In the case of general k-algebras, where left and right multiplication are not equivalent, we could call the version with the constant factors on, say, the left the "left-constant idempotence rank": (22) l-const-idp-rk(A) = inf {n (∃ x1, . . . , xn ∈ A) A = x1A + · · · + xnA} = inf {dimk(V ) A = V A}. This function is examined in [5]. 11.5. Other literature on homomorphisms from infinite products. Restrictions on homomorphisms from infinite direct products to "small" objects have been noted in other areas of algebra. In [30, Corollary 9], it is shown that a homomorphic image of a direct power of a finite nonabelian simple group G, if countable, must be finite; general finite groups G with that property are investigated in [7] and [8]. This situation has a similar flavor to that of the present note; e.g., note that simple nonabelian groups satisfy the analogs of the conditions of our Lemma 5. (We remark, however, that the groups characterized in the above papers all have trivial centers. Perhaps if one considers groups with nontrivial centers, analogs of the results of this note showing that homomorphisms Q Ai → B acquire stronger properties on composition with the natural map B → B/Z(B) will turn up.) An area of investigation with a different flavor begins with the result of [34], that every homomorphism of abelian groups ZN → Z factors through the projection onto finitely many coordinates. It can be deduced from this that the same factorization property holds for homomorphisms from any countable product of abelian groups Qi∈N Ai to Z; this is expressed by saying that Z is a slender abelian group. More generally, slenderness has been studied in abelian monoids, in modules over general rings, and in objects of general preadditive categories. Note that for these abelian groups, abelian monoids, etc., unlike the algebras of this note and unlike nonabelian groups, any finite family of morphisms can be added; hence in mapping a finite product A1 × · · · × An to an object B, one can form sums of homomorphisms Ai → B. Thus, the restrictions that turn out to hold on homomorphisms from infinite products of these objects cannot arise from restrictions on homomorphisms from finite products, like those of our Lemma 5, but must come in in a more mysterious way; roughly, it seems, from completeness-like properties of infinite products, which cannot be duplicated in a "slender" B. Examples of abelian groups that are not slender include all abelian groups 20 GEORGE M. BERGMAN AND NAZIH NAHLUS with torsion, all nonzero injective abelian groups, the additive group of p-adic integers, and, of course, ZN. For a sampling of work in this area, see, for abelian groups, [21, §94] and [33], for modules, [1] and [20], and for abelian monoids and objects of preadditive categories, [14] and [15]. Related conditions have been considered on nonabelian groups, in some cases again defined in terms of ho- momorphisms from direct products [22] [23], in others, in terms of homomorphisms from certain completions of free products [24] [17] [18]. We remark that for abelian groups and other structures whose morphisms can be added, the class of slender objects would not change if, in the definition, we restricted attention to surjective homomorphisms Q Ai → B, since if f : Q Ai → B is a nonsurjective map witnessing the failure of B to be slender, there is an obvious surjective map B × Q Ai → B which does the same. A similar observation applies to the version of slenderness in nonabelian groups defined using the "complete free products" of [17] -- but not to the one defined using direct products [22]. Hence if one defines a condition like slenderness for nonabelian groups, but based on surjective maps from direct products, one can expect to find a larger class of examples than the ordinary slender groups, and probably techniques and results close to those of this note; cf. next to last paragraph of §11.3. (It is not clear to us whether the class of groups defined similarly in terms of maps from the "unrestricted free products" of [24], [18] would similarly grow if one imposed this condition only on surjective maps from those groups.) In [19], some implications among conditions on homomorphisms AI → B are studied for algebras A and B in the general sense of universal algebra, the cases of slender abelian groups on the one hand, and of discriminator algebras on the other, being noted. In this section we have, for simplicity, limited the results quoted to the countable-index-set case; though in the works cited, what is in question is generally whether the index set is smaller than all uncountable measurable cardinals. 12. Examples. In earlier sections, we noted some examples in passing. Here we give further, mostly lengthier examples, for which we did not want to interrupt the development of our earlier results. As in §11, the subsections below are independent of one another. The only dependence on that section is that §12.5 below assumes §11.4 above. 12.1. Idempotent algebras with Z(A) 6= {0}. We noted following Proposition 9 that a unital algebra A necessarily satisfies Z(A) = {0}. Is the same true of idempotent algebras -- perhaps subject to some additional conditions? An easy example shows that this need not even hold in finite-dimensional algebras of idempotence rank 1. Let H be the R-algebra of quaternions, let Im : H → H be the "imaginary part" map, a + bi + cj + dk 7→ bi + cj + dk, and let A be H under the nonassociative multiplication x ∗ y = Im(x) Im(y) (where the right- hand side is evaluated using the ordinary multiplication of H). Note that if we call the real and imaginary parts of an element of H its "scalar" and "vector" components, then x ∗ y has for scalar component the negative of the dot product of the vector components of x and y, and for vector component the cross product of those same vectors. Now it is not hard to see geometrically that for any scalar a and vector bi + cj + dk, one can find two vectors with dot product −a and cross product bi + cj + dk. This gives the asserted idempotence of our algebra. On the other hand, clearly, Z(A) = R 6= {0}. One can get an infinite-dimensional example, again of idempotence rank 1, that is associative and com- mutative: Let V be any commutative valuation ring with nondiscrete valuation; thus, its maximal ideal m is idempotent of idempotence rank 1. Take a nonzero element x ∈ m, and let A = m/x m. Then A has idempotence rank 1, but the image of x lies in Z(A). We shall also see in §12.4 below, where we give examples of finite-dimensional idempotent associative algebras A of arbitrarily large finite idempotence rank, that such algebras can have Z(A) 6= {0}. 12.2. On the chain condition on almost direct factors. Our next example will show that a finitely generated (unital or nonunital) associative algebra need not satisfy the chain condition on almost direct factors; and thus that the Boolean algebra of central idempotents of a finitely generated unital associative algebra can be infinite. HOMOMORPHISMS ON INFINITE DIRECT PRODUCT ALGEBRAS, ESPECIALLY LIE ALGEBRAS 21 We begin by constructing a family of unital associative algebras Ai (i = 0, 1, . . . ) over any field k. For each i, let Ai be presented by three generators x, y, z, and the infinite family of relations (23) xynz = (cid:26) 1 if n = i 0 otherwise (n = 0, 1, . . . ). If we regard (23) as a system of "reduction rules" for expressions in x, y, z, we find that, in the terminology of [3], these have no "ambiguities" (roughly, there is no way to write down a word having subwords xymz and xynz which "overlap", and so force us to worry whether the two competing reductions may fail to lead ultimately to the same expression). Moreover, application of one of these reduction rules to a word in x, y and z yields at most a shorter word, so the process of recursively applying these rules always terminates. Hence, by [3, Theorem 1.2], each Ai has for k-basis the set of words in x, y, z (including the empty word, 1, since at the moment we are considering unital k-algebras) having no subwords of the form xynz. In particular, the empty word 1 belongs to this basis, so 1 6= 0 in each of these algebras. Now within the product algebra Qi=0,1,... Ai, let x be the element whose coordinate in each Ai element x ∈ Ai, define y and z analogously, and let A ⊆ Q Ai be the nonunital subalgebra generated by these three elements. Then for each n, the element xynz ∈ A will have 1 in the n-th coordinate and 0 in all others, and so be a central idempotent. It follows that the elements xz, xz + xyz, . . . , Pn m=0 xymz, . . . constitute an infinite ascending chain of central idempotents, yielding an infinite ascending chain of almost direct factors. is the If, instead, we take for A the unital algebra generated by these same three elements, we get a finitely generated unital associative algebra whose Boolean algebra of central idempotents is infinite. We remark that in (23), for conceptual simplicity, we used algebras Ai such that the sets {n xynz = 1} were singletons; but for every subset J of the natural numbers, the same structure result applies to the algebra AJ obtained by setting xynz to equal 1 if n ∈ J, and 0 otherwise. Applying the above construction to appropriate families of these algebras, we can get a 3-generator algebra A whose Boolean algebra of central idempotents is any countable Boolean algebra. 12.3. The need for k to be a field in Lemma 17. In Lemma 17, assuming k a field, we gave a necessary and sufficient condition for a homomorphism f : A → B of k-algebras to be approximable modulo Z(B) by a homomorphism f1 : A → B annihilating a given ideal C ⊆ A. The following example shows that the condition given there fails to be sufficient if, instead, k is any integral domain that is not a field (or more generally, if k is a commutative ring that is not von Neumann regular). Lemma 30. Let k be a commutative ring having an element c such that c /∈ c2k. Let B be the free k/c2k-module on one generator x, with the (associative, commutative) multiplication defined by x2 = cx; let A0 be the free k/c2k-module on one generator y, with the zero multiplication; let A = B × A0 as k-algebras, and let f : A → B be the projection onto the first factor. Then the k-submodule C of A generated by cx − cy is an ideal that satisfies condition (ii) of Lemma 17, but not condition (i). Proof. Observe first that cx − cy ∈ Z(A), hence the k-submodule C that it generates is indeed an ideal, and its image under f lies in Z(B). Note also that if an element d(cx − cy) (d ∈ k) of this ideal lies in AA, then its A0-component dcy must lie in A0A0 = {0}, hence since the subrings A0 = ky and B = kx are isomorphic as k-modules, we also have dcx = 0, so the given element d(cx − cy) is zero. Thus C ∩ AA = {0}, so our example satisfies condition (ii) of Lemma 17. Now suppose there were a homomorphism f1 as in condition (i) of that lemma. Lemma 4(ii) tells us that f − f1 annihilates AA, which contains x2 = cx, so f1 agrees with f at cx, i.e., it fixes that element. But by assumption, f1 annihilates cx − cy ∈ C, so we must also have f1(cy) = cx. Now writing f1(y) = ax ∈ B, the above equation becomes c(ax) = cx. Applying this twice, we get cx = acx = a2cx = a2x2 = (ax)2 = f1(y)2 = f1(y2) = f1(0) = 0, a contradiction. (Intuitively, the algebra structures on kx and ky are too different for there to be a nice choice of f1 annihilating cx − cy.) (cid:3) 12.4. Unbounded idempotence rank. Lemma 23 tells us that a product A = QI Ai of finite-dimensional idempotent algebras will fail to be idempotent if the idempotence ranks of those algebras are unbounded; but we have seen that in most characteristics, finite-dimensional simple Lie algebras all have idempotence 22 GEORGE M. BERGMAN AND NAZIH NAHLUS rank ≤ 2, and even in the remaining two characteristics, one may hope that the the same is true. Can we get any examples of finite-dimensional idempotent algebras with arbitrarily large finite idempotence ranks? We give below three classes of such examples: for associative algebras, for Lie algebras, and for (nonas- sociative non-Lie) simple algebras, respectively. (Our descriptions of the first two constructions also record the fact that Z(A) is nontrivial, giving examples mentioned in the last paragraph of §12.1. let A be the k-algebra with underlying vector space Lemma 31. For any field k and positive integer i, the space Mi(k) of i × i matrices over k, and multiplication " ∗" expressed, in terms of the ordinary multiplication of these matrices, by a ∗ b = a e11 b. Then A is an associative idempotent algebra with idp-rk(A) = i. Here Z(A) is spanned over k by {emn 2 ≤ m, n ≤ i}, hence is nonzero if i ≥ 2. Proof. It is easy to verify that for any element e of any associative algebra A0, the operation a ∗ b = a e b is again associative. (When e is not a right zero divisor, the resulting algebra A is isomorphic as an algebra to the right ideal A0e ⊆ A0, via the map a 7→ ae.) So our A is an associative k-algebra. To verify idempotence, we note that each basis element emn is a product, em1 ∗ e1n. Now recall that the rank, as a matrix, of any product matrix S T in Mi(k) is less than or equal to each of rank(S) and rank(T ). Hence a product a ∗ b = a e11 b has rank ≤ rank(e11) = 1 as a matrix. Thus, to get a matrix of rank i, such as the identity matrix, we need at least i summands. To show that i summands always suffice, recall that every matrix of rank 1 can be written uv for some column matrix u and row matrix v. If we embed u as the first column of a matrix u′ ∈ A, and v as the first row of some v′ ∈ A, we see that u′ ∗ v′ = uv. Since every i × i matrix w is a sum of at most i rank-1 matrices (e.g., the i matrices that agree in one column with w, and have zeroes everywhere else), every matrix w is the sum of i products in A. It follows immediately from the definition of our multiplication that the elements emn with 2 ≤ m, n ≤ i are in Z(A), and it is easy to see that any element not in the span of these elements is sent to a nonzero value either by left or right multiplication in A by e11, giving the asserted description of Z(A). (cid:3) Here is the closely related Lie example, though we will not attempt to determine its idempotence rank and total annihilator ideal as precisely as in the above case. Lemma 32. For any field k of characteristic not 2, and any integer i ≥ 2, let A be the k-algebra with underlying k-vector-space the subspace of Mi(k) consisting of all matrices in which the coefficients of e11 and of e22 sum to zero, and with operation given by (24) [a, b] = a (e11 + e22) b − b (e11 + e22) a. Then A is an idempotent Lie algebra with i/4 ≤ idp-rk(A) ≤ i + 3. (Thus, taking i sufficiently large, we get arbitrarily large idempotence ranks.) Z(A) contains all elements {emn 3 ≤ m, n ≤ i}, hence is nonzero if i ≥ 3. Proof. By the general observation with which we began the proof of the preceding lemma, Mi(k) is an associative algebra under the multiplication a ∗ b = a (e11 + e22) b; hence it becomes a Lie algebra under the corresponding commutator bracket operation (24). It is easy to check that the set of basis elements emn in which one or both of m, n are ≥ 3 spans a 2-sided ideal in the above associative algebra structure. (In a product a∗b, the only way such a basis element occurring in a or b can lead to a nonzero term of the product is when any index ≥ 3 is "facing away from" the factor e11 + e22 in the definition of our multiplication; hence such an index survives in every term of the product.) Consequently, in examining the coefficients of e11 and e22 in a commutator [a, b] = a ∗ b − b ∗ a, we can without loss of generality assume that all elements emn occurring in the expressions for a and b have both subscripts in {1, 2}. Thus, we are reduced to computing in M2(k), and there our multiplication is the ordinary multiplication, hence our brackets are ordinary commutator brackets, and we know that the value of any such bracket has trace zero. So the range of our bracket operation on Mi(k) contains only matrices in which the coefficients of e11 and e22 sum to zero, hence the set of matrices with that property indeed forms a Lie algebra A. This Lie algebra contains the simple, hence idempotent, Lie subalgebra sl2(k), so to show A is idempotent, it will suffice to show that the range of the bracket also contains all matrix units emn with at least one of HOMOMORPHISMS ON INFINITE DIRECT PRODUCT ALGEBRAS, ESPECIALLY LIE ALGEBRAS 23 m, n ≥ 3. For n ≥ 3, we have (25) e1n = [e11 − e22, e1n]. Elements e2n, em1 and em2 are obtained by obvious variants of this calculation, while if both m and n are ≥ 3, we have (26) emn = [em1, e1n], completing the proof of idempotence. Every bracket [a, b] is by definition a difference of two matrices each of which, under the ordinary matrix multiplication of Mi(i), has an internal factor e11 + e22, hence both of which have rank ≤ 2. Thus, [a, b] has rank ≤ 4, so at least i/4 summands are needed to get an element of rank i. This gives our lower bound on idp-rk(A). To get the upper bound, note first that if in (26) we hold n fixed, and taken an arbitrary k-linear combination of the resulting equations for all m ≥ 3, we get, as a single commutator [x, e1n], an arbitrary column in position n ≥ 3 having top two components zero. Thus, summing i − 2 such commutators, we can get any matrix living in the lower right-hand i − 2 × i − 2 block of Mi(k). Linear combinations of the equations (25), and of the three variants mentioned following it, show that with four more commutators, we can fill in everything but the upper left 2 × 2 block of a general member of A. Since every element of sl2(k) is a commutator, we can fill in that block in one more step; so every member of A is a sum of (i − 2) + 4 + 1 = i + 3 commutators. The final sentence of the lemma follows from the observation that the elements emn with 3 ≤ m, n ≤ i lie in the total annihilator ideal of the associative multiplication a(e11 + e22)b, hence a fortiori in the total annihilator ideal of our Lie bracket. (cid:3) The final example of this group, giving nonassociative, non-Lie, finite-dimensional simple algebras of unbounded idempotence rank, plays further changes on the idea of a multiplication whose outputs are rank-1 matrices. Lemma 33. For any field k and positive integer i, let A be the k-algebra with underlying k-vector-space ki × Mi(k), and with multiplication defined as follows: -- For u, v ∈ ki, written as row vectors, u ∗ v is the matrix uT v ∈ Mi(k), where T denotes transpose. -- For S, T ∈ Mi(k), S ∗ T is the vector in ki whose m-th entry is the (m − 1)-st main-diagonal entry of the ordinary matrix product S T. Here we treat subscripts cyclically, so that for m = 1, the "(m − 1)-st" main-diagonal entry means the i-th. -- Products, in either order, of a member of ki and a member of Mi(k) are zero. Then idp-rk(A) = i, and A is simple. Proof. In the product of any two members of A, the Mi(k)-component will be a matrix uT v (u, v ∈ ki), hence will have rank ≤ 1; so at least i such products must be summed to get elements whose Mi(k)- components have rank i. To get an arbitrary element (v, S) as a sum a1 ∗ b1 + · · · + ai ∗ bi, one first selects, as the ki-components of a1, b1, . . . , ai, bi, pairs of vectors having products, under our multiplication, summing to S. In all but one of these pairs, one then takes the Mi(k)-components zero, and in the remaining pair, one takes for those components matrices S and T such that S ∗ T is the desired first component v. Thus, idp-rk(A) = i. To show that A is simple, let C be a nonzero ideal. Then C either contains an element with nonzero ki-component, or an element with nonzero Mi(k)-component. In the former case, the square of the element in question will have nonzero Mi(k)-component, so in either case C contains an element of the latter sort; say (v, S). Suppose the matrix S has nonzero (m, m′) entry, which we may assume without loss of generality is 1. Let us form the product (v, S) ∗ (0, em′m). This will have Mi(k)-component zero; to determine its ki- component, note that the only nonzero main-diagonal component of S em′m is a 1 in the m-th position. So by our description of products S ∗ T, the element S ∗ em′m ∈ ki will be the vector fm+1 with a 1 in the (m+1)-st position and zeroes elsewhere. So C contains (fm+1, 0). Squaring this, we get (0, em+1,m+1), and squaring that in turn gives (fm+2, 0). Repeating this process i times (and recalling that in this computation, subscripts are treated cyclically), we get all of (f1, 0), . . . , (fi, 0). 24 GEORGE M. BERGMAN AND NAZIH NAHLUS Taking linear combinations of these gives all elements (v, 0); multiplying pairs of such elements, and adding together families of i such products, gives all elements (0, S); adding these two sorts of elements we get all of A, completing the proof of simplicity. (cid:3) 12.5. Idempotence rank and base change. We will now give an example showing that the idempotence rank of a finite-dimensional algebra can go down (or up) under extension of base field. However, our example will be non-simple and non-Lie. We arrived at this example by looking, first, for an example of this sort for the invariant max-rank(m) of a map m : A ⊗k B → C of vector spaces over a field k, defined in (20), a generalization of the idempotence rank of an algebra. We wondered whether we could find such a map m for a general field k, which, when restricted to tensors of rank ≤ 1, would be surjective if k was the complex numbers, but such that over the real numbers, the range would be constrained by inequalities in the coordinates; so that on passing from the reals to the complexes, the value of max-rank(m) would drop from a larger value to 1. A little thought shows that for this to happen, A and B must each be at least 2-dimensional, so we tried A = B = k2. The tensors of rank ≤ 1 within k2 ⊗k k2 can be pictured as the matrices of rank ≤ 1 in M2(k), a set with three degrees of freedom, suggesting that a linear image of this set in k3 might have the desired properties. It turns out that if we map a 2 × 2 matrix ((amn)) to the 3-tuple consisting of its upper-right and lower-left entries, and its trace, this has the desired properties. Indeed, for a 3-tuple (a12, a21, t) of elements of k to arise in this way from a matrix ((amn)) of rank ≤ 1, the entries a11 and a22 of the latter matrix must have product a12a21 (to make the determinant a11a22 − a12a21 zero) and must sum to t (by definition of the trace). But two elements of k having sum t and product a12a21 must be the roots of the quadratic polynomial x2 − tx + a12a21. Over the complexes, such a polynomial always has roots, but it will have roots over the reals only when the inequality t2 − 4 a12a21 ≥ 0 holds. To embody this idea in an idempotent algebra, let k be any field of characteristic not 2, and A the k-algebra with underlying vector space k3, and multiplication (27) (a, b, c) ∗ (a′, b′, c′) = (ab′, ba′, aa′ + bb′). Note that the components of the product are the (1, 2) entry, the (2, 1) entry, and the trace, of the 2 × 2 rank-1 matrix (a, b)T (a′, b′). It thus follows from the preceding observations that an element (r, s, t) ∈ A is a product in A if and only if t2 − 4 rs is a square in k. Hence, if k is algebraically closed (or even if it is quadratically closed, i.e., if every element of k is a square), we get idp-rk(A) = 1. Conversely, we see that if k is not quadratically closed, idp-rk(A) will not be 1. Rather, it turns out that it is 2, since any element (r, s, t) can be written (r, 0, t) + (0, s, 0), and each of these summands satisfies our criterion for being a product in A. In particular, if we construct the above algebra over the field of reals, and then extend scalars to the complexes, the idempotence rank drops from 2 to 1. If, inversely, we start with a quadratically closed field k, and extend scalars to a non-quadratically closed field K ⊇ k, then idp-rk(A) will increase from 1 to 2. (If k is algebraically closed, we must, of course, take K transcendental to get a proper extension. However, a general quadratically closed field k can have finite algebraic extensions K which are not quadratically closed [27, Corollary 7.11(1)].) (Incidentally, our use of the term "quadratically closed", defined above, follows [27], but is distinct from the usage in [28, p.462, Exercises 8-9], where it means that for each c ∈ K, one of c or −c is a square. Evidently, one definition is modeled on the properties of a subfield of C closed under taking square roots, the other on the properties of a square-root-closed subfield of R.) 12.6. Lie examples based on inseparability. To build examples of simple Lie algebras in positive char- acteristic which misbehave under change of base field, let us start with examples that don't misbehave. Namely, (28) Let k0 be a field of characteristic p > 0, and L0 a finite-dimensional simple Lie algebra over k0, such that for every extension field k of k0, the Lie algebra L0 ⊗k0 k is again simple. For instance, we can take L0 = sln(k0) for any n ≥ 2 relatively prime to p. Recall now that if K is a finite inseparable extension of a field k, then the commutative ring K ⊗k K has nilpotent elements. (This is easily seen if K is purely inseparable, and so has an element x /∈ k with xp ∈ k : then (x ⊗ 1 − 1 ⊗ x)p = xp − xp = 0. To see the general case, recall [28, Proposition V.6.6, p.250] that there will exist a separable subextension F ⊆ K over which K is purely inseparable. By the preceding observation, K ⊗F K has nilpotent elements; but that ring is a homomorphic image of K ⊗k K, and if a HOMOMORPHISMS ON INFINITE DIRECT PRODUCT ALGEBRAS, ESPECIALLY LIE ALGEBRAS 25 homomorphic image of a finite-dimensional algebra over a field has nilpotents, the original algebra must also have them.) From this we get Lemma 34. For k0 and L0 as in (28), let k ⊆ K be extension fields of k0, with K a finite inseparable extension of k, and let L = L0 ⊗k0 K, regarded as an algebra over k ⊆ K. Then L is a finite-dimensional simple Lie algebra over k, but L ⊗k K is not semisimple (i.e., it has a nonzero nilpotent ideal ). Proof. L is clearly a finite-dimensional Lie algebra over k, and is simple by (28). Note that L ⊗k K = (L0 ⊗k0 K) ⊗k K ∼= L0 ⊗k0 (K ⊗k K). Since K is inseparable over k, we can find (cid:3) a nonzero nilpotent element ε ∈ K ⊗k K. Thus, L0 ⊗ εK is a nilpotent ideal of L0 ⊗k0 (K ⊗k K). Recall next that in the proof of Theorem 26, we were able to pull the property L = [x1, L] + [x2, L] down from the case of a field K to that of an infinite subfield k. The next lemma shows that the condition of being generated as an algebra by two elements, from which we proved that property, cannot be pulled down in that way. Though as is well-known, any finite separable extension field K of a field k can be generated over k by a single element (the Theorem of the Primitive Element [28, Theorem V.6.6, p.243]), we shall use in our construction the fact this fails arbitrarily badly for inseparable extensions. For instance, if we take for k , . . . , t1/p a pure transcendental extension k0(t1, . . . , tN ) of k0, then the degree-pN extension K = k(t1/p N ) cannot be generated over k by fewer than N elements [38, Theorem 8.6.4]. 1 Lemma 35. Given k0 and L0 as in (28), let d = dimk0 (L0), let n be any positive integer, and let k ⊆ K be extension fields of k0, such that K is finite over k, but cannot be generated over k by fewer than nd + 1 elements. Again, let L = L0 ⊗k0 K, regarded as a finite-dimensional simple Lie algebra over k ⊆ K. Then L cannot be generated as a Lie algebra over k by fewer than n + 1 elements. Proof. Let B = {b1, . . . , bd} be a k0-basis for L0. Then B will likewise be a K-basis for L. Given n elements x1, . . . , xn ∈ L, their expressions in terms of that basis will involve d n coefficients in K. By assumption, d n elements cannot generate K over k, so those coefficients lie in a proper subextension F ⊆ K. Thus x1, . . . , xn lie in L0 ⊗k0 F, a proper k-subalgebra of L0 ⊗k0 K = L. (cid:3) 13. Some questions, and some directions for further study. 13.1. Can our cardinality restrictions be weakened? In the main results of this paper, we have assumed the field k infinite. Some of those results remain formally true -- but become trivial -- for finite k : the hypothesis that card(I) be less than any measurable cardinal > card(k) then says that I is finite. (Recall that under the definition we are following, ℵ0 is a measurable cardinal.) In Theorem 11 and Corollary 12, we assumed, slightly more generally, that either card(I) or the supremum of the dimensions of all the Ai was less than all measurable cardinals greater than card(k). What this would say for finite k is that either card(I) is finite, or the dimensions of the Ai have a common finite bound. Under this assumption, the conclusions of those two results follow easily from Proposition 8 and the well- known fact that any ultraproduct of finite algebraic structures (with only finitely many operations), of bounded cardinalities, is isomorphic to one of those structures. What we would like to know, of course, is Question 36. Suppose k is a finite field, and f : QI Ai → B a homomorphism of k-algebras, with I infinite, and no common finite bound assumed on the k-dimensions of the Ai. Do some or all of the main results of this paper (other than Theorem 22) have versions valid for this case? (Or can some other results in the same spirit be established? ) We have excluded Theorem 22 because, as mentioned, a result on nilpotent algebras with no condition that k be infinite is indeed proved in [4]. (Using that result, and the close relationship between nilpotence and solvability for finite-dimensional Lie algebras in characteristic 0, an analog of Theorem 21 for that particular case is also obtained in [4].) Another sort of size restriction in our results on homomorphisms QI Ai → B concerned the object B. Here some restriction is needed, since if we allowed B to be QI Ai, the identity map of that algebra would be a counterexample to most of our results. But it is not clear that the conditions need to be as strong as those we have used. For instance 26 GEORGE M. BERGMAN AND NAZIH NAHLUS Question 37. Does Theorem 19 remain true if we delete the hypothesis that B satisfy chain condition on almost direct factors? In [6, Theorem 9(i-ii)] we indeed prove results like Theorem 19 without the chain condition hypothesis -- but having, instead, restrictions such as card(I) ≤ card(k). So we want to know whether we can do without either sort of condition. A result in which the condition on the codomain algebra might be weakened in a different way is Theo- rem 11 above, where the codomain is assumed both simple (much stronger than having chain condition on almost direct factors) and countable-dimensional, and we do not know whether the latter condition can be dropped. We also don't know, in that case, whether we need the restriction on the size of the algebras Ai or the index-set I relative to measurable cardinals. So we ask Question 38. If (Ai)i∈I is a family of algebras over an infinite field k, such that no Ai admits a homo- morphism onto a simple algebra, can QI Ai admit a homomorphism onto a simple algebra? Let us note that in the existing proof of Theorem 11, the dimension-restriction on the codomain can be slightly weakened: Using the full strength of Theorem 46, we see that if k has cardinality > ℵ1, we get the indicated nonexistence result (with the parenthetical generalization in the first sentence appropriately adjusted) not just for homomorphisms onto simple algebras that are countable-dimensional, but onto the larger class of simple algebras of k-dimension < card(k). Concerning the hypothesis in that result that the dimensions of the Ai be less than any measurable cardinal > card(k), we wonder whether one might be able to remove this by showing that simple algebras Ai of such large dimensions can be replaced by simple subalgebras of smaller dimensions, without affecting the desired properties. In [4, §8], examples are given of homomorphic images B of inverse limits A of nilpotent algebras in which B has various properties that inverse limits of nilpotent algebras cannot themselves have; e.g., an associative example where B contains a nonzero element y such that y ∈ ByB, and a nonassociative example where B contains an element such that y2 = y. However, the analog of Question 38 for inverse limits of nilpotent is open; it is part of [4, Question 23]. 13.2. On idempotence rank. The next question poses the problem that we skirted by obtaining our bound on idempotence ranks of simple Lie algebras using [9] instead of [11]. Question 39 (also asked in [25, pp.652-653] for k = R ). Does every finite-dimensional simple Lie algebra L over an infinite field k have idempotence rank 1 ? 13.3. On the chain condition on almost direct factors. In §12.2 we saw that a finitely generated associative algebra over a field need not have chain condition on almost direct factors. On the other hand, a finitely generated commutative associative algebra over a field is Noetherian, and so does have that chain condition. Question 40. Does every finitely generated Lie algebra over a field have chain condition on almost direct factors? 13.4. Idempotence rank, number of generators, and base change. The algebras of §12.5, whose idempotence ranks could increase and decrease under base change, were neither associative nor Lie. Also, the motivating idea of that example -- an image-set which, when the base field is R, is constrained by inequalities, but which is not so constrained when the base field is C -- only seems to lead to examples where the idempotence rank changes by 1. So we ask Question 41. (a) Do there exist finite-dimensional associative or Lie algebras over an infinite field whose idempotence ranks change under base extension? (b) Do there exist finite-dimensional algebras of any sort over an infinite field whose idempotence ranks change by more than 1 under base extension? In [5, §2], examples are given of finite-dimensional (nonassociative, non-Lie) algebras over finite fields whose idempotence ranks change by arbitrarily large amounts under base extension, and of an infinite- dimensional commutative associative algebra over R whose idempotence rank goes down (though only by 1) on extension of scalars to C. In the same vein is HOMOMORPHISMS ON INFINITE DIRECT PRODUCT ALGEBRAS, ESPECIALLY LIE ALGEBRAS 27 Question 42 (J.-M. Bois, personal communication). Let k be a finite field, K its algebraic closure, and L a finite-dimensional Lie algebra over k. If L ⊗k K can be generated over K by two elements, can L be generated over k by two elements? For instance, is sln(k) generated over k by two elements for all n ≥ 2 relatively prime to char(k) ? 13.5. Centroids to the rescue for our inseparable-extension examples? In §12.6, where we con- structed simple Lie algebras in positive characteristic that "misbehaved" under base change, the trick was to treat them as having base field k, though they were Lie algebras over a larger field K, which was inseparable over k. As K-algebras, they are well-behaved. If L is a finite-dimensional simple Lie algebra over a field k, the largest field K to which the Lie structure extends, called the centroid of L, consists of the k-linear endomorphisms ϕ of L which respect all the adjoint maps [x, − ] (x ∈ L), i.e., which satisfy ϕ([x, y]) = [x, ϕ(y)] for all x, y ∈ L [26, p.290]. If K = k, L is called a central simple Lie algebra. Thus, every finite-dimensional simple Lie algebra over a field is a central simple Lie algebra over its centroid. It is plausible that if we of look at our Lie algebras as algebras over their centroids, the kind of misbehavior obtained in §12.6 will not occur: Question 43. Let L be a finite-dimensional central simple Lie algebra over a field k of characteristic not 2 or 3. (a) Can L be generated over k by two elements? (b) Will L ⊗k K be a direct product of simple Lie algebras over K for all extension fields K of k ? (c) If either of the above questions has a positive answer, does it remain so if rather than assuming L central, we merely assume the centroid of L to be separable over k ? 13.6. Characteristics 2 and 3 . Of course, we would like to know Question 44. Does the conclusion of Theorem 27(i) hold in the excluded characteristics, 2 and 3 ? Perhaps, when the structure theory is extended to those last two characteristics, the result of [9] that we used in the proof will also go over, yielding an affirmative answer. On the other hand, a weaker result than that of [9], perhaps asserting generation by 3 or 4 elements rather than 2, might be easier to prove than the optimal result, and might not require a full structure theory. We give the last three points of this section as topics to be investigated, rather than formal questions. 13.7. Variant formulations of solvability and nilpotence. Of the two versions of our result on homo- morphic images of direct products of solvable Lie algebras, we got the first, Corollary 12, using the criterion that a finite-dimensional Lie algebra over a field of characteristic 0 is solvable if and only if it admits no homomorphism onto a simple Lie algebra, while for the second, Theorem 21, we used the characterization of solvability (in any characteristic) by a disjunction of identities. Thus, our results on Lie algebras were obtained as cases of two different results on general algebras. There are other elegant characterizations of solvability of a finite-dimensional Lie algebra L : in arbitrary characteristic, the condition that L have no nontrivial idempotent subalgebra; in characteristic 0, either the condition that the ideal [L, L] be nilpotent (which is sufficient but not necessary in general characteristic), or that L contain no simple subalgebra (necessary, but not sufficient in general characteristic). Likewise, finite-dimensional nilpotent Lie algebras L can be characterized among finite-dimensional Lie algebras in other ways than the one used in Theorem 22: as those with no nonzero ideals C such that [L, C] = C, as those with no nonzero elements x such that x belongs to the ideal generated by [L, x], and as those whose nonzero homomorphic images M all have Z(M ) 6= {0}. It might be of interest to examine how some of these conditions on general algebras behave under homo- morphic images of direct products. 13.8. Semisimple Lie algebras in positive characteristic. A Lie algebra L is called semisimple if it has no nonzero abelian ideal. If the base field k has characteristic zero, the finite-dimensional semisimple Lie algebras are just the finite direct products of simple Lie algebras, so our results on homomorphic images of direct products of simple Lie algebras imply the corresponding statements for products of semisimple Lie algebras. When the base field has positive characteristic, a finite-dimensional semisimple Lie algebra need not be a direct product of simple Lie algebras [36, p.133, top paragraph]. The present authors know nothing about their structure. 28 GEORGE M. BERGMAN AND NAZIH NAHLUS In particular, what Lie algebras are homomorphic images of finite-dimensional semisimple Lie algebras? It is conceivable that all are. (By analogy, every finite group G is indeed a homomorphic image of a finite group having no abelian normal subgroup, namely, a wreath product of G with a finite simple group.) If so, then little can be said about homomorphic images of infinite products of such algebras -- though something might be said about homomorphisms from infinite products onto semisimple Lie algebras. These seem to be questions for the expert in Lie algebras over fields of positive characteristic. An introduction to the subject is [37]. 13.9. Restricted Lie algebras, and other algebras with additional structure. For k a field of positive characteristic p, a restricted Lie algebra or p -Lie algebra over k is a Lie algebra given with an additional operation, x 7→ x(p), satisfying certain identities which, in associative k-algebras, relate the p -th power map with the k-module structure and commutator brackets [26, §5.7]. (The concept can be motivated by the observation that in characteristic p, the set of derivations of an algebra A is closed, in the associative algebra of k-vector-space endomorphisms of A, not only under the vector space operations and commutator brackets, but also under taking p -th powers.) Thus, p -Lie algebras are not algebras as we define the term in §1. Of course, they have Lie algebra structures, and one can apply our results to those structures. But a p-Lie algebra may, for instance, be simple under its p -Lie algebra structure without being simple under its ordinary Lie algebra structure. We leave to others the investigation of homomorphic images of infinite products of p -Lie algebras, and, generally, of algebras with additional operations. 14. Appendix: Review of ultrafilters and ultraproducts. We recall here some standard definitions and notation (e.g., cf. [13, p.211ff ]). A filter on a nonempty set I means a family F of subsets of I such that (29) J1 ⊇ J2 ∈ F =⇒ J1 ∈ F , J1, J2 ∈ F =⇒ J1 ∩ J2 ∈ F . In view of the first condition, a filter F is proper (not the set of all subsets of I) if and only if ∅ /∈ F . A maximal proper filter is called an ultrafilter ; by Zorn's Lemma, every proper filter is contained in an ultrafilter. It is easy to verify that a proper filter U is an ultrafilter if and only if for every J ⊆ I, either J ∈ U or I − J ∈ U. If F is a filter (in particular, if it is an ultrafilter) on I, one says that a subset J ⊆ I is F -large if J ∈ F . This does not save much ink, but does help with the intuition of the subject. If (Ai)i∈I is a family of nonempty sets, and F is a filter on the index set I, then the reduced product QI Ai / F is the factor-set of QI Ai by the equivalence relation that identifies elements (ai) and (a′ i) if {i ai = a′ i} is F -large. If all the Ai are furnished with operations making them groups, k-algebras, etc., then this structure can be seen to carry over to their reduced product, making the natural map QI Ai → it is not hard to see that QI Ai /F ∼= QJ Ai /FJ , where QI Ai /F a homomorphism. For any J ∈ F , FJ = {J ′ ∈ F J ′ ⊆ J}, a filter on J. (This observation depends on our assumption that all Ai are nonempty.) A reduced product of objects Ai with respect to an ultrafilter is called an ultraproduct of the Ai. An ultraproduct AI / U of copies of a single object A is called an ultrapower of A. Note that in this situation, the diagonal image of A in AI maps to an isomorphic copy of A within AI / U. It is known that an ultraproduct QI Ai / U satisfies every first order sentence s which holds on a U-large subfamily of the Ai; i.e., for which {i ∈ I Ai satisfies s} ∈ U [13, Theorem 4.1.9(iii)]. For every i0 ∈ I, the filter of all subsets of I containing i0 is called the principal ultrafilter determined by i0. The ultraproduct of the Ai with respect to that ultrafilter is, up to isomorphism, just Ai0 . If I is finite, these are the only ultrafilters on I; if I is infinite, on the other hand, then the cofinite subsets of I form a proper filter, so there are ultrafilters containing this filter, the nonprincipal ultrafilters. We note, for perspective, that filters on I correspond to ideals in the Boolean algebra of subsets of I, by mapping each filter to the ideal of complements of its members. The ultrafilters correspond to the maximal ideals, which in this case are the same as the prime ideals. More generally, for any family of fields (ki)i∈I , the ideals of QI ki correspond to the filters on I, each filter F yielding the ideal of all elements having F -large zero-set; in other words, the kernel of the map from QI ki to the reduced product QI ki / F . Again the ultrafilters correspond to the maximal ideals, and these coincide with the prime ideals. (The statements HOMOMORPHISMS ON INFINITE DIRECT PRODUCT ALGEBRAS, ESPECIALLY LIE ALGEBRAS 29 about Boolean algebras are essentially the cases of the statements about products of fields in which all ki are the two-element field.) 15. Appendix: κ-complete and non-κ-complete ultrafilters. Definition 45 ([13, p.227]). Let κ be an infinite cardinal. Then an ultrafilter U on a set I is said to be κ-complete if it is closed under intersections of families of fewer than κ members. An ℵ1-complete ultrafilter (i.e., one closed under countable intersections) is called countably complete. An infinite cardinal κ is called measurable if there exists a nonprincipal κ-complete ultrafilter on κ. (We follow [13] in this definition. Many authors, restrict the term "measurable cardinal" to the case where κ is uncountable, e.g., [16, p.177]. We shall, rather, explicitly write "uncountable measurable cardinal" when that is intended.) Note that the definition of an ultrafilter makes it ℵ0-complete (closed under finite intersections), so the weakest completeness condition not automatically satisfied is countable completeness. Note also that ℵ0 is, under the above definition, a measurable cardinal, since there exist nonprincipal ultrafilters on it. It is known [13, Proposition 4.2.7] that for any nonprincipal ultrafilter U on any index set I, there is a largest cardinal κ such that U is κ-complete, and that this will be a measurable cardinal. Moreover, if an uncountable measurable cardinal exists, it must be "enormous" in many respects. In particular, truncating set theory to exclude it and all larger cardinals will leave a smaller set theory that still satisfies the standard axiom system ZFC; hence, if ZFC is consistent, so is ZFC together with the statement that there are no uncountable measurable cardinals, and therefore no nonprincipal countably complete ultrafilters [16, Chapter 6, Corollary 1.8]. If uncountable measurable cardinals µ do exist, then any set I admitting a nonprincipal ultrafilter U that is µ-complete for such a µ must itself have cardinality at least µ [13, Proposition 4.2.2]. It follows that every element of U must likewise have cardinality at least µ. Thus, the reader may prefer to assume that there are no uncountable measurable cardinals, or at least that the products of algebras he or she is interested in will always be indexed by sets I of less than any uncountable measurable cardinality, and so read only the first result below, which concerns non-κ-complete ultrafilters. But the subsequent results, about κ-complete ultrafilters, show that even if these occur, things still work out fairly nicely for our purposes! Note that since the sets not in an ultrafilter U are the complements of the sets in U, the condition that U be κ-complete is equivalent to saying that if a family of < κ sets Iα ⊆ I has union in U, then at least one of the Iα lies in U. For κ a cardinal, κ+ denotes the successor of κ, so that a κ+-complete ultrafilter is one closed under κ-fold intersections; equivalently, one which always contains some member of a κ-tuple of sets if it contains their union. We follow the standard convention that every cardinal κ is the set of all ordinals of cardinality < κ, hence is itself a set of cardinality κ. The least infinite cardinal, ℵ0, looked at as the set of natural numbers, is denoted ω. As mentioned in the preceding section, ultraproducts preserve first-order sentences; hence an ultraproduct of fields is not merely a ring, but a field. Our first result below says that except in the case involving large measurable cardinals, a nonprincipal ultrapower of an infinite field k is significantly larger than that field. Theorem 46. Let k be an infinite field, let κ = card(k), I, and let K = kI / U. Then the dimension [K : k] is uncountable, and is at least card(k). let U be a non-κ+-complete ultrafilter on a set Proof. Since U is not κ+-complete, we can take a family of κ sets Iα (α ∈ κ) which are not in U, but whose union is in U. Deleting from each the union of those that precede it in our indexing, we may assume that they are disjoint; and throwing in, as one more set, the complement of their union (which is not in U because their union is in U), we may assume the Iα have union I. (Some Iα may be empty.) We shall prove first that kI / U is transcendental over k. Thus, letting t be a transcendental element, the elements (t−c)−1 (c ∈ k) will be k-linearly independent, proving [K : k] ≥ card(k). If k is uncountable, this makes [K : k] uncountable. On the other hand, for countable k, we shall show that kI / U is uncountable, again implying that [K : k] is uncountable. To show kI / U transcendental over k, write k as {cα α ∈ κ}, with the cα distinct, and let t ∈ kI/ U be the image of the element c ∈ kI which has value cα everywhere on Iα for each α ∈ κ. Any nonzero 30 GEORGE M. BERGMAN AND NAZIH NAHLUS polynomial p(x) over k has only finitely many roots in k, hence its value at c is zero only on a finite union of the Iα, hence not on a member of U; so p(t) 6= 0 in kI / U, showing that t is transcendental. On the other hand, suppose k is countable, so that our hypothesis on U is that it is not countably complete. As above, take a decomposition of I into disjoint sets In /∈ U (n ∈ ω). We shall show that given any countable list of elements of kI / U, we can find an element not in that list, proving kI / U uncountable. Let the members of our list be the images in kI / U of elements a(0), a(1), · · · ∈ kI . Then we can choose an element c ∈ kI which disagrees with a(0) at each point of I0 (because k has more than one element), with both a(0) and a(1) at each point of I1 (because k has more than two elements), and so forth. We then see that for each n, the set at which c agrees with a(n) is a subset of I0 ∪ · · · ∪ In−1 /∈ U; hence the element of kI / U that c defines is distinct from each member of the given countable list, as claimed. (cid:3) When U is κ+-complete, we have a result of the opposite sort. Theorem 47. Let k be an infinite field, let κ = card(k), and let U be a κ+-complete ultrafilter on a set I. Then the ultrapower kI / U coincides with the natural isomorphic copy of k therein. In this situation, if µ is any cardinal > κ such that U is µ-complete, and (Ai)i∈I is a family of k-algebras whose dimensions have supremum < µ, then the ultraproduct QI Ai / U is isomorphic as a k-algebra to Ai1 for some i1 ∈ I, and the canonical map QI Ai → QI Ai / U ∼= Ai1 splits (is right invertible). Proof. The first paragraph follows from the case of the second where all the Ai are one-dimensional, so it suffices to prove the assertions of the latter paragraph. We note that these are immediate if U is principal, In that case, we may take µ to be the greatest cardinal such that U is so let us assume the contrary. µ-complete. Thus, µ is a measurable cardinal. We note first that for λ any cardinal, a λ-dimensional k-algebra can, up to isomorphism, be taken to have underlying vector space Lλ k, and its algebra structure will then be determined by λ3 structure constants cαβγ (α, β, γ ∈ λ), where cαβγ ∈ k is the coefficient, in the product of the α-th and β-th basis elements, of the γ-th basis element. (These are subject to the constraint that for each α and β, there are only finitely many γ with cαβγ 6= 0; but for counting purposes, this will not matter to us.) Thus, the number of isomorphism classes of λ-dimensional algebras is ≤ κλ3 . From the fact that a measurable cardinal µ is inaccessible [13, Theorem 4.2.14(i)], it follows that for any λ < µ, the cardinality of the set of isomorphism classes of k-algebras of dimension ≤ λ is also < µ. Thus, under the hypotheses of the second paragraph of our theorem, the Ai fall into < µ isomorphism classes. Hence if we partition I according to the isomorphism class of Ai, the µ-completeness of U implies that the subset I0 corresponding to some one of these classes belongs to U. Let us assume for notational convenience that all the Ai with i ∈ I0 are equal, and call their common value A(0). We now define the homomorphism ψ(a)i = 0 otherwise. (30) (31) Composing this with the canonical homomorphism ψ : A(0) → QI Ai, where ψ(a)i = a if i ∈ I0, ϕ : QI Ai → QI Ai / U, we clearly get an embedding ϕψ : A(0) → QI Ai / U. We claim that this is surjective, and hence an isomorphism. (This is one direction of [13, Proposition 4.2.4], but quick enough to prove directly.) For A(0), being < µ-dimensional, has cardinality < µ, hence given any a = (ai)i∈I ∈ AI0 (0), we may partition I0 into < µ subsets I0,b = {i ∈ I0 ai = b} (b ∈ A(0)). Again, by µ-completeness one of these lies in U; say I0,b0 . It follows that a falls together with ψ(b0) under ϕ, proving surjectivity of ϕψ : A(0) → QI Ai/U. Thus, ϕψ is an isomorphism, so in particular, ϕ splits. (cid:3) If one deletes the bound assumed above on the dimensions of the Ai, one can still show that QI Ai / U has properties very close to those algebras, as illustrated by the next two results. These are instances of [13, Theorem 4.2.11], which says that the fact quoted earlier, that ultraproducts preserve first-order sentences satisfied on U-large sets, can be strengthened, for κ-complete ultrafilters, to refer to an extended first-order language allowing conjunctions and disjunctions of all families of fewer than κ sentences. (Incidentally, in the first result below, we could replace each identity Wm = 0 with an arbitrary set of identities; but for simplicity of statement we refer to single identities, the case needed in in §7.) HOMOMORPHISMS ON INFINITE DIRECT PRODUCT ALGEBRAS, ESPECIALLY LIE ALGEBRAS 31 In the preceding theorem, the assumption that k be a field was not essential, but a wordier statement and proof would have been needed without it. The next two results are easy to state and prove for general k, so we revert to our default assumption that k is any commutative ring. Proposition 48 (cf. [13, Theorem 4.2.11]). Let (Ai)i∈I be a family of k-algebras, and U a countably complete ultrafilter on the index set I. Suppose W1 = 0, W2 = 0, W3 = 0, . . . is a countable list of identities for k-algebras, such that each Ai satisfies at least one of these identities. Then the ultraproduct QI Ai / U satisfies at least one of those identities. Proof. For m = 1, 2, . . . , let Jm be the set of i ∈ I such that Ai satisfies the identity Wm = 0. By (cid:3) Ai satisfies Wm = 0, hence so does its image, QI Ai / U. assumption, I = Sm Jm, hence since U is countably complete, there is an m such that Jm ∈ U. Hence QJm on the index set I. Then if all Ai are simple k-algebras, so is the ultraproduct QI Ai / U. Proposition 49. Let (Ai)i∈I be a nonempty family of k-algebras, and U a countably complete ultrafilter Proof. An algebra A is simple if and only if (i) A 6= {0}, and (ii) for every nonzero a ∈ A, the ideal of A generated by the set aA + Aa is all of A. Clearly, (i) carries over from the Ai to their ultraproduct. Note that (ii) means that for every nonzero a ∈ A, every b ∈ A can be represented as a sum of products of elements of A, all of length ≥ 2, in which each product includes a factor a. (We have made no reference in this statement to coefficients in k. This was our point in using products of length ≥ 2 : After selecting an instance of a in each product, we can absorb a coefficient from k, if any, into one of the other factors.) Now there are only countably many forms such an expression as a sum of products can take. Indeed, to generate a form for such an expression, one chooses the natural number that is to be the number of summands, for each summand one chooses the natural number ≥ 2 that is to be its length, and given this length, one chooses one of the finitely many positions for the factor a to appear in, and, finally, one of the finitely many ways for the (nonassociative) product to be bracketed. Now let us be given (ai) ∈ QI Ai with nonzero image in QI Ai / U, and any (bi) ∈ QI Ai. Let J = {i ∈ I ai 6= 0}. Since the image of (ai) in our ultraproduct is assumed nonzero, we have J ∈ U. For each i ∈ J, the simplicity of Ai says that we can write bi as a sum of products of lengths ≥ 2 in elements of Ai, such that each of these products has a factor ai. Choosing for each i such an expression for bi, we can now countably many such forms. By countable completeness, for some m we have Jm ∈ U. Suppose our m-th therefore witness the condition that the image of (bi) lies in the ideal generated by the image of (ai). This (cid:3) partition J as Sm∈ω Jm according to the form of this expression, since we have seen that there are only expression involves n variables other than a. Then we can choose n elements of QI Ai which, at every i ∈ Jm, represent the values used in our expression for bi. The images of these elements in QI Ai / U will proves the simplicity of QI Ai / U. within the product of matrix rings Qi∈ω Mi(k), the set of elements (ai)i∈ω (ai ∈ Mi(k)) such that the set of Q Mi(k) is immediate; closure under addition follows from the observation that if {rank(ai) i ∈ ω} is to verify that for any nonprincipal ultrafilter U on ω, the image of this ideal in Qi Mi(k) / U is a proper nonzero ideal, so unlike the Mi(k), the ultraproduct ring is not simple. (One can show that the ideals of this ring form an uncountable chain.) bounded by m and {rank(bi) i ∈ ω} by n, then {rank(ai + bi) i ∈ ω} is bounded by m + n.) It is easy The above result is not true for non-countably-complete ultrafilters. For instance, if k is any field, then of integers {rank(ai) i ∈ ω} is bounded forms an ideal. (Closure under multiplication by arbitrary elements This finishes the material required for the preceding sections of this paper. We end by showing, for completeness's sake, how part of the proof of Theorem 46 above can be strengthened. In that proof, K was an ultrapower of a field k with respect to a non-card(k)+-complete ultrafilter, and we showed the degree [K : k] to be uncountable by different methods depending on whether k was countable or uncountable: in the former case by showing that K had uncountable cardinality; in the latter, by showing that it was transcendental over k. Note that in the former situation, it follows that K has transcendence degree over k equal to its cardinality; but our argument in the latter case only proved transcendence degree ≥ 1. Can we similarly show in the second case that kI / U has transcendence degree over k equal to its cardinality? 32 GEORGE M. BERGMAN AND NAZIH NAHLUS This is immediate if card(kI / U) > card(k). To see this, note that it is easy to verify that whenever F is a transcendental extension of a field E, one has (32) card(F ) = sup(ℵ0, card(E), tr.degE(F )). Hence, If card(F ) > sup(ℵ0, card(E)), (33) So if card(kI / U) > card(k) ≥ ℵ0, we indeed get tr.degk(kI /U) = card(kI /U). then tr.degE(F ) = card(F ). Can the contrary case, card(kI / U) = card(k) (34) occur for a non-card(k)+-complete ultrafilter U ? Yes. For instance, if we choose k so that card(k) = 2card(I), then card(kI ) = (2card(I))card(I) = 2card(I) = card(k), so card(kI / U) certainly can't be larger. We sketch below a proof that we nevertheless have tr.degk(kI / U) = card(kI / U), though under a slightly stronger hypothesis that that of Theorem 46; namely, with the condition of that theorem that U be non- card(k)+-complete strengthened to "non-countably-complete". (This is strictly stronger only if card(k) is ≥ some uncountable measurable cardinal.) Proposition 50. Let k be an infinite field, and U a non-countably-complete ultrafilter on a set I. Then tr.degk(kI / U) = card(kI / U). Sketch of proof. As noted above, the result is straightforward unless (34) holds, so assume (34). Let k0 be any countable (possibly finite) subfield of k (e.g., its prime subfield), and note that since, by (34) and Theorem 46, k is uncountable, (33) shows that tr.degk0 (k) = card(k). Let {s(α) α ∈ card(k)} be a transcendence basis for k over k0. Since U is not countably complete, we can decompose I into a countable family of disjoint U-small subsets In (n ∈ ω). For each α ∈ card(k), let t(α) be the image in kI / U of the element of kI which on each In has the constant value sn (α). We claim that these t(α) are algebraically independent over k. Briefly, if they satisfied an algebraic dependence relation over k, they would satisfy such a relation over the pure transcendental subfield k0(s(α))α∈card(k). Hence, clearing denominators, we would get a polynomial relation among the t(α) with coefficients in the polynomial ring k0[s(α)]α∈card(k). If this holds in kI / U, then the corresponding equation among elements of kI must hold at points of infinitely many In, and by choosing n larger than the degrees in the s(α) of all the coefficients of the given polynomial, one gets a contradiction. This shows that the transcendence degree of kI /U over k is at least card(k), which, by (34), equals (cid:3) card(kI / U). We do not know whether in Proposition 50 the assumption that U is not countably complete can be weakened to "not card(k)+-complete". References [1] Denis Allouch, Les modules maigres, C. R. Acad. Sci. Paris S´er. A-B 272 (1971) A517 -- A519. MR 44#3992. [2] George M. Bergman, Ranks of tensors and change of base field, J. Alg. 11 (1969) 613-621. MR 39#4177. [3] [4] , The diamond lemma for ring theory, Advances in Mathematics 29 (1978) 178-218. MR 81b:16001. , Homomorphic images of pro-nilpotent algebras, to appear, Illinois J. Math., preprint, 19pp., October 2010, readable at http://math.berkeley.edu/~ gbergman/papers. [5] [6] , The condition (∃ x1, . . . , xn) A = x1A + · · · + xnA in nonassociative algebras, and base-change, unpublished note, accessible online at http://math.berkeley.edu/~gbergman/papers/unpub. and Nazih Nahlus, Linear maps on kI , and homomorphic images of infinite direct product algebras, preprint, 14pp., October 2010, readable at http://math.berkeley.edu/~ gbergman/papers. [7] M. Billis, A note on a theorem of B. H. Neumann and S. Yamamuro, Proc. Amer. Math. Soc. 22 (1969) 439 -- 440. MR 40#239. [8] M. J. Billis, Finite groups whose powers have no countably infinite factor groups, Canad. J. Math. 21 (1969) 965 -- 969. MR 39#6970. [9] Jean-Marie Bois, Generators of simple Lie algebras in arbitrary characteristics, Mathematische Zeitschrift 262 (2009) 715-741. http://arxiv.org/pdf/0708.1711. MR 2010c:17011. [10] Nicolas Bourbaki, ´El´ements de math´ematique, Fasc. XXXVIII: Groupes et alg`ebres de Lie, chapitres VII-VIII. Translated by Andrew Pressley in Lie Groups and Lie Algebras. Chapters 7 -- 9. MR 56#12077, MR 2005h:17001. [11] Gordon Brown, On commutators in a simple Lie algebra, Proc. Amer. Math. Soc. 14 (1963) 763 -- 767. MR 27#3676. [12] P. M. Cohn, Further Algebra and Applications, Springer, 2003. MR 2003k:00001. HOMOMORPHISMS ON INFINITE DIRECT PRODUCT ALGEBRAS, ESPECIALLY LIE ALGEBRAS 33 [13] C. C. Chang and H. J. Keisler, Model Theory, 3rd ed., Studies in Logic and the Foundations of Mathematics, v.73. North- Holland, 1990. MR 91c:03026. [14] Radoslav M. Dimitri´c, Slender monoids, pp.73 -- 78 in Algebra and its Applications, Contemp. Math. 419 (2006). MR 2007m:20098. [15] [16] F. R. Drake, Set Theory: An Introduction to Large Cardinals, Studies in Logic and the Foundations of Mathematics, v.76, , Slender Objects, book, in preparation. Elsevier, 1974. Zbl 0294.02034. [17] Katsuya Eda, Free σ-products and non-commutatively slender groups, J. Algebra 148 (1992) 243 -- 263. MR 94a:20040. [18] and Saharon Shelah, The non-commutative Specker phenomenon in the uncountable case, J. Algebra 252 (2002) 22 -- 26. MR 2003g:20041. [19] Andrzej Ehrenfeucht, Siemion Fajtlowicz, and Jan Mycielski, Homomorphisms of direct powers of algebras, Fund. Math. 103 (1979) 189 -- 203. MR 81b:08002. [20] Robert El Bashir and Tom´as Kepka, On when small semiprime rings are slender, Comm. Algebra 24 (1996) no. 5, 1575 -- 1580. MR 97b:16007. [21] L´aszl´o Fuchs, Infinite Abelian Groups. Vol. II, Pure and Applied Mathematics, v.36-II. Academic Press, 1973. MR 50#2362. [22] Rudiger Gobel, On stout and slender groups, J. Algebra 35 (1975) 39 -- 55. MR 51#13054. [23] R. Gobel, S. V. Richkov and B. Wald, A general theory of slender groups and Fuchs- 44 -groups, pp. 194 -- 201 in Abelian group theory (Oberwolfach, 1981 ), SLNM v.874, 1981. MR 84d:20057. [24] Graham Higman, Unrestricted free products, and varieties of topological groups, J. London Math. Soc. 27 (1952) 73 -- 81. MR 13,623d. [25] Karl H. Hofmann and Sidney A. Morris, The Lie Theory of Connected Pro-Lie Groups. A structure theory for pro-Lie algebras, pro-Lie groups, and connected locally compact groups, EMS Tracts in Mathematics, 2. European Mathematical Society, 2007. MR 2008h:22001. [26] Nathan Jacobson, Lie Algebras, Interscience, 1962; Dover, 1979. MR 26#1345, MR 80k:17001. [27] T. Y. Lam, Introduction to Quadratic Forms over Fields, A.M.S. Graduate Studies in Math., v.67, 2005. MR 2005h:11075. [28] Serge Lang, Algebra, Addison-Wesley, third edition, 1993, reprinted as Springer G.T.M., v.211, 2002. MR 2003e:00003. [29] Nazih Nahlus, On L = [L, a] + [L, b] and x = [a, b] in split simple Lie algebras, to be written, title tentative. [30] B. H. Neumann and Sadayuki Yamamuro, Boolean powers of simple groups, J. Austral. Math. Soc. 5 (1965) 315 -- 324. MR 33#5720. [31] Alexander Premet and Helmut Strade, Simple Lie algebras of small characteristic. VI. Completion of the classification, J. Algebra 320 (2008) 3559 -- 3604. MR 2009j:17013. [32] Richard D. Schafer, An Introduction to Nonassociative Algebras, Academic Press, 1966; Dover, 1995. MR 35#1643, MR 96j:17001. [33] Saharon Shelah and Oren Kolman, Infinitary axiomatizability of slender and cotorsion-free groups, Bull. Belg. Math. Soc. Simon Stevin 7 (2000) 623 -- 629. MR 2002a:03074. [34] Ernst Specker, Additive Gruppen von Folgen ganzer Zahlen, Portugaliae Math. 9 (1950) 131 -- 140. MR 12,587b. [35] Robert Steinberg, Automorphisms of classical Lie algebras, Pacific J. Math. 11 (1961) 1119 -- 1129. MR 26#1395. [36] Helmut Strade, Simple Lie algebras over fields of positive characteristic. I. Structure theory, de Gruyter Expositions in Mathematics, 38, 2004. MR 2005c:17025. [37] and Rolf Farnsteiner, Modular Lie algebras and their representations, Monographs and Textbooks in Pure and Applied Mathematics, 116. Marcel Dekker, 1988. MR 89h:17021. [38] Gabriel Daniel Villa Salvador, Topics in the Theory of Algebraic Function Fields, Birkhauser, 2006. MR 2007i:11002. (G. Bergman) University of California, Berkeley, CA 94720-3840, USA E-mail address: [email protected] (N. Nahlus) American University of Beirut, Beirut, Lebanon E-mail address: [email protected]
1306.1615	1	1306	2013-06-07T05:31:18	Real Clifford Algebra Cl(n,0), n=2,3(mod 4) Wavelet Transform	[ "math.RA", "math.RT" ]	We show how for $n=2,3 (\mod 4)$ continuous Clifford (geometric) algebra (GA) $Cl_n$-valued admissible wavelets can be constructed using the similitude group $SIM(n)$. We strictly aim for real geometric interpretation, and replace the imaginary unit $i \in \C$ therefore with a GA blade squaring to $-1$. Consequences due to non-commutativity arise. We express the admissibility condition in terms of a $Cl_{n}$ Clifford Fourier Transform and then derive a set of important properties such as dilation, translation and rotation covariance, a reproducing kernel, and show how to invert the Clifford wavelet transform. As an example, we introduce Clifford Gabor wavelets. We further invent a generalized Clifford wavelet uncertainty principle.	math.RA	math	Real Clifford Algebra Cln,0, n = 2, 3(mod 4) Wavelet Transform Eckhard Hitzer Department of Applied Physics, University of Fukui, 910-8507 Japan Abstract. We show how for n = 2,3(mod 4) continuous Clifford (geometric) algebra (GA) Cln-valued admissible wavelets can be constructed using the similitude group SIM(n). We strictly aim for real geometric interpretation, and replace the imaginary unit i ∈ C therefore with a GA blade squaring to −1. Consequences due to non-commutativity arise. We express the admissibility condition in terms of a Cln Clifford Fourier Transform and then derive a set of important properties such as dilation, translation and rotation covariance, a reproducing kernel, and show how to invert the Clifford wavelet transform. As an example, we introduce Clifford Gabor wavelets. We further invent a generalized Clifford wavelet uncertainty principle. Keywords: Clifford geometric algebra, Clifford wavelet transform, multidimensional wavelets, continuous wavelets, similitude group PACS: AMS Subj. Class. 15A66, 42C40, 94A12 MULTIVECTOR FUNCTIONS Multivectors M ∈ Clp,q, p + q = n, have k-vector parts (0 ≤ k ≤ n): scalar part Sc(M) = hMi = hMi0 = M0 ∈ R, vector part hMi1 ∈ Rp,q, bi-vector part hMi2, . . . , and pseudoscalar part hMin ∈Vn Rp,q M = (cid:229) MAeA = hMi + hMi1 + hMi2 + . . . + hMin, with blade index A ∈ {0,1,2,3,12,23,31,123, . . . ,12 . . .n}, MA ∈ R. The Reverse of M ∈ Clp,q is defined as A (−1) k(k−1) 2 hMik, n(cid:229) k=0 eM = defined as it replaces complex conjugation and quaternion conjugation. The scalar product of two multivectors M,eN ∈ Clp,q is For M,eN ∈ Cln = Cln,0 we get M ∗eN = (cid:229) A MANA. The modulus M of a multivector M ∈ Cln is defined as (3) M ∗eN = hMeNi = hMeNi0. M2 = M ∗ eM = (cid:229) M2 A. A (1) (2) (4) For n = 2(mod 4) and n = 3(mod 4) the pseudoscalar is in = e1e2 . . .en with (also valid for Cl0,n, n = 1,2(mod 4)) A blade B describes a vector subspace Its dual blade i2 n = −1. VB = {x ∈ Rp,qx ∧ B = 0}. B∗ = Bi−1 n describes the complimentary vector subspace V ⊥ B . The pseudoscalar in ∈ Cln is central for n = 3(mod 4) in M = M in, ∀M ∈ Cln. But for even n we get due to non-commutativity [8] of the pseudoscalar in ∈ Cln inM = Meven in − Modd in , einl M = Meven einl + Modd e−inl , ∀M ∈ Cln, l ∈ R. (5) (6) (7) (8) (9) A multivector valued function f : Rp,q → Clp,q, p + q = n, has 2n blade components ( fA : Rp,q → R) A We define the inner product of Rn → Cln functions f ,g by f (x) = (cid:229) fA(x)eA. ( f ,g) =ZRn f (x)gg(x) dnx = (cid:229) A,B eAeeBZRn fA(x)gB(x) dnx, and the L2(Rn;Cln)-norm k f k2 = h( f , f )i =ZRn f (x)2dnx, L2(Rn;Cln) = { f : Rn → Cln k f k < ¥ }. (10) (11) (12) For the Clifford geometric algebra Fourier transformation (CFT) [8] the complex unit i ∈ C is replaced by some geometric (square) root of −1, e.g. pseudoscalars in, n = 2,3(mod 4). Complex functions f are replaced by multivector functions f ∈ L2(Rn;Cln). Definition 1 (Clifford geometric algebra Fourier transformation (CFT)). The Clifford GA Fourier transform F { f }: Rn → Cln, n = 2,3(mod 4) is given by for multivector functions f : Rn → Cln. F { f }(w ) = bf (w ) =ZRn f (x)e−inw ·x dnx, NB: The CFT can also be defined analogously for Cl0,n′, n′ = 1,2(mod 4). The CFT (13) is inverted by f (x) = F −1[F { f }(w )] = 1 (2p )n ZRn F { f }(w )einw ·x dnw . (13) (14) The similitude group G = SIM(n) of dilations, rotations and translations is a subgroup of the affine group of Rn G = R+ × SO(n) ⋊ Rn = {(a,rq ,b)a ∈ R+,rq ∈ SO(n),b ∈ Rn}. The left Haar measure on G is given by dadq an+1 , is the Haar measure on SO(n). We define the inner product of f ,g : G → Cln by dl = dl (a,q ,b) = dm (a,q )dnb, dm = dm (a,q ) = where dq ( f ,g) =ZG f (a,q ,b) ^g(a,q ,b) dl (a,q ,b), and the L2(G ;Cln)-norm k f k2 = h( f , f )i =ZG f (a,q ,b)2dl , L2(G ;Cln) = { f : G → Cln k f k < ¥ }. (15) (16) (17) CLIFFORD GA WAVELETS Previous works on Clifford wavelets include [1, 2, 3, 4, 5, 6, 7, 9]. We represent the transformation group G = SIM(n) by applying translations, scaling and rotations to a so-called Clifford mother wavelet y : Rn → Cln y (x) 7−→ y a,q ,b(x) = 1 an/2 y (r−1 q ( x − b a )). The family of wavelets y Lemma 1 (Norm identity). The factor a−n/2 in y a,q ,b are so-called Clifford daughter wavelets. a,q ,b ensures (independent of a,q ,b) that ky a,q ,bkL2(Rn;Cln) = ky kL2(Rn;Cln). (18) (19) The CFT spectral representation of Clifford daughter wavelets is a,q ,b}(w ) = a A Clifford mother wavelet y ∈ L2(Rn;Cln) is admissible if F {y n q 2 by (ar−1 (w ))e−inb·w . (w )) dm =ZRn eby (w )by (w ) w n dnw , (20) (21) (22) is an invertible multivector constant and finite at a.e. w ∈ Rn. We must therefore have by (w = 0) = 0 and therefore q q Cy =ZR+ZS0(n) an{by (ar−1 every Clifford mother wavelet component (w ))}∼by (ar−1 ZRn By construction Cy = fCy . Hence for n = 2,3(mod 4) [n/4] y A(x) dnx = 0. hCy i0 > 0, Cy = (hCy i4k + hCy i4k+1). k=0 The invertibility of Cy depends on its grade content, e.g. for n = 2,3, Cy is invertible, if and only if hCy i2 1 6= hCy i2 0 : C−1y = hCy i0 − hCy i1 0 − hCy i2 hCy i2 1 . (23) Definition 2 (Clifford GA wavelet transformation). For an admissible GA mother wavelet y ∈ L2(Rn;Cln) and a multivector signal function f ∈ L2(Rn;Cln) Ty : L2(Rn;Cln) → L2(G ;Cln), f 7→ Ty f (a,q ,b) =ZRn f (x) ^y a,q ,b(x) dnx. (24) NB: Because of (9) we need to restrict the mother wavelet y for n = 2(mod 4) to even or odd grades: Either we have a spinor wavelet y ∈ L2(Rn;Cl+ n ) with e = −1. n ) with e = 1, or we have an odd parity vector wavelet y ∈ L2(Rn;Cl− NB: For n = 3(mod 4), no grade restrictions exist. We then always have e = 1. We immediately see from Definition 2 that the Clifford GA wavelet transform is left linear with respect to multivector constants l 1,l 2 ∈ Cln. The spectral (CFT) representation of the Clifford wavelet transform is Ty f (a,q ,b) = 1 (2p )nZRnbf (w )a q n 2 {by (ar−1 NB: The CFT for n = 2(mod 4) preserves even and odd grades. (w ))}∼e e inb·w dnw . (25) We further have the following set of properties. Translation covariance: If the argument of Ty a constant x0 ∈ Rn then [Ty f (· − x0)](a,q ,b) = Ty f (a,q ,b − x0) . Dilation covariance: If 0 < c ∈ R then [Ty f (c ·)](a,q ,b) = Ty f (ca,q ,cb) . 1 c n 2 Rotation covariance: If r = rq , r0 = rq 0 and r′ = rq ′ = r0r = rq 0rq are rotations, then [Ty f (rq 0 ·)](a,q ,b) = Ty f (a,q ′,rq 0b) . f (x) is translated by (26) (27) (28) Now we see some differences from the classical wavelet transforms. The next property is an inner product relation: Let C ′y = (−e )nCy , and f ,g ∈ L2(Rn;Cln) arbitrary. Then we have (Ty f ,Ty g)L2(G ;Cln) = ( fC ′y ,g)L2(Rn;Cln) . (29) (cid:229) The spectral representation (25) and the CFT Plancherel theorem [8] are essential for the proof. As a corollary we get the following norm relation: kTy f k2 L2(G ;Cln) = Sc( fC ′y , f )L2(Rn;Cln) = C ′y ∗ ( f , f )L2(Rn;Cln) . (30) We can further derive the Theorem 1 (Inverse Clifford Cln wavelet transform). Any f ∈ L2(Rn;Cln) can be decomposed with respect to an admissible Clifford GA wavelet as f (x) =ZG Ty f (a,q ,b)y a,q ,bC ′ −1 y dm dnb =ZG ( f ,y a,q ,b)L2(Rn;Cln) y a,q ,bC ′ −1 y dm dnb, the integral converging in the weak sense. Next is the reproducing kernel: We define for an admissible Clifford mother wavelet y ∈ L2(Rn;Cln) a,q ,bC ′ −1 Then Ky (a,q ,b;a′,q ′,b′) is a reproducing kernel in L2(G ,dl ), i.e, Ky (a,q ,b;a′,q ′,b′) =(cid:16)y y ,y a′,q ′,b′(cid:17)L2(Rn;Cln) . (31) (32) (33) Theorem 2 (Generalized GA wavelet uncertainty principle). Let y be an admissible Clifford algebra mother wavelet. Then for every f ∈ L2(Rn;Cln), the following inequality holds Ty Ty f (a,q ,b)Ky (a,q ,b;a′,q ′,b′)dl . f (a′,q ′,b′) =ZG kq Ty f (a,q ,b)k2 L2(G ;Cln) Cy ∗(fw NB: The integrated varianceRR+RSO(n) kw F {Ty erwise the proof is similar to the one for n = 3 in [9]. For scalar admissibility constant this reduces to Corollary 1 (Uncertainty principle for GA wavelet). Let y be a Clifford algebra wavelet with scalar admissibility constant. Then for every f ∈ L2(Rn;Cln), the following inequality holds hCy ∗ ( f , f )L2(Rn;Cln)i2 (34) is independent of the wavelet parity e . Oth- f ,fw f (a,q , .)}k2 f )L2(Rn;Cln) ≥ L2(Rn;Cln) 4 dm . n(2p )n kbTy f (a,q ,b)k2 L2(G ;Cln) kw f k2 L2(Rn;Cln) ≥ nCy k f k4 L2(Rn;Cln). (35) (2p )n 4 Finally GA Gabor Wavelets are defined as (variances s k,1 ≤ k ≤ n, for n = 2(mod 4) : A ∈ Cl+ n or A ∈ Cl− n ) y c(x) = A (2p ) 2 (cid:213) n n k=1 s k − 1 2 (cid:229) k e x2 k s 2 k  einw 0·x − e− 1 k=1 {z 2 (cid:229) n s 2 k w 2 0,k constant   , } x,w 0 ∈ Rn, constant A ∈ Cln . (36) Soli deo gloria. I do thank my dear family, B. Mawardi, G. Sommer, W. Sprössig and K. Gürlebeck. ACKNOWLEDGMENTS REFERENCES 1. M. Mitrea, Cliff. Wavelets, Singular Integrals and Hardy Spaces, Lect. Notes in Math. 1575, Springer, New York, 1994. 2. Brackx al., Ghent Clifford Analysis Wavelet Publications http://cage.ugent.be/crg/cliffordpublicaties.PDF (2001-2007), et 3. L. Traversoni, Quaternion Wavelet Problems, Proc. of 8th Int. Symp. on Approx. Theory, Texas A& M University, Jan. 1995. Image Anal. Using Quat. Wavelets, edited by E. Bayro-Corrochano, G. Sobczyk, Proc. of AGACSE 1999, Birkhäuser, Basel, 2001. 4. E. Bayro-Corrochano, List of Publications, http://www.gdl.cinvestav.mx/edb/publications.html 5. J. Zhao, and L. Peng, J. of Nat. Geom., 2001. J. Zhao, Acta Sci. Nat. Univ. Pek., 41(5), (2005). 6. Kähler et al., ZAA, 24, 813 -- 824 (2005) . 7. S. Bernstein, edited by T.E. Simos et al., AIP Proc. of ICNAAM 2008, 1048, 634 -- 637 (2008). AACA, 19(2), 173 -- 189 (2009). 8. E. Hitzer and B. Mawardi, Adv. App. Cliff. Alg. Vol. 18(S3,4), 715 -- 736 (2008). 9. B. Mawardi and E. Hitzer, Int. J. of Wavelets, Multires. and Inf. Proces., 5(6), 997 -- 1019 (2007).
0904.3772	2	0904	2012-10-01T17:12:19	Tamely ramified subfields of division algebras	[ "math.RA", "math.NT" ]	For any number field K, it is unknown which finite groups appear as Galois groups of extensions L/K such that L is a maximal subfield of a division algebra with center K (a K-division algebra). For K=Q, the answer is described by the long standing Q-admissibility conjecture. We extend a theorem of Neukirch on embedding problems with local constraints in order to determine for every number field K, what finite solvable groups G appear as Galois groups of tame maximal subfields of K-division algebras, generalizing Liedahl's theorem for metacyclic G and Sonn's solution of the Q-admissibility conjecture for solvable groups.	math.RA	math	TAMELY RAMIFIED SUBFIELDS OF DIVISION ALGEBRAS DANNY NEFTIN Abstract. For any number field K, it is unknown which finite groups appear as Galois groups of extensions L/K such that L is a maximal subfield of a division algebra with center K (a K-division algebra). For K = Q, the answer is described by the long standing Q-admissibility conjecture. We extend a theorem of Neukirch on embedding problems with local constraints in order to determine for every number field K, what finite solvable groups G appear as Galois groups of tame maximal subfields of K-division algebras, generalizing Liedahl's theorem for metacyclic G and Sonn's solution of the Q-admissibility conjecture for solvable groups. 1. Introduction A division algebra D which is finite dimensional over its center K (a K-division algebra), is called a G-crossed product if there exists a Galois extension L/K with Galois group G (a G-extension) such that L is a maximal subfield of D. Crossed products are fundamental in the study of division algebras and are accompanied by a structure which explicitly describes them (see [20, Chp. 14-19]). A group G is called K-admissible if there exists a G-crossed product K-division algebra; a field extension L/K is called adequate if L is a maximal subfield of a K-division algebra 1. It is known by the Brauer-Hasse-Noether theorem that over a number field K, all K-division algebras are crossed products with respect to a cyclic group. However, it is unknown for which groups G there exists a G-crossed product K-division algebra, i.e. what groups are K-admissible? Over Q, Schacher observed ([21]) that the Sylow subgroups P of a Q-admissible group are metacyclic, that is P has a cyclic normal subgroup C ✁ P such that P/C is also cyclic. The converse of this observation is known as the Q-admissibility conjecture: Conjecture 1.1. Every group with metacyclic Sylow subgroups is Q-admissible. This conjecture was studied extensively (e.g. [4],[5],[6],[10],[11],[21]) and proven by Sonn for solvable groups in a series of papers ([3], [23] and [24]). Recently, analogs of this conjecture were proved by Harbater, Hartmann and Krashen over function fields of curves over complete discretely valued fields with algebraically closed residue fields ([13], cf. [12]), by Paran and the author over two dimensional com- plete local domains with algebraically closed residue fields ([16]), and by Surendranath and Suresh over function fields of curves over complete discretely valued fields which contain enough roots of unity ([25]). However, the situation over number fields is far from being understood. 1In fact by [21], L/K is adequate if and only if L is a subfield of a K-division algebra. Thus, the maximality requirement can be omitted. 1 2 DANNY NEFTIN Schacher's observation extends to number fields under an additional assumption of tameness as follows. Let µn denote the set of n-th roots of unity and σt,n be the au- tomorphism of Q(µn) for which σt,n(ζ) = ζ t for all ζ ∈ µn. Using a similar argument to Liedahl's [14, Theorem 28], we observe that if G appears as a Galois group of a tamely ramified adequate extension of a number field K then its Sylow subgroups are metacyclic, and furthermore for every l G, the l-Sylow subgroups G(l) of G admit a presentation: (1.1) G(l) ∼= M(m, n, i, t) := hx, yxm = yi, yn = 1, x−1yx = yti such that σt,n fixes K ∩ Q(µn) (see "only if part" of Theorem 1.3). This observation suggests the following natural generalization of Conjecture 1.1: Question 1.2. Let K be a number field and G a group whose l-Sylow subgroups admit a presentation M(m, n, i, t) such that σt,n fixes K ∩Q(µn), for every l G. Is G necessarily K-admissible? Furthermore, is there necessarily a tamely ramified adequate G-extension of K? The first part of this question is known to have an affirmative answer for metacyclic G ([14, Theorem 27]) and for some small order groups: A5 ([11]), the central extension SL2(5) of A5 ([9]), A6, A7 ([22]), the double covers of A6 and A7 ([8]), PSL2(7) ([1]) and PSL2(11) ([7]). In this paper we give a positive answer to Question 1.2 for solvable groups, generalizing Liedahl's [14, Theorem 27] and Sonn's [24, Theorem 1]: Theorem 1.3. Let K be a number field and G a solvable group. Then there exists a tamely ramified adequate G-extension L/K if and only if for every lG, the l-Sylow subgroups of G admit a presentation M(m, n, i, t) such that σt,n fixes K ∩ Q(µn). We note that since the proof of Sonn's theorem ([24]) over Q is based on Neukirch's [17, Main Theorem] which makes an assumption on the absence of roots of unity in K, Sonn's proof does not apply over arbitrary number fields. A key ingredient in our proof is an extension of [18, Korollar 6.4]. Neukirch's Korollar 6.4 is a highly useful tool that under the assumption of at least one of six conditions on a finite set S of primes of the base field, allows to change solutions of embedding problems to satisfy any prescribed local conditions at S (generalizing the Grunwald- Wang theorem). We extend Korollar 6.4 by showing that under the assumption of at least one of four of these six conditions on S, it is possible to change a solution to satisfy prescribed conditions at S leaving the solution unchanged at any given finite set of primes T . We use this extension to strengthen Sonn's proof of [24, Theorem 1] in order to obtain tamely ramified adequate G-extensions of Q with prescribed local behavior at given finite sets of primes. This gives us a strong control over the ramification of G-crossed product Q-division algebras, allowing us to lift these to division algebras over a given number field and by that prove Theorem 1.3. This work is partially based on the author's Ph.D. thesis ([15]). I would like to thank my thesis advisor Jack Sonn for investing time and effort into teaching and guiding me, and for helpful comments on this manuscript. TAMELY RAMIFIED SUBFIELDS OF DIVISION ALGEBRAS 3 2. Embedding problems and local Galois groups 2.1. Embedding problems. The theory of embedding problems is central in the study of the inverse Galois problem and is a key ingredient in our proof of Theorem 1.3. We shall describe a setup for these problems, recall Neukirch's [18, Korollar 6.4] and extend it. 2.1.1. Setup. Embedding problems are a strong generalization of the inverse Galois prob- lem which ask whether a Galois extension can be embedded into a larger Galois extension with a given Galois group. The precise setup is as follows. A (finite) embedding problem over a number field K consists of a finite Galois extension L/K and an epimorphism of finite groups π : E → G := Gal(L/K). For our purposes it suffices to consider embedding problems with abelian kernel A := ker(π). Let GK denote the absolute Galois group of K. Two homomorphisms ψ1, ψ2 : GK → E are called equivalent if there is an a ∈ A such that a−1ψ1(g)a = ψ2(g) for all g ∈ GK. A solution for π is an equivalence class of homomorphisms ψ : GK → E (not necessarily surjective) for which π◦ψ is the restriction map resL : GK → G. For a surjective solution ψ, the fixed field M = K contains L and has Galois group Gal(M/K) ∼= E. ker(ψ) The epimorphism π defines an action of G on A and hence induces a GK-module structure on A via resL. For every crossed homomorphism χ ∈ H1(GK, A) and solution ψ : GK → E, the map ψ′ = χ · ψ given by ψ′(σ) = χ(σ)ψ(σ) for all σ ∈ GK, is also a solution (see [19, Chp. IX, §4]). In fact, for every two solutions ψ, ψ′ of π, there is a unique χ ∈ H1(GK, A) such that ψ′ = χ·ψ. We think of χ as the element that "changes" ψ to ψ′. 2.1.2. Embedding problems with prescribed local conditions. By a prime p of K we mean a finite or infinite prime. Fix an algebraic closure K of K, an algebraic closure K p of the completion Kp, and an inclusion of K into K p for every prime p of K. In particular, the embedding problem π induces a local embedding problem πp : π−1(Gp) → Gp where Gp = Gal(Lp/Kp), Lp := LKp. Moreover, the restriction ψp of a solution ψ : GK → E to the subgroup GKp is a solution for πp. Let S be a finite set of primes of K and for every p ∈ S fix (prescribe) a solution ψ(p) to πp, assuming such (local) solutions exist. Similarly to the Grunwald-Wang theorem, one is interested in solutions ψ of π such that ψp = ψ(p) for all p ∈ S. Assume π has a solution φ. Then for every p ∈ S there is χ(p) ∈ H1(GKp, A) such that ψ(p) = χ(p) · φp. If the element (χ(p))p∈S has a source χ under the restriction map: ρS : H1(GK, A) → Y p∈S H1(GKp, A) then ψ := χ · φ is a solution for π which restricts to ψ(p) = χ(p) · φp at all p ∈ S. Thus, if the map ρS is surjective, every solution for π can be "changed" to a solution with prescribed local conditions at S. 2.1.3. Neukirch's Korollar. [18, Korollar 6.4] is a highly useful criteria for the map ρS to be surjective. Let A be a GK-module and n = exp(A). Let A′ = Hom(A, µn) be the dual GK-module and K(A′) the fixed field of the centralizer of A′ in GK. Let G′ p := Gal(K(A′)p/Kp). Denote Γ(G, A) := ker (cid:16)H1(G, A) → Qg∈G H1(hgi, A)(cid:17) . := Gal(K(A′)/K) and for a prime p of K, let G′ 4 DANNY NEFTIN Theorem 2.1. (Neukirch [18, Korollar 6.4]) Let S be a finite set of primes of K. Then the map ρS is surjective in each of the following cases: (a) Γ(G′ (b) for every p ∈ S, the group G′ p, A′) = 0 for all p ∈ S, p is cyclic or a semidirect product of two cyclic groups of relatively prime orders, (c) H1(G′, A′) = 0, (d) G′ = lcm{G′ pp 6∈ S}, (e) A is cyclic of odd order, (f) the action of GK on A is trivial and (K, exp(A), S) does not fall into a special case. In (f), when exp(A) = 2tm, m odd, one says that the triple (K, exp(A), S) falls into a special case if K(µ2t)/K is noncyclic and S contains all primes p for which Kp(µ2t)/Kp is noncyclic. Thus, under each of these conditions one can change a solution to satisfy arbitrary prescribed local conditions at S. Furthermore, we show that under each of conditions (a), (b), (c) or (e) it is possible to change a solution to satisfy prescribed local conditions at S leaving the solution unchanged at a given finite set of primes T . Proposition 2.2. Let A be a finite GK-module. Assume that conditions (a) or (b) hold for a finite set S. Then the subgroup is in the image of ρS∪T for every finite set T disjoint from S. Y p∈S H1(GKp, A) × Y p∈T {0} Proof. Since by [18, Satz 6.2] condition (b) implies (a), it suffices to prove the assertion p, A′) = 0 for all p ∈ S. Let P be the set of all primes when (a) holds. Assume that Γ(G′ of K and Q′ un(GKp, A). p∈P H1(GKp, A) the restricted product over the subgroup Qp∈P H1 Recall that the Poitou-Tate theorem gives a non-degenerate bilinear map β : Y p∈P ′ H1(GKp, A) × Y p∈P ′ H1(GKp, A′) → Q/Z which is defined as the product of local bilinear maps βp : H1(GKp, A) × H1(GKp, A′) → Q/Z for every p ∈ P . Following [18], for a finite set U of primes of K we let: U : H1(GK, A′) → Y ρ′ p6∈U ′ H1(GKp, A′) be the restriction map, ∆ = coker(ρS∪T ) and ∇ = ker(ρ′ ∅). By [18, Satz 4.4], β induces a non-degenerate bilinear form β0 : ∆ × ∇ → Q/Z, which is given on p∈P H1(GKp, A) χ := (χp)p∈S∪T ∈ ∆ and λ ∈ ∇ by β0(χ, λ) := β( χ, ρ′ is any element whose p-th component is χp at all p ∈ S ∪ T . S∪T )/ ker(ρ′ ∅(λ)) where χ ∈ Q′ Let χ = Qp∈S∪T χp be an element of ∆ such that χp = 0 for all p ∈ T . We claim that χ is orthogonal to ∇ and therefore it is the zero element in ∆, proving the proposition. Letting χ = ( χp)p∈P ∈ Q′ p∈P H1(GKp, A) where χp = χp for p ∈ S and χp = 0 for ∅(∇)). Since χp = 0 for p 6∈ S, it suffices to show that p 6∈ S, we have β0(χ, ∇) = β( χ, ρ′ TAMELY RAMIFIED SUBFIELDS OF DIVISION ALGEBRAS 5 βp(χp, ρ′ factor. But [18, Satz 6.3] implies that the image of ∇ under the restriction map ∅(∇)p) = 0 for all p ∈ S, where ρ′ ∅(∇)p is the projection of ρ′ ∅(∇) to the p-th ρS∪T,A′ : H1(GK, A′) → Y p∈S∪T p, A′). Since by assumption Γ(G′ H1(GKp, A′) lies in Qp∈S∪T Γ(G′ ρS∪T,A′(∇)p = 0 and hence βp(χp, πpρ′ p, A′) = 0 for p ∈ S, we get ρ′ ∅(∇)) = 0 for all p ∈ S, proving the claim. ∅(∇)p = (cid:3) From Proposition 2.2 and the discussion above it we get: Corollary 2.3. Let π : E → Gal(L/K) be an embedding problem with solution φ. Fix solutions ψ(p) for πp at all primes p in a finite set S and let T be a finite set of primes disjoint from S. Assume that at least one of conditions (a),(b),(c) or (e) hold for S. Then there exists a solution ψ such that ψp = ψ(p) for all p ∈ S and ψp = φp for all p ∈ T . Proof. Since conditions (c) and (e) are independent of S, the image of ρS∪T contains Qp∈S H1(GKp, A) × Qp∈T {0} under these conditions as well. For p ∈ S, let χ(p) ∈ H1(GKp, A) be the element for which ψ(p) = χ(p) · ψp. By Proposition 2.2, the element (χ(p))p∈S × (0)p∈T has a source χ ∈ H1(GK, A) under the map ρS∪T . Then the solution ψ := χ · φ restricts to χ(p) · φp = ψ(p) at all p ∈ S and to 0 · φp = φp at all p ∈ T . (cid:3) Remark 2.4. (1) Proposition 2.2 need not hold under conditions (d) or (f). For example, let K be a quadratic extension of Q in which 2 splits and let p1, p2 be the primes above it. Let S = {p1}, T = {p2} and let A = Z/8 be the trivial GK-module. Then A′ ∼= µ8 as GK-modules and K(A′) = K(µ8). Both conditions (d) and (f) hold for S and hence ρS is surjective. However, since K(A′)p/Kp is cyclic for all p 6= p1, p2, conditions (d) and (f) fail for S ∪ T . Furthermore, the Grunwald-Wang theorem shows that Y p∈S H1(GKp, A) × Y p∈T {0} 6⊆ Im ρS∪T . Indeed, letting ψ(p2) = 0 and ψ(p1) ∈ Hom(GKp1 , A) be such that the fixed field of ker(ψ(p1)) is the unramified Z/8-extension of K, [2, Chp. X, Theorem 5] shows that (ψ(p1), ψ(p2)) 6∈ Im(ρS∪T ). (2) Let A be a trivial GK-module of exponent 2tm′ where m′ is odd. If K(µ2t)/K is cyclic then condition (f) holds for all finite sets S. 2.2. Tame Galois groups of local fields. We shall make use of a few well known facts about Galois groups of tame local extensions, all of which can be found in [26] and [19, Chp. VII, §5]. Let L/K a tamely ramified G-extension of p-adic fields, I its inertia group, n := I, and q the cardinality of the residue field of K. The subfield LI contains µn and L/LI is a (cyclic) Kummer extension. The Galois group of LI/K is generated by the Frobenius automorphism σL which acts on µn by raising each element to the power q. In particular the restriction of σL to Q(µn) is σq,n and fixes Q(µn) ∩ K. The action of σ on I via conjugation in G is equivalent to its action on µn. Thus, G = M(m, n, i, t) = hx, yxm = yi, yn = 1, x−1yx = yti, 6 DANNY NEFTIN where t ≡ q (mod n), I = hyi and x = σ (mod I). In particular, one has the following observation which is the basis to [14, Theorem 28]: Lemma 2.5. Let p, l be two distinct rational primes and K a p-adic field. Then ev- ery group G that appears as a Galois group over K has a metacyclic l-Sylow subgroup M(m, n, i, t) such that σt,n fixes K ∩ Q(µn). Proof. Let L/K be a G-extension, M the fixed subfield of an l-Sylow subgroup H of G and t the cardinality of the residue field of M. Then H ∼= M(m, n, i, t) and σt,n fixes M ∩ Q(µn). In particular σt,n fixes K ∩ Q(µn). (cid:3) Consider the converse problem of realizing M(m, n, i, t) over K and assume t ≡ q (mod n) so that M(m, n, i, t) = M(m, n, i, q). The Galois group Gtr K of the maximal tamely ramified extension of K is profinitely generated by two automorphisms σ and τ and one relation σ−1τ σ = τ q, where τ is of order prime to q and σ is the Frobenius automorphism. Letting M be the (unique) unramified degree m extension of K, σ restricts to the Frobenius automorphism of M/K. Thus, an embedding problem π : M(m, n, i, t) → Gal(M/K) with kernel hyi has a surjective solution whose corresponding field is a tamely ramified M(m, n, i, t)-extension of K. 3. Galois groups of tamely ramified adequate extensions 3.1. Proof of Theorem 1.3. We consider a refined notion of adequacy. For a number field K and a finite set S of primes of K, we say that L/K is S-adequate if L is a maximal subfield of a K-division algebra that is unramified outside S. Let D(L/K, P) denote the decomposition group of a prime P of L. The same proof as of Schacher's criterion ([21, Proposition 2.6]) gives the following criterion for S-adequacy: Proposition 3.1. Let L/K be a G-extension of number fields and S a finite set of primes of K. Then L/K is S-adequate if and only if for every rational prime l G, there are two primes p1, p2 ∈ S for which D(L/K, Pi) contains an l-Sylow subgroup of G, where Pi is a prime of L which divides pi, i = 1, 2. Note that the condition of containing an l-Sylow subgroup of G is independent of the choice of prime Pi dividing pi. A key ingredient in our proof of Theorem 1.3 is the following generalization of Sonn's theorem ([24, Theorem 1]) which asserts the existence of S-adequate G-extensions for prescribed sets S. Since M(m, n, i, t) is realizable over Qp when p ≡ t (mod n) (see Section 2.2), we consider the following sets S: Definition 3.2. We call a set S of distinct odd rational primes p(l) , i = 1, 2, lG which i are prime to G, a tame supporting set for G if for every lG, the l-Sylow subgroups of G admit a presentation M(m, n, i, t) such that p(l) 2 ≡ t (mod n). 1 , p(l) Theorem 3.3. Let G be a solvable group with metacyclic Sylow subgroups. Let S be a tame supporting set for G and T a finite set of rational primes which is disjoint from S. Then there exists an S-adequate G-extension L/Q in which the primes of T split completely. The proof of this theorem is based on Corollary 2.3 and on Sonn's proof of [24], and is given in Section 4. We now use this theorem to prove Theorem 1.3: TAMELY RAMIFIED SUBFIELDS OF DIVISION ALGEBRAS 7 Proof of Theorem 1.3. "Only if part": Let l be a prime that divides G. By Proposition 3.1, there is a prime p of K such that D(L/K, P), P p, contains an l-Sylow subgroup P of G. If p l, p is unramified in L and hence P is cyclic. Otherwise, Lemma 2.5 implies that P has a presentation M(m, n, i, t) such that σt,n fixes Kp ∩ Q(µn) and hence K ∩ Q(µn). In both cases, P has the required presentation. "If part": Let l G be a rational prime and let P (l) be the set of primes of K that are unramified over Q with residue degree 1, and whose restriction p to Q satisfies p ≡ tl (mod nl). We first claim that P (l) is infinite. Let M be the Q-normal closure of K. Since σtl,nl fixes K ∩ Q(µnl), σtl,nl extends to an automorphism of K(µnl) that fixes K and hence lifts to an automorphism τl ∈ Gal(M(µnl)/K). By Chebotarev's density theorem there are infinitely many primes P of M(µnl) that are unramified over Q, and whose Frobenius automorphism in M(µnl)/Q is τl. In particular, the restriction p of such P to Q has Frobenius σtl,nl in Q(µnl)/Q, and hence p ≡ tl (mod nl). Since τl fixes K, the restriction of each such P to K has residue degree 1 over Q and hence is in P (l), proving the claim. Since P (l) is infinite, we can choose two primes p(l) 2 ∈ P (l) such that the restrictions p(l) i of p(l) to Q, i = 1, 2, l G, are all distinct rational primes which are prime to G. Thus, the set S := {p(l) i = 1, 2, lG} is a tame supporting set for G and by Theorem i 3.3 there exists an S-adequate G-extension L/Q in which all of the primes l dividing G split completely. 1 , p(l) i We claim that N := LK is an adequate extension of K. This proves the theorem i has residue degree 1 over Q, since L/Q and hence N/K is tamely ramified. Since p(l) we have Kp ∼= Qpi and hence: (l) i (3.1) [NP (l) i : Kp (l) i ] = [LP (l) i ∩L : Q p(l) i ] for P(l) i p(l) i , i = 1, 2, l G. Letting lαl be the maximal power of l dividing G, (3.1) shows that lαl [N : K] for every l G and hence that Gal(N/K) ∼= G. Furthermore, (3.1) shows that D(P(l) , N/K) i contains an l-Sylow subgroup of G for i = 1, 2, l G, showing that N/K is adequate, as required. (cid:3) Remark 3.4. (1) In [14], Liedahl uses Lemma 2.5, similarly to the "only if part" of Theorem 1.3, to show that under the assumption that l does not decompose in K, the l-Sylow subgroups of a K-admissible group admit a presentation M(m, n, i, t) such that σt,n fixes K ∩ Q(µn). He also uses the flexibility of [23, Theorem 1] to prove that if G itself has a presentation M(m, n, i, t) such that σt,n fixes K ∩ Q(µn), then G is K-admissible. (2) Note that the proof of the "only if part" of Theorem 1.3 applies more generally without the assumption that G is solvable. (3) The proof of Theorem 3.3 gives furthermore that lαl [LP l G, i = 1, 2. (l) i ∩L : Q p(l) i ] for all 3.2. Consequences. For solvable groups G we get the following characterization of K-admissibility under the assumption that every l G does not decompose in K: Corollary 3.5. Let K be a number field and G a solvable group. Assume that every prime l that divides G does not decompose in K. Then the following conditions are equivalent: 8 DANNY NEFTIN (1) There exists a tamely ramified K-adequate G-extension; (2) G is K-admissible; (3) for every lG, the l-Sylow subgroups of G admit a presentation M(m, n, i, t) such that σt,n fixes K ∩ Q(µn). Proof. The implication (1) ⇒ (2) is immediate and (2) ⇒ (3) follows from Remark 3.4 ([14, Theorem 28]). The implication (3) ⇒ (1) is the "if part" of Theorem 1.3. (cid:3) Recall that a group G is called infinitely often K-admissible if there exist infinitely many adequate G-extensions Li/K, i ∈ N, such that Lr+1 ∩ (L1 · · · Lr) = K (cf. [1]). Corollary 3.6. Let K be a number field and G a solvable group such that for every lG, the l-Sylow subgroups admit a presentation M(m, n, i, t) such that σt,n fixes K ∩ Q(µn). Then G is infinitely often K-admissible. Furthermore, there exists a K-division algebra D that has infinitely many disjoint maximal subfields Li, i ∈ N, such that Gal(Li/K) ∼= G. The following lemma is useful in the proof of Theorem 3.3 and is used here to prove Corollary 3.6. Given a Galois extension N/Q, define the following condition on a finite set of rational primes T : (AN ) The decomposition groups D(N/Q, p), p ∈ T, p p, generate Gal(N/Q). Lemma 3.7. Assume T satisfies (AN ). Then every finite extension K/Q in which the primes of T split completely is disjoint from N. Proof. Let H := Gal(N/N ∩ K) and assume on the contrary that H 6= G. By condition (AN ) there exists a prime p ∈ T and p p such that D := D(N/Q, p) 6⊆ H. In particular, [Np : Kp∩K] = D ∩ H < D = [Np : Qp] and hence [Kp∩K : Qp] > 1 contradicting the assumption that p splits completely in K. (cid:3) Note that by Chebotarev's density theorem for every cyclic subgroup C ≤ G there are infinitely many primes p of N for which D(N/Q, p) = C. Thus, we can always choose a finite set T which satisfies (AN ). Proof of Corollary 3.6. Let S and lαl be as in the proof of Theorem 1.3. Define D := D0 ⊗Q K where D0 is the Q-division algebra with Hasse invariants 1/lαl at p(l) 1 , −1/lαl at p(l) 2 for l G, and 0 at all other primes. It suffices to show that given r disjoint G-extensions L1, ..., Lr of K which are maximal subfields of D there exists a maximal subfield Lr+1 of D such that Gal(Lr+1/K) = G and Lr+1 ∩ (L1 · · · Lr) = K. Let M be the Q-normal closure of K and N := L1 · · · LrM. Let T be a finite set which is disjoint from S, contains all primes l G, and for which condition (AN ) holds. As remarked in 3.4.(3), Theorem 3.3 gives a maximal subfield L of D0 in which the primes of T split completely. By Lemma 3.7, L ∩ N = Q and hence Lr+1 := LK is a G-extension of K and Lr+1 ∩ (L1 · · · Lr) = K. Since in addition Lr+1 splits D, Lr+1 is a maximal subfield of D. (cid:3) 4. Proof of Theorem 3.3 In this section we prove Theorem 3.3. In Sections 4.1 and 4.2 we treat the cases of 2-groups and {2, 3}-groups (groups of order 2a3b), respectively. We first show how the theorem follows from the latter case. TAMELY RAMIFIED SUBFIELDS OF DIVISION ALGEBRAS 9 As in Section 2, we fix an embedding of an algebraic closure of Q into an algebraic closure of each of its completions. We shall say that a set of primes T splits completely in L if every prime in T splits completely in L. Proof of Theorem 3.3. Let n = G. By [3, Lemma 1.4], there is a normal subgroup N ✁ G of order prime to 2 and 3 and a {2, 3}-subgroup H such that G = NH. Extend T to a finite set T0 disjoint from S which satisfies condition (AQ(µn)). By the case of {2, 3}-groups (Section 4.2), there exists a {p(2) 2 }-adequate H- extension K/Q in which T0 ∪ {p(l) 2 l > 3} splits completely. Since by Lemma 3.7, K ∩ Q(µn) = Q, and since the embedding problem G → Gal(K/Q) splits, we may apply [17]. It follows that K/Q embeds into a G-extension L/Q such that T splits completely ), i = 1, 2, is an l-Sylow subgroup of N for all lN. In particular, in L and Gal(Lp(l) L/Q is an S-adequate G-extension in which T splits completely, as required. (cid:3) 1 , p(2) 1 , p(l) 2 , p(3) 1 , p(3) i /Q p(l) i 4.1. 2-groups. The Q-admissibility of metacyclic 2-group was proved in [23] using The- orem 2.1. We use Corollary 2.2 in order to prove Theorem 3.3 for 2-groups, generalizing [23]: Proof of Theorem 3.3 for 2-groups. Let G be a metacyclic 2-group with presentation G ∼= M(m, n, i, t) such that σt,n fixes K ∩ Q(µn), and let k be the order of x in G. Let S = {p1, p2} be a tame supporting set for G such that pi ≡ t (mod n), i = 1, 2. Since S consists of odd primes, the Grunwald-Wang theorem (see Theorem 2.1.(f)) implies that there exists a Z/k-extension K/Q in which the primes of S are inert and T splits completely. We identify Gal( K/Q) with hxi and let K/Q be the unique Z/m- extension inside K. The embedding problem π : G → Gal(K/Q) with kernel A := hyi has a solution φ : GQ → hxi ⊆ G which is given by the restriction map to Gal( K/Q). Let πi : G → Gal(Kpi/Qpi) be the corresponding local embedding problem at pi, i = 1, 2. Since pi ≡ t (mod n), πi has a surjective solution ψ(i) : GQpi → G whose fixed field L(i) is totally ramified over Kpi and in particular µn ⊆ Kpi, for i = 1, 2 (see Section 2.2). pi := Gal(K ′ In order to change φ to a solution with the desired properties, we apply Corollary 2.3. Let A′ = Hom(A, µn) be the dual GQ-module, K ′ := Q(A′), G′ = Gal(K ′/Q) and G′ pi/Qpi), i = 1, 2. Since every automorphism in GQ that fixes A and µn also fixes A′, we have K ′ ⊆ K(µn) and hence K ′ pi is cyclic and condition 2.1.(b) holds. By Corollary 2.3, there exists a solution ψ : GQ → G of π, whose restriction at pi is ψ(i), i = 1, 2, and the restriction remains the trivial solution at each p ∈ T . Since ψ(1), ψ(2) are surjective, ψ is also surjective. Thus, the fixed field L of ker ψ is an S-adequate G-extension of Q in which T splits completely, as required. pi ⊆ Kpi(µn) = Kpi, for i = 1, 2. Thus, G′ Remark 4.1. Note that we use Corollary 2.2 since Theorem 2.1 cannot be applied for the set S ∪ T . In fact, the Grunwald-Wang theorem shows that the map ρS∪T need not be surjective if 2 ∈ T . 4.2. {2, 3}-groups. Let G be a {2, 3}-group and G(3) a 3-Sylow subgroup of G. If G(3) is normal in G then Theorem 3.3 essentially follows from the 2-groups case by applying [17]: (cid:3) 10 DANNY NEFTIN Proof of Theorem 3.3 for {2, 3}-groups when G(3) ✁ G. Let S = {p(2) 1 , p(2) 2 , p(3) 1 , p(3) 2 } be a tame supporting set for G. Extend T to a finite set T0 disjoint from S which satisfies condition (AQ(µn)) where n = G. As shown in Section 4.1, there exists a {p(2) 1 , p(2) 2 }-adequate G/G(3)-extension M/Q in which T0 ∪ {p(3) 1 , p(3) 2 } splits completely. The embedding problem G → Gal(M/Q) splits and by Lemma 3.7, M ∩ Q(µn) = Q. Thus, we may apply [17] and embed M/Q into a G-extension L/Q such that T splits completely in L and Gal(Lp(3) i /Q p(3) i ) ∼= G(3) for i = 1, 2. Therefore L/Q is an S-adequate G-extension in which T splits completely, as required. (cid:3) For {2, 3}-groups G that do not have a normal 3-Sylow subgroup, we show that the proof of [24, Theorem 1] can be adjusted to give Theorem 3.3. Let F = F (G) denote the Fitting subgroup of G and F (2) and F (3) its 2-Sylow and 3-Sylow subgroups, respectively. The approach of [24] is to construct an adequate G/F - extension N/Q and embed it into an adequate G/F (2)-extension E/Q and an adequate G/F (3)-extension L/Q. Since [EL : L] and [EL : E] are coprime, the compositum EL is an adequate G-extension of Q. (4.1) EL E F (2) ④④④④④④④④ ❈❈❈❈❈❈❈❈ F (3) ❇❇❇❇❇❇❇❇ ⑤⑤⑤⑤⑤⑤⑤⑤ F L N G/F (2) G/F G/F (3) Q When G(3) is not a normal subgroup of G, [24] shows that G/F is isomorphic either to (1) S3 or to (2) Z/3 and that in these cases we have the following partition into subcases: Case G/F 1.1 1.2 2.1 2.2 F (2) G/F (3) 2-Sylow Z/3 Z/2u × Z/2u Z/3 ⋉ (Z/2u × Z/2u) Z/2u × Z/2u Z/3 S3 S3 Z/2 × Z/2 Q16 or D∗ 16 S∗ 4 or S∗∗ 4 SL2(3) Q8 D8 Q8 Q8 S4 Here Q8 is the quaternions group, D8 the dihedral group of order 8, S∗ two central extensions of S4 with kernel Z/2, and 4 and S∗∗ 4 are the Q16 = hx, yx2 = y4, y8 = 1, x−1yx = y7i, D∗ 16 = hx, yx2 = y8 = 1, x−1yx = y3i are their 2-Sylow subgroups, respectively. In all of the above cases the 2-Sylow subgroups have unique parameters m, n and t 2. 2The parameter i is also unique up to multiplication by an odd number. TAMELY RAMIFIED SUBFIELDS OF DIVISION ALGEBRAS 11 Lemma 4.2. Let G ∼= M(m, n, i, t). (a) If G ∼= Z/2u × Z/2u then m = 2u, n = 2u, t = 1. (b) If G ∼= Q8 then m = 2, n = 4, t = 3. (c) If G ∼= D8 then m = 2, n = 4, t = 3. (d) If G ∼= D∗ 16 then m = 2, n = 8, t = 3. (e) If G ∼= Q16 then m = 2, n = 8, t = 7. Proof. (1) Let x, y be the generators of a presentation M(m, n, i, t). Since m, n2u and mn = G = 22u, one has m = n = 2u. For 1 < t < 2u the group M(2u, 2u, i, t) is non-abelian and hence t = 1. (b) -- (e) are conclusions from [14, Theorem 22, Case 3]. In this theorem, Liedahl gives necessary and sufficient conditions on a presentation M(m, n, i, t) of a group as in (b) -- (e) to have an equivalent presentation with other parameters. However, these conditions require m ≥ 4 3 which fails for the presentations in (b) -- (e). (cid:3) Proof of Theorem 3.3 for {2, 3}-groups when G(3) is not normal. We claim that the fields N, L, E in diagram (4.1) can be in fact chosen to be S- adequate extensions of Q in which T splits completely. This will imply that EL/Q is an S-adequate G-extension in which T splits completely, as required. We first construct the field E and let N = EF/(F (2)). In Case (1), G/F (2) ∼= G(3) is of odd order and therefore [17] gives a {p(3) 1 , p(3) 2 }- adequate G(3)-extension E/Q in which T ∪ {p(2) 1 , p(2) 2 } splits completely. 2 q ≡ 1 (mod p) for all p ∈ T0 ∪ {p(3) In Case (2), let q ≡ 1(mod 8) be a prime which is not in S ∪ T and such that p(2) 1 p(2) 2 q) and let q be the prime of k which lies above q. Note that k is {p(2) 2 }-adequate and T splits completely in k. Since the embedding problem G/F (2) → Gal(k/Q) splits, we may apply [17] and embed k into an S-adequate G/F (2)-extension E/Q in which T and q split completely. In both Cases (1) and (2), E/Q is S-adequate and T splits completely in E. 2 }. Let k = Q(qp(2) 1 , p(2) 1 , p(3) 1 p(2) The construction of the field L is the same as in [24] with few modifications. Since the construction in [24] is involved and long, we do not repeat it here. A self contained version of the modified construction can be found in the author's thesis ([15]). For the reader to whom [24] is available, we give below the list of required modifications. 1 , p(2) Note that our field N was denoted in Case (1) of [24] by k and in Case (2) by K. (1) Replace the primes p1, p2 (resp. p, q) in Case (1) (resp. Case (2)) of [24] by the 2 , respectively. Since S is a supporting set, Lemma 4.2 implies satisfy the congruence relations required in [24] from split completely in N(µn) 2 are prime to G, they are greater primes p(2) that the prime p(2) p1, p2, p, q. Note that in Case (1), the primes p(2) as required in [24]. Also note that since p(2) than 3 as required in Case (2) of [24]. 1 , p(2) 1 , p(2) 1 , p(2) 2 2 (2) In Case (2), we add the prime q to the modulus m and require that the ele- ment γ is congruent to 1 mod q. The field M constructed in [24] then satisfies Gal(Mq/Qq) ∼= Z/2Z. Since q ≡ 1 (mod 8), Gal(Mq/Qq) can be embedded into a Z/4Z extension and therefore the embedding problem G/(F (3)) → Gal(M/Q) is solvable at q as well. As shown in [24] it is solvable at all other primes and hence globally solvable. 3Note that our m is denoted as 2m in the notation of [14]. 12 DANNY NEFTIN With these changes the field L constructed in [24] gives an S-adequate G/F (3)-extension. In order to make the set T split completely in L, we make the following additional modifications: (1) In Case (1.1), the embedding problem G/(F (3)) → Gal(N/Q) splits and hence has the trivial solution φ. Instead of applying Theorem 2.1, we apply Corollary 2.3 insuring the same prescribed conditions at S but in addition that the solution remains trivial at primes of T . (2) In Cases (1.2) and (2), we add the primes of N that lie over primes of T to the modulus m and require that γ ≡ 1 (mod p) for every p ∩ Q ∈ T . This insures that T splits completely in K (resp. in M) in Case (1.2) (resp. Case (2)). Let φ be the solution obtained in Case (1.2) (resp. Case (2)) of [24] to the embedding problem G/F (3) → Gal(K/Q) (resp. G/F (3) → Gal(M/Q)). We apply Theorem 2.1 in order to change φ to a solution ψ which is trivial at primes of T . Since the local embedding problem at p(2) is Frattini, ψ is surjective at p(2) , i = 1, 2. Thus, the fixed field L of ker ψ is S-adequate and T splits completely in L. i i (cid:3) TAMELY RAMIFIED SUBFIELDS OF DIVISION ALGEBRAS 13 References [1] E. Allman, M. Schacher, Division algebras with P SL(2, q)-Galois maximal subfields. J. Algebra 240 (2001), 808 -- 821. [2] E. Artin, J. Tate, Class field theory. Addison-Wesley Publishing Company, Redwood City, CA (1990). [3] D. Chillag, J. Sonn, Sylow-metacyclic groups and Q-admissibility. Israel J. Math. 40 (1981), 307 -- 323. [4] B. Fein, M. Schacher, Q-admissibility questions for alternating groups. J. Algebra 142 (1991), 360 -- 382. [5] W. Feit, SL(2, 11) is Q-admissible. J. Algebra 257 (2002), 244 -- 248. [6] W. Feit, The Q-admissibility of 2A6 and 2A7. Algebraic geometry and its applications (West Lafayette, IN, 1990), Springer, New York (1994), 197-202. [7] W. Feit, PSL2(11) is admissible for all number fields. Algebra, arithmetic and geometry with applications, Springer, Berlin (2004), 295-299. [8] W. Feit, The K-admissibility of 2A6 and 2A7. Israel J. Math. 82 (1993), 141-156. [9] P. Feit, W. Feit, The K-admissibility of SL(2, 5). Geom. Dedicata 36 (1990), 1-13. [10] W. Feit, P. Vojta, Examples of some Q-admissible groups. J. Number Theroy 26 (1987), 210 -- 226. [11] B. Gordon, M. Schacher, The admissibility of A5. J. Number Theroy 11, (1979), 498 -- 504. [12] D. Harbater, J.Hartmann, Division algebras and patching. Lecture notes for the 2012 Arizona Winter School (2012). [13] D. Harbater, J. Hartmann, D. Krashen, Patching subfields of division algebras. Transactions of the AMS, 363 (2011), 3335 -- 3349. [14] S. Liedahl, Presentations of metacylic p-groups with applications to K-admissibility questions. J. Algebra 169 (1994), 965 -- 983. [15] D. Neftin, Admissibility of finite groups over number fields. Ph.D. Thesis, Technion (2011). [16] D. Neftin, E. Paran, Patching and admissibility over two dimensional complete local domains. Algebra and Number theory 4 (2010), 743 -- 762. [17] J. Neukirch, On solvable number fields. Invent. Math. 53 (1979), 135 -- 164. [18] J. Neukirch, Uber das Einbettungsproblem der algebraischen Zahlentheorie. Invent. Math. 21 (1973), 59 -- 116. [19] J. Neukirch, A. Schmidt, K. Wingberg, Cohomology of number fields. Grundlehren der Math- ematischen Wissenschaften [Fundamental Principles of Mathematical Sciences]. Springer-Verlag, Berlin (2000). [20] R.S. Pierce, Associative Algebras. Springer-Verlag (1982). [21] M. Schacher, Subfields of division rings. I. J. Algebra 9 (1968), 451 -- 477. [22] M. Schacher, J. Sonn, K-admissibility of A6 and A7. J. Algebra 145 (1992), 333-338. [23] J. Sonn, Rational division algebras as solvable crossed products. Israel J. Math. 37 (1980), 246 -- 250. [24] J. Sonn, Q-admissibility of solvable groups. J. Algebra 84 (1983), 411 -- 419. [25] B. Surendranath Reddy, V. Suresh, Admissibility of groups over function fields of p-adic curves, arXiv:1201.1938. [26] E. Weiss, Algebraic number theory. McGraw-Hill Book Co., Inc., New York-San Francisco- Toronto-London (1963). Department of Mathematics, 530 Church St., University of Michigan, Ann Arbor 48109. E-mail address: [email protected]
1005.3193	1	1005	2010-05-18T13:15:16	Associative Geometries. II: Involutions, the classical torsors, and their homotopes	[ "math.RA" ]	For all classical groups (and for their analogs in infinite dimension or over general base fields or rings) we construct certain contractions, called "homotopes". The construction is geometric, using as ingredient involutions of associative geometries. We prove that, under suitable assumptions, the groups and their homotopes have a canonical semigroup completion.	math.RA	math	ASSOCIATIVE GEOMETRIES. II: INVOLUTIONS, THE CLASSICAL TORSORS, AND THEIR HOMOTOPES WOLFGANG BERTRAM AND MICHAEL KINYON Abstract. For all classical groups (and for their analogs in infinite dimension or over general base fields or rings) we construct certain contractions, called homo- topes. The construction is geometric, using as ingredient involutions of associative geometries. We prove that, under suitable assumptions, the groups and their ho- motopes have a canonical semigroup completion. Introduction: The classical groups revisited The purpose of this work is to explain two remarkable features of classical groups: (1) every classical group is a member of a "continuous" family interpolating between the group and its "flat" Lie algebra; put differently, there is a geo- metric construction of "contractions" (in this context also called homotopes), (2) every classical group and all of its homotopes admit a canonical completion to a semigroup; the underlying (compact) space of all of these "semigroup hulls" is the same for all homotopes. In fact, these results hold much more generally. The key property of classical groups is that they are closely related to associative algebras: either they are (quotients of) unit groups of such algebras, or they are (quotients of) ∗-unitary groups (0.1) U(A, ∗) := {u ∈ A uu∗ = 1} for some involutive associative algebra (A, ∗). This way of characterizing classical groups suggests to consider as "classical" also all other groups given by these con- structions, including infinite-dimensional groups and groups over general base fields or rings K, obtained from general involutive associative algebras (A, ∗) over K. On an algebraic, or "infinitesimal", level, features (1) and (2) are supported by simple observations on associative algebras: as to (1), associative algebras really are families of products (x, y) 7→ xay (the homotopes, see below), and as to (2), it is obvious that an associative algebra forms a semigroup and not a group with respect to multiplication. Our task is, then, to "globalize" these simple observations, and at the same time to put them into the form of a geometric theory: we have to free them from choices of base points (such as 0 and the unit 1 in an associative algebra). Just as in classical geometry, this means to proceed from a "linear" to a "projective" formulation, with an "affine" formulation as intermediate step. For classical groups of the "general linear type" (An), this has already been achieved in Part I of this 2000 Mathematics Subject Classification. 20N10, 17C37, 16W10. Key words and phrases. classical groups, homotope, associative triple systems, semigroup com- pletion, involution, linear relation, adjoint relation, complemented lattice, orthocomplementation, generalized projection, torsor. 1 2 WOLFGANG BERTRAM AND MICHAEL KINYON work ([BeKi09]). In the present article we look at the remaining families (Bn, Cn, Dn) and their generalizations. They correspond to associative algebras with involution, so that the geometric theory of involutions will be a central topic of this work. Let us start by describing the "infinitesimal" situation (i.e., the Lie algebra level), before explaining how to "globalize" it. 0.1. Homotopes of classical Lie algebras. The concept of homotopy is at the base of Part I of this work: an associative algebra A should be seen rather as a family of associative algebras (A, (x, y) 7→ xay), parametrized by a ∈ A. This gives rise to a family of Lie brackets [x, y]a = xay − yax also called homotopes, interpolating between the "usual" Lie bracket (a = 1) and the trivial one (a = 0). In particular, taking for A the matrix space M(n, n; K) with Lie bracket [X, Y ]A for A ∈ M(n, n; K) we get a Lie algebra which will be denoted by gln(A; K). For abstract Lie algebras, there is no such construction; however, there is a variant that can be applied to all classical Lie algebras: let us add an involution ∗ (antiautomorphism of order 2) as a new structural feature to our associative algebra A, and write A = Herm(A, ∗) ⊕ Aherm(A, ∗) = {x ∈ A x∗ = x} ⊕ {x ∈ A x∗ = −x} for the eigenspace decomposition. If we fix a ∈ Herm(A, ∗), then ∗ : A → A is an an- tiautomorphism of the homotopic bracket [·, ·]a, and therefore (Aherm(A, ∗), [·, ·]a), a ∈ Herm(A, ∗), is a family of Lie algebra structures on Aherm(A, ∗), again called homotopes. Remarkably, the construction works also in the other direction: if we fix a ∈ Aherm(A, ∗), then A → A, x 7→ x∗ is an automorphism of the homotope bracket [·, ·]a, and hence (Herm(A, ∗), [·, ·]a), a ∈ Aherm(A, ∗), is a family of "homo- topic" Lie algebra structures on Herm(A, ∗). For instance, taking for A the matrix algebra M(n, n; K) with involution X ∗ := X t (transposed matrix; in this case we write Sym(n; K) and Asym(n; K) for the eigenspaces), we get contractions of the orthogonal Lie algebras, denoted by on(A; K) := Asym(n; K) with bracket [X, Y ]A for symmetric matrices A. For A = 1, we get the usual Lie algebra o(n); for A = Ip,q (diagonal matrix of signature (p, q) with p + q = n) we get the pseudo-orthogonal algebras o(p, q), but for p + q < n we get a new kind of Lie algebras: they are not Lie algebras defined by a form since the Lie algebra of a degenerate form has bigger dimension than the one of a non-degenerate form, whereas our contractions preserve dimension. Likewise, for skew-symmetric A, we get homotopes of "symplectic type" spn/2(A; K) := Sym(n; K) with bracket [X, Y ]A. If n is even and A invertible, then this algebra is isomorphic to the usual symplectic algebra sp(n, K), and if A is not invertible, we get "degenerate" homotopes; as in the orthogonal case, these algebras are not Lie algebras defined by a degenerate skewsymmetric form. If n is odd, then the family contains only "degenerate members", which we call half-symplectic. Summing up, looking at homotope Lie brackets on Aherm(A, ∗) not only serves to imbed the usual Lie bracket into a family, but also to restore a remarkable formal du- ality between Aherm(A) and Herm(A) which usually gets lost. An algebraic setting that takes account of this duality from the outset is the one of an associative pair (see Appendix A and [BeKi09]). For instance, the square matrix algebras gln(A; K) are generalized by the rectangular matrix algebras glp,q(A; K) := M(p, q; K) with ASSOCIATIVE GEOMETRIES. II 3 bracket [X, Y ]A where A now belongs to the "opposite" matrix space M(q, p; K). In the pair setting, a map φ is an involution if and only if so is −φ, and hence Herm(φ) and Aherm(φ) simply interchange their roles if we replace φ by −φ. It is only the consideration of unit or invertible elements that may break this symmetry: they may exist in one space but not in the other. The following table summarizes the definition of classical Lie algebras and their homotopes. In the general linear cases, K may be any ring (in particular, the quaternions H are admitted); in the orthogonal and symplectic families K has to be a commutative ring, and for the unitary families we use an involution of K: if K = C, we use usual complex conjugation, and for K = H we use the following conventions: if nothing else is specified, we use the "usual" conjugation λ 7→ λ (minus one on the imaginary part imH and one on the center R ⊂ H). If we consider H with its "split" involution λ 7→eλ := jλj−1, then we write eH. For instance, Herm(n;eH) is the space of quaternionic matrices such that eX = X t, and un(1;eH) is the Lie algebra often denoted by so∗(2n). In all cases, the Lie bracket is [X, Y ]A = XAY − Y AX. Note finally that the trace map does not behave well with respect to our contractions, and therefore we do not define homotopes of special linear or special unitary algebras. label and space gln(A; K) := M(n, n; K) parameter space Lie bracket family name general linear (square) A ∈ M(n, n; K) general linear (rectan.) glp,q(A; K) := M(p, q; K) A ∈ M(q, p; K) orthogonal on(A; K) := Asym(n; K) A ∈ Sym(n; K) spn/2(A; K) := Sym(n; K) A ∈ Asym(n; K) [half-] symplectic un(A; C) := Aherm(n; C) A ∈ Herm(n; C) C-unitary H-unitary un(A; H) := Aherm(n; H) A ∈ Herm(n; H) H-unitary split [X, Y ]A [X, Y ]A [X, Y ]A [X, Y ]A [X, Y ]A [X, Y ]A [X, Y ]A un(A;eH) := Aherm(n;eH) A ∈ Herm(n;eH) The expert reader will certainly have remarked that everything we have said so far holds, mutatis mutandis, for "Lie" replaced by "Jordan": Herm(A, ∗) is a Jordan algebra, and in the Jordan pair setting the roles of Herm(A) and Aherm(A) become more symmetric. Indeed, a conceptual and axiomatic theory will use the Jordan- and Lie-aspects of an associative product in a crucial way -- see remarks in Chapter 6 and in [Be08c]. In order to keep this paper accessible for a wide readership, no use of Jordan theory will be made in this work. 0.2. Homotopes of classical groups. Now let us explain the main ideas serving to "globalize" the Lie algebra situation just described. First of all, for the classical Lie algebras introduced above it is easy to define explicitly a corresponding algebraic group: in the setting of an abstract unital algebra A with Lie bracket [x, y]a = xay − yax, one defines the set and checks that G(A, a) := {x ∈ A 1 − xa ∈ A×} x ·a y := x + y − xay is a group law on G(A, a) with neutral element 0 and inverse of x given by ja(x) := −(1 − xa)−1x. 4 WOLFGANG BERTRAM AND MICHAEL KINYON It is easily seen (cf. Lemma 1.2) that the Lie algebra of this group is given by the bracket [x, y]a. Next, observe that an involution ∗ of A induces an isomorphism from G(A, a) onto the opposite group of G(A, a∗). Therefore, if a is Hermitian, ∗ induces a group antiautomorphism of order 2, and we can define the a-unitary group as usual to be the subgroup of elements g ∈ G(A, a) such that g∗ = ja(g). If a is skew-Hermitian, the a-symplectic group is defined similarly by the condition −g∗ = ja(g). Specializing to the classical matrix algebras, we get the following list of classical groups: label GLn(A; K) GLp,q(A; K) On(A; K) Spn/2(A; K) Un(A; C) Un(A; H) parameter space product underlying set := {X ∈ M(n, n; K)1 − AX invertible} A ∈ M(n, n; K) X ·A Y := {X ∈ M(p, q; K)1 − AX invertible} A ∈ M(q, p; K) X ·A Y := {X ∈ GLn(A, K)X + X t = X tAX} A ∈ Sym(n; K) X ·A Y := {X ∈ GLn(A, K)X − X t = X tAX} A ∈ Asym(n; K) X ·A Y := {X ∈ GLn(A, K)X + X AX} A ∈ Herm(n; C) X ·A Y AX} A ∈ Herm(n; H) X ·A Y := {X ∈ GLn(A, H)X + X = X = X t t t t On(A;eH) := {X ∈ GLn(A, H)X + eX t = eX tAX} A ∈ Herm(n;eH) X ·A Y Finally, one may observe that this realization of classical groups has the advantage of leading to a natural "semigroup hull": e.g., if At = A, a direct computation shows that the set On(A; K) := {X ∈ M(n, n; K) X t + X = X tAX} is stable under the product ·A, which turns it into a semigroup with unit element 0, and similarly in all other cases. 0.3. "Projective" theory of classical torsors. The definition of the classical groups given above is useful for calculating their Lie algebras and for starting to analyze their group structure (and their topological structure if K is a topological field or ring), but also has several drawbacks: firstly, note that the product X ·A Y is affine in both variables, and hence our groups are realized as subgroups of the affine group of the matrix space M(n, n; K). The corresponding linear representation in a space of dimension n2 + 1 is not very natural, and one may wish to realize these groups in more natural linear representations. Secondly, whereas the general linear groups are, for all A, realized as (Zariski-dense) parts of a common ambient space (M(n, n; K), resp. M(p, q; K)), this is not the case for the other classical groups: the underlying set depends on A, and hence the realization is not adapted to the point of view of deformations or contractions. Finally, and related to the preceding item, one has the impression that the "semigroup hull" On(A; K) depends on the realization, and that it should rather be part of some maximal semigroup hull intrinsically associated to the group On(A; K) . In the present work, we will give another realization of the classical groups (and, much more generally, of the groups attached to abstract involutive algebras) having none of these drawbacks: it is a sort of projective realization, as opposed to the affine picture just given. In a first step, we get rid of base points in groups by considering them as torsors, that is, we work with the ternary product (xyz) := xy−1z of a group. By classical torsor we simply mean a classical group from the preceding table equipped with this ternary law, i.e., by forgetting their base points. For the general linear family, we have seen in Part I of this work that there is a common ASSOCIATIVE GEOMETRIES. II 5 realization of all groups GLp,q(A, K) inside the Grassmannian X := Gras(Kp+q) in such a way that they are realized as subgroups of the projective group PGL(p+q, K). The parameter space is again the complete space X , and "space" and "parameter" variables are incorporated into a single object (called an associative geometry, given by a pentary product map Γ : X 5 → X ) having surprising properties. In the present work we show that, for the other families, there is a more refined construction, relying on the existence of involutions (antiautomorphisms of order 2) of associative geometries. For the classical groups, these involutions are orthocomplementation maps, so that the fixed point spaces are varieties of Lagrangian subspaces. We will realize all orthogonal groups as (Zariski dense) subsets of the Lagrangian variety of a quadratic form of signature (n, n), and the [half-] symplectic groups in the Lagrangian variety of a symplectic form on K2n. The underlying Lagrangian variety plays the role of a "projective completion" of these groups (also called "projective compactification" if K = R or K = C since it is compact in these cases), and in particular we will show that the group law extends to a semigroup law on the projective completion, thus defining the intrinsic and maximal (compact) semigroup hull for all classical groups and their homotopes. As in the general linear case, this achieves a realization in which all "deformations" or "contractions" are globally defined on the space level. In contrast to the general linear case, the parameter space now is different from the underlying Lagrangian variety of the group spaces: it is another Lagrangian variety which we call the dual Lagrangian. This duality reflects the duality between Herm(A, ∗) and Aherm(A, ∗) mentioned in Section 0.1. 0.4. Contents. The contents of this paper is as follows: in Chapter 1 we recall ba- sic facts on the "general linear construction"; in Chapter 2 we define and construct involutions of associative geometries: in Theorem 2.2 we prove that orthocomple- mentation maps of non-degenerate forms are involutions; in Chapter 3 we describe the "projective" construction of torsors and groups associated to (restricted) invo- lutions of associative geometries (Lemma 3.1), their tangent objects with respect to various choices of base points (Theorems 3.6 and 3.6) as well as the link with the "affine" realization given above (Theorem 3.3). In Chapter 4 we present the classi- fication of homotopes of classical groups (over K = R or C, the case of general base fields or rings being at least as complicated as the problem of classifying involutive associative algebras, see [KMRS98]). In Chapter 5 we describe the semi-group com- pletion of classical groups (Theorem 5.6); the main difficulty here is to prove that non-degenerate forms induce involutions of geometries in a "strong" sense. This requires some investigation of the linear algebra of linear relations, complementing those from Chapter 2 of Part I of this work, and which may be of interest in its own right. Finally, in Chapter 6 we give some brief comments on a possible axiomatic approach, involving both the Jordan- and the Lie side of the whole structure, and Appendix A contains the relevant definitions on involutions of associative pairs. 0.5. Related work. Finally, let us add some words on related literature. It seems to be folklore in symplectic geometry that the group law of Sp(m, R) extends to the whole Lagrangian variety if we interpret it via composition of linear relations: the composition of two Lagrangian linear relations is again Lagrangian (see appendix on "linear symplectic reduction" in [CDW87] or Theorem 21.2.14 in [Ho85]). In 6 WOLFGANG BERTRAM AND MICHAEL KINYON a case-by-case way, Y. Neretin ([Ner96]) has given similar constructions for other families of complex or real Lagrangrian varieties ("categories B, C, D", see loc. cit., p. 85 ff and loc. cit. Appendix A for their real analogs). It would be very interesting to investigate further the relationship between our work and Neretin's, in particular in view of applications in harmonic analysis and quantization. Note that Neretin in loc. cit. p. 59 uses a modified composition law of linear relations in order to obtain a jointly continuous operation; since we do not consider topologies here, we leave the [important] topic of joint continuity for later work. Notation. Throughout this work, K denotes a commutative unital ring and B an associative unital K-algebra, and we will consider right B-modules V, W, . . .. We think of B as "base ring", and the letter A will be reserved for other associative K-algebras such as EndB(W ). If V = a ⊕ b is a direct sum decomposition of a vector space or module, we denote by P a b : V → V the projection with kernel a and image b. 1. The general linear family 1.1. Groups and torsors living in Grassmannians. We are going to recall the basic construction from [BeKi09] which realizes groups like GLn(A, K) inside a Grassmannian manifold. Let W be a right B-module and X = Gras(W ) be the Grassmannian of all right B-submodules of W . A pair (x, a) ∈ X 2 is called transversal (denoted by a⊤x or x⊤a) if W = x ⊕ a. The set of all complements of a is denoted by Ca, so that Cab := Ca ∩ Cb is the set of common complements of a and b. One of the main results of [BeKi09] says that the set Cab carries two canonical torsor-structures. More precisely, we define, for (x, a, b, z) ∈ X 4 such that a⊤x, b⊤z, the endomorphism of W (1.1) Mxabz := P a x − P z b = P a x − 1 + P b z . By a direct calculation (see [BeKi09], Prop. 1.1), one sees that (1.2) Mxabz = Mzbax, Mxabz = −Maxzb, and, if x, z ∈ Uab, then Mxabz is invertible with inverse (Mxabz)−1 = Mzabx = Mxbaz. (1.3) Recall (see, e.g., [BeKi09]) that a torsor is the base point-free version of a group (a set G with a ternary map G3 → G, (xyz) 7→ (xyz) such that (xyy) = x = (yyx) and (xy(zuv)) = ((xyz)uv)). Then ([BeKi09], Th. 1.2): Theorem 1.1. i ) For a, b ∈ X fixed, Cab with product (xyz) := Γ(x, a, y, b, z) := Mxabz(y) is a torsor (which will be denoted by Uab). In particular, for all y ∈ Cab, the set Cab is a group with unit y and multiplication xz = Γ(x, a, y, b, z). ii ) Uab is the opposite torsor of Uba (same set with reversed product): Γ(x, a, y, b, z) = Γ(z, b, y, a, x) In particular, the torsor Ua := Uaa is commutative. (1.4) (cid:0)HomB(o+, o−), HomB(o−, o+)(cid:1). ASSOCIATIVE GEOMETRIES. II 7 iii ) The commutative torsor Ua is the underlying additive torsor of an affine space: Ua is an affine space over K, with additive structure given by (sum of x and z with respect to the origin y), and action of scalars given by x +y z = Γ(x, a, y, a, z), a + P a (multiplication of y by s with respect to the origin x). Πs(x, a, y) := sy + (1 − s)x = (sP x x )(y) Definition. (The restricted multiplication map) We call restricted multipli- cation map the map Γ : D5 → X , defined on the set of admissible 5-tuples D5 := {(x, a, y, b, z) ∈ X 5 x, y, z ∈ Cab}, by the formula from part i) of the preceding theorem. Definition. (Base points and tangent spaces) A base point in X is a fixed transversal pair, usually denoted by (o+, o−). The tangent space at (o+, o−) is the pair (A+, A−) := (Co−, Co+). Note that (A+, A−) is a pair of K-modules (with origin o± in A±), isomorphic to This tangent space carries the structure of an associative pair given by trilinear products (see [BeKi09], Th. 1.5) A± × A∓ × A± → A±, (u, v, w) 7→ hu, v, wi± := Γ(u, o+, v, o−, w). (1.5) Definition. (Transversal triples) A transversal triple is a triple of mutually If we fix such a triple, we usually denote it by (o+, e, o−). transverse elements. In this case, A := Co− carries the structure of an associative algebra with origin o := o+ and unit e, called the tangent algebra at o+ corresponding to the base triple (o+, e, o−), with product (1.6) A × A → A, (u, v) 7→ Γ(u, o+, e, o−, v). In a dual way, Co+ is turned into an algebra with origin o−. Both algebras are canonically isomorphic via the inversion map j = Meo+o−e. 1.2. Lie algebra and structure of the torsors Uab. We explain the link between the torsors Uab and the groups GLp,q(A; B) defined in the Introduction, as well as the computation of their "Lie algebra". Lemma 1.2. Choose an origin o+ in Uab and an element o−⊤o+. Then the Lie algebra (in a sense to be explained in the following proof ) of the group (Uab, o+) is the "tangent space" A+ = HomB(o+, o−) with Lie bracket [X, Y ] = X(a − b)Y − Y (a − b)X (note that o+ ∈ Uab means that a, b ∈ Co+ = A−, so that a − b ∈ A−). In particular, choosing o− = b, we get the Lie algebra of UA0: [X, Y ] = XAY − Y AX. 8 WOLFGANG BERTRAM AND MICHAEL KINYON Proof. The Lie algebra can be defined in a purely algebraic way, without using ordinary differential calculus, as follows. Let T K := K[ε] := K[X]/(X 2), ε2 = 0 be the ring of dual numbers over K and T T K := T (T K) := (K[ε1])[ε2] be the "second order tangent ring". Then (X , Γ) admits scalar extensions from K to T K and to T T K, and the commutator in the second scalar extension of the group Uab gives rise to the Lie bracket in the way described in [Be08], Chapter V. This construction is intrinsic and does not depend on "charts". Therefore we may choose o− := b in order to simplify calculations (the first formula from the claim then follows from the second one). Then Uab = Ca ∩ Co− = Ca ∩ A+, and according to [BeKi09], Section 1.4 we have the following "affine picture" of the group (Uab, o): if, under the isomorphism (1.4), a corresponds to the element A ∈ A− = HomB(o−, o+), then Uab corresponds to the set (1.7) UA0 = {X ∈ HomB(o+, o−) 1 − AX is invertible in EndB(o+)} with group law given by the product Z ·A X defined in the Introduction: (1.8) X · Z = X + Z − ZAX. Since Formulas (1.7) and (1.8) are algebraic, we may now determine explicitly the tangent group of U0A via scalar extension by dual numbers: the operator 1 − (A + εA′)(X + εX ′) = 1 − AX + ε(A′X + AX ′) is invertible iff so is 1−AX, hence the tangent bundle T (U0A) is U0A×εHomB(o+, o−), with semidirect product group structure (1.9) (X, εX ′) · (Z, εZ ′) =(cid:0)X + Z − ZAX, ε(X ′ + Z ′ + Z ′AX + ZAX ′)(cid:1). Repeating the construction, we obtain the second tangent bundle T T (U0A) by scalar extension from K to the ring T T K. As explained in [Be08], the Lie bracket [X, Y ] arises from the commutator in the second tangent group via ε1ε2[X, Y ] = (ε1X)(ε2Y )(ε1X)−1(ε2Y )−1. A direct calculation, based on (1.9), yields (ε1X)(ε2Y ) = ε1X + ε2Y + ε1ε2Y AX, which, after a short calculation using that (ε1X)−1 = ε1(−X), (ε1Y )−1 = ε1(−Y ), implies the claim. (cid:3) As is easily seen from the explicit formulas given above by choosing for A special (idempotent) elements (cf. [Be08b]), the groups Uab and their Lie algebra have a double fibered structure. These and related features for symmetric spaces will be investigated in [BeBi]. 2. Construction of involutions 2.1. Definition of (restricted) involutions. Whenever in a category we have for each object X a canonical notion of an "opposite object" X op, there is a natural notion of involution. This is the case for groups, torsors or associative geometries. Definition. A restricted involution of the Grassmannian geometry X = Gras(W ) is a bijection f : X → X of order two and such that ASSOCIATIVE GEOMETRIES. II 9 (1) f preserves transversality: for all a, x ∈ X : a⊤x iff f (a)⊤f (x), (2) f is an isomorphism onto the opposite restricted product map: for all 5- tuples (x, a, y, b, z) such that x, y, z ∈ Uab, f(cid:0)Γ(x, a, y, b, z)(cid:1) = Γ(f x, f b, f y, f a, f z) = Γ(f z, f a, f y, f b, f x). (3) f induces affine maps on affine parts: for all 3-tuples (x, a, y) such that x, y⊤a, and r ∈ K, f(cid:0)Πr(x, a, y)) = Πr(cid:0)f x, f a, f y(cid:1). In other words, by (1), f induces well-defined restrictions Uab → Uf (a),f (b) and Ua → Uf (a), which induce, by (2), anti-isomorphisms of torsors Uab → Uf (a),f (b), and by (3), isomorphisms of affine spaces Ua → Uf (a). The fixed point space Y := X τ of an involution τ will be called the Lagrangian type geometry of (X , τ ) (if it is not empty). In general, nothing guarantees existence of restricted involutions. Before turning to the general theory (next chapter), we will show that under certain conditions one can construct them by using bilinear or sesquilinear forms. In these cases, Y will be indeed realized as a geometry of Lagrangian subspaces. 2.2. Non-degenerate forms and adjoinable pairs. We assume that our B- module W admits a non-degenerate sesquilinear form β : W × W → B. By sesquilinearity we mean β(vr, w) = rβ(v, w), β(v, wr) = β(v, w)r for v, w ∈ W , r ∈ B, where B → B, z 7→ z is some fixed involution (antiautomorphism of order 2) of B, and non-degeneracy means that β(v, W ) = 0 or β(W, v) = 0 implies v = 0. Of course, for B = K and z = z we get bilinear forms. Moreover, we assume that β is Hermitian or skew-Hermitian: ∀v, w ∈ W : β(v, w) = β(w, v), resp. ∀v, w ∈ W : β(v, w) = −β(w, v). As usual, the orthogonal complement of a subset S ⊂ W will be denoted by S ⊥. The orthogonal complement of a right submodule is again a right submodule, but, unfortunately, it is in general not true that the orthocomplementation map ⊥: X → X satisfies the properties of a (restricted) involution: in general, it does not even preserve transversality, nor is it of order two. Definition. A pair (x, a) ∈ X × X is called adjoinable if W = x ⊕ a and W = x⊥ ⊕ a⊥. Lemma 2.1. A pair (x, a) ∈ X × X is adjoinable if and only if the projection P := P a x is adjoinable; i.e., there exists a linear operator P ∗ : W → W such that (2.1) ∀v, w ∈ W : β(v, P w) = β(P ∗v, w). Moreover, in this case we have (x⊥)⊥ = x and (a⊥)⊥ = a. 10 WOLFGANG BERTRAM AND MICHAEL KINYON Proof. Assume P ∗ exists. If two operators f, g are adjoinable, then we have (gf )∗ = f ∗g∗, and hence P ∗ is again idempotent. Moreover, the kernel of P ∗ is ker P ∗ = (imP )⊥ = x⊥. Now, P is adjoinable if and only if so is Q := 1 − P , whence imP ∗ = ker Q∗ = a⊥, and thus W = x⊥ ⊕ a⊥. Moreover, this shows that (2.2) (P a x )∗ = P x⊥ a⊥ . Reversing these arguments, we see that, if (x, a) is adjoinable, equation (2.2) defines an operator P ∗, and a direct check shows that then (2.1) holds. Moreover, from (P ∗)∗ = P the relations (x⊥)⊥ = x and (a⊥)⊥ = a follow. (cid:3) The lemma shows that, in the general case, we should not work with the full Grassmannian, but only with its adjoinable elements. For simplicity, let us first look at a case where the Grassmannian is well-behaved, namely the case W = Bn: Theorem 2.2. (Construction of involutions: case of Bn) Let W = Bn and X be the Grassmannian of all right submodules that admit some complementary right submodule, and let β be a non-degenerate Hermitian or skew-Hermitian form on B. Then the orthocomplementation map ⊥β: X → X , x 7→ x⊥ is a restricted involution of X . Proof. For W = Bn, every non-degenerate sesquilinear form is given by β(x, y) = nXi,j=1 xibijyj with some invertible matrix B = (bij). By assumption, B is Hermitian or skew- Hermitian. As can be checked by a direct matrix calculation, in this case every linear operator X : W → W is adjoinable, with adjoint given by the adjoint matrix X ∗ of (Xij): X ∗ = B−1X t B where X t is the transposed matrix of X. In particular, if x is an arbitrary com- plemented right-submodule of Bn with complement a, then P := P a x is adjoinable. Thus every transversal pair (x, a) is adjoinable, and moreover x⊥ = im(P )⊥ = ker(P ∗). We have thus shown that the orthocomplementation map is of order two and pre- serves transversalilty. In order to prove the crucial property (2.3) we observe that, for all x ∈ Gras(W ) and all linear maps F : W → W Γ(z⊥, a⊥, y⊥, b⊥, x⊥) =(cid:0)Γ(x, a, y, b, z)(cid:1)⊥ (F x)⊥ = (F ∗)−1(x⊥) (2.4) (inverse image), and if F is bijective, (F ∗)−1 = (F −1)∗ (inverse map). We apply this to the bijective map F = Mxabz (for x, z ∈ Cab) whose inverse is F −1 = Mzabx = Mxbaz and whose adjoint can be computed using (2.2): for x, z ∈ Cab, the operator Mxabz has an adjoint given by (Mxabz)∗ = (P a (2.5) b )∗ = Ma⊥x⊥z⊥b⊥ = −Mx⊥a⊥b⊥z⊥. x − P z ASSOCIATIVE GEOMETRIES. II 11 Now let a, b ∈ X and x, y, z ∈ Cab. Then, with F = Mxabz, (cid:0)Γ(x, a, y, b, z)(cid:1)⊥ = (F ∗)−1y⊥ = (cid:0)F (y)(cid:1)⊥ = M −1 x⊥a⊥b⊥z⊥(y⊥) = Mx⊥b⊥a⊥z⊥(y⊥) = Γ(x⊥, b⊥, y⊥, a⊥, z⊥) . This proves (2.3). Finally, property (3) of an involution can be proved in the same way as (2.3) (and this property is already known since it depends only on the underlying Jordan structure, see, e.g., [Be04]). (cid:3) The cases n = 1 and n = 2 of the preceding result deserve special interest. For n = 1, we work with the form β(u, v) = u v, and we consider the Grassmannian of complemented right ideals in B with involution ker e 7→ im e (where e ∈ B is an idempotent, ker e = (1 − e)B, ime = eB). The case n = 2 enters in the proof of Theorem 3.7 (next chapter). 2.3. The adjoinable Grassmannian. As we will see in Theorem 3.7, the case n = 2 is already suitable to treat all seemingly more general cases. Returning thus to the case of a general B-module W with a non-degenerate Hermitian or skew-Hermitian form β, we may proceed as follows: let A := {f ∈ EndB(W ) ∃f ∗ ∈ EndB(W ) : ∀v, w ∈ W : β(v, f w) = β(f ∗v, w)} the set of all adjointable linear operators. Then A is a subalgebra of EndB(W ), and ∗ is an involution on A. Now define the adjoinable Grassmannian of β to be Xβ := {im P P ∈ A, P 2 = P }, the set of all submodules x admitting a complement a such that the projection P := P a x is adjointable. (In general, not all submodules have this property -- consider e.g. a dense proper subspace x in a Hilbert space.) Let X := {P A P ∈ A, P 2 = P } be the Grassmannian of all complemented right modules in A. Then the map X → Xβ, P A 7→ imP is well-defined, bijective and compatible with the structure maps Γ. We use it to push down τ to an involution of Xβ, so that we can carry out all preceding constructions on the adjoinable Grassmannian. 3. Groups and torsors associated to involutions We assume, for all of this chapter, that τ : X → X is a restricted involution of the Grassmannian geometry X = GrasB(W ) and write Y for its fixed point space. There are two different ways to construct groups and torsors associated to (X , τ ). Here is the first construction, which simply mimics the usual definition of unitary and orthogonal groups: Definition. Fix three points a, o, b ∈ Y such that o ∈ Uab, considered as origin in the group (Uab, o), and let x−1 := Moabo(x) be inversion in this group. Then τ induces an antiautomorphism of this group: τ (xy) = τ Γ(x, a, o, b, y) = Γ(τ (y), τ (a), τ (o), τ (b), τ (x)) = Γ(τ (y), a, o, b, τ (x)) = τ (y)τ (x), 12 WOLFGANG BERTRAM AND MICHAEL KINYON and hence is a subgroup, called the τ -unitary group (located at (a, o, b)). U(τ ; a, o, b) := {x ∈ Uab τ (x) = x−1} This group is not a subset of the Lagrangien geometry Y, but rather is "tangent" to the "antifixed space of τ ": indeed, the differential of inversion at o is the negative of the identity, and hence the tangent space of U(τ ; a, o, b) at the identity should be the minus one eigenspace of τ . This will be made precise below (Theorem 3.3). Next, we describe a second construction of groups having the advantage that it directly leads to torsors living in the Lagrangian geometry: Lemma 3.1. Let τ : X → X be a restricted involution of the Grassmannian geom- etry X = GrasB(W ) and denote by Y := X τ its Lagrangian type geometry. Then i ) for any a ∈ X , τ induces a torsor-automorphism of the torsor Ua,τ (a). In particular, the fixed point set G(τ ; a) := (Ua,τ (a))τ = Ua,τ (a) ∩ Y is a subtorsor of Ua,τ (a). ii ) As a set, G(τ, a) = Ua ∩ Y. iii ) G(τ, τ (a)) is the opposite torsor of G(τ, a). If a ∈ Y, then the torsor G(τ, a) is abelian, and it is the underlying additive torsor of an affine space over K. Proof. (i) Note first that x ∈ Ca,τ (a) if and only if τ (x) ∈ Cτ (a),τ 2(a) = Ca,τ (a) since τ preserves transversality and is of order 2. Next we show that τ preserves the torsor law (xyz)a = Γ(x, a, y, τ (a), z) of Ua,τ (a): τ (((xyz)a) = τ (Γ(x, a, y, τ (a), z)) = Γ(τ z, τ a, τ y, a, τ x) = Γ(τ x, a, τ y, τ a, τ z) = (τ x τ y τ z)a. Clearly, the fixed point space Ua,τ (a) ∩ Y is then a subtorsor. (ii) If x ∈ Y, i.e., τ (x) = x, then x⊤a is equivalent to x⊤τ (a), whence Y ∩ Ua = Y ∩ Ua ∩ Uτ (a) = Y ∩ Ua,τ (a). (iii) Ua,τ (a) is the opposite torsor of Uτ (a),a. If a = τ (a), then the arguments given above show that τ is an automorphism of order 2 of the affine space Ua and hence its fixed point space is an affine subspace. (cid:3) In order to compare both constructions, we have to to study the behaviour of involutions with respect to basepoints. 3.1. Basepoints, and the dual involution. Let us fix a base point (o+, o−) in X . Recall from [BeKi09], Th. 1.3, that the middle multiplication operator Mo+o−o−o+ is an automorphism of Γ. By (1.3), it is invertible and equal to its own inverse. Moreover, Mo+o−o−o+(o±) = Γ(o+, o−, o±, o−, o+) = o±. Thus Mo+o−o−o+ is a base point preserving automorphism of the Grassmannian geometry. Its effect on the additive groups A± is simply inversion, that is, multi- plication by the scalar −1. Definition. A (restricted) involution τ of X is called ASSOCIATIVE GEOMETRIES. II 13 • base point preserving if τ (o+) = o+ and τ (o−) = o−, and • base point exchanging if τ (o+) = o− and τ (o−) = o+. Lemma 3.2. Assume τ is a base point preserving or base point exchanging involu- tion of X . Then τ commutes with the automorphism Mo+o−o−o+, and τ ′ := Mo+o−o−o+ ◦ τ = τ ◦ Mo+o−o−o+ is again of the same type (base point preserving, resp. exchanging involution) as τ . We call τ ′ the dual involution (denoted by −τ in a context where (o+, o−) is fixed). Proof. Thanks to the symmetry relation Mxabz = Maxzb we get in either case τ ◦ Mo+o−o−o+ ◦ τ = Mτ o+,τ o−,τ o−,τ o+ = Mo+o−o−o+. Therefore τ ′ is again of order 2, and it is an antiautomorphism having the same effect on o± as τ since Mo+o−o−o+ is base point preserving. (cid:3) Recall from [BeKi09] that, with respect to a fixed base point (o+, o−) and a ∈ A−, ta := Mo+ao−o+ ◦ Mo+o−o−o+ = Mao+o+o− ◦ Mo−o+o+o− = Lao+o−o+ is the (left) translation operator defined by a in the abelian group Uo+ ∼= A−. It acts rationally on A+ by the so-called quasi inverse map. Theorem 3.3. Assume τ is a base point preserving involution of X and let a ∈ Y ∩ Uo− = (A+)τ . Then the groups G(−τ ; a) and U(τ ; 2a, o+, o−) are isomorphic (the multiple 2a = a + a taken in A+). An isomorphism is induced by ta. Proof. Having fixed the base point, we use the notation −id := Mo+o−o−o+. We have to show that the group Ua,−a with its automorphism τ ′ is conjugate to the group U2a,o− with its automorphism i2aτ where i2a := Mo+2a o−o+ is inversion in the group (U2a,o−, o+). First of all, ta(a) = a + a = 2a, ta(−a) = a + (−a) = o− (sums in (A−, o−)), hence ta induces a torsor isomorphism from Ua,−a onto U2a,o− preserving the base point o+. Next, observe that i2a ◦ (−id) = Mo+2ao−o+ ◦ Mo+o−o−o+ = t2a whence, using that τ ′ ◦ ta = tτ ′a ◦ τ ′ = t−a ◦ τ ′, t−a ◦ i2aτ ◦ ta = t−a ◦ t2a ◦ (−id) ◦ τ ◦ ta = t−a ◦ t2aτ ′ ◦ ta = t−at2at−a ◦ τ ′ = τ ′ where the last equality follows from the relation tbtc = tb+c. (cid:3) In the affine chart A+, ta acts as a birational map, transforming the affine real- ization U(τ ; 2a, o+, o−) to a rational realization that is Zariski-dense in (A+)−τ . If 2 is invertible in K, all τ -unitary groups U(τ ; b, o, c) have such a realization G(τ ′; a) (just choose the base point (o+, o−) = (o, c) and let a := b/2). If 2 is not invertible in K, such a realization is not always possible. Concerning involutions of associative pairs and associative triple systems, to be used in the following result, see Appendix A. Theorem 3.4. Assume τ is a restricted involution of the Grassmannian geometry X , and let (A+, A−) be the associative pair corresponding to a base point (o+, o−). 14 WOLFGANG BERTRAM AND MICHAEL KINYON i ) If τ : X → X is base-point preserving, then by restriction τ induces K-linear maps τ ± : A± → A± which form a type preserving involution of (A+, A−). ii ) If τ : X → X is base-point exchanging, then by restriction τ induces K-linear maps τ ± : A± → A∓ which form a type exchanging involution of (A+, A−). In this case A := A+ becomes an associative triple system of the second kind when equipped with the product hxyzi := Γ(x, o+, τ (y), o−, z). iii ) Assume τ : X → X is base-point preserving, and let a ∈ Y ′ such that o+⊤a (i.e., a ∈ A− and τ (a) = −a). Then the Lie algebra of the group (G(τ ; a), o+) is the space (A+)τ + with Lie bracket [x, z]a = 2(hxazi − hzaxi). Proof. (i), (ii): All claims are simple applications of the functoriality of associating an associative pair to an associative geometry with base pair, [BeKi09], Theorem 3.5. For convenience, let us just spell out the computation proving the property of an associative triple system in part ii): huhxyziwi = Γ(cid:16)u, o+, τ(cid:0)Γ(x, o+, τ (y), o−, z)(cid:1), o−, w(cid:17) = Γ(cid:16)u, o+, Γ(τ z, τ o+, y, τ o−, τ x), o−, w(cid:17) = Γ(cid:16)u, o+, Γ(τ z, o−, y, o+, τ x), o−, w(cid:17) = Γ(cid:16)Γ(u, o+, τ z, o−, y), o+, τ x, o−, w(cid:17) = hhuzyixwi (If we had used a base point preserving automorphism instead of an involution, a similar calculation shows that we would get an associative triple system of the first kind, see Appendix A.) (iii): Using Lemma 1.2, with b = τ (a) = −a (since the effect of τ on A− is (cid:3) multiplication by −1), we get the Lie bracket [x, z] = hx(2a)zi − hz(2a)xi. Putting the preceding two results together, we obtain an explicit description of the groups G(−τ ; b/2) ∼= U(τ ; b, o+, o−) in terms of the associative pair (A+, A−): U(τ ; b, o+, o−) = {x ∈ A+ 1 − xb invertible, τ (x) = jb(x)} with jb(x) = −(1 − xb)−1x, so that the condition −τ (x) = jb(x) is equivalent to x + τ (x) = hxbτ (x)i. This formulation is valid for an arbitrary associative pair with base-point preserving involution. In practice, all known examples arise for associative pairs corresponding to unital associative algebras, to be discussed next. 3.2. Base triples, unitary groups, and Cayley transform. Next let us assume that W admits a transversal triple (o+, e, o−). Then W = o+ ⊕ o−, and saying that e is transversal to o+ and o− amounts saying that e is the graph of a linear isomorphism o+ → o−. We may consider this isomorphism as an identification, so that e becomes the diagonal ∆+ in W = o+ ⊕ o− = o+ ⊕ o+. Then the element −e := Mo+o−o−o+(e) ASSOCIATIVE GEOMETRIES. II 15 becomes the antidiagonal ∆− in o+ ⊕ o+. In this situation, we may let the group GL(2, K) act by block-matrices on W = o+ ⊕o+ in the usual way. Let G ⊂ GL(2, K) by the group generated by (cid:18)1 1 0 1(cid:19) , (cid:18)λ 0 0 1(cid:19) , (cid:18)0 1 1 0(cid:19) with λ ∈ K×. The first matrix describes left translation by e, the second multiplication by the scalar λ, Leo−o+o− := 1 − P e o−P o− o+ , o+o− = λP o+ δλ o− + P o− o+ , and the third describes a map j whose effect on the associative algebra A is inversion: j := Meo+o−e = Mo+eeo−. All of these operators are (inner) automorphisms of the geometry (X , Γ). ¿From (1.1) it follows that j is an automorphism of order 2, but this time it exchanges the points o+ and o−: Meo+o−e(o+) = Γ(e, o+, o+, o−, e) = Γ(o+, e, o+, e, o−) = o−. Moreover, j(e) = Meo+o−e(e) = Γ(e, o+, e, o−, e) = e. Definition. If (o+, e, o−) is a transversal triple, we call τ a • unital base point preserving involution if τ (o+) = o+, τ (o−) = o−, τ (e) = e, • unital base point exchanging involution if τ (o+) = o−, τ (o+) = o−, τ (e) = e. Note that, if τ is of one of these two types, then the dual involution τ ′ no longer preserves e. Indeed, Mo+o−o−o+(e) = −e is the antidiagonal, which is different from the diagonal (if W has no 2-torsion). Thus the roles of τ and τ ′ are no longer completely symmetric in the unital case. Lemma 3.5. Assume τ is a unital base point preserving involution of X . Then τ commutes with the automorphism j = Meo+o−e, and τ := jτ = τ j is a unital base-point exchanging involution. Moreover, if 2 is invertible in K, there exists an automorphism ρ : X → X ("the real Cayley transform") such that ρ ◦ τ ◦ ρ−1 = τ, ρ ◦ τ ◦ ρ−1 = τ ′. Proof. As in the proof of Lemma 3.2, we see that τ jτ = τ Meo+o−eτ = Meo+o−e = j, hence jτ is of order two, and it exchanges base points and is again an involution. The automorphism ρ is constructed as follows: let ρ ∈ G be given by the matrix R :=(cid:18)1 −1 1 1 (cid:19) =(cid:18)1 1 0 1(cid:19)(cid:18)−2 0 1(cid:19)(cid:18)0 1 1 0(cid:19)(cid:18)1 1 0 1(cid:19) . Then ρ commutes with τ : indeed, τ commutes with all generators of the group G mentioned above (since these operators are partial maps of Γ involving only the 0 16 WOLFGANG BERTRAM AND MICHAEL KINYON τ -fixed elements o+, o−, e, −e and hence commute with τ ), hence τ commutes with R. Since R sends the 4-tuple (o−, e, o+, −e) to (e, o+, −e, o−), it follows that ρjρ = ρMeo+o−eρ = Mo+(−e)eo+ = Mo+o−o−o+ (the last equality follows since Mo+(−a)ao+ = M(−a)o+o+a = Mo−o+o+o− is the map x 7→ (−a) − x + a = −x for all a ∈ V −). Together, this implies ρ ◦ τ ◦ ρ−1 = ρ ◦ τ j ◦ ρ−1 = τ ρ ◦ j ◦ ρ−1 = τ ◦ Mo+o−o−o+ = τ ′. (Note that R is not uniquely determined by the property from the lemma, but the given form corresponds of course to the well-known "real" version of the Cayley transform which enjoys further nice properties.) (cid:3) Theorem 3.6. Assume τ is a unital base-point preserving involution of the Grass- mannian geometry (X ; o+, e, o−), and let A = Co− be the corresponding unital asso- ciative algebra with origin o+ and A− = Co+ the one with origin o−, let τ ′ the dual involution of τ , τ = jτ , Y := X τ and Y ′ := X τ ′ . Let a ∈ X such that o+⊤a, i.e., a ∈ A−. i ) By restriction, τ induces an involutive antiautomorphism of A. This defines a functor from the category of unital involutive associative geometries to the category of involutive associative algebras. ii ) If a ∈ Y ′, then the Lie algebra of the group G(τ ; a) is the space Herm(A, τ ) = Aτ with Lie bracket [x, z]a = 2(hxazi − hzaxi). Identifying A and A− via the canonical isomorphism j, a is identified with the element j(a) ∈ Aherm(A, τ ) and the Lie bracket is expressed in terms of A as [x, z]a = 2(xaz − zax). iii ) If a ∈ Y, then the Lie algebra of the group G(τ ′; a) is the space Aherm(A, τ ) = with Lie bracket [x, z]a = 2(hxazi − hzaxi). With similar identifications Aτ ′ as above, this can be rewritten as [x, z]a = 2(xaz − zax). If, moreover, a is invertible in A, then the group G(τ ′; a) is isomorphic to the unitary group U(Aa, ∗) = {x ∈ A xax∗ = 1} of the involutive algebra (Aa, τ ) with product x ·a y = xay and involution τ . Proof. (i) We show that τ induces an algebra involution: τ (xz) = τ Γ(cid:0)x, o+, e, o−, z(cid:1) = Γ(cid:0)τ z, o+, e, o−, τ x(cid:1) = (τ z)(τ x) Functoriality follows from [BeKi09], Theorem 3.4. (ii) The fixed point space of τ in A is, by definition, Herm(A, ∗), and by Lemma 3.1, τ is an automorphism of Uaτ (a). The formula from the Lie bracket follows from Theorem 3.4. Finally, in order to relate the associative pair to the algebra formulation, recall from [BeKi09] that, for all a ∈ A− and x, z ∈ A+, hxayi+ = x · j(a) · z, where on the right hand side products are taken in the algebra A. Since the K-linear isomorphism j : A+ → A− commutes with τ , the formulas from the claim follow. (iii) The statement on the Lie algebra is proved in the same way as (ii), with signs changed. Now let a be invertible. Assume first a = 1. Note that the condition xx∗ = 1 is equivalent to x = (x∗)−1 = jτ (x), and hence U(A, ∗) is precisely the ASSOCIATIVE GEOMETRIES. II 17 fixed point set of τ in A. Its group structure is induced from A× = Uo+o−. Now, the setting (A×, τ ) = (Uo+o−, jτ ) is conjugate, via the Cayley transform ρ, to the setting (Ue,−e, τ ′) = (Ue,τ ′(e), τ ′), showing that the Cayley transform ρ induces the desired isomorphism. In these arguments, the fixed element e ∈ (Y ∩ Uo+o−) may be replaced by any other element a of this set; this simply amounts to replacing A by its isotope algebra Aa. (cid:3) Theorem 3.7. Consider the following classes of objects: IG: associative geometries with base triple and base triple preserving involutions, IA: involutive unital associative algebras. There are maps F : IG → IA and G : IA → IG such that G ◦ F is the identity. Proof. The map F is defined by part (i) of the preceding theorem. We define the map G: given an involutive associative algebra (A, ∗), let X be the Grassmannian of complemented right A-submodules in A2. We define on A2 the skew-Hermitian ("symplectic") form β(x, y) = x1y2 − x2y1. and consider the involution τ given by the orthocomplementation map with respect to this form. Let o+ = A ⊕ 0 (first factor), o− = 0 ⊕ A (second factor) and e = ∆ (diagonal in A2). Then (o+, e, o−) is a transversal triple, preserved by τ . This defines G. The associative algebra Co− associated to these data is the algebra A we started with (cf. [BeKi09], Theorem 3.5). It remains to prove that restriction of τ to Co− = A gives back the involution ∗ we started with. Let a ∈ A and identify it with the graph {(v, av) v ∈ A}. Then the graph of the adjoint operator a∗ is the orthogonal complement of this graph with respect to β, whence τ (a) = a∗. (cid:3) We have seen above that F is a functor; for G, this is less clear -- cf. remarks in [BeKi09], Section 3.4. We will not pursue here further the discussion of functoriality, nor will we state an analog of the theorem for the non-unital case. Constructions are similar in that case, but are more complicated (since one has to use some algebra- imbedding of an associative pair, see [BeKi09]), and practically less relevant than the unital case. 4. The classical torsors Putting together the results from the preceding two chapters, the "projective" description of the classical groups (Table given in the Introduction) is now straight- forward: we just have to restate Theorems 3.3 and 3.6 for involutions given by orthocomplementation (Theorem 2.2). In the following, we list the results, first for the case of bilinear forms, then for sesquilinear forms. 4.1. Orthogonal and (half-) symplectic groups. We specialize Theorem 3.6 to the case B = K, W = K2n = Kn ⊕ Kn. Let (o+, e, o−) be the canonical base triple (Kn ⊕0, ∆, 0⊕Kn) and β the standard symplectic form on K2n. By Theorem 2.2, we have the three (restricted) involutions τ , τ ′, τ : they are the orthocomplementation maps with respect to the three forms given by the matrices (4.1) Ωn :=(cid:18) 0 −1n 1n 0(cid:19) , Fn :=(cid:18) 0 1n 1n 0(cid:19) , In,n :=(cid:18)1n 0 −1n(cid:19) . 0 18 WOLFGANG BERTRAM AND MICHAEL KINYON Note that o+, o− and ∆ are maximal isotropic for β, hence τ is a unital base point preserving involution. The involutive algebra corresponding to the unital base point preserving involution τ is A = M(n, n; K) with involution X ∗ = X t (usual transpose). The fixed point spaces of the three involutions are the classical Lagrangian varieties corresponding to the three forms, and the tangent space of Y at o+ is Sym(n, K) and the one of Y ′ at o+ is Asym(n, K). Note that Sym(n, K) is imbedded in Y, and Asym(n, K) in Y ′, the subsets of elements of Y (resp. of Y ′) that are transversal to o−. Therefore the elements a parametrizing the torsors G(τ ; a) (resp. G(τ ′; a)) will be chosen in these subsets. From Theorem 3.3 we get: Proposition 4.1. For a = A ∈ Sym(n, K), the group G(τ ; a) with origin o+ is isomorphic to the group On(2A, K), and for a = A ∈ Asym(n, K), the group G(τ ′; a) with origin o+ is isomorphic to the group Spn/2(2A; K). If 2 is invertible in K, then these groups are isomorphic to On(A, K), resp. Spn/2(A; K). Having established the link of the projective torsors G(τ ; a) with the affine re- alization of the classical torsors from the Introduction, it is now relatively easy to classify them (in finite dimension over K = C or R; the case of general base fields is much more difficult, and for general base rings and arbitrary dimension, classification results can only be expected under rather special assumptions). Proposition 4.2. A complete classification of the homotopes of complex or real orthogonal, resp. (half-)symplectic groups is given as follows: (1) (half-)symplectic case: for K = R, C, all homotopes are isomorphic to one of the groups Spm(Ωr; K) for r = 1, . . . , m (with n = 2m or n = 2m + 1), where Ωr denotes the normal form of a skew-symmetric matrix of rank 2r, for K = C, all homotopes are isomorphic to one of the groups On(1r; C) for r = 1, . . . , n, where 1r denotes the n×n-diagonal matrix of rank r having first r diagonal elements equal to one, (2) orthogonal case: for K = R, all homotopes are isomorphic to one of the groups On(Ir,s; R), where Ir,s denotes the n × n-diagonal matrix of rank r + s (r ≤ s, r + s ≤ n) having first r diagonal elements equal to one and s diagonal elements equal to minus one. Proof. One can prove the classification from a "projective" point of view: clearly, if a and b belong to the same Aut(X , τ )-orbit in X , then G(τ ; a) and G(τ ; b) are isomorphic, and it is enough to consider orbits of subspaces a ⊂ W such that a and τ (a) have same dimension n (otherwise Ua,τ (a) is empty). Classifying such orbits is done by elementary linear algebra using Witt's theorem: a and b are conjugate iff the restriction of the given forms to a, resp. b are isomorphic. In particular, the totally isotropic subspaces form one orbit (the Lagrangian Y). The list of orbits then gives rise to the given list of homotopes. Alternatively, an "affine" version of these arguments goes as follows: using the explicit description of the classical groups given in the Introduction, one notices that, e.g., On(A; K) and On(gAgt; K) are isomorphic for all g ∈ GL(n; K); hence it suffices to to consider the classification of GL(n; K)-orbits in Sym(n; K). This leads to the same result (note, however, that different orbits may give rise to isomorphic ASSOCIATIVE GEOMETRIES. II 19 groups: e.g., On(λA; K) and On(A; K) are isomorphic whenever the scalar λ is invertible, be it a square or not in K). Similarly for the symplectic case. (cid:3) 4.2. Unitary groups. The following classification of real classical torsors associ- ated to involutive algebras of Hermitian type is established in the same way as above: Proposition 4.3. Homotopes of complex and quaternionic unitary groups are clas- sified as follows (see Introduction for the notation eH): t τ (X) := X A = M(n, n; H), a) Aτ = Herm(n, H) b) Aτ ′ homotopes of O∗(2n) = Aherm(n, H) Un(Ir,s, H) (r ≤ s, r + s ≤ n) homotopes of Sp(p, q) Un(i1r;eH) (r ≤ n) A = M(n, n; C), τ (X) := X t a) Aτ = Herm(n, C) Un(iIr,s; C) (r ≤ s, r + s ≤ n) homotopes of U(p, q) b) Aτ ′ homotopes of U(p, q) = iHerm(n, C) Un(Ir,s; C) (r ≤ s, r + s ≤ n) Over more general base fields or rings the classification of non-degenerate torsors is essentially equivalent to the classification of involutions of associative algebras -- see [KMRS98] for this vast topic. 4.3. Hilbert Grassmannian. A fairly straightforward infinite dimensional gener- alization of the preceding situation is the following: W = H ⊕ H, where H is a Hilbert space W over B = C or R, and β corresponding to the matrix B = ΩH =(cid:18) 0 −1H 1H 0 (cid:19) or B =(cid:18) 0 1H 1H 0 (cid:19) . In this case we may work with the Grassmannian of all closed subspaces of W , and it easily seen that all arguments from the proof of Theorem 2.2 go through, showing that the orthocomplementation map of β defines an involution of this geometry. We get infinite dimensional analogs of the classical groups, imbedded, together with their homotopes, in Hilbert-Lagrangian manifolds. Variants of these constructions can be applied to restricted Grassmannians and restricted unitary groups in the sense of [PS86]. 5. Semitorsors In this chapter we extend our theory from restricted involutions to "globally defined" involutions. Roughly speaking, the restricted product map Γ and the cor- responding restricted involutions deal with connected geometries (the "restricted" theory developed so far is, in spite of its algebraic flavor, analoguous to the cor- respondence between Lie algebras and connected Lie groups), whereas the global product map Γ and its global involutions rather correspond to replacing connected Lie groups by algebraic groups. 20 WOLFGANG BERTRAM AND MICHAEL KINYON 5.1. Semigroup completion of general linear groups. Let W be a right B- module and X its Grassmannian. In [BeKi09] we have shown that the torsors Uab ⊂ X admit a "semitorsor completion": the ternary law (xyz) from Uab extends to the whole of X , given by the formula (5.1) Γ(x, a, y, b, z) :=(cid:26)ω ∈ W (cid:12)(cid:12)(cid:12) ∃ξ ∈ x, ∃α ∈ a, ∃η ∈ y, ∃β ∈ b, ∃ζ ∈ z : ω = ζ + α = ζ + η + ξ = ξ + β (cid:27) . This formula defines a quintary "product map" Γ : X 5 → X having the following re- markable properties: for any fixed pair (a, b), the partial map (xyz) := Γ(x, a, y, b, z) satisfies the para-associative law (5.2) (xy(zuv)) = (x(uzy)v) = ((xyz)uv), and it is invariant under the Klein 4-group acting on (x, a, b, z): (5.3) Γ(x, a, y, b, z) = Γ(a, x, y, z, b) = Γ(z, b, y, a, x). We say that, for a, b fixed, X with (xyz) = Γ(x, a, y, b, z) is a semitorsor, denoted by Xab (for fixed y, it is in particular a semigroup), and Xba is its opposite semitorsor. For simplicity, we are not going to consider here the globally defined dilation maps Πr from [BeKi09]; in other words, for the moment we look at X as an associative geometry defined over Z (in fact, one has to be very careful with the globally defined maps Πr as soon as r or 1 − r is not invertible; in order to keep this work in reasonable bounds we postpone a more detailed discussion of these problems). Definition. An involution of the Grassmannian geometry X = Gras(W ) is a bijec- tion τ : X → X of order 2 such that, for all x, a, y, b, z ∈ X , without any restriction by transversality conditions, τ (Γ(x, a, y, b, z) = Γ(τ (z), τ (a), τ (y), τ (b), τ (x)) . The following lemma is proved exactly as Lemma 3.1: Lemma 5.1. Let τ : X → X be an involution of the Grassmannian geometry X = GrasB(W ), let Y = X τ and a ∈ X . Then τ induces a semitorsor-automorphism of Xa,τ (a). In particular, the fixed point set Y is a subsemitorsor of Xa,τ (a). If a ∈ Y, then the semitorsor Xa,τ (a) ∩ Y is abelian. Since the globally defined product map Γ encodes the lattice structure of Gras(W ), an involution τ induces an involution of the underlying lattice ([BeKi09], Theorem 2.4 and Section 3.1). Hence the condition that τ is a lattice involution is necessary, and thus orthocomplementation maps are the natural candidates. Our tool for proving that they indeed define involutions is the notion of generalized projection, which might be of independent interest for the theory of linear relations. 5.2. Generalized projections. Linear operators f ∈ EndB(W ) are generalized by linear relations in W , i.e., submodules F ⊂ W ⊕W . Following standard terminology (see, e.g., [Ner96], [Cr98]), domain, image, kernel and indefiniteness of F are the subspaces defined by domF := pr1F, imF := pr2F, ker F := F ∩ (W × 0), indefF := F ∩ (0 × W ) ASSOCIATIVE GEOMETRIES. II 21 with pri : F → W the two projections. For any a, b ∈ X , define the linear relation P a x ⊂ W ⊕ W , called a generalized projection, by (5.4) Note that P a x :=(cid:8)(ζ, ω) ω ∈ x, ω − ζ ∈ a(cid:9). imP a x = x, ker P a x is the graph of the projection denoted previously by P a and that, if a⊤x, then P a x , so there should be no confusion with preceding notation. We denote the space of generalized projections by x = x ∨ a, x = a ∧ x, indefP a domP a x = a, The map P := {P a x x, a ∈ X } ⊂ Gras(W ⊕ W ). X × X → P, (a, x) 7→ P a x is a bijection with inverse P 7→ (ker P, imP ). Transversal pairs (x, a) correspond to "true" operators (single valued and everywhere defined). Lemma 5.2. The linear relation P a x is idempotent: P a x ◦ P a x = P a x . Proof. By definition of composition, P a x ◦ P a x = {(u, w)∃v ∈ W : v ∈ x, u − v ∈ a, w ∈ x, v − w ∈ a}. x ⊂ P a Since w ∈ x and w − u = (w − v) + (v − u) ∈ a, we have P a other inclusion, let (u′, w′) ∈ P a then v, w ∈ x and u − v = u′ − w′ ∈ a, v − w = 0 ∈ a, whence (u′, w′) ∈ P a x . For the x , so w′ ∈ x, w′ − u′ ∈ a. Let u := u′, w := v := w′; x . (cid:3) x ◦ P a x ◦ P a Lemma 5.3. The set P of generalized projections is stable under "conjugation" by linear relations in the following sense: for all linear relations F ⊂ W ⊕ W and all c, z ∈ X , we have Proof. By definition of composition and inverse, F ◦ P c z ◦ F −1 = P F (c) F (z) . F ◦ P c z ◦ F −1 = {(α, δ) ∃β, γ ∈ W : (α, β) ∈ F −1, (β, γ) ∈ P c z , (γ, δ) ∈ F } = {(α, δ) ∃β ∈ W, γ ∈ z : (β, α) ∈ F, (γ, δ) ∈ F, γ − β ∈ c} These conditions imply that δ ∈ F z and (β, α) − (γ, δ) ∈ F ; since (β − γ) ∈ c, this implies also (α − δ) ∈ F c. It follows that (α, δ) ∈ P F (c) F (z) . Conversely, let (α, δ) ∈ P F (c) F (z) , i.e., δ ∈ F (z), α − δ ∈ F (c), so there exists γ ∈ z with (γ, δ) ∈ F and η ∈ c with (η, α − δ) ∈ F . Let β := γ − η, so γ − β ∈ c and (β, α) = (γ, δ) − (η, δ − α) ∈ F, whence (α, δ) ∈ F ◦ P c z ◦ F −1. (cid:3) For the next statements, recall ([Ar61], [Cr98]) the following general definitions concerning linear relations. For a linear relation F ⊂ W ⊕ W and z ∈ X , the image of z under F is F z := F (z) := {δ ∈ W ∃γ ∈ z : (γ, δ) ∈ F } = pr2(pr1)−1(z), 22 WOLFGANG BERTRAM AND MICHAEL KINYON and the difference of linear relations F, G ⊂ Gras(W ⊕ W ), is F − G := {(ξ, ω)∃α, β ∈ W : (ξ, α) ∈ F, (ξ, β) ∈ G, ω = α − β} Remark: This difference can also be written in our language in terms of the asso- ciative geometry (Gras(W ⊕ W ), Γ), with its usual base points o+, o−, as F − G := Γ(F, o−, G, o−, o+), the difference of F and G in the linear space (Co−, o+). Lemma 5.4. For all a, x ∈ X , 1 − P a x = P x a . Proof. ω = u − ω′ with ω′ ∈ x, ω′ − u ∈ a is equivalent to ω ∈ a with ω − u ∈ x. (cid:3) Theorem 5.5. Let Γ be the multiplication map of the Grassmann geometry X . (1) For all (x, a, y, b, z) ∈ X 5, Γ(x, a, y, b, z) = (1 − P x a P b y )(z) = (P a x − P z b )(y). In other words, the left multiplication operator Lxayb in the geometry (X , Γ) is induced by the linear relation 1 − P x y , and the middle multiplication operator Mxabz is induced by the linear relation P a b . Thus we can (and will) define, extending the operator notation from Chapter 1, the linear relations x − P z a P b Lxayb := 1 − P x a P b y , Mxabz := P a x − P z b . (2) For all (x, a, z) ∈ X 3, P a x (z) = Lxaax(z) = Γ(x, a, a, x, z) = x ∧ (a ∨ z). (3) For all a, b, x, y ∈ X , using Notation from part (1), L−1 xayb(z) = Lyaxb(z), M −1 xabz(y) = Mzabx(y). In particular (P a x )−1(z) = L−1 xaax(x) = Laaxx(z) = Γ(a, a, x, x, z) = a ∨ (x ∧ z). Proof. (1) Note that, under certain transversality conditions ensuring that the linear relations in question are indeed graphs of linear operators, the claim has already been proved in [BeKi09]. Let us prove it now in the general situation. a ◦ P b P x a P b 1 − P x y = {(ζ, ω)∃η ∈ y : ζ − η ∈ b, ω − η ∈ x, ω ∈ a}, y = {(ζ, ω′)∃ω ∈ W : (ζ, ω) ∈ P x y , ω′ = ζ − ω} a P b = {(ζ, ω′)∃ω ∈ W, ∃η ∈ y : ζ − η ∈ b, ω − η ∈ x, ω ∈ a, ω′ = ζ − ω} whence (1 − P x a P b y )(z) = {ω′ ∈ W ∃ζ ∈ z, ∃α ∈ a, ∃η ∈ y : ζ − η ∈ b, α − η ∈ x, ω′ = ζ − α} = {ω′ ∈ W ∃α ∈ a, ∃η ∈ y : ω′ + α − η ∈ b, α − η ∈ x, ω′ + α ∈ z} According to the "(a, y)-description" from [BeKi09], this set is indeed equal to Γ(x, a, y, b, z). Similarly, P a x − P z b = {(η, ω)∃u, v : (η, u) ∈ P a x , (η, v) ∈ P z b , ω = u − v} = {(η, ω)∃u ∈ x, ∃v ∈ b : u − η ∈ a, v − η ∈ z, ω = u − v} ASSOCIATIVE GEOMETRIES. II 23 so that (P a x − P z b )(y) = {ω∃u ∈ x, ∃v ∈ b, ∃η ∈ y : u − η ∈ a, v − η ∈ z, ω = u − v} = {ω∃β ∈ b, ∃η ∈ y : β + ω − η ∈ a, β − η ∈ z, β + ω ∈ x} Again, by the (y, b)-description, this equals Γ(x, a, y, b, z). (2) Using Lemmas 5.2 and 5.4, Γ(x, a, a, x, z) = (1 − P x a )(z) = x (z), proving the first equality. The second equality is proved in [BeKi09], Theo- a )(z) = (1 − P x a P x P a rem 2.4 (vi). (3) This is a restatement of Theorem 2.5 from [BeKi09]. (cid:3) 5.3. Orthocomplementation maps and adjoints. Theorem 5.6. Assume β is a non-degenerate Hermitian or skew-Hermitian form on the right B-module W , and let X = Gras(W ). (1) For all x, a, y, b, z ∈ X , we have the inclusion Γ(x⊥, b⊥, y⊥, a⊥, z⊥) ⊂(cid:0)Γ(x, a, y, b, z)(cid:1)⊥ . (2) Assume that B is a skew-field and W = Bn. Then equality holds in (1), and the orthocomplementation map is an involution of X . Proof. We define the adjoint relation of a linear relation F ⊂ W ⊕ W by F ∗ := {(v′, w′)∀(v, w) ∈ F : β(v′, w) = β(w′, v)} ⊂ W ⊕ W. This is the orthocomplement of F with respect to the "symplectic form" Ω on V ⊕V associated to β, Ω((u, v), (u′, v′)) = β(u, v′) − β(v, u′) (see [Ar61], [Cr98], Ch. III). Note that ∗ and inversion commute. Lemma 5.7. For all F ∈ Gras(W ⊕ W ) and all z ∈ Gras(W ), we have (F z)⊥ ⊃ (F ∗)−1z⊥. Proof. Assume v ∈ (F ∗)−1z⊤. This means there is u ∈ W with (v, u) ∈ F ∗ and β(u, z) = 0. Hence, for all (ζ, ζ ′) ∈ F with ζ ∈ z, we have 0 = β(u, ζ) = β(v, ζ ′). Thus, whenever ζ ′ ∈ F (z), we have β(v, ζ ′) = 0, that is, v ∈ (F z)⊥. (cid:3) Lemma 5.8. For any non-degenerate form β, we have (P a x )∗ ⊃ P x⊥ a⊥ and (P x a P b y )∗ ⊃ (P b y )∗(P x a )∗ ⊃ P y⊥ b⊥ P a⊥ x⊥ with equality in all cases under the assumptions of part (2) of the theorem. Proof. By definition of the adjoint, (P a x )∗ = {(v′, w′)∀(v, w) ∈ P a x : β(v′, w) = β(w′, v)} = {(v′, w′)w ∈ x, v − w ∈ a ⇒ β(v′, w) = β(w′, v)} Now assume (v′, w′) ∈ P x⊥ with v − w ∈ a: β(v′, w) = β(v′ − w′, w) + β(w′, w) = β(w′, w) = β(w′, w − v) + β(w′, v) = β(w′, v), a⊥ , that is, w′ ⊥ a, w′ − v′ ⊥ x. Then, for all w ∈ x and v 24 WOLFGANG BERTRAM AND MICHAEL KINYON whence (v′, w′) ∈ (P a x )∗, proving the first inclusion. The inclusion (5.5) (G ◦ F )∗ ⊃ F ∗ ◦ G∗ holds for all linear relations F, G, see [Ar61], Lemma 3.5, where it is also proved that equality always holds in the case of finite dimension over a field. In order to finish the proof, it only remains to show that (P a a⊥ under the assumptions of part (2) of the theorem. In view of the inclusion just proved, it es enough to prove that both subspaces in question have the same dimension over B. First of all, for every linear relation F , since pr2F induces an exact sequence 0 → ker F → F → imF → 0, x )∗ = P x⊥ hence dim F = dim(ker F ) + dim(imF ) dim P a x = dim(a) + dim(x), dim P x⊥ x )∗ is the orthogonal complement of P a a⊥ = dim(a⊥) + dim(x⊥). x with respect to a non-degenerate Since (P a form on W ⊕ W , dim(P a x )∗ = dim(W ⊕ W ) − dim P a x = dim W − dim x + dim W − dim a = dim P x⊥ a⊥ , proving the claim. (cid:3) Lemma 5.9. For all linear linear relations F ⊂ W ⊕ W : (1 + F )∗ = 1 + F ∗, (1 − F )∗ = 1 − F ∗. Proof. One checks easily that the following two linear isomorphisms of W ⊕ W A(v, w) = (v, v + w), D(v, w) = (v, v − w) preserve the form Ω, and hence they are compatible with orthocomplements with respect to Ω. The claim follows by observing that 1 + F = A.F and 1 − F = D.F (where the dot denotes the canonical push-forward action of GL(W ⊕ W ) on linear subspaces). (cid:3) From the preceding two lemmas it follows that (5.6) (Lxayb)∗ = (1 − P x ⊃ 1 − (P b ⊃ 1 − P a⊥ a P b y )∗(P x x⊥ P y⊥ a )∗ y )∗ = 1 − (P x a P b y )∗ b⊥ = Ly⊥b⊥x⊥a⊥. with equality under the assumptions of part (2). Now we prove part (1): Γ(a, x, b, y, z)⊤ = (Lxaybz)⊥ = ((1 − P x ⊃ ((1 − P x ⊃ (Ly⊥b⊥x⊥a⊥)−1z⊥ = Lx⊥b⊥y⊥a⊥z⊥ = Γ(x⊥, b⊥, y⊥, a⊥, z⊥) y )z)⊥ y )∗)−1z⊥ a P b a P b Next assume that B is a skew-field and W = Bn. Then the second inclusion becomes an equality, but we do not know whether the inclusion from Lemma 5.7 always becomes an equality. Therefore we will invoke Lemma 5.3: Choose an auxiliary ASSOCIATIVE GEOMETRIES. II 25 element c ∈ X . Then, using the fact that equality holds in (5.5), along with Lemma 5.3 and (Lxayb)∗ = Ly⊥b⊥x⊥a⊥, we get, on the one hand, z L−1 xayb)∗ = (P (Γ(x,a,y,b,c)) (Γ(x,a,y,b,z)) )∗ = P (Γ(x,a,y,b,z))⊥ (Γ(x,a,c,b,z))⊥ z )∗(Lxayb)∗ = (LxaybP c xayb)∗(P c (L−1 and on the other hand, (L−1 xayb)∗(P c xayb)−1(P c z )∗(Lxayb)∗ z )∗(Lxayb)∗ = (L∗ y⊥b⊥x⊥a⊥P z⊥ = L−1 c⊥ Ly⊥b⊥x⊥a⊥ = Lx⊥b⊥y⊥a⊥P z⊥ c⊥ L−1 = P Γ(x⊥,b⊥,y⊥,a⊥,z⊥) Γ(x⊥,b⊥,y⊥,a⊥,c⊥) . x⊥b⊥y⊥a⊥ Comparing images and kernels of these projections yields the desired equality. (cid:3) With Lemma 5.1, the theorem implies Corollary 5.10. All classical groups over fields or skew-fields, and all of their homotopes G(τ ; a), admit a canonical semigroup completion Xa,τ a ∩ Y (which is a compactification if K = R or C). Remark. The classification of classical semitorsors Xa,τ a is only slightly more com- plicated than the one of the torsors G(τa) from Chapter 4: it suffices to classify all orbits of Aut(X ) in X × X , resp. all Aut(Y)-orbits in X . However, the internal structure of the semitorsors may be very complicated! In other words, the classifica- tion of semigroups is much more difficult than the one of semitorsors (a semitorsor contains many semigroups). Finally, we conjecture that part (2) of Theorem 5.6 still holds in the context of Hilbert Lagrangians (see Subsection 4.3), providing semitorsor-completions of infinite-dimensional classical groups and their homotopes. This conjecture is sup- ported by the fact that the orthocomplementation map of a Hilbert Grassmannian is a lattice involution. However, our proof uses finite-dimensionality at several places, and thus does not generalize directly to this setting. 6. Towards an axiomatic theory In a way similar to the intrinsic-axiomatic description of associative geometries from Chapter 3 of Part I, we would like to describe axiomatically the Lagrangian geometries Y with their torsor and semitorsor structures -- so far they are only de- fined by construction and not by intrinsic properties. What are these properties? Certainly, on the one hand, the various group and torsor structures seem to be the most salient feature. But, on the other hand, there is an underlying "projective" structure playing an important role -- indeed, Lagrangian geometries are special instances of generalized projective geometries as defined in [Be02]. This is most ob- vious on the "infinitesimal" level of the corresponding algebraic structures: besides the Lie algebra structures [x, y]a which correspond to the groups and torsors, there 26 WOLFGANG BERTRAM AND MICHAEL KINYON are also Jordan algebra structures x•a y = (xay +yax)/2, and Jordan structures cor- respond precisely to generalized projective geometries. There are purely algebraic concepts combining these two structures ("Jordan-Lie" and "Lie-Jordan algebras"; cf. [E84], [Be08c]), and the Lagrangian geometries considered here should be their geometric counterparts. In particular, the infinite dimensional Hilbert Lagrangian geometry then is the geometric analog of the Jordan-Lie algebra of observables in Quantum Mechanics -- see [Be08c] for a discussion of some motivation coming from physics. Appendix A: Associative pairs and their involutions Recall (e.g., from [BeKi09], Appendix B, or [Lo75]) that an associative pair (over K) is a pair (A+, A−) of K-modules together with two trilinear maps h·, ·, ·i± : A± × A∓ × A∓ → A± such that hxyhzuvi±i± = hhxyzi±uvi± = hxhuzyi∓vi±. Definition. A type preserving involution of (A+, A−) is a pair (τ + : A+ → A+, τ − : A− → A−) of K-linear mappings such that τ ± are of order 2 and τ ±huvwi± = hτ ±w, τ ∓v, τ ±ui±. A type exchanging involution of (A+, A−) is a pair (τ + : A+ → A−, τ − : A− → A+) of K-linear mappings such that τ + is the inverse of τ − and τ ±huvwi± = hτ ∓w, τ ±v, τ ∓ui±. In other words, a type preserving involution is an isomorphism onto the opposite pair of (A+, A−), and a type exchanging involution is an isomorphism onto the dual of the opposite pair, where the opposite pair is obtained by reversing orders in products, and the dual pair is obtained by exchanging the roles of A+ and A−. Clearly, for any involution τ = (τ +, τ −), the pair τ ′ := (−τ +, −τ −) is again an involution (type preserving, resp. exchanging iff so is τ ); we call it the dual involution. For a type preserving involution, the pairs of 1-eigenspaces or of −1- eigenspaces in general do not form associative pairs (but they are Jordan pairs, see [Lo75]). For type exchanging involutions, there is an equivalent description in terms of triple systems: recall that an associative triple system of the second kind is a K-module A together with a trilinear map A3 → A, (x, y, z) 7→ hxyzi satisfying the preceding identity obtained by omitting superscripts (see [Lo72]), and an associative triple system of the first kind, or ternary ring, is a K-module A together with a trilinear map A3 → A, (x, y, z) 7→ hxyzi satisfying the identity hxyhzuvii = hhxyziuvi = hxhyzuivi (see [Li71]). It is easily checked that, if (τ +, τ −) is a type exchanging involution, the space A := A+ with hx, y, zi := hx, τ +y, zi+ becomes an associative triple system of the second kind. Conversely, from an asso- ciative triple system of the second kind we may reconstruct an associative pair with ASSOCIATIVE GEOMETRIES. II 27 type exchanging involution: A+ := A =: A−, hx, τ +y, zi± := hx, y, zi, τ ± given by the identity map of A± → A∓. In the same way, automorphisms of order two from (A+, A−) onto the opposite pair (A−, A+) correspond to associative triple systems of the first kind. Examples. 1. Every associative algebra A with hxyzi = xyz is an associative triple system of the first kind. It is equivalent to the associative pair (A, A) with the exchange automorphism (which is not an involution, in our terminology). 2. The space of rectangular matrices M(p, q; K) with hXY Zi = XY tZ forms an associative triple system of the second kind. It is equivalent to the as- sociative pair (A+, A−) = (M(p, q; K), M(q, p; K)) with type exchanging involution X 7→ X t. 3. For any involutive algebra (A, ∗), the map (x, y) 7→ (x∗, y∗) is a type preserving involution of the associative pair (A, A). Remark. We do not have an example of an associative pair with a type preserving involution which is not obtained via Example 3 above. In finite dimension over a field the existence of such examples seems rather unlikely, but there might exist infinite dimensional examples which are "very close", but not isomorphic, to pairs of the type (A, A), and admit a type-preserving involution. References [Ar61] Arens, R., "Operational calculus of linear relations", Pac. J. Math. 11 (1961), 9 -- 23 [Be00] Bertram, W., The geometry of Jordan- and Lie structures, Springer Lecture Notes in Mathematics 1754, Berlin 2000 [Be02] Bertram, W., "Generalized projective geometries: general theory and equivalence with Jordan structures", Adv. Geom. 2 (2002), 329 -- 369 (electronic version: preprint 90 at http://homepage.uibk.ac.at/ c70202/jordan/index.html) [Be04] Bertram, W., "From linear algebra via affine algebra to projective algebra", Linear Algebra and its Applications 378 [Be08] Bertram, W., Differential Geometry, Lie Groups and Symmetric Spaces over General Base Fields and Rings. Memoirs of the AMS 192 , no. 900, 2008. arXiv: math.DG/ 0502168 [Be08b] Bertram, W., "Homotopes and conformal deformations of symmetric spaces." J. Lie The- ory 18 (2008), 301 -- 333. arXiv: math.RA/0606449 [Be08c] Bertram, W., "On the Hermitian projective line as a home for the geometry of quantum theory." In Proceedings XXVII Workshop on Geometrical Methods in Physics, Bia lowieza 2008 ; arXiv: math-ph/0809.0561. [BeBi] Bertram, W. and P. Bieliavsky, "Structure variety and homotopes of classical symmetric spaces" (work in progress) [BeKi09] Bertram, W. and M. Kinyon, "Associative geometries. I: torsors, linear relations and Grassmannians", arXiv : math.RA/0903.5441 [BeNe05] Bertram, W. and K.-H. Neeb, "Projective completions of Jordan pairs. II: Manifold structures and symmetric spaces" Geom. Dedicata 112 (2005), 73 -- 113; math.GR/0401236 [BlM04] Blecher, D. P. and C. Le Merdy, Operator algebras and their modules -- an operator space approach, Clarendon Press, Oxford 2004 [CDW87] Coste, A, P. Dazord and A. Weinstein, Groupoıdes Symplectiques, lecture notes, Publi- cation du D´epartement de Math´ematiques de Lyon 1987 (available eletronically at: http:// www.math.berkeley.edu/ alanw/cdw.pdf ) [Cr98] Cross, R., Multivalued Linear Operators, Marcel Dekker, 1998 28 WOLFGANG BERTRAM AND MICHAEL KINYON [E84] Emch, G., Mathematical and Conceptual Foundations of 20th-century Physics, North- Holland Mathematics Studies 100, Amsterdam 1984 [Ho85] Hormander, L., The Analysis of Partial Differential Operators. Vol. III: Pseudo- Differential Operators, Grundlehren Springer, New York 1985 [KMRS98] Knus, M.-A., A. Merkurjew, M. Rost and J.-P. Tignol, The Book of Involutions, AMS Coll. Publ. 44, AMS, Providence 1998 [Li71] Lister, W. G., Ternary rings, Trans. Amer. Math. Soc. 154 (1971) 37 -- 55. [Lo72] Loos, Ottmar, Assoziative Tripelsysteme. Manuscripta Math. 7 (1972), 103 -- 112. [Lo75] Loos, O., Jordan Pairs, Springer LNM 460, New York 1975 [Ner96] Neretin, Y., Categories of Symmetries and Infinite-Dimensional Groups, Oxford LMS 16, Oxford 1996 [PS86] Pressley, A. and G. Segal, Loop Groups, Oxford University Press, Oxford 1986 Institut ´Elie Cartan Nancy, Nancy-Universit´e, CNRS, INRIA, Boulevard des Aiguillettes, B.P. 239, F-54506 Vandoeuvre-l`es-Nancy, France E-mail address: [email protected] Department of Mathematics, University of Denver, 2360 S Gaylord St, Denver, Colorado 80208 USA E-mail address: [email protected]
1607.04425	2	1607	2017-07-12T16:17:37	Algebras whose right nucleus is a central simple algebra	[ "math.RA" ]	We generalize Amitsur's construction of central simple algebras over a field $F$ which are split by field extensions possessing a derivation with field of constants $F$ to nonassociative algebras: for every central division algebra $D$ over a field $F$ of characteristic zero there exists an infinite-dimensional unital nonassociative algebra whose right nucleus is $D$ and whose left and middle nucleus are a field extension $K$ of $F$ splitting $D$, where $F$ is algebraically closed in $K$. We then give a short direct proof that every $p$-algebra of degree $m$, which has a purely inseparable splitting field $K$ of degree $m$ and exponent one, is a differential extension of $K$ and cyclic. We obtain finite-dimensional division algebras over a field $F$ of characteristic $p>0$ whose right nucleus is a division $p$-algebra.	math.RA	math	ALGEBRAS WHOSE RIGHT NUCLEUS IS A CENTRAL SIMPLE ALGEBRA S. PUMPL UN Abstract. We generalize Amitsur's construction of central simple algebras over a field F which are split by field extensions possessing a derivation with field of constants F to nonassociative algebras: for every central division algebra D over a field F of character- istic zero there exists an infinite-dimensional unital nonassociative algebra whose right nucleus is D and whose left and middle nucleus are a field extension K of F splitting D, where F is algebraically closed in K. We then give a short direct proof that every p-algebra of degree m, which has a purely inseparable splitting field K of degree m and exponent one, is a differential extension of K and cyclic. We obtain finite-dimensional division algebras over a field F of characteristic p > 0 whose right nucleus is a division p-algebra. Introduction In 1954, Amitsur [2] observed that all associative central division algebras over a field F of characteristic zero can be constructed using differential polynomials. His construction method can be considered as an analogue to the the well known crossed product construction, except that he uses splitting fields K of the algebras, where the base field F is algebraically closed in K, instead of their algebraic splitting fields. Some of his results also work for p-algebras, i.e. over base fields of characteristic p > 0. In this paper, we consider algebras which are also obtained from differential polynomials, but which are nonassociative. These algebras are constructed using the differential polynomial ring K[t; δ], where K is a field and δ a derivation on K and were defined by Petit [14]: given a differential polynomial f ∈ K[t; δ] of degree m, the set of all differential polynomials of degree less than m, together with the addition given by the usual addition of polynomials, can be equipped with a nonassociative ring structure using right division by f to define the multiplication as g ◦ h = gh modrf . The resulting nonassociative unital ring Sf , also denoted by K[t; δ]/K[t; δ]f , is an algebra over the field of constants F = Const(δ) of δ. If f generates a two-sided ideal in K[t; δ], then Sf is the (associative) quotient algebra obtained by factoring out the two-sided principal ideal generated by f . If f is not two-sided and δ not trivial, then the nuclei of Sf are larger than the center F = Const(δ). In that case the left and middle nucleus are always given by K, whereas the right nucleus reflects both the choice of f and the structure of the ring K[t; δ]. 2010 Mathematics Subject Classification. Primary: 17A35; Secondary: 17A60, 16S36. Key words and phrases. Differential polynomial ring, skew polynomial, differential polynomial, differen- tial operator, differential algebra, nonassociative algebra, right nucleus. 1 2 S. PUMPL UN We proceed as follows: The basic terminology and notation we use can be found in [2] and Section 1. Section 2 rephrases some of Amitsur's results for those algebras Sf which have a central simple algebra as their right nucleus. For this we employ Amitsur's A-polynomials. In Sections 3 and 4 we show how to construct algebras Sf with a given central simple algebra as right nucleus, first for base fields of characteristic zero, then for base fields of characteristic p > 0: for every central simple algebra B of degree m over a field F of characteristic zero which is split by a field extension K/F in which F is algebraically closed, there exists an infinite-dimensional unital algebra Sf = K[t; δ]/K[t; δ]f over F with right nucleus B (and left and middle nucleus K), see Theorem 8. In particular, for every central division algebra D over F there exists an infinite-dimensional unital algebra Sf over F with right nucleus D (Corollary 9). We present a short proof that every p-algebra B of degree m over a field F of characteristic p which is split by a purely inseparable field extension K/F of exponent one and degree m is isomorphic to a differential extension (K, δ, d0) of K (Theorem 13), only invoking a result on the structure of Sf and Amitsur's [2, Lemma 20']. Thus it is cyclic by [9, Main Theorem]. For every division p-algebra D of degree m over a field F of characteristic p which is split by a purely inseparable field extension K/F of exponent one such that m < [K : F ], there is a unital division algebra Sf = K[t; δ]/K[t; δ]f over F of dimension mpe with right nucleus D and left and middle nucleus K. The smallest possible dimension l of such a division algebra containing D as right nucleus is bounded via m2 < l ≤ mpm−1 and connected to the number of cyclic algebras that are needed when expressing D as a product of cyclic algebras of degree p in the Brauer group Br(F ) (Corollary 18). 1. Preliminaries 1.1. Nonassociative algebras. Let F be a field and let A be an F -vector space. A is an algebra over F if there exists an F -bilinear map A × A → A, (x, y) 7→ x · y, denoted simply by juxtaposition xy, the multiplication of A. An algebra A is called unital if there is an element in A, denoted by 1, such that 1x = x1 = x for all x ∈ A. We will only consider unital algebras from now on without explicitly saying so. An algebra A 6= 0 is called a division algebra if for any a ∈ A, a 6= 0, the left multiplication with a, La(x) = ax, and the right multiplication with a, Ra(x) = xa, are bijective. If A has finite dimension over F , A is a division algebra if and only if A has no zero divisors [17, pp. 15, 16]. Associativity in A is measured by the associator [x, y, z] = (xy)z − x(yz). The left nucleus of A is defined as Nucl(A) = {x ∈ A [x, A, A] = 0}, the middle nucleus of A is Nucm(A) = {x ∈ A [A, x, A] = 0} and the right nucleus of A as Nucr(A) = {x ∈ A [A, A, x] = 0}. Nucl(A), Nucm(A), and Nucr(A) are associative subalgebras of A. Their intersection Nuc(A) = {x ∈ A [x, A, A] = [A, x, A] = [A, A, x] = 0} is the nucleus of A. Nuc(A) is an associative subalgebra of A containing F 1 and x(yz) = (xy)z whenever one of the elements x, y, z is in Nuc(A). The center of A is C(A) = {x ∈ Nuc(A) xy = yx for all y ∈ A}. ALGEBRAS WHOSE RIGHT NUCLEUS IS A CENTRAL SIMPLE ALGEBRA 3 1.2. Differential polynomial rings. Let K be a field and δ : K → K a derivation, i.e. an additive map such that δ(ab) = aδ(b) + δ(a)b for all a, b ∈ K. The differential polynomial ring K[t; δ] is the set of polynomials a0 + a1t + · · · + antn with ai ∈ K, where addition is defined term-wise and multiplication by ta = at + δ(a) (a ∈ K). For f = a0 + a1t + · · · + antn with an 6= 0 define deg(f ) = n and deg(0) = −∞. Then deg(f g) = deg(f ) + deg(g). An element f ∈ R is irreducible in R if it is not a unit and if it has no proper factors, i.e if there do not exist g, h ∈ R with deg(g), deg(h) < deg(f ) such that f = gh. R = K[t; δ] is a left and right principal ideal domain and there is a right division algorithm in R: for all g, f ∈ R, g 6= 0, there exist unique r, q ∈ R with deg(r) < deg(f ), such that g = qf + r. There is also a left division algorithm in R [11, p. 3 and Prop. 1.1.14]. (Our terminology is the one used by Petit [14]; Jacobson's is vice versa.) Two non-zero elements f, g ∈ R are called similar (f ∼ g) if and only if there exist h, q, u ∈ R such that 1 = hf + qg and u′f = gu for some u′ ∈ R. Equivalently, f and g are similar if R/Rf and R/Rg are isomorphic as R-modules [11, p. 11]. Obviously, f ∼ g implies that deg(f ) = deg(g). 1.3. The characteristic p > 0 case. Let K be a field of characteristic p and R = K[t; δ], then (t − b)p = tp − Vp(b), Vp(b) = bp + δp−1(b), (t − b)pe = tpe − Vpe (b) for all b ∈ K with Vpe (b) = V e p (b) = Vp(. . . (Vp(b)) . . . ) [11, p. 17ff]. For any p-polynomial f (t) = a0tpe + a1tpe−1 + · · · + aet + d ∈ D[t; δ] we thus have f (t) − f (t − b) = a0Vpe (b) + a1Vpe−1 (b) + · · · + aeb for all b ∈ K and define Vf (b) = a0Vpe (b) + a1Vpe−1 (b) + · · · + aeb. 1.4. Nonassociative algebras obtained from differential polynomial rings. Let K be a field and f ∈ R = K[t; δ] of degree m. Let modrf denote the remainder of right division by f . Define F = Cent(δ) = {a ∈ K δ(a) = 0}. Definition 1. (cf. [14, (7)]) The vector space Rm = {g ∈ K[t; δ] deg(g) < m} together with the multiplication g ◦ h = gh modrf 4 S. PUMPL UN is a unital nonassociative algebra Sf = (Rm, ◦) over F0 = {a ∈ K ah = ha for all h ∈ Sf }. F0 is a subfield of K [14, (7)] and it is easy to check that F0 = Cent(δ). The algebra Sf is also denoted by R/Rf [14, 16] if we want to make clear which ring R is involved in the construction. In the following, we call the algebras Sf Petit algebras and denote their multiplication simply by juxtaposition. Without loss of generality, we may assume that f is monic, since Sf = Sg for all g = af with a ∈ K ×. Using left division by f and the remainder modlf of left division by f instead, we can define the multiplication for another unital nonassociative algebra on Rm over F , called f S or R/f R. We will only consider the Petit algebras Sf , however, since every algebra f S is the opposite algebra of some Petit algebra (cf. [14, (1)]). Right multiplication with 0 6= g ∈ Sf is given by Rg : Sf −→ Sf , h 7→ hg, and is a left K-module endomorphism. Left multiplication Lg : Sf −→ Sf , h 7→ gh is an F -module endomorphism [14], and if we view Sf as a right module over Nucr(Sf ), a right Nucr(Sf )- module endomorphism. Clearly Sf has no zero divisors if and only if Rg and Lg are injective. Theorem 1. (cf. [14, (2), p. 13-03, (5), (6), (7), (9), (14)]) Let f ∈ R = K[t; δ]. (i) If Sf is not associative then Nucl(Sf ) = Nucm(Sf ) = K and Nucr(Sf ) = {g ∈ Rm f g ∈ Rf }. The right nucleus of Sf is Amitsur's invariant ring of f . (ii) The powers of t are associative if and only if tmt = ttm if and only if t ∈ Nucr(Sf ) if and only if f t ∈ Rf. (iii) If f is irreducible then Nucr(Sf ) is an associative division algebra. (iv) Let f ∈ R be irreducible and Sf a finite-dimensional F -vector space or free of finite rank as a right Nucr(Sf )-module. Then Sf is a division algebra. Conversely, if Sf is a division algebra then f is irreducible. (v) Sf is associative if and only if f is a two-sided element (i.e., generates a two-sided ideal Rf ). In that case, Sf is the usual quotient algebra K[t; δ]/(f ). (vi) f is irreducible if and only if Sf is a right division algebra over F (i.e., each non-zero element in Sf has a left inverse: there is z ∈ Sf such that zh = 1), if and only if Sf has no zero divisors. Recall that a polynomial f ∈ R = K[t; δ] is bounded if there exists 0 6= f ∗ ∈ R, such that Rf ∗ = f ∗R is the largest two-sided ideal of R contained in Rf . If f ∈ R is bounded then f is irreducible if and only if Nucr(Sf ) has no zero divisors if and only if Nucr(Sf ) is an associative division algebra (cf. [8, Proposition 4] which sums up classical results from [10]). [5, Theorem 4] yields: Theorem 2. Let f ∈ R be irreducible. Then f is bounded if and only if Sf is free of finite rank as a Nucr(Sf )-module. In this case, Sf is a division algebra. ALGEBRAS WHOSE RIGHT NUCLEUS IS A CENTRAL SIMPLE ALGEBRA 5 Proof. The first part of the statement is [5, Theorem 4]. Since f irreducible, Sf is a right division algebra and Lh is injective for all h ∈ Sf , h 6= 0, as observed in [14, Section 2., (7)]. The second part then follows from the fact that Sf is free of finite rank as a Nucr(Sf )- module, which means the injective Nucr(Sf )-linear map Lh is also surjective. (cid:3) R = K[t; δ] has finite rank over its center if and only if K is of finite rank over Ct = {a ∈ K at = ta} if and only if all polynomials of R are bounded and if for all f of degree non-zero, deg(f ∗)/deg(f ) is bounded in Q (f ∗ being the bound of f ) [6, Theorem IV]. Since here Ct = Const(δ) = F , we conclude: Proposition 3. Assume that one of the two following equivalent conditions hold: (i) R = K[t; δ] has finite rank over its center; (ii) K/F is a finite field extension. Then every f ∈ R is bounded. In particular, if f is irreducible then Sf is a division algebra. Note that if K/F is a finite field extension then the derivation δ is trivial, or K has characteristic p > 0. We will assume throughout the paper that f ∈ K[t; δ] has deg(f ) = m ≥ 2 (if f has degree m = 1 then Sf ∼= K) and that δ 6= 0. Without loss of generality, we could only look at monic f , but will do so only when explicitly mentioned. 2. Nonassociative algebras whose right nucleus is a central simple algebra We use the terminology from [2] with the only exception that that in our definition of K[t; δ], we look at polynomials with the coefficients written on the left, not on the right- hand-side as in [2]. All results, however, work analogously in this case. By [13, Theorem 4.2], given a field extension K/F in characteristic zero, F is the field of constants of a derivation of K if and only if F is algebraically closed in K. In this section, let K be a field of characteristic 0. Let δ be a derivation of K with F = Const(δ) and f ∈ R = K[t; δ]. The finite-dimensional associative F -algebra Nucr(Sf ) is called the invariant ring of f by Amitsur [2, p. 260], in recent literature it is also referred to as the eigenspace of f . Let V be an K-vector space. An additive map T : V −→ V , such that T (αv) = αT (v) + δ(α)v for all v ∈ V and α ∈ K, is called a pseudo-linear transformation on V . Given a basis of V , a pseudo-linear transformation T on V is given by a matrix. Moreover, (V, T ) is isomorphic to K[t; δ]/f (t)K[t; δ] for some f (t) ∈ K[t; δ] which is called the characteristic polynomial of T [2, p. 250]. The characteristic polynomial is uniquely determined up to similarity and any polynomial f (t) is the characteristic polynomial of some pseudo-linear transformation (V, T ) (simply define V = K[t; δ]/K[t; δ]f (t) and T (p(t) + K[t; δ]f (t)) = tp(t) + K[t; δ]f (t)). Let (V, T ) and (V ′, T ′) be two pseudo-linear transformations with characteristic poly- nomials f, g ∈ K[t; δ] where deg(f ) = m and deg(g) = n. Then there is a pseudo-linear transformation T × T ′ on the tensor product V ⊗ V ′ defined via (T × T ′)(u) = X T (vi) ⊗ wi + X vi ⊗ T ′(wi) i i 6 S. PUMPL UN for all u = Pi vi ⊗ wi ∈ V ⊗ V ′. Furthermore, let f, g ∈ K[t; δ] where deg(f ) = m and deg(g) = n, and T and T ′ be the pseudo-linear transformation defined using f and g. Then the resultant f × g of f and g is any characteristic polynomial of T × T ′, so that f × g is a polynomial of degree nm uniquely determined up to similarity [2, p. 255]. A differential polynomial f ∈ K[t; δ] of degree m is called an A-polynomial if there is some ef ∈ K[t; δ] of degree n such that the resultant f × ef is similar to emn, the characteristic polynomial of the pseudo-linear transformation corresponding to the zero mn × mn matrix [2, p. 263]. Amitsur's results tell us when Nucr(Sf ) is a central simple algebra: Theorem 4. [2, Lemma 17, 18, 19, Theorem 17, Corollary, Lemma 22] Let f, g ∈ K[t; δ] with deg(f ) = m ≥ 2 and deg(g) = n ≥ 2. (i) Nucr(Sf ) has dimension m2 if and only if f is an A-polynomial. (ii) If f is an A-polynomial then Nucr(Sf ) is a central simple algebra of degree m which is split by K. (iii) If f and g are A-polynomials then so is h = f × g and Nucr(Sh) = Nucr(Sf ) ⊗F Nucr(Sg). (iv) If f and g are A-polynomials then Nucr(Sf ) ∼= Nucr(Sg) if and only if f ∼ g(t + a) ∼ g(t) × t + a for some a ∈ K. In particular, Sf ∼= Sg implies that f ∼ g(t + a) ∼ g(t) × t + a for some a ∈ K. (v) Suppose f is an A-polynomial. Then if and only if one of the following holds: Nucr(Sf ) ∼= Matm(F ) • f ∼ em × t + c for some c ∈ K; • f decomposes into irreducible factors and at least one factor is linear of the form t + c for some c ∈ K (then f ∼ em × t + c). In particular, then the irreducible factors of f are all similar to t + c. Let L/K be a field extension such that δ extends to L. Then L[t; δ] is an Ore extension of K[t; δ] and the constant field F = Const(δK ) of δ = δK is contained in the constant field C = Const(δ). If L = K · C is the composite field of K and C, we say L is a constant extension of K. It is clear that for f ∈ K[t; δ], Nucr(K[t; δ]/K[t; δ]f ) ⊂ Nucr(L[t; δ]/L[t; δ]f ). Theorem 5. Let f ∈ K[t; δ] be of degree m and L/K a field extension such that δ extends to L and C = Const(δ). Suppose that L is a constant extension of K. (i) If f is an A-polynomial then f ∈ L[t; δ] is an A-polynomial and Nucr(L[t; δ]/L[t; δ]f ) ∼= Nucr(K[t; δ]/K[t; δ]f ) ⊗F C. ALGEBRAS WHOSE RIGHT NUCLEUS IS A CENTRAL SIMPLE ALGEBRA 7 (ii) Suppose B = Nucr(Sf ) is a central simple algebra of degree m over F with f ∈ K[t; δ]. Then C splits B if and only if f has a left or right root in L, i.e. f = (t − a)g(t) ∈ L[t; δ] or f = g(t)(t − a) ∈ L[t; δ]. In particular, then Sf ⊗F C has right nucleus Matm(C). This follows from [2, Theorem 20] and [2, Corollary, p. 270]. Remark 6. Since every automorphism of a nonassociative algebra maps the right nucleus onto itself, for every A-polynomial f which is not two-sided, each H ∈ AutF (Sf ) satisfies HB ∈ AutF (B) when restricted to the central simple algebra B = Nucr(Sf ), thus HB is an inner automorphism of B. By an analogous argument, also HK ∈ AutF (K). 3. Algebras whose right nucleus is split by an extension in which F is algebraically closed Let F be a field of characteristic 0. Theorem 7. [2, Lemma 20] (i) Every central simple algebra B of degree m over F which is split by a field extension K/F in which F is algebraically closed, is isomorphic to Nucr(Sf ) for some f ∈ K[t; δ] of degree m and a suitable δ with F = Const(δ). The differential polynomial f is an A-polynomial. (ii) Every central division algebra D of degree m over F is isomorphic to Nucr(Sf ) for some f ∈ K[t; δ] of degree m and a suitable differential field (K, δ). Note that (ii) follows from (i), since for every central division algebra D over F , the function field K(X) of the Severi-Brauer variety X of D splits D ([2, p. 245] or [3]), and we can always find a derivation δ on K(X) with F = Const(δ), as F is algebraically closed in K(X). As an immediate consequence of Theorem 7 and Remark 6, we now get the following results: Theorem 8. For every central simple algebra B of degree m over F which is split by a field extension K/F in which F is algebraically closed, there is a derivation δ on K with field of constants F and a differential polynomial f ∈ K[t; δ] of degree m, such that Sf = K[t; δ]/K[t; δ]f is an infinite-dimensional algebra over F with right nucleus B and left and middle nucleus K. Every automorphism H ∈ AutF (Sf ) extends an inner automorphism of B. We conclude from [2, p. 246]: Corollary 9. For every central division algebra D of degree m over F , there exists a field extension K/F in which F is algebraically closed, a derivation δ on K with field of constants F , and a differential polynomial f ∈ K[t; δ] of degree m, such that Sf = K[t; δ]/K[t; δ]f is an infinite-dimensional algebra over F with right nucleus D, and left and middle nucleus K. K splits D and every automorphism H ∈ AutF (Sf ) extends an inner automorphism of D. 8 S. PUMPL UN The fact that D is a division algebra does not imply that f is irreducible, so Sf might not be a right division algebra. Corollary 10. If the differential polynomial f in Corollary 9 is irreducible, then Sf is an infinite-dimensional right division algebra over F and therefore does not have zero divisors. If f is an irreducible A-polynomial, it is not bounded by Theorem 2. Example 11. Suppose F = R. The only central division algebra over R is D = (−1, −1)R. The function field K of the projective real conic given by x2 + y2 + z 2 = 0 is a field extension of R in which R is algebraically closed and that splits D. There exists a derivation δ on K with R = Const(δ). Thus there is an A-polynomial f ∈ K[t; δ] of degree 2, such that Sf = K[t; δ]/K[t; δ]f = K ⊕ Kt is an infinite-dimensional unital algebra over R with right nucleus (−1, −1)R, and left and middle nucleus K. For B = Matm(R) and any field extension K ′ of R in which R is algebraically closed, with a derivation δ on K ′ such that R = Const(δ), there is a reducible A-polynomial f ∈ K ′[t; δ] of degree m, such that Sf = K ′[t; δ]/K ′[t; δ]f is an infinite-dimensional unital algebra over R with right nucleus B and left and middle nucleus K ′. 4. Algebras whose right nucleus is a p-algebra Let now K be a field of characteristic p > 0 together with a derivation δ on K. Put R = K[t; δ] and F = Const(δ). There are two cases which can occur: either δ is an algebraic derivation, or δ is transcendental which means [K : F ] = ∞. We assume that δ is an algebraic derivation of degree pe with minimum polynomial g(t) = tpe + c1tpe−1 + · · · + cet ∈ F [t] of degree pe. Then K = F (u1, . . . , ue) = F (u1) ⊗F · · · ⊗F F (ue) with up i = ai ∈ F for all i ∈ {1, . . . , e}, and [K : F ] = pe, that is K is a finite purely inseparable field extension of exponent one and K p ⊂ F ⊂ K. The center C(R) of R is F [z] with z = g(t) − d0, d0 ∈ F , and the two-sided elements in R have the form uh(t) with u ∈ K ×, h(t) ∈ C(R). Recall that a central simple algebra B = Matr(D) over a field F of characteristic p is a p-algebra if it has index pn, equivalently, if its exponent is a power of p [11, p. 154]. Note that for f (t) = g(t) − d ∈ F [t] (so f (t) is two-sided in this case), (K, δ, d) = K[t; δ]/K[t; δ]f (t) is an associative central simple F -algebra called a differential extension of K and treated in [11, p. 23]. K is a maximal subfield of (K, δ, d). Theorem 12. [2, Lemma 20'] Let B be a p-algebra of degree m over F which is split by a purely inseparable extension K of exponent one (i.e., has exponent p), such that m ≤ [K : F ]. Then B ∼= Nucr(Sf ) ALGEBRAS WHOSE RIGHT NUCLEUS IS A CENTRAL SIMPLE ALGEBRA 9 for some f ∈ K[t; δ] of degree m and a suitable δ with F = Const(δ). We start by looking at the case that m = [K : F ] = pe and immediately obtain (i) and (ii) in the following result on p-algebras by employing only Theorem 1 (v) from Petit [14] and Amitsur's Theorem 12 (only the fact that then B is cyclic uses Hood's Main Theorem [9, Main Theorem]): Theorem 13. Let B be a p-algebra of degree m over F which is split by a purely inseparable field extension K of exponent one with m = [K : F ]. (i) There is an algebraic derivation δ on K of degree m with minimum polynomial g(t) such that the center of K[t; δ] is F [z] with z = g(t) − d0, d0 ∈ F , and B = (K, δ, d0) with f (t) = g(t) − d0. B is a cyclic algebra. (ii) F [t]/(f ) is a subfield of B of degree pe over F if and only if f is irreducible in F [t]. (iii) f ∈ K[t; δ] is irreducible if and only if B is a division algebra. (iv) B ∼= Matpe (F ) if and only if there is b ∈ K such that d0 = Vg(b) = Vpe (b) + c1Vpe−1 (b) + · · · + ceb. Proof. (i) If m = [K : F ] then there is a differential polynomial f ∈ K[t; δ] of degree m and a suitable δ such that B ∼= Nucr(Sf ) by Theorem 12. Here B is an associative subalgebra of Sf of dimension m2 and Sf has dimension m2 as well. Therefore Sf = B is associative and f ∈ K[t; δ] must be a two-sided differential polynomial of degree m, i.e. B = K[t; δ]/(f ) is a quotient algebra (Theorem 1 (v)). Without loss of generality we may assume f is monic. Thus f ∈ C(R) and since f has degree m = pe, we obtain that f (t) = g(t) − d0 and so B = (K, δ, d0). K is a purely inseparable field extension of F which is an (even maximal) subfield of B splitting B, therefore B is cyclic [9, Main Theorem]. (ii) Since here f (t) ∈ F [t], we know that F [t]/(f ) is a subfield of B of degree pe over F if and only if f is irreducible in F [t]. (iii) is [8, Proposition 4] and (iv) is a consequence from (i) together with Theorem [11, Theorem 1.3.27]. (cid:3) Remark 14. Let us briefly put the previous result into context: (i) Let A be a central simple p-algebra of degree pn over F . It is a well known classical result that A is cyclic over F if and only if A has a subfield K such that K is a purely inseparable extension of F and K is a splitting field for A (this is [9, Main Theorem], which removed Albert's restriction that K be simple from [1, Theorem (7.27)]). (ii) Mammone characterized the central simple algebras split by a purely inseparable field extension K of exponent one in [12]: in particular, if B is a central simple algebra over F of degree m = pe containing K where [K : F ] = m, then B is a differential crossed product, that means B contains a K-basis of the form {zi1 n 0 ≤ ik ≤ p − 1} satisfying a kind of commutativity law with elements of K which involves a set of n F -derivations of K. The algebra B then yields elements bi = zp i and uij = zizj − zjzi in K. Conversely, given sets B = {bi i = 1, . . . , n} and U = {uij : i, j = 1, . . . , n} satisfying certain relations involving F -derivations of K, then (U, B) arises from such a differential crossed product. 1 · · · zin 10 S. PUMPL UN In case m < [K : F ] = pe we obtain a nonassociative algebra of dimension mpe containing B as right nucleus: Theorem 15. Let B be a p-algebra of degree m over F which is split by a purely inseparable extension K of exponent one such that m < [K : F ]. (i) There is an algebraic derivation δ and a differential polynomial f ∈ K[t; δ] of degree m such that Sf = K[t; δ]/K[t; δ]f is an algebra over F of dimension mpe with right nucleus B, left and middle nucleus K, and nucleus Nuc(Sf ) = B ∩ K an intermediate field of K/F , unequal to K. (ii) f is irreducible if and only if B is a division algebra, if and only if Sf is a division algebra. (iii) Every automorphism H ∈ AutF (Sf ) extends an inner automorphism of B and an automorphism of K. Proof. (i) The existence of a suitable f follows from Theorem 12 and the statements on the left and middle nuclei from Theorem 1. Since f is not two-sided, K is not contained in the right nucleus of Sf , i.e. not contained in B [15, Theorem 9]. Thus Nuc(Sf ) = B ∩ K is properly contained in K, so that it is an intermediate field of the field extension K/F . (ii) By Proposition 3 and Theorem 1, f is irreducible if and only if B is a division algebra, if and only if Sf is a division algebra. (iii) An automorphism of Sf extends both an inner automorphism of B and an automorphism of K by Remark 6. (cid:3) Corollary 16. Let D be a division p-algebra of degree m over F which is split by a purely inseparable extension K of exponent one such that m < [K : F ]. Then there is an irreducible polynomial f ∈ K[t; δ] of degree m such that Sf is a division algebra over F of dimension mpe with right nucleus D, left and middle nucleus K, and nucleus D ∩ K an intermediate field of K/F , unequal to K. The fact that f is irreducible in Corollary 16 follows from Proposition 3. Note that every division p-algebra over F split by K has degree m ≤ [K : F ], so that Theorem 13 (iii) and Corollary 16 cover all possible cases for a division p-algebra. We could ask for the algebra Sf of smallest possible dimension which contains a given central simple algebra B as a right nucleus. This is equivalent to asking for a purely in- separable extension K of exponent one splitting B of smallest possible degree [K : F ] = pe satisfying m < [K : F ], which in turn is connected to the question how many cyclic algebras are needed when saying that B is similar to a product of cyclic algebras of degree p in the Brauer group Br(F ). Theorem 17. Let B be a p-algebra over F of degree m, index d = pn and exponent p, such that m = r2pn < pd−1. Then there is a purely inseparable extension K of exponent one with [K : F ] = pd−1, and a differential polynomial f ∈ K[t; δ] of degree m such that Sf = K[t; δ]/K[t; δ]f ALGEBRAS WHOSE RIGHT NUCLEUS IS A CENTRAL SIMPLE ALGEBRA 11 is an algebra over F of dimension mpd−1 with right nucleus B and the properties listed in Theorem 15. Proof. Let B be a p-algebra of index pn and exponent p. Then there is a purely inseparable field extension K/F of exponent one with K = F (u1, . . . , ud−1), up i = ai ∈ F , and [K : F ] = pd−1, which splits B [7, Theorem 1.1.]. We have m = r2d = r2pn for some r ≥ 1. We need m = r2pn ≤ [K : F ] = pd−1 to be able to apply Theorem 12. By Theorem 12 this implies that B ∼= Nucr(Sf ) for some f ∈ K[t; δ] of degree m and a suitable δ with F = Const(δ). Since each f ∈ K[t; δ] is bounded by Proposition 3, B = Nucr(Sf ) is a division algebra if and only if f is irreducible [8, Proposition 4], if and only if Sf is a division algebra. (cid:3) We obtain that for a division algebra D, the smallest possible dimension l of a division algebra Sf containing D as right nucleus satisfies m2 < l = mpe ≤ mpm−1: Corollary 18. Let D be a division p-algebra of degree m and exponent p over F . Then there is a purely inseparable extension K of exponent one with [K : F ] = pm−1, and an irreducible differential polynomial f ∈ K[t; δ] of degree m such that Sf = K[t; δ]/K[t; δ]f is a division algebra over F of dimension mpm−1 with right nucleus D and the properties listed in Theorem 15. Proof. There is a purely inseparable field extension K = F (u1, . . . , um−1) of exponent one, up i = ai ∈ F , and [K : F ] = pm−1, which splits D [7, Theorem 1.1.]. We need m = pn ≤ [K : F ] = pm−1 to be able to apply Theorem 12. This holds for all prime p and n ≥ 1 as it is equivalent to n ≤ pn − 1, i.e. to n + 1 ≤ pn, which is true for all prime p and n ≥ 1. Therefore there is a purely inseparable field extension K/F of exponent one with m ≤ [K : F ] = pm−1 which splits D. By Theorem 12 this implies that B ∼= Nucr(Sf ) for some f ∈ K[t; δ] of degree m and a suitable δ with F = Const(δ). Since D is a division algebra and f bounded, f is irreducible and Sf is a division algebra. (cid:3) References [1] A. A. Albert, "Structure of algebras." Revised printing, Amer. Math. Soc., Providence, R.I., 1961. [2] A. S. Amitsur, Differential polynomials and division algebras. Annals of Mathematics, Vol. 59 (2) (1954) 245-278. [3] A. S. Amitsur, Generic splitting fields of central simple algebras. Ann. of Math. 62 (2) (1955), 8-43. [4] M. Boulagouaz, A. Leroy, (σ, δ)-codes. Adv. Math. Commun. 7 (4) (2013), 463-474. [5] J. Carcanague, Id´eaux bilat`eres d'un anneau de polynomes non commutatifs sur un corps. J. Algebra (18) 1971, 1-18. [6] J. Carcanague, Quelques r´esultats sur les anneaux de Ore. C. R. Acad. Sci. Paris Sr. A-B 269 (1969), A749-A752. [7] M. Florence, On the symbol length of p-algebras. Compositio Mathematica 149 (8) (2013), 1353-1363. [8] J. G`omez-Torrecillas, Basic module theory over non-commutative rings with computational aspects of operator algebras. With an appendix by V. Levandovskyy. Lecture Notes in Comput. Sci. 8372, Algebraic and algorithmic aspects of differential and integral operators, Springer, Heidelberg (2014) 23-82. [9] J. M. Hood, Central simple p-algebras with purely inseparable subfields. J. Algebra 17 (1971) 299-301. 12 S. PUMPL UN [10] N. Jacobson, The Theory of Rings, AMS, Providence, RI, 1943 [11] N. Jacobson, "Finite-dimensional division algebras over fields." Springer Verlag, Berlin-Heidelberg-New York, 1996. [12] P. Mammone, Remarques sur les produits crois´es diff´erentiels. Acad. Roy. Belg. Bull. Cl. Sci. (5) 68 (1982), no. 10, 651-664. [13] A. Nowicki, Rings and fields of constants for derivations in characteristic zero. J. Pure Appl. Algebra 96 (1) (1994), 47-55. [14] J.-C. Petit, Sur certains quasi-corps g´en´eralisant un type d'anneau-quotient. S´eminaire Dubriel. Alg`ebre et th´eorie des nombres 20 (1966 - 67), 1-18. [15] S. Pumplun, Nonassociative differential extensions of characteristic p. Results in Mathematics, DOI 10.1007/s00025-017-0656-x [16] J.-C. Petit, Sur les quasi-corps distributifs `a base monog`ene. C. R. Acad. Sc. Paris 266 (1968), S´erie A, 402-404. [17] R. D. Schafer, "An Introduction to Nonassociative Algebras." Dover Publ., Inc., New York, 1995. E-mail address: [email protected] School of Mathematical Sciences, University of Nottingham, University Park, Nottingham NG7 2RD, United Kingdom
1805.02001	1	1805	2018-05-05T05:20:27	Isomorphism problem and homological properties of DG free algebras	[ "math.RA" ]	A differential graded (DG for short) free algebra $\mathcal{A}$ is a connected cochain DG algebra such that its underlying graded algebra is $$\mathcal{A}^{\#}=\k\langle x_1,x_2,\cdots, x_n\rangle,\,\, \text{with}\,\, \|x_i\|=1,\,\, \forall i\in \{1,2,\cdots, n\}.$$ We prove that the differential structures on DG free algebras are in one to one correspondence with the set of crisscross ordered $n$-tuples of $n\times n$ matrixes. We also give a criterion to judge whether two DG free algebras are isomorphic. As an application, we consider the case of $n=2$. Based on the isomorphism classification, we compute the cohomology graded algebras of non-trivial DG free algebras with $2$ generators, and show that all those non-trivial DG free algebras are Koszul and Calabi-Yau.	math.RA	math	ISOMORPHISM PROBLEM AND HOMOLOGICAL PROPERTIES OF DG FREE ALGEBRAS X.-F. MAO, J.-F. XIE, Y.-N. YANG, AND ALMIRE. ABLA Abstract. A differential graded (DG for short) free algebra A is a connected cochain DG algebra such that its underlying graded algebra is A# = khx1, x2, · · · , xni, with xi = 1, ∀i ∈ {1, 2, · · · , n}. We prove that the differential structures on DG free algebras are in one to one correspondence with the set of crisscross ordered n-tuples of n × n matrixes. We also give a criterion to judge whether two DG free algebras are isomorphic. As an application, we consider the case of n = 2. Based on the isomorphism classification, we compute the cohomology graded algebras of non-trivial DG free algebras with 2 generators, and show that all those non-trivial DG free algebras are Koszul and Calabi-Yau. 1. introduction Throughout this paper, k is an algebraically closed field of characteristic 0. The construction of some interesting DG algebras is important in DG homological algebra. One always depends on the computations of some specific examples to deduce general rules on DG algebras. Recall that a cochain DG algebra is a Z- graded k-algebra A together with a degree one k-linear map ∂A from A to itself such that ∂A ◦ ∂A = 0 and ∂A(ab) = ∂A(a)b + (−1)aa∂A(b), for all graded elements a, b ∈ A. By definition, the graded algebra structure and the differential structure are two essential factors of a DG algebra. If one regard a DG algebra A as a living thing, then the underlying graded algebra A# and the differential ∂A are its body and soul, respectively. An efficient way to create meaningful DG algebras is to select some well known regular graded algebras as bodies, and then inject reasonable differential structures into it. In the literature, there has been some attempt for this. Especially, we have the following list: Reference [Mao] [MHLX] [MGYC] the chosen underlying graded algebra Artin-Schelter regular algebra of global dimension 2 graded down-up algebra polynomial algebra In this paper, we choose free graded associative algebras as the underlying graded algebras. We say that a cochain DG algebra A is a DG free algebra if A# is the free algebra khx1, x2,··· , xni with each xi = 1. A cochain DG algebra A is called non-trivial if ∂A 6= 0, and A is said to be connected if its underlying graded algebra A# is a connected graded algebra. Obviously, any DG free algebra is a connected cochain DG algebra. In order to study DG free algebras systematically, we describe all possible differential structures on DG free algebras by the following theorem (see Theorem 2.3). 2010 Mathematics Subject Classification. Primary 16E45, 16E65, 16W20,16W50. Key words and phrases. free algebra, DG algebra, isomorphism problem, Calabi-Yau, Koszul, homologically smooth. 1 2 X.-F. MAO, J.-F. XIE, Y.-N. YANG, AND ALMIRE. ABLA Theorem A. Let (A, ∂A) be a connected cochain DG algebra such that A# is a free graded algebra khx1, x2,··· , xni with xi = 1, for any i ∈ {1, 2,··· , n}. Then there exist a crisscross ordered n-tuple (M 1, M 2,··· , M n) of n × n matrixes such that ∂A is defined by ∂A(xi) = (x1, x2,··· , xn)M i  x1 x2 ... xn .   x1 x2 ... xn ,∀i ∈ {1, 2,··· , n} Conversely, given a crisscross ordered n-tuple (M 1, M 2,··· , M n) of n× n matrixes, we can define a differential ∂ on khx1, x2,··· , xni by   ∂(xi) = (x1, x2,··· , xn)M i  such that (khx1, x2,··· , xni, ∂) is a cochain DG algebra. One sees the definition of crisscross ordered n-tuple (M 1, M 2,··· , M n) of n × n matrixes in Definition 2.1. Theorem A indicates that DG free algebras have plenty of differential structures. To study DG free algebras systematically, we should consider the isomorphism problem first. This paper gives a criterion by the following theorem (see Theorem 3.1). Theorem B. Let A and B be two DG free algebras such that A# = khx1, x2,··· , xni, B# = khy1, y2,··· , yni, ... .   AT M nA N1 N2 ... Nn AT M 1A AT M 2A with each xi = yi = 1. Assume that ∂A and ∂B are defined by crisscross ordered n-tuples (M 1,··· , M n) and (N 1,··· , N n), respectively. Then A ∼= B if and only if there exists A = (aij)n×n ∈ GLn(k) such that  =   (aij En)n2×n2  In general, the properties of a DG algebra are determined by the joint effects of its underlying graded algebra structure and differential structure. However, it is feasible, at least in some special cases, to judge some properties of a DG algebra A from A#. For example, it is shown in [Mao] that a connected cochain DG algebra B is Gorenstein if its underlying graded algebra B# is an Artin-Schelter regular algebra of global dimension 2. In [MHLX], all non-trivial Noetherian DG down-up algebras are proved Calabi-Yau. Recently, DG polynomial algebras with degree one generators are systematically studied in [MGYC]. It is proved that any non- trivial DG polynomial algebra is Calabi-Yau and a trivial DG polynomial algebra is Calabi-Yau if and only if it is generated by odd number of generators. In this paper, we attempt to figure out homological properties of DG free algebras. We show that any trivial DG free algebra is not Gorenstein but homologically smooth (see Proposition 6.2). When it comes to non-trivial cases, things become more complicated since it seems not feasible to classify all the isomorphism classes of DG free algebras when n ≥ 3. As a consolation, we completely solve the case of n = 2. We classify the isomorphism classes of DG free algebras with 2 degree one generators (see Proposition 4.1, Proposition 4.2 and Proposition 4.4). And we reach the following interesting conclusion (see Theorem 6.6). ISOMORPHISM PROBLEM AND HOMOLOGICAL PROPERTIES OF DG FREE ALGEBRAS 3 Theorem C. Let A be a DG free algebra with 2 degree one generators. Then A is a Koszul Calabi-Yau DG algebra if and only if ∂A 6= 0. 2. differential structures on dg free algebras In this section, we will study the differential structures on DG free algebras. For this, we introduce the definition of crisscross ordered n-tuple of n× n matrixes first. Definition 2.1. Let (M 1, M 2,··· , M n) be an ordered n-tuple of n × n matrixes with each ri 1 ri 2 ... ri n We say that (M 1, M 2,··· , M n) is crisscross if n) =  2,··· , ci M i = (ci 1, ci n , i = 1, 2,··· , n.   [ck j ri k − ci krk j ] = (0)n×n,∀i, j ∈ {1, 2,··· , n}. Xk=1 In the rest of this section, we will reveal the close relations between crisscross ordered n-tuples of n×n matrixes and the differential structure of DG free algebras. The following lemma will be used in subsequent computations. Lemma 2.2. Let N 1, N 2,··· , N n be n × n matrixes such that x1 x2 ... xn in khx1, x2,··· , xni. Then each N i = (0)n×n,∀i ∈ {1, 2,··· , n}. Proof. Let N i = (ai (x1, x2,··· , xn)[N 1x1 + N 2x2 + ··· + N nxn]  jk)n×n, i = 1, 2,··· , n. By the assumption, we have   = 0 0 =(x1, x2,··· , xn)[N 1x1 + N 2x2 + ··· + N nxn]  n n n n x1 x2 ... xn   Pi=1 ··· ... Pi=1 ··· ... Pi=1 ··· jnxjxi)  n x1 x2 ... xn       ai 1nxi ... ai jnxi ... ai nnxi x1 x2 ... xn   ai 1kxi ... ai jkxi ... ai nkxi  n n  Pi=1 Pi=1 Pi=1 Xi=1 Xj=1 ai 11xi ... ai j1xi ... ai n1xi ··· ... ··· ... ··· Pi=1 Pi=1 Pi=1 Xj=1 ai j2xjxi,··· , n n n n n ai n Xi=1 =(x1, x2,··· , xn) n n =( Xj=1 Xi=1 ai j1xjxi, = n n n Xk=1 Xj=1 Xi=1 ai jkxjxixk in khx1, x2,··· , xni. 4 X.-F. MAO, J.-F. XIE, Y.-N. YANG, AND ALMIRE. ABLA This implies that ai jk = 0,∀i, j, k ∈ {1, 2,··· , n}. So N i = (0)n×n, i = 1, 2,··· , n. (cid:3) Theorem 2.3. Let (A, ∂A) be a connected cochain DG algebra such that A# is a free graded algebra khx1, x2,··· , xni with xi = 1, for any i ∈ {1, 2,··· , n}. Then there exist a crisscross ordered n-tuple (M 1, M 2,··· , M n) of n × n matrixes such that ∂A is defined by ∂A(xi) = (x1, x2,··· , xn)M i  x1 x2 ... xn .   Conversely, given a crisscross ordered n-tuple (M 1, M 2,··· , M n) of n×n matrixes, we can define a differential ∂ on khx1, x2,··· , xni by ∂(xi) = (x1, x2,··· , xn)M i  x1 x2 ... xn   ,∀i ∈ {1, 2,··· , n} such that (khx1, x2,··· , xni, ∂) is a cochain DG algebra. Proof. Since the differential ∂A of A is a k-linear map of degree 1, we may let ∂A(xi) = (x1, x2,··· , xn)M i  x1 x2 ... xn ,   where M i = (ci 1, ci 2,··· , ci n) =  ri 1 ri 2 ... ri n   ,∀i ∈ {1, 2,··· , n}. Since (A, ∂A) is a cochain DG algebra, ∂A satisfies the Leibniz rule and (1) ∂A ◦ ∂A(xi) = 0,∀i ∈ {1, 2,··· , n}. ISOMORPHISM PROBLEM AND HOMOLOGICAL PROPERTIES OF DG FREE ALGEBRAS 5 Hence ∂A(x1) ∂A(x2) ∂A(xn) ... x1 x2 ... xn     x1 x2 ... xn x1 x2 ... xn x1 x2 ... xn ]   (x1, x2,··· , xn)M n   x1 x2 ... xn x1 x2 ... xn (x1, x2,··· , xn)M 1 (x1, x2,··· , xn)M 2 0 = ∂A ◦ ∂A(xi) = ∂A[(x1, x2,··· , xn)M i  = (∂A(x1), ∂A(x2),··· , ∂A(xn))M i  − (x1, x2,··· , xn)M i    ]M i ,··· , M n  , M 2  = (x1, x2,··· , xn)[M 1          − (x1, x2,··· , xn)M i      j xj)        j ]     Xj=1 Pj=1 Pj=1 Pj=1 Xk=1 Xj=1 j )xj ]  Xj=1 = (x1, x2,··· , xn)( x1 x2 ... xn ri 1 ri 2 ... ri n − (x1, x2,··· , xn)(ci ,∀i ∈ {1, 2,··· , n}. = (x1, x2,··· , xn)[ = (x1, x2,··· , xn)[ xj r2 j ... xj rn j 1, ci 2,··· , ci n) (ck j ri k − ci krk Xk=1 Xj=1 Xj=1 Xk=1     c2 j xj,··· ,     ck j xjri k − x1 x2 ... xn x1 x2 ... xn n n n n x1 x2 ... xn n c1 j xj, n Xj=1 n cn xj r1 j x1 x2 ... xn ci kxjrk ... n n n n n By Lemma 2.2, we have n (ck j ri k − ci krk j ) = (0)n×n,∀i, j ∈ {1, 2,··· , n}. Xk=1 So (M 1, M 2,··· , M n) is a crisscross ordered n-tuple of n × n matrixes. 6 X.-F. MAO, J.-F. XIE, Y.-N. YANG, AND ALMIRE. ABLA Conversely, if (M 1, M 2,··· , M n) is a crisscross ordered n-tuple of n×n matrixes with each M i = (ci 1, ci 2,··· , ci n) =  ri 1 ri 2 ... ri n ,   then we have n (ck j ri k − ci krk j ) = (0)n×n,∀i, j ∈ {1, 2,··· , n}. Xk=1 We can define a differential ∂ on khx1, x2,··· , xni by   ∂(xi) = (x1, x2,··· , xn)M i  x1 x2 ... xn , i = 1, 2,··· , n, such that (khx1, x2,··· , xni, ∂) is a cochain DG algebra, since one can check as above that if ∂ satisfies the Leibniz rule. ∂ ◦ ∂(xi) = 0,∀i ∈ {1, 2,··· , n}, (cid:3) Remark 2.4. Theorem 2.3 indicates that there is a one to one in correspondence between {AA is a DG free algebra with A# = khx1, x2,··· , xni} and the set {(M 1,··· , M n)(M 1,··· , M n) is a crisscross ordered n-tuple, M i ∈ Mn(k)}. 3. Isomorphism problems for DG free algebras Remark 2.4 implies that DG free algebras have abundant differential structures. For future systematically studies, one should consider the isomorphism problem of DG free algebras first, since two isomorphic DG free algebras have same homological properties. A successful classification work on the isomorphism classes of DG free algebras will efficiently simplify our research. We have the following theorem. Theorem 3.1. Let A and B be two DG free algebras such that A# = khx1, x2,··· , xni, B# = khy1, y2,··· , yni, with each xi = yi = 1. By Theorem 2.3, ∂A and ∂B are defined by crisscrossed n × n matrixes M 1, M 2,··· , M n and N 1, N 2,··· , N n respectively. Then A ∼= B if and only if there exists A = (aij )n×n ∈ GLn(k) such that AT M 1A AT M 2A (aij En)n2×n2  N1 N2 ... Nn   =  ... AT M nA .   ISOMORPHISM PROBLEM AND HOMOLOGICAL PROPERTIES OF DG FREE ALGEBRAS 7 Proof. If the DG algebras A ∼= B, then there exists an isomorphism f : A → B of DG algebras. Since f 1 : A1 → B1 is a k-linear isomorphism, we may let f (x1) f (x2) ... f (xn)     = A  y1 y2 ... yn   for some A = (aij )n×n ∈ GLn(k). Since f is a chain map, we have f ◦ ∂A = ∂B ◦ f . For any i ∈ {1, 2,··· , n}, we have (Eq1) and (Eq2) n aijyj) Xj=1 ∂B ◦ f (xi) = ∂B( aij[(y1, y2,··· , yn)N j  n n y1 y2 ... yn ]   nj stysyt] Xt=1 aijnj stysyt aijnj stysyt aij[ n n Xs=1 Xt=1 n Xs=1 Xt=1 Xj=1 n = = = = n Xj=1 n n Xj=1 Xj=1 Xs=1 n f ◦ ∂A(xi) = f [(x1, x2,··· , xn)M i  n n x1 x2 ... xn ]   = f [ mi klxkxl] = klf (xk)f (xl) n n n n kl( mi mi Xl=1 Xk=1 Xk=1 Xl=1 Xl=1 Xt=1 Xt=1 Xk=1 Xk=1 Xs=1 Xl=1 Xl=1 n n n n n n n aksys)( Xt=1 altyt) mi klaksaltysyt aksmi klaltysyt. n n = Xk=1 Xs=1 Xs=1 So ∂B ◦ f (xi) = f ◦ ∂A(xi) implies that = = n aijnj st = n Xj=1 n n Xk=1 Xl=1 aksmi klalt,∀i, s, t ∈ {1, 2,··· , n}. 8 X.-F. MAO, J.-F. XIE, Y.-N. YANG, AND ALMIRE. ABLA  =    =   = A  f (x1) f (x2) ... f (xn)     y1 y2 ... yn .   Then we have aijN j = AT M iA, for any i = 1, 2,··· , n. Therefore, n Pj=1 AT M 1A AT M 2A N1 N2 ... Nn   Conversely, if there exists A = (aij)n×n ∈ GLn(k) such that   (aij En)n2×n2  (aij En)n2×n2  AT M 1A AT M 2A N1 N2 ... Nn AT M nA AT M nA ... ... . , we should show that A ∼= B. Define a k-linear map f : A1 → B1 by Obviously, f is invertible since A ∈ GLn(k). Extend f to a morphism of graded algebras between A# and B#. We still denote it by f . For any i ∈ {1, 2,··· , n}, we still have (Eq1) and (Eq2). Since (aijEn)n2×n2  N 1 N 2 ... N n   =  AT M 1A AT M 2A ... AT M nA ,   we have f ◦ ∂A(xi) = ∂B ◦ f (xi), for any i ∈ {1, 2,··· , n}. So f is an isomorphism of DG algebras. (cid:3) Corollary 3.2. Let A and B be two DG free algebras such that A# = khx1, x2,··· , xni, B# = khy1, y2,··· , yni, with each xi = yi = 1. If A ∼= B and ∂A and ∂B are defined respectively by two crisscross ordered n-tuples (M 1, M 2,··· , M n) and (N 1, N 2,··· , N n) of n × n matrixes, then M 1 M 2 ... M n (1) r    = r  N 1 N 2 ... N n ;   (2) N 1, N 2,··· , N n are symmetric matrixes whenever M 1, M 2,··· , M n are symmetric matrixes. Proof. (1) By Theorem 3.1, there exists A = (aij )n×n ∈ GLn(k) such that (2) (aijEn)n2×n2  N1 N2 ... Nn   =  AT M 1A AT M 2A ... AT M nA   ISOMORPHISM PROBLEM AND HOMOLOGICAL PROPERTIES OF DG FREE ALGEBRAS 9 since A ∼= B. We have AT M 1A AT M 2A   ... AT M nA Since A ∈ GLn(k) and   we have   =  AT 0n×n 0n×n AT ... ... 0n×n 0n×n ··· ··· . . . ··· 0n×n 0n×n ... AT M 1 M 2 ... M n       A. AT 0n×n 0n×n AT ... ... 0n×n 0n×n ··· ··· . . . ··· 0n×n 0n×n ... AT   ∈ GLn2 (k), r  AT M 1A AT M 2A ... AT M nA   = r  M 1 M 2 ... M n .   On the other hand, let A−1 = (bij)n×n, then (bijEn)n2×n2 (aij En)n2×n2 = En2 , which implies that (aij En)n2×n2 ∈ GLn2 (k) and hence AT M 1A  AT M 2A  N1 N2 ... Nn AT M nA ... r      . N 1 N 2 ... N n  = r    = r   = (bijEn)n2×n2  n (2)By (2), we have by (2). Therefore, r    M 1 M 2 ... M n N1 N2 ... Nn AT M 1A AT M 2A ... AT M nA   =   n Pj=1 Pj=1 Pj=1 n b1jAT M jA b2jAT M jA ... bnjAT M jA .   Hence, N i = bijAT M jA and N i is a symmetric matrix when M 1, M 2,··· , M n are symmetric matrixes, i = 1, 2,··· , n. (cid:3) n Pj=1 10 X.-F. MAO, J.-F. XIE, Y.-N. YANG, AND ALMIRE. ABLA Corollary 3.3. Let A be the DG free algebra such that A# = khx1, x2,··· , xni with xi = 1, i = 1, 2,··· , n. Assume that ∂A is determined by a crisscross ordered n-tuple (M 1, M 2,··· , M n) of n × n matrixes. Then Autdg(A) = {A = (aij)n×n ∈ GLn(k)(aij En)n2×n2  M 1 M 2 ... M n   =  AT M 1A AT M 2A ... AT M nA   }. 4. isomorphism classes of dg free algebras with two generators By Remark 2.4, the set of crisscross ordered 2-tuples of 2 × 2 matrixes are in one to one correspondence with the set of DG free algebras with two degree one generators. Hence we should describe all crisscross ordered 2-tuples of 2×2 matrixes first in order to figure out the isomorphism classes of DG free algebras with two generators. For this, let M 1 and M 2 be two 2 × 2 matrixes such that M 1 = (c1 1, c1 M 2 = (c2 1, c2 1 r1 2) =(cid:18) r1 2) =(cid:18) r2 2 (cid:19) =(cid:18) m1 2 (cid:19) =(cid:18) m2 11 m1 12 21 m1 11 m2 12 21 m2 22 (cid:19) , 22 (cid:19) . 1 r2 m1 m2 By Definition 2.1, (M1, M2) is a crisscross ordered 2-tuple of 2 × 2 matrixes if and only if   if and only if 2r2 2 − c1 1r1 c2 1r1 2r1 1 − c1 c1 1r1 1r2 1 − c2 c1 1 = (0)2×2 2 + c2 1 + c2 2r2 2r1 2 − c1 2r2 1r2 2 − c2 2 = (0)2×2 1 = (0)2×2 (1) (2) (3), (3)   m2 m2 m2 m2 m1 (m1 m1 m1 m2 m1 m1 m1 21 − m1 11m1 12m2 22 − m1 11m1 12m2 21 − m1 21m1 22m2 22 − m1 21m1 22m2 12m1 11 − m1 11m1 12)2 − m1 11m1 22m1 11 − (m1 12 − m1 22m1 21 − m2 11m2 12 − m2 11m2 21m2 11 − m2 12 − m2 21m2 11 = 0 12 = 0 11 = 0 12 = 0 21 + m2 22 + m2 21)2 + m1 21m1 22 = 0 12m2 11 = 0 12 + m2 11m1 21m1 11 + (m2 12 + m2 21m1 (1.1) (1.2) (1.3) (1.4) 21 = 0 (2.1) (2.2) 22 = 0 (2.3) 21 = 0 (2.4) (3.1) 12)2 = 0 (3.2) (3.3) 11 = 0 12 = 0 (3.4), 12m1 12m1 21m2 21 − m1 22 − m1 22 − m1 12m2 12m2 22m2 11m2 22 − (m2 22m2 21)2 − m2 22m2 22 − m2 21m2 ISOMORPHISM PROBLEM AND HOMOLOGICAL PROPERTIES OF DG FREE ALGEBRAS11 where equations (i.1), (i.2), (i.3), (i.4) are equivalent to the equation (i), i = 1, 2, 3. When m2 22 = 0, the equations (3) are equivalent to 11 = m1 (4)   12 = m2 0 = m1 m1 12(m1 m1 12(m1 21(m2 m1 12(m1 m2 m2 21(m1 12(m1 m2 12m2 11 − m2 12 − m2 22 − m1 11 − m2 11 − m2 21 − m2 21m1 21 21) − m1 21(m1 22) = 0 21) = 0 12) = 0 21) = 0 22) − m2 21(m1 11 − m2 12) = 0 12 − m2 22) = 0. By computations, we have the following classification when m1 Cases for m1 M 1 M 2 22 = m2 11 = 0. Parameters 11 = 0 22 = m2 21 = 0 21 6= 0 21 = 0, 21 6= 0 21 = 0, 21 = 0 21 = 0 21 = 0 21 = 0 21 = 0 m2 m1 m2 m2 1.(m1 2.(m1 3.(m1 4.(m1 5.(m1 6.(m1 7.(m2 8.(m2 9.(m2 10.(m2 11.(m2 12.(m2 12 = 0, m1 12 6= 0, m2 m2 12 = 0, m1 12 = 0, m2 12 = 0, m1 12 6= 0, m2 m2 12 = 0, m1 12 = 0, m2 12 = 0, m2 21 6= 0 12 = 0, m2 12 6= 0 12 = 0, m2 11 6= 0, m1 m1 12 = 0, m2 11 = 0, m1 12 = 0, m2 11 = 0, m1 12 = 0, m2 11 = 0, m1 12 = 0, m1 12 6= 0 12 = 0, m1 21 6= 0 m1 m1 m2 m1 m1 21 = 0 21 = 0 12 6= 0 21 = 0 12 6= 0, m1 21 = 0 12 6= 0, m1 21 = 0 12 = 0, m1 21 = 0 21 = 0 21 6= 0 21 6= 0 0 0 0 ν 0 ν 0 0 0 0 0 0 0 ν ν µ λ (cid:19) µ ∈ k×, λ ∈ k 0 (cid:19) (cid:18) 0 µ (cid:18) µ 0 0 (cid:19) (cid:18) 0 (cid:18) ν 0 (cid:19) (cid:18) 0 (cid:18) ν (cid:18) λ 0 0 (cid:19) (cid:18) 0 µ 0 (cid:19) (cid:18) 0 (cid:18) ν (cid:18) ν µ 0 (cid:19) (cid:18) 0 µ 0 (cid:19) (cid:18) 0 (cid:18) λ µ (cid:18) 0 µ 0 (cid:19) (cid:18) 0 0 (cid:19) (cid:18) 0 (cid:18) 0 µ 0 (cid:19) (cid:18) 0 (cid:18) 0 (cid:18) ν µ 0 (cid:19) (cid:18) 0 0 (cid:19) (cid:18) 0 (cid:18) ν µ 0 (cid:19) ν ∈ k× 0 (cid:19) ν ∈ k× 0 µ (cid:19) λ, µ ∈ k 0 µ (cid:19) µ ∈ k×, ν ∈ k 0 µ (cid:19) µ ∈ k, ν ∈ k× 0 µ (cid:19) λ, µ ∈ k× 0 µ (cid:19) µ ∈ k× ν (cid:19) ν ∈ k× 0 µ (cid:19) µ ∈ k× ν µ (cid:19) µ ∈ k×, ν ∈ k ν µ (cid:19) µ ∈ k, ν ∈ k× 0 0 0 0 0 0 0 0 0 0 0 ν ν 12 X.-F. MAO, J.-F. XIE, Y.-N. YANG, AND ALMIRE. ABLA When (m1 22, m2 Cases 11 6= 0, m1 22 6= 0 m2 11 = 0, m1 22 6= 0, m2 12 = 0 m2 11 = 0, m1 22 6= 0, m1 12 = 0 m2 11 6= 0, m1 22 = 0, m1 12 = 0 m2 11 6= 0, m1 22 = 0, m2 12 = 0 13.(m2 14.  15.  16.  17.  µ ν ν 0 µ µ λ µ M 2 M 1 Parameters λ ∈ k×, µ, ω ∈ k µ, ν, λ ∈ k×, ω ∈ k 11) 6= (0, 0), we have the following classification by computations. (cid:18) ν + λ(µ − ω) µ (cid:18) µ(µ−ω) λ (cid:19) (cid:18) ω 0 0 λ (cid:19) (cid:18) ω 0 0 (cid:19) (cid:18) µ ω ω 0 (cid:19) λ (cid:19) (cid:18) λν ω (cid:19) (cid:18) 0 0 ω (cid:19) (cid:18) 0 ω ω µ (cid:19) (cid:18) λ (cid:18) λ 0 0 ω (cid:19) (cid:19) λ ∈ k×, µ, ω ∈ k λ ∈ k×, µ, ω ∈ k λ ∈ k×, µ, ω ∈ k µ µ(µ−ω) µ 0 λ Now, lets come back to our concerned isomorphism problem. Let A and B be two DG free algebras such that A# = khx1, x2i, B# = khy1, y2i, with each xi = yi = 1. Assume that ∂A and ∂B are defined by crisscross ordered 2-tuples (M 1, M 2) and (N 1, N 2) of 2 × 2 matrixes, respectively. Let 22 (cid:19) , i = 1, 2. 22 (cid:19) , N i =(cid:18) ni M i =(cid:18) mi By Theorem 3.1, A ∼= B if and only if there exists A = (aij )2×2 ∈ GL2(k) such that (5) 11 ni 12 21 ni ni 11 mi 12 21 mi (aij E2)4×4(cid:18) N 1 N 2 (cid:19) =(cid:18) AT M 1A AT M 2A (cid:19) , mi i.e., 11m1 a11n1 a11n1 a11n1 a11n1 a21n1 a21n1 a21n1 a21n1 11 + a12n2 12 + a12n2 21 + a12n2 22 + a12n2 11 + a22n2 12 + a22n2 21 + a22n2 22 + a22n2 11 = a2 12 = a11a12m1 21 = a12a11m1 22 = a2 11 = a2 12 = a11a12m2 21 = a11a12m2 22 = a2 12m1 11m2 12m2 11 + a11a21m1 21 + a11a21m1 12 + a2 21m1 22 11 + a12a21m1 11 + a11a22m1 21 + a11a22m1 21 + a12a21m1 12 + a21a22m1 22 12 + a21a22m1 22 11 + a12a22m1 11 + a11a21m2 21 + a12a22m1 21 + a11a21m2 12 + a2 12 + a2 22m1 22 21m2 22 11 + a12a21m2 11 + a11a22m2 21 + a11a22m2 21 + a12a21m2 12 + a21a22m2 22 12 + a21a22m2 22 11 + a12a22m2 21 + a12a22m2 12 + a2 22m2 22.   N 2 (cid:19) = r(cid:18) M 1 If A ∼= B, then by Corollary 3.2, we have (1)r(cid:18) N 1 Proposition 4.1. Assume that A is a DG free algebra such that M 2 (cid:19); (2)N 1, N 2 are symmetric matrixes whenever M 1, M 2 are symmetric matrixes. A# = khx1, x2i,x1 = x2 = 1 and ∂A is defined by a crisscross ordered 2-tuple (M 1, M 2) of 2 × 2 matrixes. If M 1 and M 2 are not both symmetric matrixes, then A is isomorphic to either of the following two DG free algebras: ISOMORPHISM PROBLEM AND HOMOLOGICAL PROPERTIES OF DG FREE ALGEBRAS13 (1) B1 whose differential ∂B1 is defined by (cid:18) 1 (2) B2 whose differential ∂B2 is defined by (cid:18) 1 0 0 0 1 0 (cid:19) and (cid:18) 0 0 (cid:19) and (cid:18) 0 0 0 0 0 (cid:19) ; 0 (cid:19). 1 Proof. Since M 1 and M 2 are not both symmetric matrixes, we have m2 22 = 0, and the crisscross ordered 2-tuple (M 1, M 2) of 2 × 2 matrixes belongs to one of following cases: 11 = m1 Case M 1 M 2 Parameters 2. 3. 5. 6. 8. 10. 11. 12. 0 0 0 0 0 0 (cid:18) ν 0 (cid:19) (cid:18) ν 0 (cid:19) (cid:18) ν µ 0 (cid:19) (cid:18) ν µ 0 (cid:19) (cid:18) 0 µ 0 (cid:19) (cid:18) 0 µ 0 (cid:19) (cid:18) ν µ 0 (cid:19) (cid:18) ν µ 0 (cid:19) 0 0 0 0 ν ν ν ν 0 0 (cid:18) 0 0 (cid:19) (cid:18) 0 0 (cid:19) (cid:18) 0 0 µ (cid:19) (cid:18) 0 0 µ (cid:19) (cid:18) 0 0 µ (cid:19) (cid:18) 0 0 µ (cid:19) (cid:18) 0 ν µ (cid:19) (cid:18) 0 ν µ (cid:19) 0 0 0 0 ν ∈ k× ν ∈ k× µ ∈ k×, ν ∈ k µ ∈ k, ν ∈ k× µ ∈ k× µ ∈ k× µ ∈ k×, ν ∈ k µ ∈ k, ν ∈ k× by the classification above. Let (M 1, M 2) and (N 1, N 2) are two crisscross ordered 2-tuples of 2 × 2 matrixes belong to one of the 8 cases listed above. We want to check whether there exists A ∈ GL2(k) such that (5) holds. By computations, (M 1, M 2) (N 1, N 2) Cases Cases Cases 2 8 3 Cases 10 Cases 3 8 11 6 5 10 1 0 ν µ µ ν ∃A (cid:18) 0 0 (cid:19) (cid:18) 1 1 (cid:19) 1 (cid:19) (cid:18) 1 1 + ν 2 (cid:18) 0 0 (cid:19) . µ ν µ ν ν µ µ2 0 1 ν µ 1 ! And it is easy to see that the crisscross ordered 2-tuple (M 1, M 2) of 2× 2 matrixes in Case 12 belongs to Case 11 and Case 2 when µ ∈ k× and µ = 0, respectively. By Theorem 3.1, A ∼= B1 when the crisscross ordered 2-tuple (M 1, M 2) of 2 × 2 matrixes belongs to any one of the following cases: Case 2, Case 8, Case 11, and Case 12. And A ∼= B2 when (M 1, M 2) belongs to Case 3, Case 5, Case 6 and Case 14 X.-F. MAO, J.-F. XIE, Y.-N. YANG, AND ALMIRE. ABLA 10. Since 1 0 0 1 we have B1 6∼= B2 by Corollary 3.2. 1 = r  0 0 0 0   6= r  1 0 0 0 0 1 0 0   = 2, (cid:3) Proposition 4.2. Assume that A is a DG free algebra such that A# = khx1, x2i,x1 = x2 = 1 and ∂A is defined by crisscrossed 2 × 2 matrixes M 1 and M 2. If M 1 and M 2 are both symmetric matrixes with m1 11 = 0, then A is isomorphic to one of the following four DG free algebras: 22 = m2 (1) B3 whose differential ∂B3 is defined by (cid:18) 1 (2) B4 whose differential ∂B4 is defined by (cid:18) 1 (3) B5 whose differential ∂B5 is defined by (cid:18) 1 (4) B0 whose differential ∂B0 = 0. 0 0 0 0 1 0 0 (cid:19) and (cid:18) 0 0 (cid:19) and (cid:18) 0 0 (cid:19) and (cid:18) 0 0 0 0 0 1 0 (cid:19) ; 1 (cid:19); 0 (cid:19) ; 0 Proof. Since M 1 and M 2 are both symmetric matrixes with m1 crisscrossed 2 × 2 matrixes pair (M 1, M 2) belongs to one of following cases: 22 = m2 11 = 0, the Case M 1 M 2 Parameters 1. 4. 7. 9. 0 0 (cid:18) µ 0 0 (cid:19) (cid:18) λ 0 0 (cid:19) (cid:18) λ µ µ 0 (cid:19) (cid:18) 0 0 (cid:19) ν ν 0 (cid:18) 0 µ µ λ (cid:19) (cid:18) 0 0 µ (cid:19) (cid:18) 0 0 µ (cid:19) (cid:18) 0 ν (cid:19) 0 0 0 µ ∈ k×, λ ∈ k λ, µ ∈ k λ, µ ∈ k× ν ∈ k× by the classification above. By computations, for 0 M 1 =(cid:18) µ 0 N 1 =(cid:18) λ µ 0 (cid:19) ,M 2 =(cid:18) 0 µ µ λ (cid:19) 0 µ (cid:19) , λ, µ ∈ k× µ 0 (cid:19) ,N 2 =(cid:18) 0 0 and 0 0 M 1 =(cid:18) λ 0 N 1 =(cid:18) λ µ 0 (cid:19) ,M 2 =(cid:18) 0 µ 0 (cid:19) ,N 2 =(cid:18) 0 0 µ (cid:19) 0 µ (cid:19) , λ, µ ∈ k×, 0 ISOMORPHISM PROBLEM AND HOMOLOGICAL PROPERTIES OF DG FREE ALGEBRAS15 there exists A =(cid:18) 0 1 1 0 (cid:19) and A =(cid:18) 0 On the other hand, for any λ, µ ∈ k×, and λ µ µ λ 1 (cid:19), respectively, such that (5) holds. 0 0 0 0 (cid:19) ,M 2 =(cid:18) 0 0 (cid:19) ,N 2 =(cid:18) 0 M 1 =(cid:18) λ 0 N 1 =(cid:18) 1 µ (cid:19) such that (5) holds. Therefore, A is isomorphic to B4, 0 µ (cid:19) 1 (cid:19) , 0 1 0 0 0 there exists A =(cid:18) 1 λ 0 when M 1 and M 2 belong to each of the following cases: (1) Case 7; (2) Case 1, λ 6= 0; (3) Case 4, λ 6= 0, µ 6= 0. If the parameter λ = 0 in Case 1, then M 1 =(cid:18) µ 0 0 (cid:19) , M 2 =(cid:18) 0 µ µ 0 (cid:19) , µ ∈ k×. 0 Let N 1, N 2 belong to Case 9, i.e., N 1 =(cid:18) 0 ν ν 0 (cid:19) , N 2 =(cid:18) 0 0 ν (cid:19) . 0 Then there exists A = (cid:18) 0 B3, when M 1 and M 2 belong to either of the following two cases: 0 (cid:19) such that (5) holds. Hence A is isomorphic to 1 ν µ (1) Case 1, λ = 0; (2) Case 9. If the parameter λ 6= 0 and µ = 0 in Case 4, then M 1 =(cid:18) λ 0 0 (cid:19) , M 2 =(cid:18) 0 0 0 0 (cid:19) . 0 Let N 1, N 2 belong to Case 4 with λ = 0, µ 6= 0, i.e., 0 0 (cid:19) , N 2 =(cid:18) 0 0 (cid:19) such that (5) holds. So A is isomorphic to B5, N 1 =(cid:18) 0 0 0 µ (cid:19) . when M 1 and M 2 belong to either of the following two cases: µ λ 0 Then there exists A = (cid:18) 0 (1) Case 4, λ 6= 0, µ = 0; (2) Case 5, λ = 0, µ 6= 0. 1 (cid:3) Remark 4.3. We have B3 6∼= B4 since one can't find A = (aij)2×2 ∈ GL2(k) satisfying (5),i.e., a11 a12 a12 0 a21 a22 a22 0     =  a2 11 a12a11 a2 21 a21a22 a11a12 a2 12 a21a22 a2 22 .   Hence B3,B4,B5 and B6 are 4 different isomorphism classes by Corollary 3.2. 16 X.-F. MAO, J.-F. XIE, Y.-N. YANG, AND ALMIRE. ABLA Proposition 4.4. Assume that A is a DG free algebra such that A# = khx1, x2i,x1 = x2 = 1 and ∂A is defined by a crisscross ordered 2-tuple (M 1, M 2) of 2× 2 matrixes. If M 1 and M 2 are both symmetric matrixes with (m1 11) 6= (0, 0), then A is isomorphic to one of the following DG free algebras: 22, m2 0 0 0 0 0 0 0 1 1 0 (1) B5, where ∂B5 is defined by (cid:18) 1 (2) B6, where ∂B6 is defined by (cid:18) 0 (3) B7, where ∂B7 is defined by (cid:18) 0 (4) B8, where ∂B8 is defined by (cid:18) 0 (5) B9, where ∂B9 is defined by (cid:18) 1 (6) B10, where ∂B10 is defined by (cid:18) 1 (7) B11, where ∂B11 is defined by (cid:18) 1 (8) B(s, t), where ∂B(s,t) is defined by (cid:18) 1 + s − st 0 (cid:19) and (cid:18) 0 1 (cid:19) and (cid:18) 0 1 (cid:19) and (cid:18) 0 1 (cid:19) and (cid:18) 0 0 (cid:19) and (cid:18) 0 0 0 − 1 0 0 (cid:19); 0 (cid:19); 1 (cid:19); 1 (cid:19) ; 1 (cid:19) ; 4 (cid:19) and (cid:18) 0 1 1 1 (cid:19) ; 1 (cid:19) and (cid:18) 0 1 1 0 (cid:19); s (cid:19) and (cid:18) s 1 1 0 0 0 0 0 1 0 1 0 0 0 1 k×, t ∈ k. 1 t (cid:19) , s ∈ 1 Proof. Since M 1 and M 2 are both symmetric matrixes with (m1 11) 6= (0, 0), the crisscross ordered 2-tuple (M 1, M 2) of 2 × 2 matrixes belongs to one of the following cases: 22, m2 Case M 1 M 2 Parameters 13. 14. 15. 16. 17. µ µ µ λ µ λ (cid:19) (cid:18) ν + λ(µ − ω) µ (cid:18) µ(µ−ω) λ (cid:19) (cid:18) ω 0 0 λ (cid:19) (cid:18) ω 0 0 (cid:19) (cid:18) µ ω ω 0 (cid:19) 0 ν ν 0 (cid:18) λν ω (cid:19) (cid:18) 0 0 ω (cid:19) (cid:18) 0 ω ω µ (cid:19) (cid:19) (cid:18) λ 0 0 ω (cid:19) µ λ µ µ(µ−ω) (cid:18) λ µ, ν, λ ∈ k×, ω ∈ k λ ∈ k×, µ, ω ∈ k λ ∈ k×, µ, ω ∈ k λ ∈ k×, µ, ω ∈ k λ ∈ k×, µ, ω ∈ k by the classification above. By computations, for M 1 =(cid:18) µ(µ−ω) λ µ N 1 =(cid:18) ω 0 0 µ λ (cid:19) ,M 2 =(cid:18) 0 0 (cid:19) ,N 2 =(cid:18) λ 0 0 ω (cid:19) µ µ(µ−ω) µ λ (cid:19) , λ ∈ k×, µ, ω ∈ k ISOMORPHISM PROBLEM AND HOMOLOGICAL PROPERTIES OF DG FREE ALGEBRAS17 and M 1 =(cid:18) ω 0 N 1 =(cid:18) µ ω 0 λ (cid:19) ,M 2 =(cid:18) 0 ω ω µ (cid:19) ω 0 (cid:19) ,N 2 =(cid:18) λ 0 0 ω (cid:19) , λ ∈ k×, µ, ω ∈ k, 0 (cid:19) such that (5) holds. By Theorem 3.1, we only need to 1 there exists A =(cid:18) 0 1 check the isomorphism classes of Case 13, Case 14 and Case 15 one by one. For Case 14, we divide it into the following 5 cases: Cases M 1 M 2 Parameters 0 µ λ µ 14.2 ω = 0, µ = 0 14.1 ω = 0, µ 6= 0 (cid:18) µ2 λ (cid:19) (cid:18) 0 0 λ (cid:19) (cid:18) 0 µ µ λ (cid:19) 14.3 µ = ω 6= 0 (cid:18) 0 0 λ (cid:19) 14.4 ω 6= 0, µ = 0 14.5 ω 6= 0, µ 6= ω, µ 6= 0 (cid:18) µ(µ−ω) λ µ 0 0 0 0 0 (cid:18) 0 0 (cid:19) λ, µ ∈ k× (cid:18) 0 0 (cid:19) λ ∈ k× (cid:18) 0 0 µ (cid:19) λ, µ ∈ k× (cid:18) 0 0 ω (cid:19) λ, ω ∈ k× λ (cid:19) (cid:18) 0 0 ω (cid:19) λ, µ, ω ∈ k× 0 0 0 µ . For the cases listed above, we can choose corresponding N 1, N 2, such that there exists A ∈ GLk(2) such that (5) holds. We have the following tabular: Cases N 1 N 2 A 0 0 Case 14.1 (cid:18) 1 Case 14.2 (cid:18) 0 Case 14.3 (cid:18) 0 Case 14.4 (cid:18) 0 Case 14.5 (cid:18) 1 1 0 1 0 0 1 0 0 0 (cid:19) (cid:18) 0 1 (cid:19) (cid:18) 0 1 (cid:19) (cid:18) 0 1 (cid:19) (cid:18) 0 0 (cid:19) (cid:18) 0 1 0 0 0 0 0 0 0 0 0 (cid:19) (cid:18) λ µ (cid:19) µ2 −λ 0 0 (cid:19) (cid:18) λ 0 1 (cid:19) µ ! 1 (cid:19) λ 1 (cid:19) (cid:18) λ ω (cid:19) ω2 0 µ(µ−ω) − λ 1 (cid:19) (cid:18) µ2 0 0 1 0 1 0 0 λ 0 µω 1 ω (cid:19) . By Theorem 3.1, A is isomorphic to B5,B6,B7,B8 and B9, when M 1, M 2 belong to Case 14.1, Case 14.2, Case 14.3, Case 14.4 and Case 14.5, respectively. For Case 15, we divide it into the following 5 cases: Cases M 1 M 2 Parameters 15.1 ω 6= 0, µ 6= 0, ωλ + µ2 15.2 ω 6= 0, µ 6= 0, ωλ + µ2 15.3 ω = 0, µ = 0 15.4 µ = 0, ω 6= 0 15.5 µ 6= 0, ω = 0 4 6= 0 (cid:18) ω 0 4 = 0 (cid:18) ω 0 0 0 λ (cid:19) (cid:18) 0 ω 0 λ (cid:19) (cid:18) 0 ω (cid:18) 0 0 λ (cid:19) (cid:18) 0 0 (cid:18) ω 0 0 λ (cid:19) (cid:18) 0 ω (cid:18) 0 0 λ (cid:19) (cid:18) 0 ω µ (cid:19) λ, ω, µ ∈ k× ω µ (cid:19) λ, ω, µ ∈ k× 0 0 (cid:19) λ ∈ k× ω 0 (cid:19) λ, ω ∈ k× 0 µ (cid:19) λ, µ ∈ k× 0 0 . 18 X.-F. MAO, J.-F. XIE, Y.-N. YANG, AND ALMIRE. ABLA For the cases listed above, we can choose corresponding N 1, N 2, such that there exists A ∈ GLk(2) such that (5) holds. We have the following tabular: Cases N 1 N 2 A −µ 1 ω 0 0 0 −1 ω√4ωλ+µ2 Case 15.1 (cid:18) 1 0 0 1 (cid:19) Case 15.2 (cid:18) 1 Case 15.3 (cid:18) 0 0 0 1 (cid:19) Case 15.4 (cid:18) 1 0 0 1 (cid:19) Case 15.5 (cid:18) 0 0 0 1 (cid:19) 2√4ωλ+µ2  1 0 (cid:19)  (cid:18) 0 1   1 1 (cid:19) (cid:18) 1 4 (cid:19) (cid:18) 0 1 µ (cid:19) (cid:18) 0 0 0 0 (cid:19) (cid:18) λ 0 1 (cid:19) 1 0 (cid:19) (cid:18) 1 1√λω (cid:19) (cid:18) 0 1 µ ! 0 1 (cid:19) λ (cid:18) 0 0 µ2 0 0 1 0 1 ω 0 ω 0 0 0 . By Theorem 3.1, A is isomorphic to B11,B10,B6,B11 and B8, when M 1, M 2 belong to Case 15.1, Case 15.2, Case 15.3, Case 15.4 and Case 15.5, respectively. For Case 13, we have M 1 =(cid:18) ν + λ(µ − ω) µ λ (cid:19) , M 2 =(cid:18) λν µ ν µ ν ω (cid:19) with λ, µ, ν ∈ k×, ω ∈ k. Let µ (cid:19) . ν 1 ν 0 ν 1 1 ν 0 1 1 w N 1 = 1 + λ(µ−w) There exists A =(cid:18) 1 λµ ! , N 2 =(cid:18) µλ µ (cid:19) such that (5) holds. Let s = λµ t (cid:19) with s ∈ k×, t ∈ k. By Theorem 3.1, A is isomorphic to B(s, t). Proposition 4.5. For any s ∈ k×, we have N 1 =(cid:18) 1 + s − st s (cid:19) , N 2 =(cid:18) s 1 1 1 1 1 (1) B(s, s−1) ∼= B8, if s 6= −1; (2) B(s, s−1) ∼= B6, if s = −1. ν , t = ω µ . Then (cid:3) Proof. By definition, B8 and B(s, s−1) are defined by crisscrossed ordered 2-tuples (M 1, M 2) and (N 1, N 2), where 1 1 s−1 (cid:19) = N 2. 1 s + 2 s+1 s (cid:19) in GLk(2) such 0 0 that (5) hold, i.e., 1 (cid:19) = M 2 M 1 =(cid:18) 0 and N 1 =(cid:18) s When s 6= −1, we can choose A = (aij)2×2 = (cid:18) s =  a11s + a12s a11 + a12 a21s + a22s a21 + a22 By Theorem 3.1, B(s, s−1) ∼= B8, when s 6= −1. a11s−1 + a12s−1 a21s−1 + a22s−1     a11 + a12 a21 + a22 s + 1 a21a22 a2 21 a2 21 a21a22 a21a22 a2 22 a21a22 a2 22 .   ISOMORPHISM PROBLEM AND HOMOLOGICAL PROPERTIES OF DG FREE ALGEBRAS19 When s = −1, B(s, s−1) = B(−1,−1) is defined by the crisscross ordered 2-tuple (N 1, N 2) of 2 × 2 matrixes, where N 1 =(cid:18) −1 1 −1 (cid:19) = N 2. 1 By definition, B6 is defined by the crisscrossed ordered 2-tuple (M 1, M 2) with M 1 =(cid:18) 0 0 We can choose A = (aij )2×2 =(cid:18) −1 0 1 (cid:19) , M 2 =(cid:18) 0 1 −1 (cid:19) in GLk(2) such that (5) hold, i.e., 0 (cid:19) . 0 0 0 −a11 − a12s a11 + a12 −a21 − a22s a21 + a22   a11 + a12 −a11 − a12 a21 + a22 −a21 − a22 a2 21 a21a22 0 0 a21a22 a2 22 0 0   =  .   (cid:3) By Theorem 3.1, B(−1,−1) ∼= B6. 5. cohomology graded algebras of DG free algebras In this section, we will compute the cohomology graded algebra of DG free algebras with two degree one generators. Due to the classifications finished in the previous section, we only need to compute H(B1), H(B2),··· , H(B11), H(B(s, t)) with st 6= 1, s ∈ k×, by Proposition 4.1, Proposition 4.2, Proposition 4.4 and Proposition 4.5. For any cochain DG algebra A and cocycle element z ∈ ker(∂i ), we write ⌈z⌉ as the coho- A mology class in H(A) represented by z. The following proposition gives all possible cohomology graded algebras for non-trivial DG free algebras with two degree one generators. Proposition 5.1. We have (1) H(B1) = k; (2) H(B2) = k; (3) H(B3) = k; (4) H(B4) = k; (5) H(B5) = k[⌈x2⌉]; (6) H(B6) = k[⌈x2⌉,⌈x1x2 + x2x1⌉]/(⌈x2⌉2); (7) H(B7) = k; (8) H(B8) = k[⌈x1 − x2⌉]; (9) H(B9) = k; (10) H(B10) = k; (11) H(B11) = k; (12) H(B(s, t)) = k, when st 6= 1 and s ∈ k×. Proof. (1)It is easy to see that H 0(B1) = k and H 1(B1) = 0 since ∂B1(x1) = x2 and ∂B1 (x2) = x2x1. One only need to prove that H i(B1) = 0, for any i ≥ 2. For 1, we may write it as a1x1 + a2x2 for some a1, a2 ∈ Bi−1 any cocycle element in Bi . We have 1 1 0 =∂B1 [a1x1 + a2x2] =∂B1 (a1)x1 + (−1)ia1x2 1 + ∂B1 (a2)x2 + (−1)ia2x2x1 =[∂B1 (a1) + (−1)ia1x1 + (−1)ia2x2]x1 + ∂B1(a2)x2. 20 X.-F. MAO, J.-F. XIE, Y.-N. YANG, AND ALMIRE. ABLA Hence ∂B1(a1) + (−1)ia1x1 + (−1)ia2x2 = 0 and ∂B1 (a2) = 0. Then ∂B1 [(−1)i−1a1] = a1x1 + a2x2. (2)Since ∂B2(x1) = x2 So H i(B1) = 0, for any i ≥ 2. Hence H(B1) = k. 1 and ∂B2(x2) = x1x2, one sees that H 0(B2) = k and H 1(B2) = 0. It suffices to show that H i(B2) = 0 for any i ≥ 2. For any cocycle element in Bi 2, we may write it as x1a1 + x2a2 for some a1, a2 ∈ Bi−1 . We have 2 0 =∂B2[x1a1 + x2a2] =x2 1a1 − x1∂B2(a1) + x1x2a2 − x2∂B2(a2) =x1[x1a1 − ∂B2(a1) + x2a2] − x2∂B2 (a2). Hence ∂B2 (a1) = x1a1 + x2a2 and ∂B2 (a2) = 0. So H i(B2) = 0, for any i ≥ 2. Hence H(B2) = k. 1, ∂B3(x2) = x1x2 + x2x1. Obviously, we have H 0(B3) = k and H 1(B3) = 0. For any cocycle element in Bi 3, we may write it as x1a1 + x2a2 for some a1, a2 ∈ Bi−1 (3)The differential of B3 is defined by ∂B3(x1) = x2 . Since 1 0 =∂B3[x1a1 + x2a2] =x2 1a1 − x1∂B3(a1) + (x1x2 + x2x1)a2 − x2∂B3(a2) =x1[x1a1 − ∂B3(a1) + x2a2] + x2(x1a2 − ∂B3 (a2)), (4)Since ∂B4(x1) = x2 we have ∂B3(a1) = x1a1 + x2a2 and ∂B3 (a2) = x1a2. So H i(B3) = 0, for any i ≥ 2. Hence H(B3) = k. 2, one sees that H 0(B2) = k and H 1(B2) = 0. We should prove that H i(B4) = 0, for any i ≥ 2. For any cocy- cle element in Bi . We have 4, we may write it as x1a1 + x2a2 for some a1, a2 ∈ Bi−1 1 and ∂B4 (x2) = x2 4 0 =∂B4 [x1a1 + x2a2] 1a1 − x1∂B4 (a1) + x2 =x2 =x1[x1a1 − ∂B4 (a1)] + x2[x2a2 − ∂B4 (a2)]. 2a2 − x2∂B4(a2) (5)The differential of B5 is defined by ∂B5(x1) = x2 Hence ∂B4(a1) = x1a1 and ∂B4(a2) = x2a2. Since ∂B4(a1 + a2) = x1a1 + x2a2, we have H i(B4) = 0 for any i ≥ 2. So H(B4) = k. 1, ∂B5(x2) = 0. Obviously, we have H 0(B5) = k and H 1(B5) = k⌈x2⌉. We want to show inductively that H j(B5) = k⌈xj 2⌉ for any j ≥ 1. Suppose that we have proved that H i(B5) = k⌈xi 2⌉. For any cocycle element in Bi+1 5. Since , we may write it as x1a1 + x2a2, a1, a2 ∈ Bi 5 0 =∂B5 [x1a1 + x2a2] =x2 1a1 − x1∂B5 (a1) − x2∂B5 (a2) =x1[x1a1 − ∂B5 (a1)] − x2∂B5 (a2), we have x1a1 = ∂B5 (a1) and ∂B5(a2) = 0. By the induction hypothesis, we have H i(B5) = k⌈xi 2⌉. Hence there exists some l ∈ k such that ⌈a2⌉ = l⌈xi 2⌉. So ⌈x1a1 + x2a2⌉ = ⌈∂B5(a1) + x2a2⌉ = ⌈x2a2⌉ = ⌈x2⌉ · ⌈a2⌉ = l⌈xi+1 ⌉. 2 2 Then H i+1(B5) = k⌈xi+1 (6)The differential of B6 is defined by ∂B6(x1) = x2 to see that H 0(B6) = k and H 1(B6) = k⌈x2⌉. Since B2 any cocycle element in B2 6 can be written as l1x2 ⌉. By the induction above, we get H(B5) = k[⌈x2⌉]. 2, ∂B6(x2) = 0. It is easy for one 1⊕kx1x2⊕kx2x1⊕kx2 6 = kx2 2, 2 for some 1 + l2x1x2 + l3x2x1 + l4x2 For any cocycle element in B2i+1 have 6 , we may write it as x1a1 + x2a2, a1, a2 ∈ B2i 6 . We H 2i+1(B6) = k⌈x2(x1x2 + x2x1)i⌉ and H 2i+2(B6) = k⌈(x1x2 + x2x1)i+1⌉. 0 =∂B6 (x1a1 + x2a2) =x2 2a1 − x1∂B6 (a1) − x2∂B6 (a2) =x2[x2a1 − ∂B6 (a2)] − x1∂B6 (a1). ISOMORPHISM PROBLEM AND HOMOLOGICAL PROPERTIES OF DG FREE ALGEBRAS21 l1, l2, l3, l4 ∈ k. We have 1 + l2x1x2 + l3x2x1 + l4x2 2) 0 = ∂B6(l1x2 = l1x2 = l1(x2 2x1 − l1x1x2 2x1 − x1x2 2 + l2x3 2 − l3x3 2 2) + (l2 − l3)x3 2. We have l1 = 0 and l2 = l3. Hence ker(∂2 2. By the definition of ∂B6, one sees that im(∂1 2. So H 2(B6) = k⌈x1x2 + x2x1⌉. B6) = kx2 Assume that we have proved H 2i−1(B6) = k⌈x2(x1x2 + x2x1)i−1⌉ and H 2i(B6) = k⌈(x1x2 + x2x1)i⌉, i ≥ 1. We should show that B6) = k(x1x2 + x2x1) ⊕ kx2 Then ∂B6(a2) = x2a1 and ∂B6 (a1) = 0. Since Z 2i(B6) ∼= H 2i(B6) ⊕ B2i(B6), we may let a1 = l(x1x2 + x2x1)i + ∂B6(χ) for some l ∈ k and χ ∈ B2i−1 . We have ∂B6 (a2) = lx2(x1x2 + x2x1)i + x2∂B6 (χ). This implies that l = 0 and a2 = −x2χ + t(x1x2 + x2x1)i + ∂B6(λ), for some t ∈ k and λ ∈ B2i−1 . Then 6 6 ⌈x1a1 + x2a2⌉ = ⌈x1∂B6(χ) − x2 2χ + tx2(x1x2 + x2x1)i + x2∂B6(λ)⌉ = ⌈∂B6 (−x1χ − x2λ) + tx2(x1x2 + x2x1)i⌉ = t⌈x2(x1x2 + x2x1)i⌉. So H 2i+1(B6) = k⌈x2(x1x2 + x2x1)i⌉. Now, let x1b1 + x2b2 be an arbitrary cocycle element in B2i+2 6 . Then we have 0 =∂B6 (x1b1 + x2b2) =x2 2b1 − x1∂B6 (b1) − x2∂B6 (b2) =x2[x2b1 − ∂B6 (b2)] − x1∂B6 (b1). So ∂B6(b2) = x2b1 and ∂B6(b1) = 0. Since Z 2i+1(B6) ∼= H 2i+1(B6) ⊕ B2i+1(B6), we may let b1 = rx2(x1x2 + x2x1)i + ∂B6(ω) for some r ∈ k and ω ∈ B2i 6 . Then ∂B6(b2) = rx2 2(x1x2 + x2x1)i + x2∂B6(ω). So b2 = rx1(x1x2 + x2x1)i − x2ω + sx2(x1x2 + x2x1)i + ∂B6(ϕ) for some s ∈ k and ϕ ∈ B2i 6 Therefore, ⌈x1b1 + x2b2⌉ = ⌈r(x1x2 + x2x1)i+1 + ∂B6[−x1ω + sx1(x1x2 + x2x1)i − x2ϕ]⌉ = r⌈(x1x2 + x2x1)i+1⌉. Thus H 2i+2(B6) = k⌈(x1x2 + x2x1)i+1⌉. By the induction above, we obtain that H 2n−1(B6) = k⌈x2(x1x2 + x2x1)n−1⌉ and H 2n(B6) = k⌈(x1x2 + x2x1)n⌉,∀n ≥ 1. Since x2(x1x2 + x2x1) − (x1x2 + x2x1)x2 = x2 2x1 − x1x2 2 = ∂B6(x2 1), we have ⌈x2⌉ · ⌈x1x2 + x2x1⌉ = ⌈x1x2 + x2x1⌉ · ⌈x2⌉ in H(B6). Hence H(B6) = k[⌈x2⌉,⌈x1x2 + x2x1⌉]/(⌈x2⌉2). 22 X.-F. MAO, J.-F. XIE, Y.-N. YANG, AND ALMIRE. ABLA (7)Since ∂B7(x1) = x1x2 +x2x1 +x2 2, one sees that H 0(B7) = k and H 1(B7) = 0. It suffices to show that H i(B7) = 0 for any i ≥ 2. For any cocycle element in Bi 2 and ∂B7 (x2) = x2 . We have 7, we may write it as x1a1 + x2a2 for some a1, a2 ∈ Bi−1 0 =∂B7 [x1a1 + x2a2] =(x1x2 + x2x1 + x2 2a2 − x2∂B7(a2) =x1[x2a1 − ∂B7 (a1)] + x2[x1a1 + x2a1 + x2a2 − ∂B7 (a2)]. 2)a1 − x1∂B7 (a1) + x2 7 Hence ∂B7(a1) = x2a1 and ∂B2 (a2) = x1a1 + x2a1 + x2a2. Then ∂B2(a2 − a1) = x1a1 + x2a2. (8)The differential of B8 is defined by ∂B8(x1) = x2 So H i(B2) = 0, for any i ≥ 2. Hence H(B2) = k. 2. It is easy for one to see that H 0(B8) = k and H 1(B8) = k⌈x2 − x1⌉. We want to show inductively that H j(B8) = k⌈(x2 − x1)j⌉ for any j ≥ 1. Suppose that we have proved that H i(B8) = k⌈(x2 − x1)i⌉. For any cocycle element in Bi+1 , we may write it as x1a1 + x2a2, a1, a2 ∈ Bi 2, ∂B8(x2) = x2 8. Since 8 0 =∂B8[x1a1 + x2a2] =x2 = − x1∂B8(a1) + x2[x2a1 + x2a2 − ∂B8(a2)], 2a1 − x1∂B8(a1) + x2 2a2 − x2∂B8 (a2) we have x2a1 + x2a2 = ∂B8(a2) and ∂B8(a1) = 0. By the induction hypothesis, we have H i(B8) = k⌈(x2 − x1)i⌉. So ⌈a1⌉ = l⌈(x2 − x1)i⌉, for some l ∈ k. So ⌈x1a1 + x2a2⌉ = ⌈x2a1 + x2a2 + x1a1 − x2a1⌉ = ⌈∂B8 (a2) + (x1 − x2)a1⌉ = ⌈(x1 − x2)a1⌉ = l⌈(x2 − x1)i+1⌉ and H i+1(B8) = k⌈(x2 − x1)i+1⌉. By the induction above, we have H(B8) = k[⌈x2 − x1⌉]. (9)We have ∂B9(x1) = x2 and H 1(B9) = 0. Any cocycle element in Bi some a1, a2 ∈ Bi−1 , i ≥ 2. We have 9 1+x1x2+x2x1 and ∂B9 (x2) = x2 2. Obviously, H 0(B9) = k 9 can be written by x1a1 + x2a2 for 0 =∂B9 [x1a1 + x2a2] 1 + x1x2 + x2x1)a1 − x1∂B9 (a1) + x2 =(x2 2a2 − x2∂B9(a2) =x1[x1a1 + x2a1 − ∂B9 (a1)] + x2[x1a1 + x2a2 − ∂B9 (a2)]. Then ∂B9(a1) = x1a1 + x2a1 and ∂B9 (a2) = x1a1 + x2a2, which implies that H i(B9) = 0 and hence H(B9) = k. (10)The differential of B10 is defined by ∂B10 (x1) = x2 1 − x2 2, ∂B10(x2) = x1x2 + x2x1 + x2 2. 1 4 Obviously, H 0(B10) = k and H 1(B10) = 0. For any x1a1 + x2a2 ∈ Z i(B10), i ≥ 2, we have 0 =∂B10[x1a1 + x2a2] =(x2 1 4 1 − 2)a1 − x1∂B10 (a1) + [x1x2 + x2x1 + x2 x2 2]a2 − x2∂B10(a2) =x1[x1a1 + x2a2 − ∂B10 (a1)] + x2[− 1 4 x2a1 + x2a2 − ∂B10 (a2)]. ISOMORPHISM PROBLEM AND HOMOLOGICAL PROPERTIES OF DG FREE ALGEBRAS23 Then ∂B10(a1) = x1a1 + x2a2 and ∂B10 (a2) = − 1 2 and ∂B11(x2) = x1x2 + x2x1. It is easy to check that H 0(B11) = k and H 1(B11) = 0. For any x1a1 + x2a2 ∈ Z i(B11), i ≥ 2, we have 4 x2a1 + x2a2. So H i(B10) = 0. (11)We have ∂B11(x1) = x2 1 + x2 0 =∂B11[x1a1 + x2a2] 1 + x2 =(x2 2)a1 − x1∂B11 (a1) + [x1x2 + x2x1]a2 − x2∂B11(a2) =x1[x1a1 + x2a2 − ∂B11 (a1)] + x2[x2a1 + x1a2 − ∂B11 (a2)]. Thus ∂B11 (a1) = x1a1 + x2a2 and ∂B11(a2) = x2a1 + x1a2. Then H i(B11) = 0 and hence H(B11) = k. (12)The differential of Bst is defined by (∂Bst (x1) = (1 + s − st)x2 ∂Bst (x2) = sx2 1 + x1x2 + x2x1 + tx2 2, 1 + x1x2 + x2x1 + 1 s x2 2 where s ∈ k×, t ∈ k and st 6= 1. Obviously, H 0(Bst) = k. Let l1x1 + l2x2 be a cocycle element in B1 st. Then 0 =∂Bst(l1x1 + l2x2) =l1[(1 + s − st)x2 =[l1(1 + s − st) + l2s]x2 1 + x1x2 + x2x1 + 1 s x2 2] + l2[sx2 1 + x1x2 + x2x1 + tx2 2] 1 + (l1 + l2)(x1x2 + x2x1) + ( l1 s + l2t)x2 2. We have l1(1 + s − st) + l2s = 0 l1 + l2 = 0 l1 s + l2t = 0.   Then l1 = l2 = 0 and hence H 1(Bst) = 0. For any i ≥ 2, any element in Z i(Bst) can be written by x1a1 + x2a2, for some a1, a2 ∈ Bi−1 . And we have st 0 = ∂Bst(x1a1 + x2a2) = [(1 + s − st)x2 1 + x1x2 + x2x1 + 1 s x2 2]a1 − x1∂Bst(a1) + [sx2 1 + x1x2 + x2x1 + tx2 2]a2 − x2∂Bst(a2) = x1[(1 + s − st)x1a1 + x2a1 − ∂Bst (a1) + sx1a2 + x2a2] Hence + x2[x1a1 + x2a1 + x1a2 + tx2a2 − ∂Bst (a2)]. ((1 + s − st)x1a1 + x2a1 − ∂Bst (a1) + sx1a2 + x2a2 = 0 s x2a1 + x1a2 + tx2a2 − ∂Bst(a2) = 0 x1a1 + 1 (1) (2) . (1) − (2) × s implies that (1 − st)x1a1 + (1 − st)x2a2 − ∂Bst(a1 − sa2) = 0. Then 1 s a1 1 − st − Thus H i(Bst) = 0 and H(Bst) = k. ∂Bst[ sa2 1 − st ] = x1a1 + x2a2. (cid:3) 6. Homological properties of DG free algebras For any k-vector space V , we write V ∗ = Homk(V, k). Let {eii ∈ I} be a basis of a finite dimensional k-vector space V . We denote the dual basis of V by {e∗ii ∈ I}, i.e., {e∗i i ∈ I} is a basis of V ∗ such that e∗i (ej) = δi,j. For any graded vector space W and j ∈ Z, the j-th suspension ΣjW of W is a graded vector space defined by (ΣjW )i = W i+j. 24 X.-F. MAO, J.-F. XIE, Y.-N. YANG, AND ALMIRE. ABLA In this section, we will study homological properties of DG free algebras. Now, let us review some fundamental homological properties for DG algebras. Definition 6.1. Let A be a connected cochain DG algebra. Gam]); (1) If dimk H(RHomA(k,A)) = 1, then A is called Gorenstein (cf. [FHT1, FM, (2) If Ak, or equivalently AeA, has a minimal semi-free resolution with a semi- (3) If Ak, or equivalently the DG Ae-module A is compact, then A is called (4) If A is homologically smooth and basis concentrated in degree 0, then A is called Koszul (cf. [HW]); homologically smooth (cf. [MW3, Corollary 2.7]); RHomAe (A,Ae) ∼= Σ−nA in the derived category D((Ae)op) of right DG Ae-modules, then A is called an n-Calabi-Yau DG algebra (cf. [Gin, VdB]). A natural question is whether DG free algebras have the properties listed in Definition 6.1. It is reasonable for us to consider the cases from easy to difficult. We have the following proposition for trivial DG free algebras. Proposition 6.2. Let A be a connected cochain DG algebra such that H(A) = kh⌈y1⌉,··· ,⌈yn⌉i, for some degree 1 cocycle elements y1,··· , yn in A. Then A is not a Gorenstein DG algebra but a Koszul and homologically smooth DG algebra. Proof. The graded module H(A)k has the following minimal graded free resolution: n 0 → H(A) ⊗ ( keyi) d1→ H(A) ε→ k → 0, Mi=1 where µ and ∂1 are defined by µ(⌈a⌉ ⊗ ⌈b⌉) = ⌈ab⌉, ∀⌈a⌉ ∈ H(A),⌈b⌉ ∈ H(A)op; d1(eyi) = yi, i = 1, 2,··· , n. Applying the constructing procedure of Eilenberg- Moore resolution, we can construct a minimal semi-free resolution F of the DG A-module k. We have F # = A# ⊕ [A# ⊗ ( kΣeyi)] n Mi=1 and ∂F is defined by ∂F (Σeyi) = yi, i = 1, 2,··· , n. Hence A is a Koszul and homologically smooth DG algebra. The DG Aop-module HomA(F,A) is a minimal semi-free DG module whose underlying graded module is {k1∗ ⊕ [ k(Σeyi)∗]} ⊗ A#. n Mi=1 Hence HomA(F,A) is concentrated in degrees ≥ 0. On the other hand, we have Pi=1−(Σeyi)∗yi, since the ∂Hom[(Σeyi)∗] = 0,∀i ∈ {1, 2,··· , n} and ∂Hom(1∗) = differential ∂Hom of HomA(F,A) is defined by n ∂Hom(f ) = ∂A ◦ f − (−1)jf ◦ ∂F ,∀f ∈ [HomA(F,A)]j . k(Σeyi)∗ and hence A is not Gorenstein. n So H 0(HomA(F,A)) = (cid:3) Remark 6.3. By Theorem 6.2, any trivial DG free algebra is not Gorenstein but a Koszul and homologically smooth DG algebra. The following proposition indicates that is not Calabi-Yau. Li=1 ISOMORPHISM PROBLEM AND HOMOLOGICAL PROPERTIES OF DG FREE ALGEBRAS25 Proposition 6.4. Any Calabi-Yau connected cochain DG algebra A is a Gorenstein DG algebra. Proof. Since A is a Calabi-Yau connected cochain DG algebra, A is homologically smooth and Ω = RHomAe (A,Ae) ∼= ΣiA in D(Ae), for some i ∈ Z. Let F be the minimal semi-free resolution of Ak. Then F admits a finite semi-basis. Let I be a semi-injective resolution of the DG A- module A. One sees that Homk(F, I) is a homotopically injective DG Ae-module. By [MW3, Lemma 2.4], we have HomA(F, I) ∼= HomAe (A, Homk(F, I)). Since A ∈ Dc(Ae), it admits a minimal semi-free resolution G, which has a finite semi-basis. We have (a) [HomA(F, I)]∗ ∼= [HomAe (A, Homk(F, I))]∗ ≃ [HomAe (G, Homk(F, I))]∗ ∼= [HomAe (G,Ae) ⊗Ae Homk(F, I)]∗ ≃ [HomAe (G,Ae) ⊗Ae (F ∗ ⊗ I)]∗ ≃ [F ∗ ⊗A HomAe (G,Ae) ⊗A I]∗ ∼= HomA(HomAe(G,Ae) ⊗A I, (F ∗)∗) ≃ HomA(HomAe(G,Ae) ⊗A A, (F ∗)∗) ∼= HomA(HomAe(G,Ae), (F ∗)∗) ≃ HomA(ΣiA, (F ∗)∗) ∼= Σ−i(F ∗)∗ ≃ Σ−i k, where (a) is obtained by [Shk, A1]. So H(RHomA(k,A)) = H(HomA(F,A)) ∼= H(HomA(F, I)) ∼= Σi and A is Gorenstein. k (cid:3) It remains to consider the homological properties of non-trivial DG free algebras. We have classified all the isomorphism classes of DG free algebras with two degree one generators. Hence, we only need to check case by case when n = 2. In [MH], it is proved that a connected cochain DG algebra A is a Kozul Calabi-Yau DG algebra if H(A) belongs to either of the following cases: (a)H(A) ∼= k; (b)H(A) = k[⌈z⌉], z ∈ ker(∂1 A). So Proposition 5.1 indicates that B1,B2,···B11 and B(s, t) with s ∈ k× and st 6= 1 are all Koszul Calabi-Yau DG algebras except B6. For B6, we have the following proposition. Proposition 6.5. The connected cochain DG algebra B6 is a Koszul Calabi-Yau DG algebra. Proof. By definition, B# 6 = khx1, x2i and the differential ∂B6 is defined by ∂B6(x1) = x2 2, ∂B6(x2) = 0. According to the constructing procedure of the minimal semi-free resolution in [MW1, Proposition 2.4], we get a minimal semi-free resolution f : F ≃→ B6 k, where F is a semi-free DG B6-module with ∂F (Σex2) = x2, ∂F (Σez) = x1 + x2Σex2; F # = B# 6 ⊕ B# 6 Σex2 ⊕ B# 6 Σez, 26 X.-F. MAO, J.-F. XIE, Y.-N. YANG, AND ALMIRE. ABLA We should prove that f is a quasi-isomorphism. It suffices to show H(F ) = k. and f is defined by fB6 = ε, f (Σex2) = 0 and f (Σez) = 0. For any graded cocycle element azΣez + ax2Σex2 + a ∈ Z 2k(F ), we have 0 = ∂F (azΣez + ax2Σex2 + a) = ∂B6(az)Σez + az(x1 + x2Σex2) + ∂B6(ax2 )Σex2 + ax2x2 + ∂B6 (a) = ∂B6(az)Σez + [azx2 + ∂B6 (ax2)]Σex2 + azx1 + ax2x2 + ∂B6 (a). This implies that   and ∂B6(az) = 0 azx2 + ∂B6 (ax2) = 0 azx1 + ax2x2 + ∂B6(a) = 0. Since Z 2k(B6) = H 2k(B6) ⊕ B2k(B6) we have az = ∂B6 (c) + t(x1x2 + x2x1)k for some c ∈ B2k−1 6 , t ∈ k. Then H(B6) = k[⌈x2⌉,⌈x1x2 + x2x1⌉]/(⌈x2⌉2), [∂B6(c) + t(x1x2 + x2x1)k]x2 + ∂B6 (ax2) = 0. Hence t = 0 and ∂B6(c)x2 + ∂B6(ax2 ) = 0. Then for some µ ∈ B2k−1 6 ax2 = −cx2 + ∂B6(µ) + s(x1x2 + x2x1)k and s ∈ k. We have ∂B6(c)x1 + [−cx2 + ∂B6(µ) + s(x1x2 + x2x1)k]x2 + ∂B6(a) = 0, which implies that ∂B6 (c) = 0, s = 0, and ∂B6(a) = cx2 ax2 = −cx2 + ∂B6 (µ) and 2 − ∂B6(µ)x2. Then az = 0, for some λ ∈ B2k−1 let c = ∂B6 (χ) + ω(x1x2 + x2x1)k−1x2, for some χ ∈ B2k−2 a = −cx1 − µx2 + ∂B6(λ) + τ (x1x2 + x2x1)k , τ ∈ k. Since c ∈ Z 2k−1(B6) ∼= H 2k−1(B6)⊕B2k−1(B6), we may and ω ∈ k. Therefore, 6 6 azΣez + ax2Σex2 + a =[−cx2 + ∂B6(µ)]Σex2 − cx1 − µx2 + ∂B6(λ) + τ (x1x2 + x2x1)k ={−[∂B6(χ) + ω(x1x2 + x2x1)k−1x2]x2 + ∂B6(µ)}Σex2 − [∂B6(χ) + ω(x1x2 + x2x1)k−1x2]x1 − µx2 + ∂B6(λ) + τ (x1x2 + x2x1)k =∂F [(ω − τ )(x1x2 + x2x1)k−1x2Σez] + ∂F{[−χx2 + µ − τ (x1x2 + x2x1)k−1x1]Σex2 − χx1}. Hence H 2k(F ) = 0, for any k ∈ N. k ∈ N. Let azΣez + ax2Σex2 + a ∈ Z 2k−1(F ), we have It remains to show H 2k−1(F ) = 0, for any 0 = ∂F (azΣez + ax2Σex2 + a) = ∂B6(az)Σez − az(x1 + x2Σex2) + ∂B6(ax2 )Σex2 − ax2x2 + ∂B6 (a) = ∂B6(az)Σez + [−azx2 + ∂B6(ax2 )]Σex2 − azx1 − ax2x2 + ∂B6(a). This implies that   and ∂B6(az) = 0 −azx2 + ∂B6(ax2 ) = 0 −azx1 − ax2x2 + ∂B6 (a) = 0. Since Z 2k−1(B6) = H 2k−1(B6) ⊕ B2k−1(B6) H(B6) = k[⌈x2⌉,⌈x1x2 + x2x1⌉]/(⌈x2⌉2), ISOMORPHISM PROBLEM AND HOMOLOGICAL PROPERTIES OF DG FREE ALGEBRAS27 we have az = ∂B6 (c) + t(x1x2 + x2x1)k−1x2 for some c ∈ B2k−2 6 , t ∈ k. Then −[∂B6(c) + t(x1x2 + x2x1)k−1x2]x2 + ∂B6 (ax2) = 0. Then ax2 = cx2 + t(x1x2 + x2x1)k−1x1 + ∂B6(µ) + s(x1x2 + x2x1)k−1x2 for some µ ∈ B2k−2 6 and s ∈ k. We have 0 = − [cx2 + t(x1x2 + x2x1)k−1x1 + ∂B6 (µ) + s(x1x2 + x2x1)k−1x2]x2 − [∂B6 (c) + t(x1x2 + x2x1)k−1x2]x1 + ∂B6 (a), which implies that t = 0, az = ∂B6(c), ax2 = cx2 + ∂B6 (µ) + s(x1x2 + x2x1)k−1x2 and a = cx1 + µx2 + s(x1x2 + x2x1)k−1x1 + ∂B6(λ) + τ (x1x2 + x2x1)k−1x2, for some λ ∈ B2k−2 , τ ∈ k. Hence 6 azΣez + ax2Σex2 + a =∂B6(c)Σez + [cx2 + ∂B6 (µ) + s(x1x2 + x2x1)k−1x2]Σex2 + cx1 + µx2 + s(x1x2 + x2x1)k−1x1 + ∂B6 (λ) + τ (x1x2 + x2x1)k−1x2 =∂F{[c + s(x1x2 + x2x1)k−1]Σez + [µ + τ (x1x2 + x2x1)k−1]Σex2 + λ}. Hence H 2k−1(F ) = 0, for any k ∈ N. Therefore, f is a quasi-isomorphism. homologically smooth DG algebra. By the minimality of F , we have Since F has a semi-basis {1, Σex2, Σez} concentrated in degree 0, B6 is a Koszul H(HomB6(F, k)) = HomB6(F, k) = k · 1∗ ⊕ k · (Σex2)∗ ⊕ k · (Σez)∗. So the Ext-algebra E = H(HomB6(F, F )) is concentrated in degree 0. On the other hand, HomB6 (F, F )# ∼= (k · 1∗ ⊕ k · (Σex2)∗ ⊕ k · (Σez)∗) ⊗k F # is concentrated in degree ≥ 0. This implies that E = Z 0(HomB6 (F, F )). Since F # is a free graded B# 6 -module with a basis {1, Σey, Σez} concentrated in degree 0, the elements in HomB6(F, F )0 are in one to one correspondence with the matrices in M3(k). Indeed, any f ∈ HomB6(Fk, Fk)0 is uniquely determined by a matrix Af = (aij )3×3 ∈ M3(k) with   Af ·   = Af ·  0   =  We see f ∈ Z 0(HomB6 (F, F ) if and only if ∂F ◦ f = f ◦ ∂F , if and only if   . 0   · Af , f (Σez)  0 0 x2 0 x1 x2 0 0 x2 0 x1 x2 Σex2 Σez 0 0 f (1) f (Σex2) 0 0 1 which is also equivalent to by a direct computation. Let   E = {  a12 = a13 = a23 = 0 a11 = a22 = a33 a21 = a32 a 0 0 b a 0 c b a   a, b, c,∈ k} 28 X.-F. MAO, J.-F. XIE, Y.-N. YANG, AND ALMIRE. ABLA be the subalgebra of the matrix algebra. Then one sees E ∼= E. Set 0 0  0 0 0 0  . e1 =  1 0 0 1 0 0 0 0 1 Then {e1, e2, e3} is a k-linear bases of the k-algebra E. The multiplication on E is defined by the following relations 0 0 0 0 0 0 1 0 0 1 1   ,e2 = 0   , e3 =   (e1 · ei = ei · e1 = ei, i = 1, 2, 3 e2 2 = e3, e2 · e3 = e3 · e2 = 0 . So E is a local commutative k-algebra isomorphic to k[X]/(X 3). Hence E ∼= k[X]/(X 3) is a symmetric Frobenius algebra concentrated in degree 0. This implies B6(kB6 , B6 k) ∼= E∗ is a symmetric coalgebra. By [HM1, Theorem 4.2], B6 that Tor0 is a Koszul Calabi-Yau DG algebra. (cid:3) By Proposition 4.1, Proposition 4.2, Proposition 4.4, Proposition 4.5, Proposition 5.1 and Proposition 6.5, we reach the following conclusion. Theorem 6.6. Let A be a DG free algebra with 2 degree one generators. Then A is a Koszul Calabi-Yau DG algebra if and only if ∂A 6= 0. Acknowledgments The first author is supported by NSFC (Grant No.11001056), the China Post- doctoral Science Foundation (Grant Nos. 20090450066 and 201003244), the Key Disciplines of Shanghai Municipality (Grant No.S30104) and the Innovation Pro- gram of Shanghai Municipal Education Commission (Grant No.12YZ031). References [AFH] L. L. Avramov, H.-B. Foxby and S. Halperin, Differential graded homologcial algebra, In preparation, 2004. Version from 02.07.2004. [AT] W. Andrezejewski and A. Tralle, Cohomology of some graded differential algebras, Fund. Math. 145 (1994), 181 -- 203. [Bez] R. Bezrukavnikov, Koszul DG-algebras arising from configuration spaces, Geom. Funct. Anal, 4 (1994), 119 -- 135. [FHT1] Y. F´elix, S. Halperin, and J. C. Thomas, Gorenstein spaces, Adv. Math. 71 (1988), 92 -- 112. [FHT2] Y. F´elix, S. Halperin and J. C. Thomas, "Rational Homotopy Theory", Grad. Texts in Math. 205, Springer, Berlin, 2000. [FIJ] A. Frankild and P. Jørgensen, Dualizing Differential Graded modules and Gorenstein Dif- ferential Graded Algebras, J. London Math. Soc. (2) 68 (2003), 288 -- 306. [FJ1] A. Frankild and P. Jørgensen, Gorenstein Differential Graded Algebras, Israel J. Math. 135 (2003), 327-353. [FJ2] A. Frankild and P. Jørgensen Homological properties of cochain differential graded algebras, J. Algebra, 320 (2008),3311 -- 3326. [FM] Y. F´elix and A. Murillo, Gorenstein graded algebras and the evaluation map, Canad. Math. Bull. Vol. 41 (1998), 28 -- 32. [Gam] H. Gammelin, Gorenstein space with nonzero evaluation map, Trans. Amer. Math Soc. 351 (1999), 3433 -- 3440. [Gin] V. Ginzberg, Calabi-Yau algebra, arxiv: math. AG/0612.139 v3. [HM1] J.-W. He and X.-F. Mao, Connected cochain DG algebras of Calabi-Yau dimension 0, Proc. Amer. Math. Soc. 145 (2017), 937 -- 953. [HW] J.-W. He and Q.-S. Wu, Koszul differential graded algebras and BGG correspondence, J. Algebra 320 (2008), 2934 -- 2962. [Jorg1] P. Jørgensen, Auslander-Reiten theory over topological spaces. Comment. Math. Helv., 79 (2004), 160-182. [Jorg2] P. Jørgensen, Duality for cochain DG algebras. Sci. China Math.,56 (2013), 79 -- 89. [Kal] D. Kaledin, Some remarks on formality in families, Moscow Math. J., 7 (2007), 643 -- 652. ISOMORPHISM PROBLEM AND HOMOLOGICAL PROPERTIES OF DG FREE ALGEBRAS29 [Lunt] V. A. Lunts, Formality of DG algebras (after Kaledin), J. Algebra, 323 (2010), 878 -- 898. [Mao] X.-F. Mao, DG algebra structures on AS-regular algebras of dimension 2, Sci. China Math., 54, (2011) 2235 -- 2248. [MGYC] X.-F. Mao, X.-D. Gao, Y.-N. Yang and J.-H. Chen, DG polynomial algebras and their homological properties, Sci. China Math. http://engine.scichina.com/doi/10.1007/s11425- 017-9182-1 [MH] X.-F. Mao and J.-W. He, A special class of Koszul Calabi-Yau DG algebras, Acta Math. Sinica, Chinese series, 60 (2017), 475 -- 504 [MHLX] X.-F. Mao, J.-W. He, M. Liu and J.-F. Xie, Calabi-Yau properties of non-trivial Noe- therian DG down-up algebras, J. Algebra Appl. 17, no.5 (2018), 1850090-45 [MW1] X.-F. Mao and Q.-S. Wu, Homological invariants for connected DG algebra, Comm. Al- gebra, 36 (2008), 3050 -- 3072. [MW2] X.-F. Mao and Q.-S. Wu, Compact DG modules and Gorenstein DG algebra, Sci. China Ser. A, 52 (2009), 711 -- 740. [MW3] X.-F. Mao and Q.-S. Wu, Cone length for DG modules and global dimension of DG algebras, Comm. Algebra, 39 (2011), 1536-1562. [Sch] K. Schmidt, Families of Auslander-Reiten theory for simply connected differential graded algebras, Math. Z. 264 (2010), 43 -- 62. [Shk] D. Shklyarov, On Serre duality for compact homologically smooth DG algebras, arxiv: math. RA/0702590 v1. [VdB] M. Van den Bergh, Calabi-Yau algebras and superpotentials, Sel. Math. New Ser. 21 (2015), 555 -- 603. Department of Mathematics, Shanghai University, Shanghai 200444, China E-mail address: [email protected] Department of Mathematics, Shanghai University, Shanghai 200444, China E-mail address: [email protected] Department of Mathematics, Shanghai University, Shanghai 200444, China E-mail address: [email protected] Department of Mathematics, College of Mathematics and Statistics, Kashgar Uni- versity, Kashgar, Xinjiang, 844006, China E-mail address: [email protected]
1208.5579	1	1208	2012-08-28T07:57:30	Minimal quasivarieties of semilattices over commutative groups	[ "math.RA" ]	We continue some recent investigations of W. Dziobiak, J. Jezek, and M. Maroti. Let G=(G,\cdot) be a commutative group. A semilattice over G is a semilattice enriched with G as a set of unary operations acting as semilattice automorphisms. We prove that the minimal quasivarieties of semilattices over a finite abelian group G are in one-to-one correspondence with the subgroups of G. If G is not finite, then we reduce the description of minimal quasivarieties to that of those minimal quasivarieties in which not every algebra has a zero element.	math.RA	math	Minimal quasivarieties of semilattices over commutative groups Ildik´o V. Nagy Abstract. We continue some recent investigations of W. Dziobiak, J. Jezek, and M. Mar´oti. Let G = hG, ·i be a commutative group. A semilattice over G is a semilattice enriched with G as a set of unary operations acting as semilattice automorphisms. We prove that the minimal quasivarieties of semilattices over a finite abelian group G are in one-to-one correspondence with the subgroups of G. If G is not finite, then we reduce the description of minimal quasivarieties to that of those minimal quasivarieties in which not every algebra has a zero element. 1. Introduction Let τ be a homomorphism from an abelian group G = hG; ·, idi to the automorphism group of a semilattice hA; ∧i. Then the elements g ∈ G become unary operations g(x) = τ (g)(x) on A, and the algebra A = hA; ∧, Gi obtained this way is a G-semilattice, also called a semilattice over G. Semilattices over abelian groups or their term equivalent variants were investigated in several papers, to be mentioned soon. In particular, W. Dziobiak, J. Jezek, and M. Mar´oti [2] described the minimal quasivarieties of semilattices over the infinite cyclic group. Our goal is to extend their result to other abelian groups. Clearly, each minimal quasivariety R of semilattices over an abelian group G is determined by its 1-generated free algebra FR(1), provided FR(1) is nontriv- ial. Our first result, Theorem 4.1, describes these minimal R by characterizing the free algebras FR(1). If G is a finite abelian group, then Theorem 6.1 gives a much more explicit description by establishing a bijective correspondence between the minimal quasivarieties of G-semilattices and the subgroups of G. For infinite abelian groups, only a less explicit description of the minimal qua- sivarieties R is given in Theorem 7.3 since Theorem 4.1 in itself is insufficient for a complete understanding of those FR(1) that have no zero element. 1.1. Outline. Section 2 gives the basic concepts and notation, including some earlier results. Section 3 is devoted to easy statements on G-semilattices. The first result is stated and proved in Section 4. Section 5 gives two constructions that yield minimal quasivarieties. Minimal quasivarieties of semilattices over finite abelian groups are completed described in Section 6, while Section 7 is 2010 Mathematics Subject Classification: Primary 06A12, secondary 08A35. Key words and phrases: Semilattice over a group, group extension of semilattices, mini- mal quasivariety. 2 I. V. Nagy Algebra univers. devoted to the infinite case. Finally, Section 8 points out why the infinite case is much subtler than the finite one. 2. Preliminaries 2.1. Basic concepts and notation. For a second look at the key concept, an algebra A = hA; ∧, Gi is called a G-semilattice, or a semilattice over G, if G = hG; ·, idi is a commutative group, the elements of G are unary operations acting on the set A, and the following identities hold: (i) ∧ is an associative, idempotent, and commutative operation; (ii) id(x) ≈ x; (iii) f (g(x)) ≈ (f · g)(x) for every f, g ∈ G; (iv) g(x) ∧ g(y) ≈ g(x ∧ y) for every g ∈ G. This definition is due to M. Mar´oti [5]. Axioms (i) -- (iv) imply that, for every f ∈ G, the map x 7→ f (x) is an automorphism of the semilattice reduct hA; ∧i. A G-semilattice A is trivial if it is a singleton. If g(x) = x holds for all x ∈ A and g ∈ G, then A = hA; ∧, Gi is called G-trivial. Following J. Jezek [4], an algebra hA; ∧, g, g−1i is a semilattice with an automorphism if hA; ∧i is a semilattice, and the unary operations g and g−1 are reciprocal automorphisms of hA; ∧i. If G happens to be a cyclic group generated by g, then the G-semilattice A = hA; ∧, Gi is term equivalent to the semilattice hA; ∧, g, g−1i with an automorphism, that is, these two algebras have the same term functions. For n ∈ N ∪ {∞} = {1, 2, . . . , ∞}, the n- element cyclic group is denoted by grCn; in this context, g always stands for a generating element of grCn. Notice that a grCn-semilattice hA; ∧, grC ni is uniquely determined by (but, in lack of g−1, not necessarily term equivalent to) its reduct hA; ∧, gi; we often rely on this fact implicitly. A quasi-identity (also called Horn formula) is a universally quantified sen- tence of the form (p1 ≈ q1 & · · · & pn ≈ qn) ⇒ p ≈ q, where n ∈ N0 = {0, 1, 2, . . .} and p1, q1, . . . , pn, qn, p, q are terms. Quasivarieties are classes of (similar) algebras defined by quasi-identities. The least quasivariety and the least variety containing a given algebra A are denoted by Q(A) and V(A), respectively. A quasivariety is trivial if it consists of trivial algebras. A non- trivial quasivariety is minimal if it has exactly one proper subquasivariety, the trivial one. For concepts and notation not defined in the paper, the reader is referred to S. Burris and H. P. Sankappanavar [1]. 2.2. Earlier results motivating the present investigations. The sys- tematic study of semilattices with an automorphism started in J. Jezek [4], where the simple ones and the subdirectly irreducible ones were described. The simple semilattices with two commuting automorphisms, which can also be considered (grC∞ × grC∞)-semilattices (up to term equivalence), were de- scribed in J. Jezek [3]. Generalizing this result, M. Mar´oti [5] characterized the Vol. 00, XX Minimal quasivarieties of semilattices over commutative groups 3 ... ... ... . . . Figure 1. C1 ... ... . . . simple G-semilattices for every abelian group G. The minimal quasivarieties of grC∞-semilattices were described by W. Dziobiak, J. Jezek, and M. Mar´oti [2]. Their result, to be detailed soon, is equivalent to the description of minimal quasivarieties of semilattices with an automorphism. To recall the result of [2] in an economic way, we define two concepts. In general, the opposite of a group G = hG, ·i is G∗ = hG∗, ∗i, where G∗ = G and x ∗ y := y · x. Note that G∗ = G in our case since G is assumed to be commutative. For a G-semilattice A = hA; ∧, Gi, the opposite of A is A∗ = hA; ∧, G∗i, where g∗(x) = g−1(x) for g ∈ G∗ = G and x ∈ A. Then A∗ is a G∗-semilattice (even without assuming the commutativity of G), and it can be different from A (even when G is commutative). Let n ∈ N, and let A = hA; ∧, grC∞i be a grC∞-semilattice. Remember that grC∞ is generated by g. We define a new grC∞-semilattice n ×tw A , the n-fold twisted multiple of A, as follows. n ×tw A =(cid:10){o} ∪ {ha, ii : a ∈ A, 0 ≤ i < n}; ∧, grC ∞(cid:11), where ha, ii ∧ hb, ji =(ha ∧ b, ii g(ha, ii) =(ha, i + 1i if i < n − 1, if i = n − 1 g(o) = o. hg(a), 0i , ha, ii ∧ o = o ∧ o = o, , if i = j, if i 6= j o (2.1) The trivial (that is, one-element) grC∞-semilattice is denoted by o. We define the following grC∞-semilattices. (i) Ak = k ×tw o for k ∈ N. (ii) A∞ = hZ ∪ {o}; ∧, Gi, where g(o) = o, g(i) = i + 1, o is the zero element of the semilattice reduct, and i ∧ j = o for i 6= j ∈ N. (iii) B+ (iv) B+ (v) B− (vi) B− 1 = hZ; min, grC∞i, where g(i) = i + 1. k = k ×tw B+ 1 = hZ; min, grC ∞i, where g(i) = i − 1; note that B− 1 = (B+ 1 ) ∗ k = k ×tw B− k = (B+ k ) . 1 for 2 ≤ k ∈ N; note that B− 1 for 2 ≤ k ∈ N. ∗ . 4 I. V. Nagy Algebra univers. (vii) C1 = (cid:10){hx, yi ∈ Z2 : x ≤ y}; ∧, grC ∞(cid:11), where g(hx, yi) = hx + 1, y + 1i and hx1, y1i∧hx2, y2i = hmin(x1, x2), max(y1, y2)i. The Hasse diagram of the semilattice reduct is depicted in Figure 1, and g is the shift operation to the right by one unit (that is, by the unit vector given in the figure). (viii) Ck = k ×tw C1 for 2 ≤ k ∈ N. With reference to the list above, now we are ready to recall the main result of W. Dziobiak, J. Jezek, and M. Mar´oti [2]. Theorem 2.1 ([2]). The minimal quasivarieties of grC∞-algebras are precisely the quasivarieties generated by one of the algebras (i) -- (viii). These minimal quasivarieties are pairwise distinct. 3. More about G-semilattices Let us agree that G always denotes an abelian group, and G-SLat stands for the variety of G-semilattices. For A ∈ G-SLat, if (A; ∧) has a zero element (in other words, a least element), then it is unique and we denote it by o. As rule, none of the formulas A \ {o} and a 6= o implies that A has a zero. (If A has no zero, then a 6= o means no condition on a and A \ {o} = A.) A 1-generated (or cyclic) G-semilattice is a G-semilattice generated by a single element. The following two lemmas were stated for G = grC∞ in W. Dziobiak, J. Jezek, and M. Mar´oti [2]; their proofs are presented for the reader's convenience. Lemma 3.1. Assume that G is an abelian group and t is a unary G-semilattice term. Then the following assertions hold. (i) The group G has a finite nonempty subset H such G-SLat satisfies the identity t(x) ≈V{h(x) : h ∈ H}. (ii) For every A ∈ G-SLat, tA : A → A is an endomorphism. Hence, for ev- ery unary G-SLat-term s, G-SLat satisfies the identity s(t(x)) ≈ t(s(x)). (iii) If A ∈ G-SLat has a zero element o, then t(o) = o in A. (iv) If G is finite, then every finitely generated G-semilattice is finite and has a zero element. Proof. Since any two basic operations commute, (ii) is clear; it also follows from ´A. Szendrei [7, Proposition 1.1]. This implies (i) by induction on the length of t. Assume that o ∈ A ∈ G-SLat. Since o is a fixed point of every automorphism of hA; ∧i and o ∧ o = o, (iii) and (iv) follow from (i). (cid:3) Lemma 3.2. Assume A ∈ G-SLat is generated by an element a ∈ A. Let s, s1 and s2 be unary G-SLat-terms. Then (i) if s1(a) = s2(a), then the identity s1(x) ≈ s2(x) holds in V(A); (ii) if s(a) = o, the zero element of A, then the identity s(x) ∧ y ≈ s(x) holds in V(A). Vol. 00, XX Minimal quasivarieties of semilattices over commutative groups 5 Proof. Assume that s1(a) = s2(a), and let b ∈ A. Since A is generated by a, b is of the form t(a) for some unary term t. Using Lemma 3.1(ii), s1(b) = s1(t(a)) = t(s1(a)) = t(s2(a)) = s2(t(a)) = s2(b). Hence, (i) holds. To prove (ii), assume that s(a) = o in A, and let b, c ∈ A. Pick a unary term t such that b = t(a). It follows from Lemma 3.1(ii)-(iii) that s(b) = s(t(a)) = t(s(a)) = t(o) = o. Hence, s(b) ∧ c = s(b). Consequently, the identity s(x) ∧ y ≈ s(x) holds in A, and also in V(A). (cid:3) The concept of a semilattice over G is analogous to that of a vector space over a field. The following two statements indicate that this analogy is quite strong in the "one-dimensional case". If A is a G-semilattice and b ∈ A, then [b]A or simply [b] denotes the subalgebra generated by b. For B = [b], we can also write B = [b] if we consider [b] an algebra rather than a subset. Corollary 3.3. Let A be a G-semilattice generated by an element a. Then A is a free algebra in V(A), freely generated by a. Proof. Consider an arbitrary b ∈ B ∈ V(A). Define a map ϕ : A → B by s(a) 7→ s(b), where s denotes a unary term. Then ϕ is a well-defined map by Lemma 3.2(i). It is a homomorphism since ϕ(s(a)) = s(b) = s(ϕ(a)) and ϕ(s(a) ∧ r(a)) = s(b) ∧ r(b) = ϕ(s(a)) ∧ ϕ(r(a)). Clearly, ϕ extends the {a} → {b} map. Hence, {a} freely generates A. (cid:3) Lemma 3.4. Let A be a 1-generated G-semilattice, and assume that {d} is a subalgebra of A. Then d is the zero element of A. Proof. Let A = [a]A. By Lemma 3.1(i), d ≤ h(a) for some h ∈ G. Using that {d} is a subalgebra, we obtain that d = h−1(d) ≤ h−1(h(a)) = a. Hence a belongs to the order filter ↑d generated by d. Since this filter is clearly a subalgebra and contains a, we conclude that A = ↑d. Thus d = o. (cid:3) Since there is only one way, the "G-trivial way", to expand the two-element meet-semilattice h{0, 1}, ≤i into a G-semilattice, we can speak of the two- element G-semilattice. Although a 1-generated G-semilattice B ∈ Q(A) is free in V(B) and in Q(B) by Corollary 3.3, it is not necessarily free in Q(A). Lemma 3.5. Let A be a 1-generated G-semilattice that is isomorphic to each of its nontrivial 1-generated subalgebras. Then Q(A) does not contain the two-element G-semilattice. Moreover, if A = [a]A and s and t are unary terms with s(a) 6= t(a), then the quasivariety Q(A) satisfies the quasi-identity s(x) ≈ t(x) ⇒ x ≈ x ∧ y. Proof. We can assume that A = [a] is nontrivial. Let s and t be unary terms such that s(a) 6= t(a). There are such terms since A 6= 1. We claim that s(b) 6= t(b) for all b ∈ A \ {0}. (3.1) To obtain a contradiction, suppose that b ∈ A \ {0} such that s(b) = t(b). Then, by Lemma 3.2(i), the subalgebra [b] satisfies the identity s(x) ≈ t(x). 6 I. V. Nagy Algebra univers. Moreover, [b] is a nontrivial subalgebra by Lemma 3.4. Thus it is isomorphic to A by the assumption. Therefore, the identity s(x) ≈ t(x) also holds in A, which contradicts s(a) 6= t(a). This proves (3.1). Next, it follows from (3.1) that A satisfies the quasi-identity s(x) ≈ t(x) ⇒ x ≈ x ∧ y. So does Q(A). The evaluation (x, y) = (1, 0) shows that this quasi-identity fails in the two-element G-semilattice, whence this G-semilattice does not belong to Q(A). (cid:3) The largest congruence and the least congruence of an algebra A are denoted by ∇ = ∇A and ∆ = ∆A, respectively. Lemma 3.6. Let A be a nontrivial 1-generated G-semilattice such that every b ∈ A \ {o} generates a subalgebra isomorphic to A. Assume that is a congruence of A and /∈ {∇A, ∆A}. Then the quotient algebra A/ does not belong to the quasivariety Q(A). Proof. Since 6= ∆, we can pick a pair (b, c) ∈ such that b 6= c. Pick an element a ∈ A that generates A. Then there exist unary terms s and t such that b = s(a) and c = t(a). Since A/ is generated by a/ and s(a/) = s(a)/ = t(a)/ = t(a/), we obtain from Lemma 3.2(i) that the identity s(x) ≈ t(x) holds in A/. But A/ is nontrivial since 6= ∇, whence the quasi- identity s(x) ≈ t(x) ⇒ x ≈ x ∧ y fails in A/. On the other hand, this quasi- identity holds in A and also in Q(A) by Lemma 3.5. Thus A/ /∈ Q(A). (cid:3) 4. The 1-generated free algebras of minimal quasivarieties The minimal quasivarieties of G-semilattices are described by the follow- ing theorem; except for the obvious G-trivial case, it suffices to deal with 1-generated nontrivial G-semilattices. Theorem 4.1. (A) The variety S of all G-trivial G-semilattices is a minimal quasivariety, and it is generated by the two-element G-semilattice. Except for S, each minimal quasivariety of G-semilattices is generated by a 1-generated G-semilattice. Furthermore, S is the only minimal quasivariety of G-semi- lattices whose 1-generated free algebra is one-element. (B) Let A be a nontrivial 1-generated G-semilattice, and let a ∈ A be a fixed element that generates A. Then the following four conditions are equivalent. (i) Q(A) is a minimal quasivariety. (ii) For each b ∈ A, if the subalgebra B generated by b is not a singleton, then there is an isomorphism ϕ : A → B such that ϕ(a) = b. (iii) A is isomorphic to each if its nontrivial 1-generated subalgebras. (iv) Each nonzero element of A generates a subalgebra isomorphic to A. Moreover, if some (equivalently, each) of the conditions (i), (ii), (iii), and (iv) holds, then Vol. 00, XX Minimal quasivarieties of semilattices over commutative groups 7 (v) all nontrivial 1-generated algebras of Q(A) are (isomorphic to) the free algebra FQ(A)(1) of Q(A). The set of minimal quasivarieties of G-semilattices will often be denoted by MinQVar(G). (Since the quasi-identities in the language of G-semilattices form a set, so do the minimal quasivarieties.) Proof of Theorem 4.1. Since G-trivial G-semilattices are essentially (that is, up to term equivalence) semilattices, it belongs to the folklore that S is gen- erated by the two-element G-semilattice. Since FS(1) is one-element, it does not generate S. Assume that U ∈ MinQVar(G) such that U is not G-trivial. Then there are an algebra B ∈ U, an element b ∈ B, and a group element g ∈ G such that g(b) 6= b. Hence the subalgebra generated by b has at least two elements, which implies that FU (1) not a singleton. Therefore, by the minimality of U, FU (1) generates U. This proves part (A). Next, we deal with part (B). Clearly, (ii) implies (iii). It follows from Lemma 3.4 that (iii) and (iv) are equivalent. To prove that (i) implies (ii), assume that (i) holds. Let b ∈ A, and assume that the subalgebra [b] is not a singleton. Define a map ϕ : A → B by the rule r(a) 7→ r(b), where r ranges over the set of unary terms. Letting r be the identity map, we obtain that ϕ(a) = b. We conclude from Corollary 3.3 that ϕ is a homomorphism, and it is clearly surjective. In order to prove that ϕ is injective, assume that r and s are unary terms such that ϕ(r(a)) = ϕ(s(a)), that is, r(b) = s(b). By Lemma 3.2(i), the identity r(x) ≈ s(x) holds in V(B), whence it also holds in Q(B). Since B ∈ Q(A) and B is nontrivial, the minimality of Q(A) implies Q(B) = Q(A). Thus the identity r(x) ≈ s(x) holds in A, and we obtain that r(a) = s(a). Hence, ϕ is injective, and it is an isomorphism. Therefore, (i) implies (ii). Next, to show that (iii) implies (v), assume that (iii) holds. Let B ∈ Q(A) be a nontrivial 1-generated algebra generated by b ∈ B. By Corollary 3.3, there exists a (unique) surjective homomorphism ϕ : A → B such that ϕ(a) = b. Suppose, to derive a contradiction, that ϕ is not an isomorphism. Then ϕ is not injective, whence ker ϕ 6= ∆. Since B is nontrivial, ker ϕ 6= ∇. Thus Lemma 3.6 (together with Lemma 3.4) applies, and we obtain that B ∼= A/ker ϕ /∈ Q(A). This contradicts the assumption on B, and we conclude that (iii) implies (v). Finally, to show that (iii) implies (i), assume that (iii) holds. Let K be a nontrivial subquasivariety of Q(A). Let B denote the free algebra FK(b). For the sake of contradiction, suppose that B is one-element. Then g(b) = b for every g ∈ G, and we conclude that the identity g(x) ≈ x holds in K. Therefore, the two-element G-semilattice belongs to K ⊆ Q(A), which contradicts Lemma 3.5. Thus B ∈ K is a nontrivial 1-generated algebra in Q(A). Since we already know that (iii) implies (v), we obtain from (v) that A ∼= B ∈ K. Thus A ∈ K, yielding that Q(A) ⊆ K. This means that Q(A) ∈ MinQVar(G). Consequently, (iii) implies (i). (cid:3) 8 I. V. Nagy Algebra univers. 5. Two constructs The following construct is due to M. Mar´oti [5]. Let H be a subgroup (that is, a nonempty subuniverse) of G. The Mar´oti semilattice over G is M(H, G) = hM (H, G); ∩, Gi, where M (H, G) = ∅ ∪ {gH : g ∈ G}, ∩ is the usual intersection, and f (gH) for f, g ∈ G is defined as f gH. Note that M(H, G) consists of atoms and a zero; the atoms are the (left) cosets of H while the emptyset is o. Clearly, M(H, G) is a G-semilattice. Note that A∞ in Theorem 2.1 is (isomorphic to) M({1}, grC∞). The importance of this G-semilattice is explained by the following statement. Proposition 5.1. (A) For every subgroup H of an abelian group G, M(H, G) generates a minimal quasivariety K. If H = G, then M(H, G) is the two- element G-semilattice. Otherwise, M(H, G) ∼= FK(1). (B) Let H1 and H2 be subgroups of G. Then the algebras M(H1, G) and M(H2, G) generate the same minimal quasivarieties of G-semilattices if and only if H1 = H2. Proof. Since the case of H = G is trivial, we assume that H is a proper subgroup. Then there is an f ∈ G \ H. Since o = ∅ = 1H ∩ f H and, for any g, h ∈ G, hH = (hg−1)(gH), M(H, G) is generated by each of its nonzero elements. Thus part (A) follows from Theorem 4.1(B). To prove part (B), we can assume that both subgroups in question are proper. The "if" part is obvious. To prove the converse implication, as- sume that M(H1, G) and M(H2, G) generate the same quasivariety K. Then M(H1, G) ∼= M(H2, G) since both are isomorphic to FK(1) by part (A). Hence, there is an isomorphism ϕ : M(H1, G) → M(H2, G). Since H1 = 1H1 is an atom, so is its ϕ-image. Thus there is an f ∈ G such that ϕ(H1) = f H2. Let x denote an arbitrary element of G. Using that ϕ is an isomorphism, we obtain that x ∈ H1 ⇐⇒ xH1 = H1 ⇐⇒ ϕ(xH1) = ϕ(H1) ⇐⇒ xϕ(H1) = f H2 ⇐⇒ xf H2 = f H2 ⇐⇒ xf f −1 ∈ H2 ⇐⇒ x ∈ H2, which means that H1 = H2. (cid:3) Let Sub(G) denote the set of all subgroups (that is, nonempty subuniverses) of G. The following statement clearly follows from Proposition 5.1. Corollary 5.2. (A) For every abelian group G, there are at least Sub(G) many minimal quasivarieties of G-semilattices. (B) For each cardinal κ, there exists an abelian group G such that there are at least κ minimal quasivarieties of G-semilattices. The next construction generalizes the n-fold twisted multiple construct, see (2.1). Let K be a subgroup of G, and choose a system T of representatives of the (left) cosets of K. That is, T ⊆ G such that T ∩ gK = 1 holds for Vol. 00, XX Minimal quasivarieties of semilattices over commutative groups 9 all g ∈ G. Usually (but not in Lemma 5.3), we assume that T ∩ K = {1}. (We will prove that, up to isomorphism, our construct does not depend on the choice of T .) Assume that U is a K-semilattice. We define a G-semilattice N(K, G, U) (up to isomorphism) as follows. N(K, G, U) =(cid:10){o} ∪ (U × T ); ∧, G(cid:11), g(hu, ti) = hgtf −1(u), f i if f ∈ T ∩ gtK, where, for g ∈ G, g(o) = o, (u, t) ∧ o = o ∧ o = o, and (5.1) hu1, t1i ∧ hu2, t2i =(hu1 ∧ u2, t1i o if t1 = t2, if t1 6= t2 . Lemma 5.3. The isomorphism class of the algebra N(K, G, U) does not de- pend on the choice of T . Furthermore, N(K, G, U) is a G-semilattice. Proof. First, we prove that N(K, G, U) is a G-semilattice. Assume that hu, ti ∈ U × T and g1, g2 ∈ G. Denote by f1 and f2 the unique element of T ∩ g2tK and T ∩ g1f1K, respectively. Then f1K = g2tK yields that f2 ∈ g1f1K = g1g2tK. Using the commutativity of G at =∗, we obtain that g1(cid:0)g2(hu, ti)(cid:1) = g1(hg2tf −1 = hg1f1f −1 2 g2tf −1 1 (u), f2i =∗ hg1g2tf −1 1 (u), f1i) = hg1f1f −1 2 (g2tf −1 1 (u)), f2i 2 (u), f2i = (g1g2)(hu, ti). This implies that axiom (iii) holds. To prove the validity of (iv), let g ∈ G, u1, u2, ∈ U and t, t1, t2 ∈ T such that t1 6= t2. Let f ∈ T ∩ gtK, f1 ∈ T ∩ gt1K, and f2 ∈ T ∩ gt2K. Since t1 and t2 belong to distinct cosets of K, we obtain that gt1K 6= gt2K and f1 6= f2. Thus g(hu1, t1i) ∧ g(hu2, t2i) = hgt1f −1 1 (u1), f1i ∧ hgt2f −1 2 (u2), f2i Since we also obtain that = o = g(o) = g(cid:0)hu1, t1i ∧ hu2, t2i(cid:1). g(hu1, ti) ∧ g(hu2, ti) = hgtf −1(u1), f i ∧ hgtf −1(u2), f i = hgtf −1(u1) ∧ gtf −1(u2), f i = hgtf −1(u1 ∧ u2), f i = g(hu1 ∧ u2, ti) = g(hu1, ti ∧ hu2, ti), axiom (iv) also holds. The rest of the axioms trivially hold, whence the algebra N(K, G, U) is a G-semilattice. Second, let B be the algebra N(K, G, U) defined above with T . Let T ′ be another system of representatives of the cosets of K, and denote by B′ the algebra N(K, G, U) constructed with T ′ instead of T . For t ∈ T , the unique element of T ′ ∩ tK is denoted by t′. We claim that the map ϕ : B → B′, defined by o 7→ o, hu, ti 7→ ht′−1t(u), t′i is an isomorphism. It is clearly a semilattice isomorphism since t′−1t induces an automorphism of hU ; ∧i. Assume that g ∈ G, hu, t1i ∈ B, and hv, t′ 1i ∈ B′. 10 I. V. Nagy Algebra univers. Let T ∩ gt1K = {t2}. Then t2 ∈ (gK)(t1K) = (gK)(t′ implies that T ′ ∩ gt′ 2}. Hence, we obtain that 1K) = gt′ 1K, which 1K = {t′ gB(hu, t1i) = hgt1t−1 1t′−1 gB′(hv, t′ 1i) = hgt′ 2i in B′. t1 ∈ K. Using (5.2) with v = t′−1 (v), t′ 2 1 2 (u), t2i in B and (5.2) t1(u), we obtain that ϕ Observe that t′−1 preserves the operation g since 1 gB′ (ϕhu, t1i) = gB′ (ht′−1 1 1t′−1 t1(u), t′ t′−1 1 1i) = hgt′ 1t′−1 2 2i = ht′−1 (t′−1 t2gt1t−1 t1(u)), t′ 2i 2 (u), t′ 2i 2 1 = hgt′ = ht′−1 2 t2(gt1t−1 = ϕ(gB(hu, t1i)). t1(u), t′ 2 (u)), t′ 2 2i = ϕ(cid:0)hgt1t−1 2 (u), t2i(cid:1) (cid:3) Remark 5.4. The n-fold twisted multiple construct is indeed a particular case of (5.1), up to term equivalence. To see this, assume that G1 and G2 are abelian groups and ϕ : G2 → G1 is a surjective homomorphism. Then each G1-semilattice B1 = hB; ∧, G1i becomes a G2-semilattice B2 = hB; ∧, G2i by change of groups: B2 = hB; ∧, G2i and, for g ∈ G2 and b ∈ B, we define g(b) = (ϕ(g))(b). Clearly, B1 and B2 are term equivalent. Now, let n ∈ N, and let A = hA; ∧, grC∞i be a grC∞-semilattice, where grC∞ is generated by g. Take the subgroup K = {gnk : k ∈ Z}, and let T = {g0, g1, . . . , gn−1}. Using the group isomorphism ϕ : K → grC∞, defined by ϕ(gnk) = gk, A becomes a K-semilattice A′ by change of groups. We claim that n ×tw A is isomorphic to N(K, grC∞, A′); the easy proof is omitted since we will not use this fact. Remark 5.5. The Mar´oti semilattice over G is also a particular case of (5.1) since M(K, G) ∼= N(K, G, o), where o = h{o}; ∧, Gi. Lemma 5.6. Besides the assumptions of Lemma 5.3, assume that K is a proper subgroup of G, and that U = [a] is a 1-generated K-semilattice. Then, for every t ∈ T , ha, ti generates the G-semilattice N(K, G, U). Proof. Let B = [ha, ti]N(K,G,U). It suffices to show that U × T ⊆ B since then o ∈ B follows from T > 1. If g ∈ K, then t ∈ tK ∩ T = tgK ∩ T = gtK ∩ T , and we obtain that g(ha, ti) = hgtt−1(a), ti = hg(a), ti. Hence, it follows from Lemma 3.1(i) that hu, ti ∈ B for all u ∈ U . Now, let f ∈ T . Then f ∈ f K∩T = t−1f tK∩T , whence (t−1f )(hu, ti) = h(t−1f )tf −1(u), f i = hu, f i. Thus hu, f i ∈ B. Hence, U × T ⊆ B, as desired. (cid:3) 6. Minimal quasivarieties of semilattices over finite abelian groups Given a minimal quasivariety R of G-semilattices, we associate a subgroup HR with R as follows. Take the free algebra FR(a), and define HR as the stabilizer of a, that is, HR = {g ∈ G : g(a) = a}. The minimal quasivarieties of semilattices over finite abelian groups are satisfactorily described by the following theorem. Vol. 00, XX Minimal quasivarieties of semilattices over commutative groups 11 Theorem 6.1. Let G be a finite abelian group. Then the map α : R 7→ HR is a bijection from the set MinQVar(G) of minimal quasivarieties to Sub(G). The inverse map β : Sub(G) → MinQVar(G) is defined by H 7→ Q(cid:0)M(H, G)(cid:1). Proof. Since the (minimal) quasivariety of G-trivial G-semilattices clearly cor- responds to the case H = G and vice versa, we can assume that R is not G-trivial and that H is a proper subgroup. We know from Proposition 5.1 that β is an injective map from Sub(G) to MinQVar(G). Obviously, α is a map from MinQVar(G) to Sub(G). To prove that the composite map α ◦ β act identically on Sub(G), assume that H is a proper subgroup of G. Then β(H) is the quasivariety Q(cid:0)M(H, G)(cid:1). By Proposition 5.1(A), M(H, G) is (isomorphic to) the free algebra of rank 1 in this quasivariety. Its free generator is of the form f H for some f ∈ G. Hence α(β(H)) is the stabilizer of f H, that is α(β(H)) = {g ∈ G : gf H = f H} = {g ∈ G : gf f −1 ∈ H} = H. This proves that α ◦ β acts identically on Sub(G) (even when G is infinite). Next, we prove that β ◦ α acts identically on MinQVar(G). To do so, let R ∈ MinQVar(G), distinct from the quasivariety of G-trivial G-semilattices. Then FR(a) generates R by Theorem 4.1(A). We know from Lemma 3.1(iv) that o ∈ FR(a), FR(a) is finite, and FR(a) has an atom, b. Since every g ∈ G preserves the semilattice order, the subalgebra [b]FR(a) consists of some atoms and, possibly, of o. This subalgebra is isomorphic to FR(a) by Theorem 4.1(B). Hence, by finiteness, this subalgebra equals FR(a). It follows that FR(a) = {o} ∪ {g(a) : a ∈ G}; note that g1(a) = g2(a) may occur with distinct g1, g2 ∈ G. Let H = α(R) = {g ∈ G : g(a) = a}, the stabilizer of a. It suffices to show that FR(a) ∼= M(H, G) (6.1) since then the minimality of R implies that β(α(R)) = β(H) = Q(cid:0)M(H, G)(cid:1) = R . Define a map ϕ : FR(a) → M(H, G) by g(a) 7→ gH and o 7→ ∅. Since f (a) = g(a) iff a = f −1g(a) iff f −1g ∈ H iff f H = gH, we obtain that ϕ is indeed a map, and it is a bijection. Since the nonzero elements are atoms both in FR(a) and M(H, G), ϕ preserves the meet. Finally, ϕ is an isomorphism since, for (cid:3) any f, g ∈ G, ϕ(cid:0)f (g(a))(cid:1) = ϕ(cid:0)(f g)(a)(cid:1) = (f g)H = f (gH) = f ϕ(cid:0)g(a)(cid:1). 12 I. V. Nagy Algebra univers. 7. When G is not necessarily finite The set MinQVar(G) of minimal quasivarieties of G-semilattices splits into three disjoint subsets, MinQVar0(G) = {R ∈ MinQVar(G) : o ∈ FR(1), FR(1) ≥ 2}, MinQVar1(G) = {the class of G-trivial G-semilattices} = {R ∈ MinQVar(G) : FR(1) = 1}, and MinQVar2(G) = {R ∈ MinQVar(G) : o /∈ FR(1)}. For example, by Theorem 2.1, MinQVar0(grC∞) = {Q(Ak) : k ∈ N} ∪ {Q(A∞)} ∪{Q(B+ MinQVar2(grC∞) = {Q(B+ k ), Q(B− 1 ), Q(B− k ), Q(Ck) : 2 ≤ k ∈ N} and 1 ), Q(C1)}. Let R ∈ MinQVar(G) such that R is distinct from the quasivariety of G- trivial G-semilattices. Then, by Lemma 3.2, there are exactly two cases: either R ∈ MinQVar0(G) and all members of R have o, or R ∈ MinQVar2(G) and there are algebras in R without o. The obvious singleton set MinQVar1(G) deserves no separate attention. The target of this section is to describe the members of MinQVar0(G). As we know from Theorem 4.1, they are determined by their free algebras on one generator, that is, by the nontrivial 1-generated G-semilattices A with zero that satisfy condition (iii) (or (iv)) of Theorem 4.1. To give the main definition of this section, assume that R ∈ MinQVar0(G) and A = FR(a). Let K = KR be the set {g ∈ G : a ∧ g(a) 6= o}. (We will show that K is a subgroup of G.) Let U = {t(a) : t is a unary term in the language of K-semilattices}. Then we define U = UR = hU ; ∧, Ki. Lemma 7.1. Assume that R ∈ MinQVar0(G). Then, for K = KR and U = UR defined above, the following assertions hold. (i) K is a subgroup of G. (ii) Let f1, . . . , fn ∈ G. Then f1(a) ∧ · · · ∧ fn(a) 6= o iff f1K = · · · = fnK. (iii) U is a K-semilattice. If U is nontrivial, then U has no zero element. (iv) If U is nontrivial, then Q(U) ∈ MinQVar2(K) and U ∼= FQ(U)(1). (v) A = FR(a) is isomorphic to N(K, G, U). If G is finite, then U is trivial by (6.1). The case U = 1 is important even if finiteness is not assumed since Lemma 7.1, together with Remark 5.5, clearly implies the following corollary. Corollary 7.2. Let R ∈ MinQVar0(G). Then UR is 1-element iff FR(1) is (up to isomorphism) M(K, G) with a proper subgroup K of G. Proof of Lemma 7.1. Since A is nontrivial, a 6= o and id ∈ K. If g belongs to K, then so does g−1 since a ∧ g−1(a) = g−1(cid:0)a ∧ g(a)(cid:1) 6= o. To get a Vol. 00, XX Minimal quasivarieties of semilattices over commutative groups 13 contradiction, suppose that f, g ∈ K but f g /∈ K. Then a ∧ f (a) 6= o and o = a ∧ f g(a) ≥ a ∧ f (a) ∧ g(a) ∧ f g(a) = (a ∧ f (a)) ∧ g(a ∧ f (a)). It follows from Theorem 4.1 that a ∧ f (a) generates A = FR(a). Hence, by applying Lemma 3.2(ii) to the element a∧f (a) and to the term s(x) = x∧g(x), we obtain that the identity x ∧ g(x) ∧ y ≈ x ∧ g(x) holds in R = Q(A). Substituting ha, oi for hx, yi, we obtain that o = a ∧ g(a) ∧ o ≈ a ∧ g(x), which contradicts g ∈ K. Thus f g ∈ K, and K is a subgroup of G. This proves (i). To prove (ii), assume that f1(a) ∧ · · · ∧ fn(a) 6= o, and let i, j ∈ {1, . . . , n} with i 6= j. Then fi(a) ∧ fj(a) 6= o since fi(a) ∧ fj(a) ≥ f1(a) ∧ · · · ∧ fn(a). Thus o = f −1 i fj ∈ K, that is, fiK = fjK, for all i, j ∈ {1, . . . , n}. This proves the "only if" part of (ii). i (cid:0)fi(a) ∧ fj(a)(cid:1) = a ∧ f −1 i fj(a). Thus f −1 To prove the "if" part, we claim that, for all n ∈ N, (o) 6= f −1 i if g1, . . . , gn ∈ K, then a ∧ g1(a) ∧ · · · ∧ gn(a) 6= o. (7.1) Suppose, for contradiction, that there is a least n such that (7.1) fails. By the definition of K, 2 ≤ n. Let b = a ∧ g1(a) ∧ · · · ∧ gn−1(a). By the minimality of n, we have that b 6= o but b ∧ gn(a) = o. Since b ≤ a, we obtain that b ∧ gn(b) ≤ b ∧ gn(a), that is, b ∧ gn(b) = o. By Theorem 4.1(B), there is a G-semilattice isomorphism ϕ from A to the subalgebra [b] such that ϕ(a) = b. By Lemma 3.4, ϕ(o) = o since {o} is the only singleton subalgebra of A. Hence ϕ(cid:0)a ∧ gn(a)(cid:1) = ϕ(a) ∧ gn(cid:0)ϕ(a)(cid:1) = b ∧ gn(b) = o = ϕ(o), which implies that a ∧ gn(a) = o. Thus gn /∈ K, which is a contradiction that proves (7.1). Next, assume that f1K = · · · = fnK. Then g2 = f2f −1 1 , . . . , gn = fnf −1 belong to K, and (7.1) yields that a ∧ g2(a) ∧ · · · ∧ gn(a) 6= o. Thus o = f1(o) 6= f1(a) ∧ f1g2(a) ∧ · · · ∧ f1gn(a) = f1(a) ∧ · · · ∧ fn(a). This proves (ii). 1 It is obvious that U is a K-semilattice, and it is generated by a. Hence, to give a proof for (iii) by contradiction, suppose that U is nontrivial but it has a zero element b, distinct from a. It follows from Lemma 3.1(i) that b is of the form f1(a) ∧ · · · ∧ fn(a) for some f1, . . . , fn ∈ K. Since f1K = · · · = fnK = K, we conclude from (ii) that b 6= 0. Applying Lemma 3.1(iii) to the K-semilattice U, we obtain that for all f ∈ K, f (b) = b. (7.2) By Lemma 3.4, the G-subsemilattice B = [b]A is not a singleton. Thus The- orem 4.1(B)(ii) yields a G-semilattice isomorphism from A onto B such that a 7→ b. Applying the inverse of this isomorphism to (7.2), we conclude that f (a) = a, for all f ∈ K. Thus U, the K-semilattice generated by a, is a singleton. This contradiction proves (iii). Next, we deal with (iv). Consider an arbitrary b ∈ U . By Theorem 4.1(B), it suffices to show that the K-semilattice B = [b]U is isomorphic to U. We know from Theorem 4.1(B) that there is a G-semilattice isomorphism ϕ : A = [a]A → [b]A such that ϕ(a) = b. This implies that, for all unary K-semilattice terms r and s, r(a) = s(a) iff r(b) = s(b). Hence we conclude that the 14 I. V. Nagy Algebra univers. restriction ϕ⌉U of ϕ to U is a K-semilattice isomorphism from U = [a]U to B = [b]U. This proves (iv). To prove (v), fix a set T of representatives of the (left) cosets of K such that T ∩ K = {1}. Consider N(K, G, U), defined in (5.1), and the map ϕ : N(K, G, U) → A, defined by hu, ti 7→ t(u) and o 7→ o. We claim that ϕ is an isomorphism. To see this, let u, u1, u2 ∈ U , t, t1, t2 ∈ T with t1 6= t2, and g ∈ G. Then ϕ(hu1, hi ∧ hu2, hi) = ϕ(hu1 ∧ u2, hi) = h(u1 ∧ u2) = h(u1) ∧ h(u2) = ϕ(hu1, hi) ∧ ϕ(hu2, hi). By Lemma 3.1(i), there are f1, . . . , fn, g1, . . . , gm ∈ K such that u1 = f1(a) ∧ · · · ∧ fn(a) and u2 = g1(a) ∧ · · · ∧ gm(a). Since t1K 6= t2K, the group elements t1f1, . . . , t1fn and t2g1, . . . , t2gm do not belong to the same coset of K. Hence, it follows from (ii) that t1(u1) ∧ t2(u2) = o. Consequently, ϕ(hu1, t1i ∧ hu2, t2i) = ϕ(o) = t1(u1) ∧ t2(u2) = ϕ(hu1, t1i) ∧ ϕ(hu2, t2i). Thus ϕ preserves the meet operation. If T ∩ gtK = {f }, then ϕ(ghu, ti) = ϕ(hgtf −1(u), f i) = f (gtf −1(u)) = gt(u) = gϕ(hu, ti). Therefore, ϕ is a homomorphism. Clearly, o is a preimage of o. To show that ϕ is surjective, let b ∈ A \ {o}. By Lemma 3.1(i), b = f1(a) ∧ · · · ∧ fn(a) for some f1, . . . , fn ∈ G. These fi belong to the same coset hK by (ii). Then h−1f1, . . . , h−1fn ∈ K, hh−1f1(a) ∧ · · · ∧ h−1fn(a), hi belongs to N(K, G, U), and ϕ(cid:0)hh−1f1(a) ∧ · · · ∧ h−1fn(a), hi(cid:1) = f1(a) ∧ · · · ∧ fn(a) = b. Hence, ϕ is surjective. Next, we assert that o, the zero element of A, does not belong to U . To obtain a contradiction, suppose that o ∈ U . Then U = 1 since otherwise U cannot have a zero element by (iii). Thus U = 1, U = {a} = {o}, which implies that A = 1, a contradiction. Hence, o /∈ U . We are now in the position to show that ϕ is injective. If hu, ti ∈ U × T , then ϕ(hu, ti) = t(u) 6= o since otherwise o = t−1(o) = u would belong to U . Hence the only preimage of o is itself. Assume that hu1, t1i, hu2, t2i ∈ U × T such that ϕ(hu1, t1i) = ϕ(hu2, t2i). Then t1(u1) = t2(u2), and o 6= ϕ(hu1, t1i) = ϕ(hu1, t1i) ∧ ϕ(hu2, t2i) = ϕ(hu1, t1i ∧ hu2, t2i). Hence, hu1, t1i ∧ hu2, t2i 6= o implies that t1 = t2, whence t1(u1) = t2(u2) entails that u1 = u2. Thus hu1, t1i = hu2, t2i, proving that ϕ is injective. (cid:3) Next, we define an auxiliary set D(G) as follows. Here Sub(G) is the set of all subgroups of G, {1} is the 1-element subgroup, and o = h{o}; ∧, Gi is the Vol. 00, XX Minimal quasivarieties of semilattices over commutative groups 15 one-element semilattice over G. D(G) =(cid:8)hK, oi : K ∈ Sub(G) \ {G}(cid:9) ∪ (7.3) (cid:8)hK, FS (1)i : K ∈ Sub(G) \ {{1}, G}, S ∈ MinQVar2(K)(cid:9). (Of course, FS(1) above and in similar situations is understood as the isomor- phism class of FS(1). However, we simply speak of free algebras rather than their isomorphism classes.) Now, we are in the position to formulate the main result of this section. Theorem 7.3. Let G be an abelian group. Define a map γ : MinQVar0(G) → D(G) by R 7→ hKR, URi, where KR and UR are given before Lemma 7.1, and a map δ : D(G) → MinQVar0(G) by hK, Ui 7→ Q(cid:0)N(K, G, U)(cid:1). Then γ and δ are reciprocal bijections. Remark 7.4. Theorem 7.3 reduces the difficulty to D(G). That is, if we could describe MinQVar2(K) for all nontrivial subgroups K ∈ Sub(G), including K = G, then we would obtain a full description of MinQVar(K). Generally, this seems to be hopeless in view of Section 8. Remark 7.5. If G is finite, then MinQVar2(G) = ∅ by Lemma 3.1(iv). Thus UR = o, N(K, G, o) ∼= M(K, G), and Theorem 7.3 reduces to Theorem 6.1. That is, Theorem 7.3, together with Lemma 3.1(iv), implies Theorem 6.1. However, the easy proof of Theorem 6.1 in Section 6 is justified by the fact that the proofs in Section 7 are much more complicated. Proof of Theorem 7.3. First we show that γ is a map from MinQVar0(G) to D(G). Let R ∈ MinQVar0(G), K = KR and U = UR. By Lemma 7.1, it suffices to show that K 6= G and that U 6= 1 implies K 6= {1}. Striving for contradiction, suppose that K = G. Since o belongs to FR(a), Lemma 3.1(i) implies the existence of f1, . . . , fn ∈ G such that o = f1(a) ∧ · · · ∧ fn(a). But this contradicts Lemma 7.1(ii) since any two cosets of K = G are equal. Thus K 6= G. Clearly, if K = {1}, then every unary K-semilattice term induces the identity map and U = 1. Hence, U 6= 1 implies that K 6= {1}. This proves that γ is a map from MinQVar0(G) to D(G). Next, we show that δ is a D(G) → MinQVar0(G) map. It follows easily from Proposition 5.1(A) and Remark 5.5 that δ(hK, oi) = M(K, G) belongs to MinQVar0(G), provided G 6= K ∈ Sub(G). So let hK, Ui := hK, FS(a)i be in D(G), where K belongs to Sub(G)\ {{1}, G} and S belongs to MinQVar2(K). We have to show that δ(hK, Ui) = Q(cid:0)N(K, G, U)(cid:1) belongs to MinQVar0(G). As in (5.1), let T be a set of representatives of the cosets of K such that 1 ∈ T . We know from Lemma 5.6 that, for every t ∈ T , A = N(K, G, U) is generated by ha, ti. (7.4) 16 I. V. Nagy Algebra univers. By Theorem 4.1, it suffices to prove that for each hb, ti of A, the subalgebra B = [hb, ti]A is isomorphic to A. Define a map ϕ : A → B by o 7→ o and hr(a), ti 7→ hr(b), ti, where r denotes an arbitrary unary K-semilattice term. Since U = FS(a) has no zero, it follows from Theorem 4.1(B) that there exists a K-semilattice isomorphism ψ from U = [a]U onto [b]U with ψ(a) = b. This implies that ϕ is a well-defined map and a bijection. By (7.4), each element of A is of the form s(ha, ti) for an appropriate unary G-semilattice term s. Clearly, each element of B is of the form s(hb, ti). Observe that, in (5.1), gtf −1 ∈ K. Thus it is clear from the definition of N(K, G, U) that the action of s on a pair hx, ti depends on two ingredients. First, it depends on how we compute within U; from this aspect, ψ allows us to replace a with b. Second, on how we compute with group elements; then x is irrelevant. Consequently, ϕ is an isomorphism. Therefore, A = N(K, G, U) satisfies condition (ii) of Theorem 4.1. Since δ(hK, Ui) = Q(A), (i), (ii), and (v) of Theorem 4.1 yield that there is an isomorphism N(K, G, U) → Fδ(hK,Ui)(d) with ha, 1i 7→ d, (7.5) and that δ(hK, Ui) ∈ MinQVar0(G). This proves that δ is a map from D(G) to MinQVar0(G). Let R ∈ MinQVar0(G). Since N(KR, G, U)R ∼= FR(1) by Lemma 7.1(v) and R is generated by any of its nontrivial algebra, we obtain that δ(cid:0)γ(R)(cid:1) = δ(hKR, URi) = Q(cid:0)N(KR, G, U)R(cid:1) = Q(cid:0)FR(1)(cid:1) = R. That is, δ ◦ γ is the identity map on MinQVar0(G). Next, to show that γ ◦ δ is the identity map on D(G), assume that hK, Ui belongs to D(G). We distinguish two cases. First, assume that U is nontrivial. Let R = δ(hK, Ui) = Q(cid:0)N(K, G, U)(cid:1). Then, by (7.5), we can compute γ(cid:0)δ(hK, Ui)(cid:1) = γ(R) = hKR, URi based on the algebra N(K, G, U) and its free generator ha, 1i. Hence KR = {g ∈ G : ha, 1i ∧ g(ha, 1i) 6= o}. Since o /∈ U , it is clear from definitions that KR = K. It is also clear that the K-semilattice generated by ha, 1i is U × {1}, which is isomorphic to U. Therefore, since now isomorphic algebras are treated as equal ones, γ(cid:0)δ(hK, Ui)(cid:1) = hK, Ui. and Proposition 5.1(A), we obtain that Second, assume that U = o. Let R = δ(hK, oi). By definitions, Remark 5.5, R = Q(cid:0)N(K, G, o)(cid:1) = Q(cid:0)M(K, G)(cid:1) and M(K, G) ∼= FR(1). Hence, instead of FR(1), we can compute γ(R) = hKR, URi from M(K, G) and its generating element K = 1K. Hence, clearly, we obtain that γ(R) = (cid:3) hK, oi, that is, γ(cid:0)δ(hK, Ui)(cid:1) = hK, Ui. Thus γ ◦ δ is the identity map. Vol. 00, XX Minimal quasivarieties of semilattices over commutative groups 17 8. An example Since MinQVar2(grC∞) is less complicated than MinQVar0(grC∞), see at the beginning of Section 7, one might hope that MinQVar2(G) can somehow be described for any abelian group G. This hope is minimized by the following example. Consider grC2 ∞, the direct square of the infinite cyclic group. It is gener- ated, in fact freely generated, by {(g, 1), (1, g)}. It has only countably many subgroups by, say, W. R. Scott [6, Theorem 5.3.5]. Hence, if we had that MinQVar2(grC∞) ≤ ℵ0, then MinQVar(grC2 ∞) = ℵ0 would follow from The- orem 7.3, and we could expect a reasonable description of MinQVar(grC2 ∞). However, we construct continuously many members of MinQVar2(grC2 ∞). Given an irrational number α, let Bα = {m + nα : m, n ∈ Z}. Define the action of hgi, gji by hgi, gji(m + nα) = m + i + (n + j)α. This way we obtain a grC2 ∞-semilattice Bα = hBα; ∧, grC 2 ∞i, where ∧ is the minimum with respect to the usual order of real numbers. Example 8.1. For each irrational number α, Q(Bα) ∈ MinQVar2(grC2 α and β are distinct irrational numbers, then Q(Bα) 6= Q(Bβ). ∞). If Proof. The first part follows from Theorem 4.1 since Bα is generated by each of its elements. To prove the second part, assume that α < β. We can pick p, q ∈ Z such that α < p/q < β. Since qα < p < qβ, the identity hgp, 1i(x) ∧ h1, gqi(x) ≈ h1, gqi(x) holds in Q(Bα) but fails in Q(Bβ). (cid:3) Acknowledgment. The author is deeply indebted to Dr. Mikl´os Mar´oti for an excellent introduction to Universal Algebra, for raising the problem the present paper deals with, and for his valuable friendly support since then. Also, the help obtained from Dr. ´Agnes Szendrei is gratefully acknowledged. References [1] Burris, S., Sankappanavar, H.P.: A Course in Universal Algebra. Graduate Texts in Mathematics, vol 78. Springer-Verlag, New York -- Berlin (1981). The Millennium Edition, http://www.math.uwaterloo.ca/~snburris/htdocs/ualg.html [2] Dziobiak, W., Jezek, J., Mar´oti, M.: Minimal varieties and quasivarieties of semilattices with one automorphism. Semigroup Forum 78, 253 -- 261 (2009) [3] Jezek, J: Simple semilattices with two commuting automorphisms. Algebra Universalis 15, 162 -- 175 (1982) [4] Jezek, J: Subdirectly irreducible semilattices with an automorphism. Semigroup Forum 43, 178 -- 186 (1991) [5] Mar´oti, M.: Semilattices with a group of automorphisms. Algebra Universalis 38, 238 -- 265 (1997) [6] Scott, W.R.: Group Theory. Dover Publ. Inc., New York, 1987 [7] Szendrei, ´A.: Clones in Universal Algebra. S´eminaire de Math´ematiques Sup´erieures, vol. 99., Les Presses de l'Universit´e de Montr´eal, Montr´eal, 1986. [Available from: Centre de Recherches Math´ematiques, Universit´e de Montr´eal] 18 I. V. Nagy Algebra univers. Ildik´o V. Nagy Szeged, Hungary, [email protected]
1406.1574	2	1406	2016-12-06T13:34:50	Triple Derivations and Triple Homomorphisms of Perfect Lie Superalgebras	[ "math.RA" ]	In this paper, we study triple derivations and triple homomorphisms of perfect Lie superalgebras over a commutative ring $R$. It is proved that, if the base ring contains $\frac{1}{2}$, $L$ is a perfect Lie superalgebra with zero center, then every triple derivation of $L$ is a derivation, and every triple derivation of the derivation algebra $ Der (L)$ is an inner derivation. Let $L,~L^{'}$ be Lie superalgebras over a commutative ring $R$, the notion of triple homomorphism from $L$ to $L^{'}$ is introduced. We proved that, under certain assumptions, homomorphisms, anti-homomorphisms, and sums of homomorphisms and anti-homomorphisms are all triple homomorphisms.	math.RA	math	Triple Derivations and Triple Homomorphisms of Perfect Lie Superalgebras Jia Zhou1, Liangyun Chen2, Yao Ma2 1 College of Information Technology, Jilin Agriculture University, 2 School of Mathematics and Statistics, Northeast Normal University, Changchun 130118, China Changchun 130024, China Abstract In this paper, we study triple derivations and triple homomorphisms of perfect Lie superalgebras over a commutative ring R. It is proved that, if the base ring contains 1 2, L is a perfect Lie superalgebra with zero center, then every triple derivation of L is a derivation, and every triple derivation of the derivation algebra Der(L) is an inner derivation. Let ′ be Lie superalgebras over a commutative ring R, the notion of triple homomorphism L, L ′ is introduced. We prove that, under certain assumptions, homomorphisms, from L to L anti-homomorphisms, and sums of homomorphisms and anti-homomorphisms are all triple homomorphisms. Keywords: Perfect Lie superalgebras; Triple derivations; Triple homomorphisms; En- veloping Lie superalgebras. 1 Introduction In studies to derivations of associative algebras ( [2], [4], [12]) appear naturally differ- ent sorts of triple derivations such as associative triple derivations, Jordan triple deriva- tions and Lie triple derivations. Lie triple derivations are interesting not only to studies of associative rings and associative algebras, but also to studies such as that of Lie groups [9] Address correspondence to Prof. Liangyun Chen, School of Mathematics and Statistics, Northeast Normal University, Changchun 130024, China; E-mail: [email protected]. Supported by NNSF of China (No. 11171055 and 11471090) and China Postdoctoral Science Foun- dation (No. 2015M581989). 1 and operator algebras( [8], [11], [13]). Triple derivation of Lie algebra is apparently a gen- eralization of derivation, and is an analogy of triple derivation of associative algebra and of Jordan algebra. It was first introduced independently [9] by Muller where it was called prederivation. It can be easily checked that, for any Lie algebra, every derivation is a Lie triple derivation, but the converse does not always hold [14]. The relations of homomorphisms, anti-homomorphisms, Jordan homomorphisms, Lie homomorphisms, and Lie triple homomorphisms became attractive questions. Bresar gave a characterization of Lie triple isomorphisms associated to certain associative algebras [1]. Jacobson and Rickart gave some conditions such that every Jordan homomorphism of a ring is either a homomorphism or an anti-homomorphism [10]. Similar problems arose in the study of operator algebras( [6], [7]). An analogous result was proved for more general perfect Lie algebras [15]. Lie superalgebras are the natural generalization of Lie algebras, and have important applications both in mathematics and physics. Lie superalgebras are also interesting from a purely mathematical point of view. The aim of this article is to generalize some results in [14]and [15]to the triple derivations and triple homomorphisms of Lie superalgebras. Throughout the following sections, L always denote a Lie superalgebra over a com- mutative ring R with 1. A Lie superalgebra L is called perfect if the derived subalgebra [L, L] = L. For a subset S of L, denote by CL(S) the centralizer of S in L, and the center of L is denoted by Z(L). L is called centerless if Z(L) = 0. Der(L) is the derivation algebra of L. Some definitions needed in this paper are as follows. Definition 1.1. [5] L = L¯0 ⊕ L¯1 is a Z2-graded algebra over a commutative ring R with 1, we call L a Lie superalgebra if the multiplication [ , ] satisfies the following identities: (1) [x, y] = −(−1)xy[y, x]; (2) [x, [y, z]] = [[x, y], z] + (−1)xy[y, [x, z]]; (graded skew − symmetry) (graded Jacobi identity) where x, y, z ∈ hg(L), hg(L) denotes the set of all Z2-homogeneous elements of L. If x occurs in some expression in this paper, we always regard x as a Z2-homogeneous element and x as the Z2-degree of x. Definition 1.2. For a subset S of L, the enveloping Lie superalgebra of S is the Lie subalgebra of L generated by S. A Lie superalgebra is called indecomposable if it cannot be written as a direct sum of two nontrivial ideals. Definition 1.3. An endomorphism D of an R-module L is called a triple derivation of L, if ∀x, y, z ∈ L, D satisfies D([[x, y], z]) = [[D(x), y], z] + (−1)Dx[[x, D(y)], z] + (−1)D(x+y)[[x, y], D(z)]. Denote by TDer(L) the set of all triple derivations of L. It is not difficult to show that TDer(L) is a Lie superalgebra under the usual bracket of endomorphisms of R-module (See Lemma 2.1 ). 2 Definition 1.4. Let L, L f : L → L ′ is called: ′ be two Lie superalgebras over R. An even R-linear mapping (i) a homomorphism from L to L (ii) an anti-homomorphism if it satisfies f ([x, y]) = (−1)xy[f (y), f (x)],∀x, y ∈ ′ if it satisfies f ([x, y]) = [f (x), f (y)], ∀x, y ∈ L. hg(L). (iii) a triple homomorphism if it satisfies f ([x, [y, z]]) = [f (x), [f (y), f (z)]], ∀x, y, z ∈ L. ′ is called a Definition 1.5. Let L, L direct sum of g1 and g2, if g = g1 + g2 and there exist ideals I1, I2 of the enveloping Lie superalgebra of g(L) such that I1 ∩ I2 = 0, and g1(L) ⊆ I1, g2(L) ⊆ I2. ′ be Lie superalgebras. A mapping g : L → L The main results of this article are the following two theorems. Theorem 1.1 Let L be a Lie superalgebra over a commutative ring R. If 1 perfect and has zero center, then we have that: 2 ∈ R, L is (1) TDer(L) = Der(L); (2) TDer(Der(L)) = ad(Der(L)). Theorem 1.2 Suppose that R is a commutative ring with 1, and 2 is invertible in R. ′, and Let L and L M the enveloping Lie superalgebra of f (L). Assume the following statements: ′ be Lie superalgebras over R, f a triple homomorphism from L to L (1) L is perfect; (2) M is centerless and can be decomposed into a direct sum of indecomposable ideals. Then f is either a homomorphism or an anti-homomorphism or a direct sum of a homomorphism and an anti-homomorphism. 2 Triple Derivations of Perfect Lie Superalgebras We proceed to prove Theorem 1.1 by the following lemmas. Lemma 2.1. For any Lie superalgerba L, TDer(L) is closed under the usual Lie bracket. Proof. Let D1, D2 ∈ TDer(L), x1, x2 ∈ hg(L), x3 ∈ L. By the definition of triple 3 derivation, we have D1D2([[x1, x2], x3]) =D1([[D2(x1), x2], x3] + (−1)D2x1[[x1, D2(x2], x3] + (−1)D2(x1+x2)[[x1, x2], D2(x3)]) =[D1D2(x1), x2], x3] + (−1)D2(x1+x2)(−1)D1(x1+x2)[[x1, x2], D1D2(x3)] + (−1)D2x1[[D1(x1), D2(x2)], x3] + (−1)D1(D2+x1)[[D2(x1), D1(x2)], x3] + (−1)D1(D2+x1+x2)[[D2(x1), x2], D1(x3)] + (−1)D2x1(−1)D1x1[[x1, D1D2(x2)], x3] + (−1)D2(x1+x2)(−1)D1x1[[x1, D1(x2)], D2(x3)] + (−1)D2(x1+x2)[[D1(x1), x2], D2(x3)] + (−1)D2x1(−1)D1(D2+x1+x2)[[x1, D2(x2)], D1(x3)], and D2D1([[x1, x2], x3]) =D2([[D1(x1), x2], x3] + (−1)D1x1[[x1, D1(x2], x3] + (−1)D1(x1+x2)[[x1, x2], D1(x3)]) =[[D2D1(x1), x2], x3] + (−1)D1(x1+x2)(−1)D2(x1+x2)[[x1, x2], D2D1(x3)] + (−1)D1x1[[D2(x1), D1(x2)], x3] + (−1)D2(D1+x1)[[D1(x1), D2(x2)], x3] + (−1)D2(D1+x1+x2)[[D1(x1), x2], D2(x3)] + (−1)D1x1(−1)D2x1[[x1, D2D1(x2)], x3] + (−1)D1(x1+x2)(−1)D2x1[[x1, D2(x2)], D1(x3)] + (−1)D1(x1+x2)[[D2(x1), x2], D1(x3)] + (−1)D1x1(−1)D2(D1+x1+x2)[[x1, D1(x2)], D2(x3)]. Then from easy computation we have [D1, D2]([[x1, x2], x3]) =D1D2([[x1, x2], x3]) − (−1)D1D2D2D1([[x1, x2], x3]) =[[[D1, D2](x1), x2], x3] + (−1)x1(D1+D2)[[x1, [D1, D2](x2)], x3] + (−1)(D1+D2)(x1+x2)[[x1, x2], [D1, D2](x3)]. Hence, [D1, D2] ∈ TDer(L). The lemma is proved. Clearly, ad(L), Der(L) are all subalgebras of TDer(L). Since L is perfect, every element x ∈ hg(L) can be written as a finite sum of Lie brackets, i.e., there exists a finite index set I such that x = P i∈I [xi1, xi2], for some xi1, xi2 ∈ L. In this article, we always put P in place of P i∈I for convenience. xi1+xi2=x xi1+xi2=x Moreover, we have the following lemma. Lemma 2.2. If L is perfect, then ad(L) is an ideal of Lie superalgebra TDer(L). Proof. Let D ∈ TDer(L), x ∈ hg(L). ∀z ∈ L, we have [D, adx](z) = Dadx(z) − (−1)Dxadx(D(z)) 4 = D[x, z] − (−1)Dx[x, D(z)] = D([X[xi1, xi2], z]) − (−1)Dx[X[xi1, xi2], D(z)] = X D([[xi1, xi2], z]) − X(−1)Dx[[xi1, xi2], D(z)] = X[[D(xi1), xi2], z] + X(−1)Dxi1[[xi1, D(xi2)], z] + X(−1)Dx[[xi1, xi2], D(z)] − X(−1)Dx[[xi1, xi2], D(z)] = ad(X[D(xi1), xi2] + X(−1)Dxi1[xi1, D(xi2)])(z). By the arbitrariness of z, [D, adx] is an inner derivation. Hence, ad(L) is an ideal of TDer(L). The lemma holds. Lemma 2.3. If L is a perfect Lie superalgebra with zero center, then there exits an R-module homomorphism δ : TDer(L) → End(L), δ(D) = δD such that ∀x ∈ L, D ∈ TDer(L), one has [D, adx] = adδD(x). Proof. From Lemma 2.2, if L is perfect and L has zero center, D ∈ TDer(L), we can define a module endomorphism δD on L such that for ∀x = P[xi1, xi2] ∈ hg(L), δD(x) = X([D(xi1), xi2] + (−1)Dxi1[xi1, D(xi2)]). In fact, the definition is independent of the form of expression of x. For proving it, let α = X([D(xi1), xi2] + (−1)Dxi1[xi1, D(xi2)]). If there exists another finite index set J and yji ∈ L such that x can be expressed in the form x = P j∈J when necessary, let [yj1, yj2], we also put P in place of P j∈J yj1+yj2=x yi1+yi2=x β = X([D(yj1), yj2] + (−1)Dyj1[yj1, D(yj2)]). Since D ∈ TDer(L), ∀z ∈ L, we have [α, z] = X([[D(xi1), xi2], z] + (−1)Dxi1[[xi1, D(xi2)], z]) = X(D([[xi1, xi2], z]) − (−1)Dx[[xi1, xi2], D(z)]) = D([x, z]) − (−1)Dx[x, D(z)] = X(D([[yj1, yj2], z]) − (−1)Dx[[yj1, yj2], D(z)]) = X([[D(yj1), yj2], z] + (−1)Dyj1[[yj1, D(yj2)], z]) = [β, z]. Hence, [α − β, z] = 0. This means that α − β ∈ Z(L). Since Z(L) = 0, α = β. Hence, δD is well-defined. The rests of the lemma follow from the proof of Lemma 2.2. 5 Using the mapping δD and the proof of Lemma 2.2, we have the following lemmas. Lemma 2.4. If L is a perfect Lie superalgebra with zero center, then for ∀D ∈ TDer(L), δD ∈ Der(L). Proof. Suppose D ∈ TDer(L), x ∈ hg(L), y ∈ L. Then [D, ad([x, y])] = adδD([x, y]). In other hand, [D, ad([x, y])] = [D, [adx, ady]] = [[D, adx], ady] + (−1)Dx[adx, [D, ady]] = [adδD(x), ady] + (−1)Dx[adx, adδD(y)] = ad([δD(x), y] + (−1)Dx[x, δD(y)]). Hence, adδD([x, y]) = ad([δD(x), y] + (−1)Dx[x, δD(y)]). Since Z(L) = 0, δD([x, y] = [δD(x), y] + (−1)Dx[x, δD(y)]. By the arbitrariness of x, y, δD ∈ Der(L). Lemma 2.5. If the base ring R contains 1 2 , L is perfect, then the centralizer of ad(L) in TDer(L)is trivial, i.e., CTDerL(ad(L)) = 0. In particular, the center of TDer(L) is zero. Proof. Let D ∈ CTDerL(ad(L)). Then for ∀x ∈ L, [D, adx] = 0. Hence, for ∀x, y ∈ hg(L), D([x, y]) − (−1)Dx[x, D(y)] = [D, adx](y) = 0. Thus, D([x, y]) = [D(x), y] = (−1)Dx[x, D(y)]. For x1, x2, x3 ∈ hg(L), we always have that D([[x1, x2], x3]) = (−1)D(x1+x2)[[x1, x2], D(x3)] = (−1)Dx1[[x1, D(x2)], x3] = [[D(x1), x2], x3]. Therefore, D([[x1, x2], x3]) =[[D(x1), x2], x3] + (−1)Dx1[[x1, D(x2)], x3] + (−1)D(x1+x2)[[x1, x2], D(x3)] =3D([[x1, x2], x3]). Hence, 2D([[x1, x2], x3]) = 0. Because 1 2 ∈ R, D([[x1, x2], x3]) = 0. Since L is perfect, every element of L can be expressed as the linear combination of elements of the form [[x1, x2], x3], we have that D = 0. This completes the proof. The next lemma is well known. Lemma 2.6. [5] For all Lie superalgebra L, if x ∈ L, D ∈ Der(L), then [D, adx] = ad(D(x)). Now we can prove the first conclusion of the theorem. 6 Lemma 2.7. If the base ring R contains 1 TDer(L) = Der(L). 2 , L is perfect and has trivial center, then Proof. Suppose x ∈ L, D ∈ TDer(L). By Lemma 2.3, [D, adx] = adδD(x). By Lemma 2.4 and Lemma 2.6, adδD(x) = [δD, adx]. Hence, D − δD ∈ CTDer(L)(ad(L)). By Lemma 2.5, D − δD = 0, i.e., D = δD ∈ Der(L). Hence, TDer(L) ⊆ Der(L). The lemma follows from Lemma 2.4. The remainder of the chapter aims to prove the second conclusion of theorem 1.1. Lemma 2.8. If L is a perfect Lie superalgebra, D ∈ TDer(Der(L)), then D(ad(L)) ⊆ ad(L). Proof. Since L is perfect, we have D(adx) = X D(ad[[xi1, xi2], xi3]) = X D([[adxi1, adxi2], adxi3]) = X([[D(adxi1), adxi2], adxi3] + (−1)Dxi1[[adxi1, D(adxi2)], adxi3] + (−1)D(xi1+xi2)[[adxi1, adxi2], D(adxi3)]. Hence, D(adx) ∈ ad(L). The lemma holds thanks to Lemma 2.2. Lemma 2.9. Suppose that R is the base ring containing 1 with zero center, D ∈ TDer(Der(L)). If D(ad(L)) = 0, then D = 0. 2, L is a perfect Lie superalgebra Proof. For ∀d ∈ Der(L), x ∈ hg(L), since L is perfect, x = P[xi1, xi2]. We have that [adx, D(d)] = [X[adxi1, adxi2], D(d)] = X((−1)DxD([[adxi1, adxi2], d]) − (−1)Dx[[D(adxi1), adxi2], d] − (−1)Dx(−1)Dxi1[[adxi1, D(adxi2)], d]). By Lemma 2.2, [adx, d] ∈ ad(L), so D([adx, d]) = 0. Hence, [adx, D(d)] = 0. Therefore, D(d) ∈ CTDer(L)(ad(L)). By Lemma 2.5, D(d) = 0. Hence, D = 0. The lemma holds. Lemma 2.10. Let L is a Lie superalgebra over commutative ring R. Suppose that 1 2 ∈ R, L is perfect and has zero center. If D ∈ TDer(Der(L)), then there exists d ∈ Der(L) such that for ∀x ∈ L, D(adx) = ad(d(x)). Proof. For ∀D ∈ TDer(Der(L)), x ∈ L, by Lemma 2.8, D(adx) ∈ ad(L). Let y ∈ L and D(adx) = ady. Since the center Z(L) is trivial, such y is unique. Clearly, the map d : L → L given by d(x) = y is a R-module endomorphism of L. Let x1, x2 ∈ hg(L), x3 ∈ L. We have add([[x1, x2], x3]) = D(ad([[x1, x2], x3])) 7 = D([[adx1, adx2], adx3]) = [[D(adx1), adx2], adx3] + (−1)Dx1[[adx1, D(adx2)], adx3] + (−1)D(x1+x2)[[adx1, adx2], D(adx3)]. = [[ad(d(x1)), adx2], adx3] + (−1)Dx1[[adx1, ad(d(x2))], adx3] + (−1)D(x1+x2)[[adx1, adx2], ad(d(x3))] = ad([[d(x1), x2], x3] + (−1)Dx1[[x1, d(x2)], x3] + (−1)D(x1+x2)[[x1, x2], d(x3)]). Since Z(L) = 0, d([[x1, x2], x3]) = [[d(x1), x2], x3] + (−1)Dx1[[x1, d(x2)], x3] + (−1)D(x1+x2)[[x1, x2], d(x3)]. That is to say, d ∈ TDer(L). By Lemma 2.7, d ∈ Der(L). Proof of Theorem 1.1. By Lemma 2.7, it remains only to prove the second asser- tion. By Lemma 2.10, for ∀D ∈ TDer(Der(L)), x ∈ L, there exists d ∈ Der(L) such that for ∀x ∈ L, D(adx) = ad(d(x)). Thanks to Lemma 2.6, ad(d(x)) = [d, adx]. Hence, D(adx) = ad(d(x)) = [d, adx] = ad(d)(adx). Thus, (D − ad(d))(adx) = 0. By Lemma 2.9, D = ad(d). Therefore, TDer(Der(L)) = ad(Der(L)). The theorem holds. Remark 2.11. [14] The condition 1 2 is necessary. For example, if the base ring is field F of characteristic 2 and L is not abelian, then the identity map is a triple derivation but not a derivation. 3 Triple Homomorphisms of Perfect Lie Superalge- bras Throughout this section, L and L ′ are Lie superalgebras over R, f is a triple homo- ′, and M is the enveloping Lie superalgebra of f (L). We always morphism from L to L assume that L is perfect and that M is centerless and can be decomposed into a direct sum of indecomposable ideals. We proceed to prove the theorem by a series of lemmas. Lemma 3.1. There exists an even R-linear mapping δf : L → L [xi1, xi2] (xi1, xi2 ∈ L), δf (x) = P i∈I with x = P i∈I xi1+xi2=x xi1+xi2=x ′ such that for ∀x ∈ hg(L) [f (xi1), f (xi2)]. Proof. It is sufficient to prove that P[f (xi1), f (xi2)] is independent of the expression of x. Suppose that x = P[xi1, xi2] = P[yj1, yj2]. 8 Let α = X[f (xi1), f (xi2)], β = X[f (yj1), f (yj2)]. For ∀z ∈ L, we have that [f (z), α − β] = [f (z), X[f (xi1), f (xi2)] − X[f (yj1), f (yj2)]] = X[f (z), [f (xi1), f (xi2)]] − X[f (z), [f (yj1), f (yj2)]] = f ([z, X[xi1, xi2]] − [z, X[yj1, yj2]]) = f ([z, x] − [z, x]) = 0. It follows α − β ∈ Z(M) and thus α = β since M is centerless. This completes the proof. Lemma 3.2. Let δf be the mapping in Lemma 3.1. Then, for ∀x ∈ L, we have that f adx = adδf (x)f. Proof. Let x = P[xi1, xi2] ∈ L. For ∀z ∈ L, we have f adx(z) = f ([x, z]) = X f ([[xi1, xi2], z]) = X[[f (xi1), f (xi2)], f (z)] = [X[f (xi1), f (xi2)], f (z)] = [δf (x), f (z)] = adδf (x)f (z). Thus f adx = adδf (x)f , and the lemma holds. Lemma 3.3. The mapping δf is a homomorphism of Lie superalgebras. Proof. For ∀x, y ∈ hg(L), z ∈ L, it follows from Lemmas 3.1 and 3.2 that [δf ([x, y]) − [δf (x), δf (y)], f (z)] = [δf ([x, y]), f (z)] − [δf (x), [δf (y), f (z)]] + (−1)xy[δf (y), [δf (x), f (z)]] = [[f (x), f (y)], f (z)] − [δf (x), adδf (y)f (z)]] + (−1)xy[δf (y), adδf (x)f (z)] = [[f (x), f (y)], f (z)] − [δf (x), f ([y, z])] + (−1)xy[δf (y), f ([x, z])] = [[f (x), f (y)], f (z)] − adδf (x)f ([y, z]) + (−1)xyadδf (y)f ([x, z]) = f ([[x, y], z]) − f ([x, [y, z]]) + (−1)xyf ([y, [x, z]]) = f ([[x, y], z] − [x, [y, z]] + (−1)xy[y, [x, z]]) = 0. The last equality is due to the Jacobi identity. Since M is the enveloping Lie superalgebra of f (L) and z is arbitrary in L, δf ([x, y]) − [δf (x), δf (y)] ∈ Z(M). Hence, δf ([x, y]) = [δf (x), δf (y)] since M is centerless. Therefore, the lemma follows from the arbitrariness of x, y ∈ L. 9 Lemma 3.4. Denote M + = Im(f + δf ), M − = Im(f − δf ). Then, M + and M − are both ideals of M. Proof. Similar with [15]. Lemma 3.5. [M +, M −] = 0. Proof. Take x, y, z ∈ hg(L), we have that [[f (x) + δf (x), f (y) − δf (y)], f (z)] = [[f (x), f (y)], f (z)] − [[f (x), δf (y)], f (z)] + [[δf (x), f (y)], f (z)] − [[δf (x), δf (y)], f (z)] = f ([[x, y], z]) + (−1)xy[adδf (y)f (x), f (z)] + [adδf (x)f (y), f (z)] − [δf (x), [δf (y), f (z)]] + (−1)xy[δf (y), adδf (x)f (z)] = f ([[x, y], z]) + (−1)xy[f ([y, x]), f (z)] + [f ([x, y]), f (z)] − [δf (x), adδf (y)f (z)] + (−1)xy[δf (y), f ([x, z])] = f ([[x, y], z]) + (−1)xy[f ([y, x]), f (z)] + [f ([x, y]), f (z)] − [δf (x), f ([y, z])] − (−1)xy(−1)y(x+z)[f ([x, z]), δf (y)] = f ([[x, y], z]) − [δf (x), f ([y, z])] − (−1)yz[f ([x, z]), δf (y)] = f ([[x, y], z]) − f ([x, [y, z]]) + (−1)yz(−1)y(x+z)f ([y, [x, z]]) = f ([[x, y], z] − [x, [y, z]] + (−1)xy[y, [x, z]]) = 0. Therefore, [f (x) + δf (x), f (y) − δf (y)] ∈ Z(M). Since Z(M) = 0, we have that [f (x) + δf (x), f (y) − δf (y)] = 0. The lemma follows. Lemma 3.6. M + ∩ M − = 0. Proof. Similar with [15]. Lemma 3.7. If M can not be decomposed into a direct sum of two nontrivial ideals. Then f is either a homomorphism or an anti-homomorphism of Lie superalgebras. 2(f (x) + δf (x)), m− = 1 Proof. For ∀x ∈ L, let m+ = 1 2(f (x) − δf (x)). Then m+ ∈ M +, m− ∈ M − and f (x) = m+ + m−. Hence, f (L) ⊆ M + + M −. Therefore, M ⊆ M + + M −. By Lemma 3.6, M = M + ⊕ M −. Since M cannot be decomposed into direct sum of two nontrivial ideals, either M + or M − must be trivial. If M + is trivial i.e. (f + δf )([x, y]) = 0, then f ([x, y]) = −δf ([x, y]) = −[f (x), f (y)] = (−1)xy[f (y), f (x)]. So f is an anti-homomorphism. If M − is trivial i.e. (f − δf )([x, y]) = 0, then f ([x, y]) = δf ([x, y]) = [f (x), f (y)]. So f is a homomorphism. Hence the lemma follows. 10 Proof of Theorem 1.2. By Lemma 3.7, it remains to prove the theorem in case M is decomposable. By the assumptions, M can be written as the sum M = M1 ⊕ M2 ⊕ ... ⊕ Ms, where each Mi is an indecomposable ideal of M. Since M is centerless, each Mi is also centerless (See Lemma 3.1 in [3]). Let pi be the projection of M into Mi. Then, f = Ps i=1 pif and pif is a triple homomorphisms from L to Mi, and Mi is the envelop- ing Lie superalgebras of pif (L) for i = 1, 2...s. Since each Mi is indecomposable, by Lemma 3.7, pif is either a homomorphism or an anti-homomorphism from L to Mi. Let P = {i pif is a homomorphism}, Q is the complementary set of P in the set {1, 2..., s}. Let M1 = Pi∈P Mi, M2 = Pi∈Q Mi. Let f1 = Pi∈P pif, f1 = Pi∈Q pif . It can be checked by direct verification that M = M1 ⊕ M2, [M1, M2] = 0. f = f1 + f2, f1 is a homomorphism and f2 is an anti-homomorphism of Lie superalgebras. The theorem is proved. Acknowledgements The authors would like to thank the referee for valuable comments and suggestions on this article. References [1] Bresar, M. (1993). Commuting traces of biadditive mappings, commutativity- preserving mappings and Lie mappings. Trans. Amer. Math. Soc. 335:525-546. [2] Beidar, K. I., Chebotar, M. A. (2001). On Lie derivations of Lie ideals of Prime algebras. Israel J. Math. 123:131-148. [3] Chun, J. H., Lee, J. S. (1996). On Complete Lie Superalgebras. Comm. Korean Math. Soc. 11:323-334. [4] Herstein, I. N., Jordan, L. (1961). Structure in simple, associative rings. Bulletin of the Amer. Math. Soc. 67:517-531. [5] Kac V. G.. (1977). Lie Superalgebras. Advances in mathematics 26:8-96. [6] Miers, C. R. (1971). Lie homomorphisms of operator algebras. Pacific J. Math. 38: 717-737. [7] Miers, C. R. (1976). Lie triple homomorphisms into von Neumann algebras. Proc. Amer. Math. Soc. 58:169-172. [8] Miers, C. R. (1978). Lie triple derivations of Von Neumann algebras. Proc. Amer. Math. Soc. 71:57-61. [9] Muller, D. (1989). Isometries of bi-invariant pseudo-Riemannian metries on Lie groups. Geom. Deicata 29:65-96. 11 [10] Jacobson, N., Rickart, C. E. (1950). Jordan homomorphisms of rings. Trans. Amer. Math. Soc. 69:479-502. [11] Ji, P. S., Wang, L. (2005). Lie triple derivations of TUHF algebras. Linear Algebra Applications 403:399-408. [12] Swain, G. A. (1996). Lie derivations of the skew elements of prime ring with involu- tion. J. Algebra 184:679-704. [13] Zhang, J. H., Wu, B. W., Cao, H. X. (2006). Lie triple derivations of nest algebras. Linear Algebra Applications 416:559-567. [14] Zhou, J. H.. (2013). Triple Derivations of Perfect Lie Algebras. Comm. Algebra 41:1647-1654. [15] Zhou, J. H.. (2014). Triple Homomorphisms of Perfect Lie Algebras. Comm. Algebra 42:3724-3730. 12
1312.4023	1	1312	2013-12-14T09:42:26	On A Class of Lifting Modules	[ "math.RA" ]	In this paper, we introduce principally $\delta$-lifting modules which are analogous to $\delta$-lifting modules and principally $\delta$-semiperfect modules as a generalization of $\delta$-semiperfect modules and investigate their properties.	math.RA	math	ON A CLASS OF LIFTING MODULES HATICE INANKIL, SAIT HALICIO GLU, AND ABDULLAH HARMANCI Abstract. In this paper, we introduce principally δ-lifting modules which are analogous to δ-lifting modules and principally δ-semiperfect modules as a generalization of δ-semiperfect modules and investigate their properties. 1. Introduction Throughout this paper all rings have an identity, all modules considered are unital right modules. Let M be a module and N, P be submodules of M . We call P a supplement of N in M if M = P + N and P ∩ N is small in P . A module M is called supplemented if every submodule of M has a supplement in M . A module M is called lifting if, for all N ≤ M , there exists a decomposition M = A ⊕ B such that A ≤ N and N ∩ B is small in M . Supplemented and lifting modules have been discussed by several authors (see [2, 4, 6]) and these modules are useful in characterizing semiperfect and right perfect rings (see [4, 7]). In this note, we study and investigate principally δ-lifting modules and princi- pally δ-semiperfect modules. A module M is called principally δ-lifting if for each cyclic submodule has the δ-lifting property, i.e., for each m ∈ M , M has a decom- position M = A ⊕ B with A ≤ mR and mR ∩ B is δ-small in B, where B is called a δ-supplement of mR. A module M is called principally δ-semiperfect if, for each m ∈ M , M/mR has a projective δ-cover. We prove that if M1 is semisimple, M2 is principally δ-lifting, M1 and M2 are relatively projective, then M = M1 ⊕ M2 is a principally δ-lifting module. Among others we also prove that for a principally δ-semiperfect module M , M is principally δ-supplemented, each factor module of M is principally δ-semiperfect, hence any homomorphic image and any direct sum- mand of M is principally δ-semiperfect. As an application, for a projective module M , it is shown that M is principally δ-semiperfect if and only if it is principally δ-lifting, and therefore a ring R is principally δ-semiperfect if and only if it is principally δ-lifting. In section 2, we give some properties of δ-small submodules that we use in the paper, and in section 3, principally δ-lifting modules are introduced and various properties of principally δ-lifting and δ-supplemented modules are obtained. In section 4, principally δ-semiperfect modules are defined and characterized in terms of principally δ-lifting modules. In what follows, by Z, Q, Zn and Z/Zn we denote, respectively, integers, ra- tional numbers, the ring of integers and the Z-module of integers modulo n. For unexplained concepts and notations, we refer the reader to [1, 4]. 2000 Mathematics Subject Classification. 16U80. Key words and phrases. lifting modules, δ-lifting modules, semiperfect modules, δ-semiperfect modules. 1 2 HATICE INANKIL, SAIT HALICIO GLU, AND ABDULLAH HARMANCI 2. δ-Small Submodules Following Zhou [9], a submodule N of a module M is called a δ-small submodule if, whenever M = N + X with M/X singular, we have M = X. We begin by stating the next lemma which is contained in [9, Lemma 1.2 and 1.3]. Lemma 2.1. Let M be a module. Then we have the following. (1) If N is δ-small in M and M = X + N , then M = X ⊕ Y for a projective semisimple submodule Y with Y ⊆ N . (2) If K is δ-small in M and f : M → N is a homomorphism, then f (K) is δ-small in N . In particular, if K is δ-small in M ⊆ N , then K is δ-small in N . (3) Let K1 ⊆ M1 ⊆ M , K2 ⊆ M2 ⊆ M and M = M1 ⊕ M2. Then K1 ⊕ K2 is δ-small in M1 ⊕ M2 if and only if K1 is δ-small in M1 and K2is δ-small in M2. (4) Let N , K be submodules of M with K is δ-small in M and N ≤ K. Then N is also δ-small in M . Lemma 2.2. Let M be a module and m ∈ M . Then the following are equivalent. (1) mR is not δ-small in M . (2) There is a maximal submodule N of M such that m 6∈ N and M/N singular. Proof. (1) ⇒ (2) Let Γ := {B ≤ M B 6= M, mR + B = M , M/B singular}. Since mR is not δ-small in M , there exists a proper submodule B of M such that mR + B = M and M/B singular. So Γ is non empty. Let Λ be a nonempty totally ordered subset of Γ and B0 := ∪B∈ΛB. If m is in B0 then there is a B ∈ Λ with m ∈ B. Then B = mR + B = M which is a contraction. So we have m /∈ B0 and B0 6= M . Since mR + B0 = M and M/B0 singular, B0 is upper bound in Γ. By Zorn's Lemma, Γ has a maximal element, say N . If N is a maximal submodule of M there is nothing to do. Assume that there exists a submodule K containing N properly. Since N is maximal in Γ, K is not in Γ. Since M = mR + N and N ≤ K, so M = mR + K. M/K is singular as a homomorphic image of singular module M/N . Hence K must belong to the Γ. This is the required contradiction. (2) ⇒ (1) Let N be a maximal submodule with m ∈ M \ N and M/N singular. We have M = mR + N . Then mR is not δ-small in M . (cid:3) Let A and B be submodules of M with A ≤ B. A is called a δ-cosmall submodule of B in M if B/A is δ-small in M/A. Let A be a submodule of M . A is called a δ-coclosed submodule in M if A has no proper δ-cosmall submodules in M . A submodule A is called δ-coclosure of B in M if A is δ-coclosed submodule of M and it is δ-cosmall submodule of B. Equivalently, for any submodule C ≤ A with A/C is δ-small in M/C implies C = A and B/A is δ-small in M/A. Note that δ-coclosed submodules need not always exist. ON A CLASS OF LIFTING MODULES 3 Lemma 2.3. Let A and B be submodules of M with A ≤ B. Then we have: (1) A is δ-cosmall submodule of B in M if and only if M = A + L for any submodule L of M with M = B + L and M/L singular. (2) If A is δ-small and B is δ-coclosed in M , then A is δ-small in B. Proof. (1) Necessity: Let M = B + L and M/L be singular. We have M/A = B/A + (L + A)/A and M/(L + A) is singular as homomorphic image of singular module M/L. Since B/A is δ-small, M/A = (L + A)/A or M = L + A. Sufficiency: Let M/A = B/A + K/A and M/K singular. Then M = B + K. By hypothesis, M = A + K and so M = K. Hence A is δ-cosmall submodule of B in M . (2) Assume that A is δ-small submodule of M and B is δ-coclosed in M . Let B = A + K with B/K singular. Since B is δ-coclosed in M , to complete the proof, by part (1) it suffices to show that K is δ-small submodule of B in M. Let M = B + L with M/L singular. By assumption, M = A + K + L = K + L since M/(K + L) is singular. By (1), K is δ-small submodule of B in M . (cid:3) Lemma 2.4. Let A, B and C be submodules of M with M = A + C and A ⊆ B. If B ∩ C is a δ-small submodule of M , then A is a δ-cosmall submodule of B in M . Proof. Let M/A = B/A + L/A with M/L singular. We have M = B + L and B = A + (B ∩ C). Then M = A + (B ∩ C) + L = (B ∩ C) + L. Hence M = L since B ∩ K is δ-small in M and M/L is singular. Hence B/A is δ-small in M/A. Thus A is δ-cosmall submodule of B in M . (cid:3) 3. Principally δ-Lifting Modules In this section, we study and investigate some properties of principally δ-lifting modules. The following definition is motivated by [9, Lemma 3.4] and Lemma 3.4. Definition 3.1. A module M is called finitely δ-lifting if for any finitely generated submodule A of M has the δ-lifting property, that is, there is a decomposition M = N ⊕ S with N ≤ A and A ∩ S is δ-small in S. In this case A ∩ S is δ-small in S if and only if A ∩ S is δ-small in M . A module M is called principally δ-lifting if for each cyclic submodule has the principally δ-lifting property, i.e., for each m ∈ M , M has a decomposition M = A ⊕ B with A ≤ mR and mR ∩ B is δ-small in B. Example 3.2. Every submodule of any semisimple module satisfies principally δ-lifting property. Example 3.3. Let p be a prime integer and n any positive integer. Then the Z-module M = Z/Zpn is a principally δ-lifting module. 4 HATICE INANKIL, SAIT HALICIO GLU, AND ABDULLAH HARMANCI Lemma 3.4 is proved in [7] and [9]. Lemma 3.4. The following are equivalent for a module M . (1) M is finitely δ-lifting. (2) M is principally δ-lifting. Let M be a module and N a submodule of M . A submodule L is called a δ-supplement of N in M if M = N + L and N ∩ L is δ-small in L(therefore in M ). Proposition 3.5. Let M be a principally δ-lifting module. The we have: (1) Every direct summand of M is a principally δ-lifting module. (2) Every cyclic submodule C of M has a δ-supplement S which is a direct summand, and C contains a complementary summand of S in M . Proof. (1) Let K be a direct summand of M and k ∈ K. Then M has a decom- position M = N ⊕ S with N ≤ kR and kR ∩ S is δ-small in M . It follows that K = N ⊕ (K ∩ S), and kR ∩ (K ∩ S) ≤ kR ∩ S is δ-small in M and so kR ∩ (K ∩ S) is δ-small in K. Therefore K is a principally δ-lifting module. (2) Assume that M is a principally δ-lifting module and C is a cyclic submodule of M . Then we have M = N ⊕ S, where N ≤ C and C ∩ S is δ-small in M . Hence M = N + S ≤ C + S ≤ M , we have M = C + S. Since S is direct summand and C ∩ S is δ-small in M , C ∩ S is δ-small in S. Therefore S is a δ-supplement of C in M . (cid:3) Theorem 3.6. The following are equivalent for a module M . (1) M is a principally δ-lifting module. (2) Every cyclic submodule C of M can be written as C = N ⊕ S, where N is direct summand and S is δ-small in M . (3) For every cyclic submodule C of M , there is a direct summand A of M with A ≤ C and C/A is δ-small in M/A. (4) Every cyclic submodule C of M has a δ-supplement K in M such that C ∩K is a direct summand in C. (5) For every cyclic submodule C of M , there is an idempotent e ∈ End(M ) with eM ≤ C and (1 − e)C is δ-small in (1 − e)M . (6) For each m ∈ M , there exist ideals I and J of R such that mR = mI ⊕ mJ, where mI is direct summand of M and mJ is δ-small in M . Proof. (1)⇒(2) Let C be a cyclic submodule of M . By hypothesis there exist N and S submodules of M such that N ≤ C, C ∩ S is δ-small in M and M = N ⊕ S. Then we have C = N ⊕ (C ∩ S). ON A CLASS OF LIFTING MODULES 5 (2) ⇒ (3) Let C be a cyclic submodule of M . By hypothesis, C = N ⊕ S, where N is direct summand and S is δ-small in M . Let π : M → M/N be the natural projection. Since S is δ-small in M , we have π(S) is δ-small in M/N . Since π(S) ∼= S ∼= C/N , C/N is δ-small in M/N . (3) ⇒ (4) Let C be a cyclic submodule of M . By hypothesis, there is a direct summand A ≤ M with A ≤ C and C/A is δ-small in M/A. Let M = A ⊕ A′ . Hence C = A ⊕ (A′ ∩ C). Let σ : M/A → A′ denote the obvious isomorphism. Then σ(C/A) = A′ ∩ C is δ-small in A′. (4) ⇒ (5) Let C be any cyclic submodule of M and K ≤ M such that C ∩K is direct summand of C, M = C + K and C ∩ K is δ-small in K . So C = (C ∩ K) ⊕ X for some X ≤ C . Then M = X +(C ∩K)+K = X ⊕K . Let e : M → X ; e(x+k) = x and (1 − e) : M → K ; e(x + k) = k are projection maps. e(M ) ≤ X ≤ C and (1 − e)C = C ∩ (1 − e)M = C ∩ K is δ-small in (1 − e)M . (5) ⇒ (6) Let mR be any cyclic submodule of M . By hypothesis, there exists an idempotent e ∈ End(M ) such that eM ≤ mR, M = eM ⊕(1−e)M and (1−e)mR is δ-small in (1 − e)M . Note that (mR)∩((1 − e)M ) = (1 − e)mR ( for if m = em1 + y, where em1 ∈ eM , y ∈ (mR) ∩ ((1 − e)M ). Then (1 − e)m = em1 + (1 − e)y = y and so (1 − e)mR ≤ (mR) ∩ ((1 − e)M ). Let mr = (1 − e)m′ ∈ (mR) ∩ ((1 − e)M ). Then mr = (1 − e)mr ∈ (1 − e)mR. So (mR) ∩ ((1 − e)M ) ≤ (1 − e)mR. Thus (mR) ∩ ((1 − e)M ) = (1 − e)mR ). So mR = eM ⊕ (1 − e)mR. Let I = {r ∈ R : mr ∈ eM } and J = {t ∈ R : mt ∈ (1 − e)mR}. Then mR = mI ⊕ mJ , mI = eM and mJ = (1 − e)mR is δ-small in (1 − e)M . (6) ⇒ (1) Let m ∈ M . By hypothesis, there exist ideals I and J of R such that mR = mI ⊕ mJ, where mI is direct summand and mJ is δ-small in M . Let M = mI ⊕ K for some submodule K. Since K ∩ mR ∼= mJ and mJ is δ-small in M , M is principally δ-lifting. (cid:3) Note that every lifting module is principally δ-lifting. There are principally δ-lifting modules but not lifting. Example 3.7. Let M be the Z-module Q and m ∈ M . It is well known that every cyclic submodule mR of M is small, therefore δ-small in M . Hence M is a principally δ-lifting Z-module. If N is a nonsmall proper submodule of M , then N is neither direct summand nor contains a direct summand of M . It follows that M is not a lifting Z-module. It is clear that every δ-lifting module is principally δ-lifting. However the converse is not true. Example 3.8. Let R and T denote the rings in [9, Example 4.1], where 6 HATICE INANKIL, SAIT HALICIO GLU, AND ABDULLAH HARMANCI ∞ ∞ R = Pi=1L Z2 + Z2.1 = {(f1, f2, . . . , fn, f, f, . . . ) ∈ o x # : x ∈ R, y ∈ Soc(R)) . Then Radδ(T ) = " 0 Soc(R) 0 # Qi=1 Z2} 0 and T = (" x y and T /Radδ(T ) is not semisimple as isomorphic to R. So T is not δ-semiperfect by [9, Theorem 3.6]. Hence T is not a δ-lifting module over T . It is easy to show that T /Radδ(T ) lift to idempotents of T , so T is a semiregular ring. Since T is a δ-semiregular ring, every finitely generated right ideal H of T can be written as H = aT ⊕ S, where a2 = a ∈ T and S ≤ Radδ(T ) by [9, Theorem 3.5]. Hence T is a principally δ-lifting module. Proposition 3.9. Let M be a principally δ-lifting module. If M = M1 + M2 such that M1 ∩ M2 is cyclic, then M2 contains a δ-supplement of M1 in M . Proof. Assume that M = M1 +M2 and M1 ∩M2 is cyclic. Then we have M1 ∩M2 = N ⊕ S, where N is direct summand of M and S is δ-small in M . Let M = N ⊕ N ′ and M2 = N ⊕ (M2 ∩ N ′). It follows that M1 ∩ M2 = N ⊕ (M1 ∩ M2 ∩ N ′) = N ⊕ S. Let π : M2 = N ⊕ (M2 ∩ N ′) → N ′ be the natural projection. It follows that π(M1 ∩ M2 ∩ N ′) = M1 ∩ M2 ∩ N ′ = π(S). Since S is δ-small in M , it is δ-small in N ′ by Lemma 2.1. Hence M = M1 + (M2 ∩ N ′), M2 ∩ N ′ ≤ M2 and M1 ∩ (M2 ∩ N ′) is δ-small in M2 ∩ N ′. M2 ∩ N ′ is contained in M2 and a δ-supplement of M1 in M2. This completes the proof. (cid:3) Let M be a module. A submodule N is called fully invariant if for each endo- morphism f of M , f (N ) ≤ N . Let S = End(MR), the ring of R-endomorphisms of M . Then M is a left S-, right R-bimodule and a principal submodule N of the right R-module M is fully invariant if and only if N is a sub-bimodule of M . Clearly 0 and M are fully invariant submodules of M . The right R-module M is called a duo module provided every submodule of M is fully invariant. For the readers' convenience we state and prove Lemma 3.10 which is proved in [5]. Lemma 3.10. Let a module M = Li∈I and let N be a fully invariant submodule of M . Then N = Li∈I (N ∩ Mi). Mi be a direct sum of submodules Mi (i ∈ I) Proof. For each j ∈ I, let pj : M → Mj denote the canonical projection and let ij : Mj → M denote inclusion. Then ijpj is an endomorphism of M and hence ijpj(N ) ⊆ N for each j ∈ I. It follows that N ⊆ Lj∈I so that N = Lj∈I (N ∩ Mj). ijpj(N ) ⊆ Lj∈I (N ∩ Mj) ⊆ N , (cid:3) One may suspect that if M1 and M2 are principally δ-lifting modules, then M1 ⊕ M2 is also principally δ-lifting. But this is not the case. ON A CLASS OF LIFTING MODULES 7 Example 3.11. Consider the Z-modules M1 = Z/Z2 and M2 = Z/Z8. It is clear that M1 and M2 are principally δ-lifting. Let M = M1 ⊕ M2. Then M is not a principally δ-lifting Z-module. Let N1 = (1, 2)Z and N2 = (1, 1)Z. Then M = N1 + N2, N1 is not a direct summand of M and does not contain any nonzero direct summand of M . For any proper submodule N of M , M/N is singular Z-module. Hence the principal submodule does not satisfy δ-lifting property. It follows that M is not principally δ-lifting Z-module. By the same reasoning, for any prime integer p, the Z-module M = (Z/Zp) ⊕ (Z/Zp3) is not principally δ-lifting. We have already observed by the preceding example that the direct sum of principally δ-lifting modules need not be principally δ-lifting. Note the following fact. Proposition 3.12. Let M = M1 ⊕ M2 be a decomposition of M with M1 and M2 principally δ-lifting modules. If M is a duo module, then M is principally δ-lifting. Proof. Let M = M1 ⊕ M2 be a duo module and mR be a submodule of M . By Lemma 3.10, mR = ((mR) ∩ M1) ⊕ ((mR) ∩ M2). Since (mR) ∩ M1 and (mR) ∩ M2 are principal submodules of M1 and M2 respectively, there exist A1, B1 ≤ M1 such that A1 ≤ (mR) ∩ M1 ≤ M1 = A1 ⊕ B1, B1 ∩ ((mR) ∩ M1) = B1 ∩ (mR) is δ-small in B1, and A2, B2 ≤ M2 such that A2 ≤ (mR) ∩ M2 ≤ M2 = A2 ⊕ B2, B2 ∩ ((mR) ∩ M2) = B2 ∩ (mR) is δ-small in B2. Then M = A1 ⊕ A2 ⊕ B1 ⊕ B2, A1 ⊕ A2 ≤ N and (mR) ∩ (B1 ⊕ B2) = ((mR) ∩ B1) ⊕ ((mR) ∩ B2) is δ-small in M1 ⊕ M2. (cid:3) Lemma 3.13. The following are equivalent for a module M = M ′ ⊕ M ′′. (1) M ′ is M ′′-projective. (2) For each submodule N of M with M = N + M ′′, there exists a submodule N ′ ≤ N such that M = N ′ ⊕ M ′′. Proof. See [7, 41.14] (cid:3) Theorem 3.14. Let M1 be a semisimple module and M2 a principally δ-lifting module. Assume that M1 and M2 are relatively projective. Then M = M1 ⊕ M2 is principally δ-lifting. Proof. Let 0 6= m ∈ M and let K = M1 ∩ ((mR) + M2). We divide the proof into two cases: Case (i): K 6= 0. Then M1 = K ⊕ K1 for some submodule K1 of M1 and so M = K ⊕ K1 ⊕ M2 = (mR) + (M2 ⊕ K1). Hence K is M2 ⊕ K1-projective. By Lemma 3.13, there exists a submodule N of mR such that M = N ⊕ (M2 ⊕ 8 HATICE INANKIL, SAIT HALICIO GLU, AND ABDULLAH HARMANCI K1). We may assume (mR) ∩ (M2 ⊕ K1) 6= 0. Note that for any submod- ule L of M2, we have (mR) ∩ (L + K1) = L ∩ ((mR) + K1). In particular (mR) ∩ (M2 + K1) = M2 ∩ (mR + K1). Then mR = N ⊕ (mR) ∩ (K1 ⊕ M2). There exist n ∈ N and m′ ∈ (mR) ∩ (K1 ⊕ M2) such that m = n + m′. Then nR = N and m′R = (mR) ∩ (K1 ⊕ M2). Since (mR) ∩ (M2 + K1) = M2 ∩ ((mR) + K1), M2 ∩ ((mR) + K1) is a principal submodule of M2 and M2 is principally δ-lifting, there exists a submodule X of M2 ∩ ((mR) + K1) = (mR) ∩ (M2 ⊕ K1) such that M2 = X ⊕ Y and Y ∩ M2 ∩ ((mR) + K1) = Y ∩ ((mR) + K1) is δ-small in M2 ∩ ((mR) + K1) and in M2. Hence M = (N ⊕ X) ⊕ (Y ⊕ K1). Since N ⊕ X ≤ mR and (mR) ∩ (Y ⊕ K1) = Y ∩ ((mR) + K1), (mR) ∩ (Y ⊕ K1) = Y ∩ ((mR) + K1) is δ-small in Y ⊕ K1. So M is δ-lifting. Case (ii): K = 0. Then mR ≤ M2. Since M2 is δ-lifting, there exists a submodule X of mR such that M2 = X ⊕ Y and (mR) ∩ Y is δ-small in Y for some submodule Y of M2. Hence M = X ⊕ (M1 ⊕ Y ). Since (mR) ∩ (M1 ⊕ Y ) = (mR) ∩ Y and (mR) ∩ (M1 ⊕ Y ) = (mR) ∩ Y is δ-small in Y . By Lemma 2.1 (3), (mR) ∩ (M1 ⊕ Y ) is δ-small in M1 ⊕ Y . It follows that M is δ-lifting. (cid:3) A module M is said to be a principally semisimple if every cyclic submodule is a direct summand of M . Tuganbayev calls a principally semisimple module as a regular module in [3]. Every semisimple module is principally semisimple. Every principally semisimple module is principally δ-lifting. For a module M , we write Radδ(M ) =P{L L is a δ-small submodule of M }. Lemma 3.15. Let M be a principally δ-lifting module. Then M/Radδ(M ) is a principally semisimple module. Proof. Let m ∈ M . There exists M1 ≤ mR such that M = M1 ⊕M2 and (mR)∩M2 is δ-small in M2. So(mR) ∩ M2 is δ-small in M . Then M/Radδ(M ) = [(mR + Radδ(M ))/Radδ(M )] ⊕ [(M2 + Radδ(M ))/Radδ(M )] because (mR+ Radδ(M )) ∩ (M2+Radδ(M )) =Radδ(M ). Hence every principal submodule of M/Radδ(M ) is a direct summand. (cid:3) Proposition 3.16. Let M be a principally δ-lifting module. Then M = M1 ⊕ M2, where M1 is a principally semisimple module and M2 is a module with Radδ(M ) essential in M2. Proof. Let M1 be a submodule of M such that Radδ(M )⊕ M1 is essential in M and m ∈ M1. Since M is principally δ-lifting, there exists a direct summand M2 of M 2 is δ-small in M . Hence mR ∩ M ′ such that M2 ≤ mR, M = M2 ⊕ M ′ 2 is a submodule of Radδ(M ) and so mR ∩ M ′ 2 = 0. Then m ∈ M2 and mR = M2. Since M2∩ Radδ(M ) = 0, M2 is isomorphic to a submodule of M/Radδ(M ). By 2 and mR ∩ M ′ ON A CLASS OF LIFTING MODULES 9 Lemma 3.15, M/Radδ(M ) is principally semisimple, M2 is principally semisimple. On the other hand, Radδ(M ) =Radδ(M ′ 2) is essential in M2 that it is clear from the construction of M ′ 2. (cid:3) A nonzero module M is called δ-hollow if every proper submodule is δ-small in M , and M is principally δ-hollow if every proper cyclic submodule is δ-small in M , and M is finitely δ-hollow if every proper finitely generated submodule is δ-small in M . Since finite direct sum of δ-small submodules is δ-small, M is principally δ-hollow if and only if it is finitely δ-hollow. Lemma 3.17. The following are equivalent for an indecomposable module M . (1) M is a principally δ-lifting module. (2) M is a principally δ-hollow module. Proof. (1)⇒(2) Let m ∈ M . Since M is a principally δ-lifting module, there exist N and S submodules of M such that N ≤ mR, mR ∩ S is δ-small in M and M = N ⊕ S. By hypothesis, N = 0 and S = M . So that mR ∩ S = mR is δ-small in M . (2)⇒(1) Let m ∈ M . Then mR = (mR) ⊕ (0). By (2) mR is δ-small and (0) is direct summand in M . Hence M is a principally δ-lifting module. (cid:3) Lemma 3.18. Let M be a module, then we have (1) If M is principally δ-hollow, then every factor module is principally δ- hollow. (2) If K is δ-small submodule of M and M/K is principally δ-hollow, then M is principally δ-hollow. (3) M is principally δ-hollow if and only if M is local or Radδ(M ) = M . Proof. (1) Assume that M is principally δ-hollow and N a submodule of M . Let m + N ∈ M/N and (mR + N )/N + K/N = M/N . Suppose that M/K is singular. We have mR + K = M . Since M/K is singular and M is principally δ-hollow, M = K. (2) Let m ∈ M . Assume that mR + N = M for some submodule N with M/N singular. Then (m + K)R = (mR + K)/K is a cyclic submodule of M/K and (mR+K)/K +(N +K)/K = M/K and M/(N +K) is singular as an homomorphic image of M/N . Hence (N + K)/K = M/K or N + K = M . By hypothesis N = M . (3) Suppose that M is principally δ-hollow and it is not local. Let N and K be two distinct maximal submodules of M and k ∈ K \ N . Then M = kR + N and M/N is a simple module, and so M/N is a singular or projective module. If M/N is singular, then M = N since kR is δ-small. But this is not possible since N is maximal. So M/N is projective. Hence N is direct summand. So M = N ⊕ N ′ 10 HATICE INANKIL, SAIT HALICIO GLU, AND ABDULLAH HARMANCI for some nonzero submodule N ′ of M , that is, N and kR are proper submodules of M . Since every proper submodule of M is contained in Radδ(M ), M = Radδ(M ). The converse is clear. (cid:3) Proposition 3.19. Let M be a module. Then the following are equivalent. (1) M is principally δ-hollow. (2) If N is submodule with M/N cyclic, then N is a δ-small submodule of M . Proof. (1) ⇒ (2) Assume that N is a submodule with M/N cyclic. Lemma 2.1 implies that M/N is principally δ-hollow since being δ-small is preserved under homomorphisms. Since M/N has maximal submodules, and by Lemma 3.18, M/N is local. There exists a unique maximal submodule N1 containing N . Hence N is small, therefore it is δ-small. (2) ⇒ (1) We prove that every cyclic submodule is δ-small in M . So let m ∈ M and M = mR + N with M/N singular. Then M/N is cyclic. By hypothesis, N is δ-small submodule of M . By Lemma 2.1, there exists a projective semisimple submodule Y of N such that M = (mR) ⊕ Y . Let Y = Li∈I simple. Now we write M = ((mR)Li6=j module as it is isomorphic to simple module Ni. By hypothesis, ((mR)Li6=j Z of ((mR)Li6=j Nj) ⊕ Ni. Then M/((mR)Li6=j small in M . Again by Lemma 2.1, there exists a projective semisimple submodule module. So M = N ⊕ N ′ for some submodule N ′. Then N ′ is projective. M/N is projective as it is isomorphic to N ′. Hence M/N is both singular and projective module. Thus M = N . (cid:3) Nj) such that M = Z ⊕ Ni. Hence M is projective semisimple Ni where each Ni is Nj) is cyclic Nj) is δ- 4. Applications In this section, we introduce and study some properties of principally δ-semiperfect modules. By [9], a projective module P is called a projective δ-cover of a module M if there exists an epimorphism f : P −→ M with Kerf is δ-small in P , and a ring is called δ-perfect (or δ-semiperfect) if every R-module (or every simple R-module) has a projective δ-cover. For more detailed discussion on δ-small submodules, δ- perfect and δ-semiperfect rings, we refer to [9]. A module M is called principally δ-semiperfect if every factor module of M by a cyclic submodule has a projective δ-cover. A ring R is called principally δ-semiperfect in case the right R-module R is principally δ-semiperfect. Every δ-semiperfect module is principally δ-semiperfect. In [9], a ring R is called δ-semiregular if every cyclically presented R-module has a projective δ-cover. Theorem 4.1. Let M be a projective module. Then the following are equivalent. (1) M is principally δ-semiperfect. ON A CLASS OF LIFTING MODULES 11 (2) M is principally δ-lifting. Proof. (1)⇒ (2) Let m ∈ M and P M π→ M/mR the natural epimorphism. f → M/mR be a projective δ-cover and M ♣ ♣ ♣ π ❄ g ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣✠ f P ✲ M/mR ✲ 0 Then there exists a map M g → P such that f g = π. Then P = g(M )+Ker(f ). Since Ker(f ) is δ-small, by Lemma 2.1, there exists a projective semisimple submodule Y of Ker(f ) such that P = g(M ) ⊕ Y . So g(M ) is projective. Hence M = K ⊕ Ker(g) for some submodule K of M . It is easy to see that g(K ∩ mR) = g(K)∩Ker(f ) and Ker(g) ≤ mR. Hence M = K + mR. Next we prove K ∩ (mR) is δ-small in K. Since Ker(f ) is δ-small in P , g(K)∩Ker(f ) = g(K ∩ mR) is δ-small in P by Lemma 2.1 (4). Hence K ∩ (mR) is δ-small in K since g−1 is an isomorphism from g(M ) onto K. (2)⇒ (1) Assume that M is a principally δ-lifting module. Let m ∈ M . There exist direct summands N and K of M such that M = N ⊕ K, N ≤ mR and mR ∩ K is δ-small in K. Let K π→ M/mR denote the natural epimorphism de- It is obvious that fined by π(k) = k + mR where k ∈ K, k + mR ∈ M/mR. Ker(π) = mR ∩ K. It follows that K is projective δ-cover of M/mR. So M is principally δ-semiperfect. (cid:3) Corollary 4.2. Let R be a ring. Then the following are equivalent. (1) R is principally δ-semiperfect. (2) R is principally δ-lifting. (3) R is δ-semiregular. Proof. (1) ⇔ (2) Clear by Theorem 4.1. (2) ⇔ (3) By Theorem 3.6 (2), R is principally δ-lifting if and only if for every principal right ideal I of R can be written as I = N ⊕ S, where N is direct summand and S is δ-small in R. This is equivalent to being R δ-semiregular since for any ring R, Radδ(R) is δ-small in R and each submodule of a δ-small submodule is δ-small. (cid:3) The module M is called principally δ-supplemented if every cyclic submodule of M has a δ-supplement in M . Clearly, every δ-supplemented module is principally δ-supplemented. Every principally δ-lifting module is principally δ-supplemented. 12 HATICE INANKIL, SAIT HALICIO GLU, AND ABDULLAH HARMANCI In a subsequent paper we investigate principally δ-supplemented modules in detail. Now we prove: Theorem 4.3. Let M be a principally δ-semiperfect module. Then (1) M is principally δ-supplemented. (2) Each factor module of M is principally δ-semiperfect, hence any homomor- phic image and any direct summand of M is principally δ-semiperfect. β → M/mR. Proof. (1) Let m ∈ M . Then M/mR has a projective δ-cover P There exists P α→ M such that the following diagram is commutative, β = πα, where M π→ M/mR is the natural epimorphism. P ♣ ♣ ♣ ♣ β ❄ M/mR ✲ 0 α ♣ ♣ ♣ ♣ π ✲ ♣ ♣ ♣✠ M Then M = α(P ) + mR, and α(P ) ∩ mR is δ-small in α(P ), by Lemma 2.1 (1). Hence M is principally δ-supplemented. f → N be an epimorphism and nR a cyclic submodule of N . Let (2) Let M → M/(mR) be a projective δ-cover. Define M/(mR) h→ N/nR m ∈ f −1(nR) and P by h(m′ + mR) = f (m′) + nR, where m′ + mR ∈ M/(mR). Then Ker(g) is con- tained in Ker(hg). By projectivity of P , there is a map α from P to N such that g hg = πα. P ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ ♣ α ♣❄ N g ✲ M/mR h π ✲ ❄ N/nR ✲ 0 It is routine to check that (nR) ∩ α(P ) = α(Ker(g)). By Lemma 2.1 (2), α(Ker(g)) is δ-small in N since Ker(g) is δ-small. Let x ∈Ker(πα). Then hg(x) = (πα)(x) = 0 or α(x) ∈ (nR) ∩ α(P ). So Ker(πα) is δ-small. Hence P is a projective δ-cover for N/(nR). (cid:3) Theorem 4.4. Let P be a projective module with Radδ(P ) is δ-small in P . Then the following are equivalent. (1) P is principally δ-lifting. (2) P/Radδ(P ) is principally semisimple and, for any cyclic submodule xR of P/Radδ(P ) that is a direct summand of P/Radδ(P ), there exists a cyclic direct summand A of P such that xR = A. ON A CLASS OF LIFTING MODULES 13 Proof. (1)⇒ (2) Since P is a principally δ-lifting module, P/Radδ(P ) is principally semisimple by Lemma 3.15. Let xR be any cyclic submodule of P/ Radδ(P ). By Theorem 3.6, there exists a direct summand A of P and a δ-small submodule B such that xR = A ⊕ B. Since B is contained in Radδ(R), xR+ Radδ(R) = A+ Radδ(R). Hence xR = A. (2)⇒ (1) Let xR be any cyclic submodule of P . Then we have P/ Radδ(P ) = [(xR+Radδ(P ))/Radδ(P )] ⊕ [U/ Radδ(P )] for some U ≤ P . By (2), there exists a direct summand A of P such that P = A ⊕ B and U = B+ Radδ(P ). Then P = A ⊕ B = A + U + Radδ(P ). Since Radδ(P ) is δ-small in P , there exists a projective and semisimple submodule Y of P such that P = A ⊕ B = (A + U ) ⊕ Y . Since P is projective, A + B is also projective and so by Lemma 3.13, we have A + B = V ⊕ B for some V ≤ A. Hence P = V ⊕ B ⊕ Y . On the other hand (xR) ∩ (B ⊕ Y ) = (xR) ∩ B ≤ (xR) ∩ U ≤ Radδ(R). Since Radδ(R) is δ-small in P , it is δ-small in B ⊕ Y by Lemma 2.1 (3). Thus P is principally δ-lifting. (cid:3) References [1] F.W. Anderson and K.R. Fuller, Rings and Categories of Modules, Springer-Verlag, New York, 1974. [2] J. Clark, C. Lomp, N. Vanaja and R. Wisbauer, Lifting modules, Brikhauser-Basel, 2006. [3] C. Lomp, Regular and Biregular Module Algebras, Arab. J. Sci. and Eng., 33(2008), 351-363. [4] S. Mohamed and B. J. Muller, Continuous and discrete modules, Cambridge University Press, 1990. [5] A. C. Ozcan , A. Harmanci and P. F. Smith, Duo Modules, Glasgow Math. J., 48(3)(2006), 533-545. [6] K. Oshiro, Semiperfect modules and quasi-semiperfect modules, Osaka J. Math., 20(1983), 337-372. [7] R. Wisbauer, Foundations of module and ring theory, Gordon and Breach , Reading, 1991. [8] M.F. Yousif and Y. Zhou, Semiregular, Semiperfect and Perfect Rings Relative To An Ideal, Rocky Mountain J. Math., 32(4)(2002), 1651-1671. [9] Y. Zhou, Generalizations of Perfect, Semiperfect and Semiregular Rings, Algebra Colloq., 7(3)(2000), 305-318. Department of Mathematics, Ankara University, 06100 Ankara, Turkey E-mail address: [email protected] Department of Mathematics, Ankara University, 06100 Ankara, Turkey E-mail address: [email protected] Department of Mathematics, Hacettepe University, 06550 Ankara, Turkey E-mail address: [email protected]
1204.3203	2	1204	2012-11-23T09:45:41	Deformation of the Hopf algebra of plane posets	[ "math.RA" ]	We describe and study a four parameters deformation of the two products and the coproduct of the Hopf algebra of plane posets. We obtain a family of braided Hopf algebras, generally self-dual. We also prove that in a particular case (when the second parameter goes to zero and the first and third parameters are equal), this deformation is isomorphic, as a self-dual braided Hopf algebra, to a deformation of the Hopf algebra of free quasi-symmetric functions.	math.RA	math	Deformation of the Hopf algebra of plane posets Loïc Foissy Laboratoire de Mathématiques, Université de Reims Moulin de la Housse - BP 1039 - 51687 REIMS Cedex 2, France e-mail : [email protected] ABSTRACT. We describe and study a four parameters deformation of the two products and the coproduct of the Hopf algebra of plane posets. We obtain a family of braided Hopf algebras, which are generically self-dual. We also prove that in a particular case (when the second param- eter goes to zero and the first and third parameters are equal), this deformation is isomorphic, as a self-dual braided Hopf algebra, to a deformation of the Hopf algebra of free quasi-symmetric functions FQSym. KEYWORDS. Plane posets; Deformation; Braided Hopf algebras; Self- duality. AMS CLASSIFICATION. 06A11, 16W30, 16S80. Contents 1 Recalls and notations 1.1 Double and plane posets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Algebraic structures on plane posets . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Pairings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4 Morphism to free quasi-symmetric functions . . . . . . . . . . . . . . . . . . . . . 2 Deformation of the products 2.1 Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Particular cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 Subalgebras and quotients . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Dual coproducts 3.1 Constructions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Particular cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Compatibilities with the products . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 Self-duality results 4.1 A first pairing on Hq . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 Properties of the pairing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 Comparison of pairings with colinear parameters . . . . . . . . . . . . . . . . . . 4.4 Non-degeneracy of the pairing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Morphism to free quasi-symmetric functions 5.1 A second Hopf pairing on H(q1,0,q,1,q4) . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Quantization of the Hopf algebra of free quasi-symmetric functions . . . . . . . . 1 3 3 4 6 6 7 7 9 11 12 12 13 15 16 16 18 20 21 23 23 26 Introduction A double poset is a finite set with two partial orders. As explained in [8], the space generated by the double posets inherits two products and one coproduct, here denoted by m, and ∆, making it both a Hopf and an infinitesimal Hopf algebra [6]. Moreover, this Hopf algebra is self dual. A double poset is plane if its two partial orders satisfy a (in)compatibility condition, see definition 1. The subspace HPP generated by plane posets is stable under the two products and the coproduct, and is self-dual [2, 3]: in particular, two Hopf pairings are defined on it, using the notion of picture. Moreover, as proved in [3], it is isomorphic to the Hopf algebra of free quasi-symmetric functions FQSym, also known as the Malvenuto-Reutenauer Hopf algebra of permutations. An explicit isomorphism Θ is given by the linear extensions of plane posets, see definition 11. We define in this text a four-parameters deformation of the products and the coproduct of HPP , together with a deformation of the two pairings and of the morphism from HPP to FQSym. If q = (q1, q2, q3, q4) ∈ K 4, the product mq(P ⊗ Q) of two plane posets P and Q is a linear span of plane posets R such that R = P ⊔ Q as a set, P and Q being plane subposets of R. The coefficients are defined with the help of the two partial orders of R, see theorem 14, and are polynomials in q. In particular: m(1,0,0,0) = , m(0,1,0,0) = op, m(0,0,1,0) = m, m(0,0,0,1) = mop.   We also obtain the product dual to the coproduct ∆ (considering the basis of double posets as orthonormal) as m(1,0,1,1), and its opposite given by m(0,1,1,1). Dually, we define a family of coassociative coproducts ∆q. For any plane poset P , ∆(P ) is a linear span of terms (P \ I) ⊗ I, running over the plane subposets I of P , the coefficients being polynomials in q. In the particular cases where at least one of the components of q is zero, only h-ideals, r-ideals or biideals can appear in this sum (definition 7 and proposition 22). We study the compatibility of ∆q with both products m and on HPP (proposition 23). For example, (HPP , m, ∆q) satisfies the axiom ∆q(xy) =XX q qy′′ q x′ 3 q y′ q x′′ 4 q (x′ qy′ q) ⊗ (x′′ q y′′ q ). If q3 = 1, it is a braided Hopf algebra, with braiding given by cq(P ⊗ Q) = qP Q Q ⊗ P ; in particular, if q3 = 1 and q4 = 1 this is a Hopf algebra, and if q3 = 1 and q4 = 0 it is an infinitesimal Hopf algebra. If q4 = 1, it is the coopposite (or the opposite) of a braided Hopf algebra. Similar results hold if we consider the second product , permuting the roles of (q3, q4) and (q1, q2). 4 We define a symmetric pairing h−, −iq such that: hx ⊗ y, ∆q(z)iq = hxy, ziq for all x, y, z ∈ HPP . If q = (1, 0, 1, 1), we recover the first "classical" pairing of HPP . We prove that in the case q2 = 0, this pairing is nondegenerate if, and only if, q1 6= 0 (corollary 36). Consequently, this pairing is generically nondegenerate. The coproduct of FQSym is finally deformed, in such a way that the algebra morphism Θ from HPP to FQSym becomes compatible with ∆q, if q has the form q = (q1, 0, q1, q4). De- forming the second pairing h−, −i′ of HPP and the usual Hopf pairing of FQSym, the map Θ becomes also an isometry (theorem 40). Consequently, the deformation h−, −i′ q is nondegenerate if, and only if, q1q4 6= 0. 2 This text is organized in the following way. The first section contains reminders on the Hopf algebra of plane posets HPP , its two products, its coproducts and its two Hopf pairings, and on the isomorphism Θ from HPP to FQSym. The deformation of the products is defined in section 2, and we also consider the compatibility of these products with several natural bijections on PP and the stability of certain families of plane posets under these products. We proceed to the dual construction in the next section, where we also study the compatibility with the two (undeformed) products. The deformation of the first pairing is described in section 4. The compatibilities with the bijections on PP or with the second product are also given, and the nondegeneracy is proved for q = (q1, 0, q3, q4) if q1 6= 0. The last section is devoted to the deformation of the second pairing and of the morphism to FQSym. 1 Recalls and notations 1.1 Double and plane posets Definition 1 1. [8] A double poset is a triple (P, ≤1, ≤2), where P is a finite set and ≤1, ≤2 are two partial orders on P . 2. A plane poset is a double poset (P, ≤h, ≤r) such that for all x, y ∈ P , such that x 6= y, x and y are comparable for ≤h if, and only if, x and y are not comparable for ≤r. The set of (isoclasses of ) plane posets will be denoted by PP. For all n ∈ N, the set of (isoclasses of ) plane posets of cardinality n will be denoted by PP(n). Examples. Here are the plane posets of cardinal ≤ 4. They are given by the Hasse graph of ≤h; if x and y are two vertices of this graph which are not comparable for ≤h, then x ≤r y if y is more on the right than x. PP(0) = {1}, PP(1) = { q}, PP(2) = { q q, q q }, PP(3) = { q q q, q q q , q q q, q q∨q q q , q , q∧qq }, q q PP(4) =   q q q q, q q q , q q q q q q, q q q, q q∨qq q q q∨qq q , q∨qq , q∨qq , q q q q∨q q , q q , q∧qq q , , q∧q q q q∨qq , q q, q q q∧q q , q q , q q, q q∧q q , q∧q q q q, q q q∧q q , q q(cid:30) , q q q q(cid:31) , q q q q(cid:30)(cid:31) , q q . q , q q q∧   q∨q The following proposition is proved in [2] (proposition 11): Proposition 2 Let P ∈ PP. We define a relation ≤ on P by: (x ≤ y) if (x ≤h y or x ≤r y). Then ≤ is a total order on P . As a consequence, substituing ≤ to ≤r, plane posets are also special posets [8], that is to say double posets such that the second order is total. For any plane poset P ∈ PP(n), we shall assume that P = {1, . . . , n} as a totally ordered set. The following theorem is proved in [3]: Theorem 3 Let σ be a permutation in the nth symmetric group Sn. We define a plane poset Pσ in the following way: • Pσ = {1, . . . , n} as a set. 3 • If i, j ∈ Pσ, i ≤h j if i ≤ j and σ(i) ≤ σ(j). • If i, j ∈ Pσ, i ≤r j if i ≤ j and σ(i) ≥ σ(j). Note that the total order on {1, . . . , n} induced by this plane poset structure is the usual one. Then for all n ≥ 0, the following map is a bijection: Examples. Ψn :(cid:26) Sn −→ PP(n) σ −→ Pσ. q q q , q , q , Ψ3((132)) = Ψ3((321)) = q q q. q q∨q Ψ2((12)) = q q , q∧q Ψ3(213)) = q , Ψ2((21)) = q q, q, Ψ3((231)) = q q Ψ3(123)) = Ψ3(312)) = q We define several bijections on PP: Definition 4 Let P = (P, ≤h, ≤r) ∈ PP. We put: ι(P ) = (P, ≤r, ≤h), α(P ) = (P, ≥h, ≤r), β(P ) = (P, ≤h, ≥r), γ(P ) = (P, ≥h, ≥r).   Remarks. 1. Graphically: • A Hasse graph of α(P ) is obtained from a Hasse graph of P by a horizontal symmetry. • A Hasse graph of β(P ) is obtained from a Hasse graph of P by a vertical symmetry. • A Hasse graph of γ(P ) is obtained from a Hasse graph of P by a rotation of angle π. 2. These bijections generate a group of permutations of PP of cardinality 8. It is described by the following array: ◦ α β γ ι ι ◦ α ι ◦ β ι ◦ γ α Id γ β β γ Id α γ β α Id ι ◦ α ι ◦ β ι ◦ γ ι ◦ γ ι ◦ β ι ◦ α ι ◦ γ ι ◦ β ι ◦ α ι ι ι ι ι ◦ α ι ◦ β ι ◦ γ ι ◦ α ι ◦ γ ι ◦ β ι ι ι ◦ β ι ◦ γ ι ◦ α ι ◦ γ ι ◦ β ι ◦ α Id β α γ β Id γ α α γ Id β ι γ α β Id This is a dihedral group D4. 1.2 Algebraic structures on plane posets Two products are defined on PP. The first is called composition in [8] and denoted by in [2]. We shall shortly denote it by m in this text. Definition 5 Let P, Q ∈ PP. 1. The double poset P Q = m(P ⊗ Q) is defined as follows: • P Q = P ⊔ Q as a set, and P, Q are plane subposets of P Q. 4 • if x ∈ P and y ∈ Q, then x ≤r y in P Q. 2. The double poset P Q is defined as follows: • P Q = P ⊔ Q as a set, and P, Q are plane subposets of P Q. • if x ∈ P and y ∈ Q, then x ≤h y in P Q. Examples. 1. The Hasse graph of P Q is the concatenation of the Hasse graphs of P and Q. q q 2. Here are examples for : q q q = q q , q q = q q q , q q q = q∨q q , q q q = q∧q q . The vector space generated by PP is denoted by HPP . These two products are linearly extended to HPP ; then (HPP , m) and (HPP , ) are two associative, unitary algebras, sharing the same unit 1, which is the empty plane poset. Moreover, they are both graded by the cardinality of plane posets. They are free algebras, as implied that the following theorem, proved in [2]: Theorem 6 1. (a) Let P be a nonempty plane poset. We shall say that P is h-irreducible if for all Q, ∈ PP, P = QR implies that Q = 1 or R = 1. (b) Any plane poset P can be uniquely written as P = P1 . . . Pk, where P1, . . . , Pk are h-irreducible. We shall say that P1, . . . , Pk are the h-irreducible components of P . 2. (a) Let P be a nonempty plane poset. We shall say that P is r-irreducible if for all Q, ∈ PP, P = Q R implies that Q = 1 or R = 1. (b) Any plane poset P can be uniquely written as P = P1 . . . Pk, where P1, . . . , Pk are r-irreducible. We shall say that P1, . . . , Pk are the r-irreducible components of P . Remark. The Hasse graphs of the h-irreducible components of H are the connected compo- nents of the Hasse graph of (P, ≤h), whereas the Hasse graphs of the r-irreducible components of H are the connected components of the Hasse graph of (P, ≤r). Definition 7 Let P = (P, ≤h, ≤r) be a plane poset, and let I ⊆ P . 1. We shall say that I is a h-ideal of P , if, for all x, y ∈ P : (x ∈ I, x ≤h y) =⇒ (y ∈ I). 2. We shall say that I is a r-ideal of P , if, for all x, y ∈ P : (x ∈ I, x ≤r y) =⇒ (y ∈ I). 3. We shall say that I is a biideal of P if it both an h-ideal and a r-ideal. Equivalently, I is a biideal of P if, for all x, y ∈ P : (x ∈ I, x ≤ y) =⇒ (y ∈ I). The following proposition is proved in [2] (proposition 29): Proposition 8 HPP is given a coassociative, counitary coproduct in the following way: for any plane poset P , ∆(P ) = XI h-ideal of P (P \ I) ⊗ I. 5 Moreover, (HPP , m, ∆) is a Hopf algebra, and (HPP , , ∆) is an infinitesimal Hopf algebra [6], both graded by the cardinality of the plane posets. In other words, using Sweedler's notations ∆(x) =P x(1) ⊗ x(2), for all x, y ∈ HPP : ∆(xy) = X x(1)y(1) ⊗ x(2)y(2), ∆(x y) = X x y(1) ⊗ y(2) +X x(1) ⊗ x(2) y − x ⊗ y.   Remarks. The following compatibilities are satisfied: 1. For all P, Q ∈ PP: ι(P Q) = ι(P ) ι(Q), α(P Q) = α(P )α(Q), β(P Q) = β(Q)β(P ), γ(P Q) = γ(Q)γ(P ), ι(P Q) = ι(P )ι(Q), α(P Q) = α(Q) α(P ), β(P Q) = β(P ) β(Q), γ(P Q) = γ(Q) γ(P ).   2. Moreover, ∆ ◦ α = (α ⊗ α) ◦ ∆op, ∆ ◦ β = (β ⊗ β) ◦ ∆, and ∆ ◦ γ = (γ ⊗ γ) ◦ ∆op. 1.3 Pairings We also defined two pairings on HPP, using the notion of pictures: Definition 9 Let P, Q be two elements of PP. 1. We denote by S(P, Q) the set of bijections σ : P −→ Q such that, for all i, j ∈ P : • (i ≤h j in P ) =⇒ (σ(i) ≤r σ(j) in Q). • (σ(i) ≤h σ(j) in Q) =⇒ (i ≤r j in P ). 2. We denote by S′(P, Q) the set of bijections σ : P −→ Q such that, for all i, j ∈ P : • (i ≤h j in P ) =⇒ (σ(i) ≤ σ(j) in Q). • (σ(i) ≤h σ(j) in Q) =⇒ (i ≤ j in P ). The following theorem is proved in [2, 3, 8]: Theorem 10 We define two pairings: h−, −i h−, −i′ : (cid:26) HPP ⊗ HPP −→ K : (cid:26) HPP ⊗ HPP −→ K P ⊗ Q −→ hP, Qi = Card(S(P, Q)), P ⊗ Q −→ hP, Qi′ = Card(S′(P, Q)). They are both homogeneous, symmetric, nondegenerate Hopf pairings on the Hopf algebra HPP = (HPP , m, ∆). 1.4 Morphism to free quasi-symmetric functions We here briefly recall the construction of the Hopf algebra FQSym of free quasi-symmetric functions, also called the Malvenuto-Reutenauer Hopf algebra [1, 7]. As a vector space, a basis of FQSym is given by the disjoint union of the symmetric groups Sn, for all n ≥ 0. By convention, the unique element of S0 is denoted by 1. The product of FQSym is given, for σ ∈ Sk, τ ∈ Sl, by: στ = Xǫ∈Sh(k,l) (σ ⊗ τ ) ◦ ǫ, 6 where Sh(k, l) is the set of (k, l)-shuffles, that is to say permutations ǫ ∈ Sk+l such that ǫ−1(1) < . . . < ǫ−1(k) and ǫ−1(k + 1) < . . . < ǫ−1(k + l). In other words, the product of σ and τ is given by shifting the letters of the word representing τ by k, and then summing all the possible shufflings of this word and of the word representing σ. For example: (123)(21) = (12354) + (12534) + (15234) + (51234) + (12543) +(15243) + (51243) + (15423) + (51423) + (54123). Let σ ∈ Σn. For all 0 ≤ k ≤ n, there exists a unique triple (cid:16)σ(k) Sh(k, n − k) such that σ = ζ −1 1 ⊗ σ(k) k ◦(cid:16)σ(k) ∆(σ) = 1 , σ(k) 2 , ζk(cid:17) ∈ Sk × Sn−k × 2 (cid:17). The coproduct of FQSym is then defined by: Xk=0 1 ⊗ σ(k) σ(k) 2 . n For example: ∆((43125)) = 1 ⊗ (43125) + (1) ⊗ (3124) + (21) ⊗ (123) +(321) ⊗ (12) + (4312) ⊗ (1) + (43125) ⊗ 1. 1 2 and σ(k) Note that σ(k) are obtained by cutting the word representing σ between the k-th and the (k + 1)-th letter, and then standardizing the two obtained words, that is to say applying to their letters the unique increasing bijection to {1, . . . , k} or {1, . . . , n − k}. Moreover, FQSym has a nondegenerate, homogeneous, Hopf pairing defined by hσ, τ i = δσ,τ −1 for all permutations σ and τ . Definition 11 [4, 8] 1. Let P = (P, ≤h, ≤r) a plane poset. Let x1 < . . . < xn be the elements of P , which is totally ordered. A linear extension of P is a permutation σ ∈ Sn such that, for all i, j ∈ {1, . . . , n}: (xi ≤h xj) =⇒ (σ−1(i) < σ−1(j)). The set of linear extensions of P will be denoted by SP . 2. The following map is an isomorphism of Hopf algebras: HPP −→ FQSym P ∈ PP −→ Xσ∈SP σ. Θ :  moreover, for all x, y ∈ HPP , hΘ(x), Θ(y)i′ = hx, yi. 2 Deformation of the products 2.1 Construction Definition 12 Let P ∈ PP and X, Y ⊆ P . We put: 1. hY X = ♯{(x, y) ∈ X × Y / x ≤h y in P }. 2. rY X = ♯{(x, y) ∈ X × Y / x ≤r y in P }. Lemma 13 Let X and Y be disjoint parts of a plane poset P . Then: hY X + hX Y + rY X + rX Y = XY . 7 Proof. Indeed, hY X + hX Y + rY X + rX Y = ♯{(x, y) ∈ X × Y x < y or x > y} = X × Y . ✷ Remark. If it more generally possible to prove that for any part X and Y of a plane poset: hY X + hX Y + rY X + rX Y = 3X ∩ Y 2 + XY . Theorem 14 Let q = (q1, q2, q3, q4) ∈ K 4. We consider the following map: HPP ⊗ HPP −→ HPP P ⊗ Q −→ X(R,I)∈PP 2 I⊆R, R\I=P, I=Q R\I hI 1 hR\I I 2 q rI R\I 3 rR\I 4 R, I q q q mq :  where P, Q ∈ PP. Then (HPP , mq) is an associative algebra, and its unit is the empty plane poset 1. Proof. Let P, Q, R ∈ PP. We put: (cid:26) P1 = (mq ⊗ Id) ◦ mq(P ⊗ Q ⊗ R), P2 = (Id ⊗ mq) ◦ mq(P ⊗ Q ⊗ R). Then:   With: P1 = P2 = X(R1,I1,R2,I2)∈E1 X(R1,I1,R2,I2)∈E2 h 1 q h 1 q I1 R1\I1 +h I2 R2\I2 R1\I1 I1 +h R2\I2 I2 h 2 q I1 R1\I1 +h I2 R2\I2 hR1\I1 I1 2 q +hR2\I2 I2 q q I1 R1\I1 r 3 +r I2 R2\I2 R1\I1 I1 r 4 q +r R2\I2 I2 R2, I1 R1\I1 r 3 +r I2 R2\I2 rR1\I1 I1 4 q +rR2\I2 I2 R2, E1 = {(R1, I1, R2, I2) ∈ PP 4 / I1 ⊆ R1, I1 = Q, R1 \ I1 = P, I2 ⊆ R2, I2 = R, R2 \ I2 = R1}, E2 = {(R1, I1, R2, I2) ∈ PP 4 / I1 ⊆ R1, I1 = R, R1 \ I1 = Q, I2 ⊆ R2, I2 = R1, R2 \ I2 = P }. We shall also consider: E = {(R, J1, J2, J3) ∈ PP 4 / R = J1 ⊔ J2 ⊔ J3, J1 = P, J2 = Q, J3 = R}. First step. Let us consider the following maps: φ : (cid:26) φ′ : (cid:26) E1 −→ E (R1, I1, R2, I2) −→ (R2, R1 \ I1, I1, I2), E −→ E1 (R, J1, J2, J3) −→ (J1 ⊔ J2, J2, R, J3). By definition of E and E1, these maps are well-defined, and an easy computation shows that φ ◦ φ′ = IdE and φ′ ◦ φ = IdE1, so φ is a bijection. Let (R1, I1, R2, I2) ∈ E1. We put φ(R1, I1, R2, I2) = (R, J1, J2, J3). Then: hI1 R1\I1 + hI2 R2\I2 = hJ2 J1 + hJ3 J1⊔J2 = hJ2 J1 + hJ3 J1 + hJ3 J2 . Similar computations finally give: P1 = X(R,J1,J2,J3)∈E J2 J1 +h J3 J1 +h J3 J2 h 1 q h 2 q J1 J2 +h J1 J3 +h J2 J3 +r J3 J1 +r J3 J2 J2 J1 r 3 q J1 J2 r 4 q +r J1 J3 +r J2 J3 R. 8 Second step. Let us consider the following maps: ψ : (cid:26) ψ′ : (cid:26) E2 −→ E (R1, I1, R2, I2) −→ (R2, R1 \ R1, R1 \ I1, I1), E −→ E2 (R, J1, J2, J3) −→ (J2 ⊔ J3, J3, R, J2 ⊔ J3). By definition of E and E2, these maps are well-defined, and a simple computation shows that ψ ◦ ψ′ = IdE and ψ′ ◦ ψ = IdE2, so ψ is a bijection. Let (R1, I1, R2, I2) ∈ E2. We put ψ(R1, I1, R2, I2) = (R, J1, J2, J3). Then: hI1 R1\I1 + hI2 R2\I2 = hJ3 J2 + hJ2⊔J3 J1 = hJ2 J1 + hJ3 J1 + hJ3 J2 . Similar computations finally give: P2 = X(R,J1,J2,J3)∈E J2 J1 +h J3 J1 +h J3 J2 h 1 q J1 J2 +h J1 J3 +h J2 J3 h 2 q +r J3 J1 +r J3 J2 J2 J1 r 3 q J1 J2 r 4 q +r J1 J3 +r J2 J3 R. So mq is associative. Last step. Let P ∈ PP. Then: P.1 = X(R,I)∈PP 2 I⊆R, R\I=P, I=1 RI hI 1 q hR\I I 2 q rI R\I 3 rR\I 4 R = q I q q h1 1 q P hP 1 2 q r1 3 q P rP 1 4 P = P. ✷ q q + (q2 q + (q2 q q + (q2 q + (q2 q 1 + q1q2 + q2 3 + q3q4 + q2 1 + q1q2 + q2 3 + q3q4 + q2 q , 2) q 4) q q q, q q , 2) q 4) q q q. q + q2(q3 + q4) q∧q q + (q1 + q2)q3 q q Similarly, for all Q ∈ PP, 1.Q = Q. Examples. mq( q ⊗ q) = q3q4 q q + q1q2 q q , q ) = q2 mq( q ⊗ q 3 q mq( q ⊗ q q) = q2 1 q ⊗ q) = q2 mq( q 3 q mq( q q ⊗ q) = q2 2 q q q + q2 4 q q∨qq + q2 2 q q q + q2 4 q q∨qq + q2 1 q + q2(q3 + q4) q∨q q∧qq + (q1 + q2)q4 q q + q1(q3 + q4) q∨q q∧qq + (q1 + q2)q4 q q + q1(q3 + q4) q∧q q + (q1 + q2)q3 q q The following result is immediate: Proposition 15 Let (q1, q2, q3, q4) ∈ K 4. Then:   2.2 Particular cases mop (q1,q2,q3,q4) = m(q2,q1,q4,q3), m(q1,q2,q3,q4) ◦ (ι ⊗ ι) = ι ◦ m(q3,q4,q1,q2), m(q1,q2,q3,q4) ◦ (α ⊗ α) = α ◦ m(q2,q1,q3,q4), m(q1,q2,q3,q4) ◦ (β ⊗ β) = β ◦ m(q1,q2,q4,q3), m(q1,q2,q3,q4) ◦ (γ ⊗ γ) = γ ◦ m(q2,q1,q4,q3). Lemma 16 Let P ∈ PP and X ⊆ P . 1. X is a h-ideal of P if, and only if, hP \X X = 0. 9 2. X is a r-ideal of P if, and only if, rP \X X = 0. 3. The h-irreducible components of P are the h-irreducible components of X and P \ X if, and only if, hX P \X = hP \X X = 0. 4. The r-irreducible components of P are the r-irreducible components of X and P \ X if, and only if, rX P \X = rP \X X = 0. 5. P = X(P \ X) if, and only if, hX X = rX P \X = 0. 6. P = X (P \ X) if, and only if, rX X = hX P \X = 0. P \X = hP \X P \X = rP \X Proof. We give the proofs of points 1, 3 and 5. The others are similar. 1. ⇐=. Let x ∈ X, y ∈ P , such that x ≤h y. As hP \X X = 0, y /∈ P \ X, so y ∈ X: X is a h-ideal. =⇒. Then, for all x ∈ X, y ∈ P \ X, x ≤h y is not possible. So hP \X X = 0. 3. =⇒. Let us put P = P1 . . . Pk, where P1, . . . , Pk are the h-irreducible components of P . By hypothesis, X is the disjoint union of certain Pi's, and P \ X is the disjoint union of the other Pi's. So hX P \X = hP \X X = 0. ⇐=. Let I be a h-irreducible component of P . If I ∩X and I ∩(P \X), then for any x ∈ I ∩X and any y ∈ I ∩ (P \ X), x and y are not comparable for ≤h: contradiction. So I is included in I or in P \ X, so is a h-irreducible component of X or P \ X. 5. =⇒. If x ∈ X and y ∈ P \ X, then x <r y, so hX ⇐=. If x ∈ X and y ∈ P \ X, by hypothesis we do not have x <h y, x >h y nor x >r y, so ✷ X = rX P \X = 0. P \X = hP \X x <r y. Hence, P = X(P \ X). Proposition 17 1. Let us assume that q3 = q4 = 0. Let P = P1 . . . Pk and P ′ = Pk+1 . . . Pk+l be two plane posets, decomposed into their r-connected components. Then: mq(P ⊗ P ′) = Xσ∈Sh(k,l) Y1≤i≤k<j≤k+l Qi,j(σ)PiPjPσ(1) . . . Pσ(k+l), where Qi,j(σ) = q1 if σ−1(i) < σ−1(j) and q2 if σ−1(i) > σ−1(j). 2. Let us assume that q1 = q2 = 0. Let P = P1 . . . Pk and P ′ = Pk+1 . . . Pk+l be two plane posets, decomposed into their r-connected components. Then: mq(P ⊗ P ′) = Xσ∈Sh(k,l) Y1≤i≤k<j≤k+l Q′ i,j(σ)PiPjPσ(1) . . . Pσ(k+l), where Q′ i,j(σ) = q3 if σ−1(i) < σ−1(j) and q4 if σ−1(i) > σ−1(j). Proof. Let us prove the first point; the proof of the second point is similar. Let us consider a plane poset R such that the coefficient of R in mq(P ⊗ P ′) is not zero. So there exists I ⊆ R such that R \ I = P and I = P ′. Moreover, as q3 = q4 = 0, rI I = 0. By lemma 16-4, the r-connected components of R are the r-components of R \ I and I. As a consequence, there exists a (k, l)-shuffle σ, such that R = Pσ(1) . . . Pσ(k+l). Then hI R\I is the sum of PiPj, where 1 ≤ i ≤ k < j ≤ k + l, such that σ−1(i) < σ−1(j); hR\I is the sum of PiPj, where 1 ≤ i ≤ k < j ≤ k + l, such that σ−1(i) > σ−1(j). This immediately implies the announced result. ✷ R\I = rR\I I Remarks. 10 1. The first point implies that: 1 • m(q1,0,0,0)(P ⊗ P ′) = qP P ′ • m(0,q2,0,0)(P ⊗ P ′) = qP P ′ • m(0,0,q3,0)(P ⊗ P ′) = qP P ′ • m(0,0,0,q4)(P ⊗ P ′) = qP P ′ 2 3 4 P P ′. In particular, m(1,0,0,0) = . P ′ P . In particular, m(0,1,0,0) = op. P P ′. In particular, m(1,0,0,0) = m. P ′P . In particular, m(1,0,0,0) = mop. 2. It is possible to define mq on the space of double posets. The same arguments prove that it is still associative. However, m(1,0,0,0) is not equal to on HPP and m(0,0,1,0) is not equal to m; for example, if we denote by ℘2 the double poset with two elements x, y, x and y being not comparable for ≤h and ≤r: m(1,0,0,0)( q ⊗ q) = q q + 2℘2, m(0,0,1,0)( q ⊗ q) = q q + 2℘2. 2.3 Subalgebras and quotients These two particular families of plane posets are used in [2, 3]: Definition 18 Let P ∈ PP. 1. We shall say that P is a plane forest if it does not contain q∧q q as a plane subposet. The set of plane forests is denoted by PF . 2. We shall say that P is WN ("without N") if it does not contain q q q(cid:30) nor q q q(cid:31) . The set of q q WN posets is denoted by WN P. Examples. A plane poset is a plane forest if, and only if, its Hasse graph is a rooted forest. PF (0) = {1}, PF (1) = { q}, PF (2) = { q q, q q }, PF (3) = { q q q, q q q , q q q, q q∨qq q , q }, q , q q q q q q, q q q, q q q∨q , q q∨q q q, q q q , q q q, q q , q q∨q q q , q q∨q , q∨qq , q q q q PF (4) = ( q q q q, q q WN P(0) = {1}, WN P(1) = { q}, WN P(2) = { q q, q q }, q∨qq q , q q q q ) , WN P(3) = { q q q, q q q , q q q, q∨qq q q , q , q∧q q }, q q q q q q, q q q q , q q q q q, q q q, q q q∨q , q q∨q q q q∨q q q , q q∨q , q q∨q , q q q q∨q q , q q , q∧q q q , q q, q q q∧q q q , q , q q∧q q , q q, q q∧q q , q∧q q q q∧qq , q q(cid:30)(cid:31) , q q WN P(4) =   Definition 19 We denote by: q , q q q, q q∧ q∨qq .   • HWN P the subspace of HPP generated by WN plane posets. • HPF the subspace of HPP generated by plane forests. • IWN P the subspace of HPP generated by plane posets which are not WN. • IPF the subspace of HPP generated by plane posets which are not plane forests. 11 Note that HWN P and HPF are naturally identified with HPP /IWN P and HPP/IPF . Proposition 20 Let q = (q1, q2, q3, q4) ∈ K 4. 1. HWN P is a subalgebra of (HPP , mq) if and only if, q1 = q2 = 0 or q3 = q4 = 0. 2. HPF is a subalgebra of (HPP , mq) if and only if, q1 = q2 = 0. 3. IWN P and IPF are ideals of (HPP , mq). Proof. 1. ⇐=. We use the notations of proposition 17-1. If q3 = q4 = 0, let us consider two WN posets P and P ′. then the Pi's are also WN, so for any σ ∈ Sk+l, Pσ−1(1) . . . Pσ−1(k+l) is WN. As a conclusion, mq(P ⊗P ′) ∈ HWN P. The proof is similar if q1 = q2 = 0, using proposition 17-2. 1. =⇒. Let us consider the coefficients of q q(cid:30) and q q(cid:31) in certain products. We obtain: q q q q q q(cid:30) q q q q(cid:31) q q mq( q∧qq ⊗ q) mq( q ⊗ q∧q q ) q1q2 3 q2q2 4 q1q2 4 q2q2 3. If HWN P is a subalgebra of (HPP , mq), then these four coefficients are zero, so, from the first row, q1 = 0 or q3 = q4 = 0 and from the second row, q2 = 0 or q3 = q4 = 0. As a conclusion, q1 = q2 = 0 or q3 = q4 = 0. 2. ⇐=. We use the notations of proposition 17-2. If q1 = q2 = 0, let us consider two plane forests P and P ′. Then the Pi's are plane trees, so for any σ ∈ Sk+l, Pσ−1(1) . . . Pσ−1(k+l) is a plane forest. As a conclusion, mq(P ⊗ P ′) ∈ HPF . 2. =⇒. Let us consider the coefficients of q∧q q in certain products. We obtain: mq( q ⊗ q q) mq( q q ⊗ q) q∧qq q2 2 q2 1. If HPF is a subalgebra of (HPP , mq), then q1 = q2 = 0. 3. Let P and P ′ be two plane posets such that P or P ′ is not WN. Let us consider a plane poset R such that the coefficient of R in mq(P ⊗ P ′) is not zero. There exists I ⊆ R, such that R \ I = P and I = P ′. As P or P ′ is not WN, I or R \ I contains q(cid:30) or of q(cid:31) : R is not WN. So mq(P ⊗ P ′) ⊆ IWN P. The proof is similar for IPF , using q(cid:30) and q q(cid:31) , so R contains instead q(cid:30) or q(cid:31) . q∧q q ✷ q q q q q q q q q q q q q q q q q 3 Dual coproducts 3.1 Constructions Dually, we give HPP a family of coproducts ∆q, for q ∈ K 4, defined for all P ∈ PP by: ∆q(P ) = XI⊆P P \I hI 1 hP \I I 2 q rI P \I 3 rP \I I 4 q q q (P \ I) ⊗ I. These coproducts are coassociative; their common counit is given by: ε :(cid:26) HPP −→ K P ∈ PP −→ δ1,P . 12 Examples. We put, for all P ∈ PP, nonempty, ∆q(P ) = ∆(P ) − P ⊗ 1 − 1 ⊗ P . ∆q( q q ) = (q1 + q2) q ⊗ q, ∆q( q q) = (q3 + q4) ⊗ q, 1 + q1q2 + q2 q ) = q2(q3 + q4) q ⊗ q q ) = q1(q3 + q4) q ⊗ q q q ) = (q2 ∆( q ∆( q∨q ∆( q∧q ∆( q q) = q2 ∆( q q ) = q2 ∆( q q q) = (q2 q q q + (q2 2) q ⊗ q q + q2(q3 + q4) q q + q1(q3 + q4) q 1 + q1q2 + q2 q ⊗ q, 2) q q ⊗ q + q2 1 q ⊗ q q + q2 2 q ⊗ q q + q2 q ⊗ q + q2 2 q q ⊗ q, 1 q q ⊗ q, q ⊗ q + (q1 + q2)q3 q ⊗ q q + (q1 + q2)q4 q q ⊗ q, q ⊗ q + (q1 + q2)q4 q ⊗ q q + (q1 + q2)q3 q q ⊗ q, 4) q ⊗ q q + (q2 3 + q3q4 + q2 4) q q ⊗ q. q + q2 4 q ⊗ q 3 q q + q2 3 q ⊗ q 4 q 3 + q3q4 + q2 Dualizing proposition 15: Proposition 21 Let (q1, q2, q3, q4) ∈ K 4. Then:   3.2 Particular cases ∆op (q1,q2,q3,q4) = ∆(q2,q1,q4,q3), (ι ⊗ ι) ◦ ∆(q1,q2,q3,q4) = ∆(q3,q4,q1,q2) ◦ ι, (α ⊗ α) ◦ ∆(q1,q2,q3,q4) = ∆(q2,q1,q3,q4) ◦ α, (β ⊗ β) ◦ ∆(q1,q2,q3,q4) = ∆(q1,q2,q4,q3) ◦ β, (γ ⊗ γ) ◦ ∆(q1,q2,q3,q4) = ∆(q2,q1,q4,q3) ◦ γ. Proposition 22 Let P ∈ PP. Then: 2. 1. ∆(q,q,q,q)(P ) = XI⊆P     3. 4. 5. If P = P1 · · · Pk, where the Pi's are h-connected, qP \II(P \ I) ⊗ I. ∆(0,q2,q3,q4)(P ) = XI h-ideal of P ∆(q1,0,q3,q4)(P ) = XI h-ideal of P ∆(q1,q2,0,q4)(P ) = XI r-ideal of P ∆(q1,q2,q3,0)(P ) = XI r-ideal of P   ∆(0,0,q3,q4)(P ) = XI⊆{1,··· ,k} ∆(0,q2,0,q4)(P ) = XI biideal of P ∆(q1,0,q3,0)(P ) = XI biideal of P qαP (I) 3 13 P \I hI 2 rP \I I 3 q rI P \I 4 q I ⊗ (P \ I), P \I hI 1 rP \I I 4 q rI P \I 3 q (P \ I) ⊗ I. q q hP \I I 1 q q P \I hI 2 rI P \I 4 q I ⊗ (P \ I), hP \I I 2 q q P \I hI 1 rI P \I 3 q (P \ I) ⊗ I. P \I hI 2 rI P \I 4 q P \I hI 1 rI P \I 3 q q q I ⊗ (P \ I), (P \ I) ⊗ I. qαP ({1,...,k}\I) 4 PI ⊗ P{1,··· ,k}\I, with, for all J = {j1, · · · , jl}, 1 ≤ j1 < · · · < jl ≤ k, PJ = Pj1 · · · Pjl, and αP (J) = PiPj. Xi∈J,j /∈J,i<j 6. If P = P1 · · · Pk, where the Pi's are r-connected, ∆(q1,q2,0,0)(P ) = XI⊆{1,··· ,k} qβP (I) 1 qβP ({1,...,k}\I) 2 I ⊗ P P {1,··· ,k}\I, with, for all J = {j1, · · · , jl}, 1 ≤ j1 < · · · < jl ≤ k, PJ = Pj1 · · · Pjl, and βP (J) = PiPj. Xi∈J,j /∈J,i<j 7. ∆(q,0,0,0)(P ) = XP1 P2=P ∆(0,q,0,0)(P ) = XP1 P2=P ∆(0,0,q,0)(P ) = XP1P2=P ∆(0,0,0,q)(P ) = XP1P2=P qP1P2P1 ⊗ P2, qP1P2P2 ⊗ P1, qP1P2P1 ⊗ P2, qP1P2P2 ⊗ P1.   8. ∆(0,0,0,0)(P ) = P ⊗ 1 + 1 ⊗ P if P 6= 1. Proof. We only prove points 1, 2, 5, 7 and 8; the others are proved in the same way. 1. Immediate, as hI I = IP \ I by lemma 13. P\I + hP \I P\I + rP \I I + rI 2. By lemma 16: ∆(q1,0,q3,q4)(P ) = XI⊆P hI P \I =0 P \I hI 1 rI P \I 3 rP \I I 4 q q q (R \ I) ⊗ I = XP \ I h-ideal of P P \I hI 1 rI P \I 3 rP \I I 4 q q q (P \ I) ⊗ I. As ∆(0,q2,q3,q4) = ∆op (0,q2,q4,q3), we obtain also the first assertion. 5. By lemma 16, point 3: ∆(0,0,q3,q4)(P ) = XI⊆P P \I =hP \I hI I =0 rI P \I 3 rP \I I 4 q q (R \ I) ⊗ I = XI⊆{1,··· ,k} q PI P \PI r 3 P \PI PI r 4 q P ⊗ I P{1,··· ,k}\I , and it is immediate that rPI P \PI = αP (I) and rP \PI PI = αP ({1, . . . , k} − I). 7. By lemma 16: ∆(0,0,q,0)(P ) = XI⊆P qrI P \I (P \ I) ⊗ I = XP1P2=P qr P2 P1 P1 ⊗ P2 = XP1P2=P qP1P2P1 ⊗ P2. P \I =hP \I hI I =rP \I I =0 8. Let I ⊆ P , such that hI P \I = hP \I I = rI P \I = rP \I I = 0. Then: IP \ I = hI P \I + hP \I I + rI P \I + rP \I I = 0, 14 so I = 1 or I = P . ✷ Remark. In particular, the coproduct defined in section 1 is ∆(1,0,1,1). The coproduct of deconcatenation, dual of m, is ∆(0,0,1,0) and the coproduct of deconcatenation, dual of , is ∆(1,0,0,0). 3.3 Compatibilities with the products the x′ q 's, y′ q's, y′′ q's, x′′ q 's homogeneous. Then: Proposition 23 Let x, y ∈ HPP. We put ∆q(x) =P x′ 1. ∆q(xy) =XX q 2. ∆q(x y) =XX q q) ⊗ (x′′ q) ⊗ (x′′ q y′′ q y′ q y′′ x′′ 2 x′′ 4 qy′′ q qy′′ q q y′ q q y′ q qy′ x′ 1 x′ 3 (x′ (x′ q ). q ). q q q ⊗ x′′ q and ∆q(y) =P y′ q ⊗ y′′ q , with Proof. We only prove the first point; the proof of the second point is similar. Let P, Q ∈ PP. Then: ∆q(P Q) = XI⊆P, J⊆Q = XI⊆P, J⊆Q q q (P Q)\(IJ ) hIJ 1 h(P Q)\(IJ ) IJ 2 q rIJ (P Q)\(IJ ) 3 r(P Q)\(IJ ) IJ 4 q q ((P Q) \ (IJ)) ⊗ (IJ) (P Q)\(IJ ) hIJ 1 hP Q\IJ IJ 2 q rIJ P Q\IJ 3 rP Q\IJ IJ 4 q q ((P \ I)(Q \ J)) ⊗ (IJ). Moreover, for all I ⊆ P , J ⊆ Q: • For all x ∈ P \ I, y ∈ J, x <r y in P Q, so hJ P \I = 0. Similarly, hI Q\J = 0. Hence: hIJ P Q\IJ = hI P \I + hJ Q\J + hJ P \I + hI Q\J = hI P \I + hJ Q\J . • In the same way, hP Q\IJ J + hP \I • For all x ∈ P \ I, y ∈ J, x <r y in P Q, so rJ I + hQ\J = hP \I IJ J + hQ\J I = hP \I I + hQ\J J . P \I = P \ IJ. For all x ∈ Q \ J, y ∈ I, x >r y, so rI Q\J = 0. Hence: rIJ P Q\IJ = rI P \I + rJ Q\J + rJ P \I + rI Q\J = hI P \I + hJ Q\J + P \ IJ. • In the same way, rP Q\IJ IJ = rP \I I + rQ\J J + rP \I J + rQ\J I = rP \I I + rQ\J J + IQ \ J. So: which is the announced formula. ∆q(P Q) =XX q qQ′′ q P ′ 3 q Q′ q P ′′ 4 q (P ′ qQ′ q) ⊗ (P ′′ q Q′′ q ), ✷ Examples. 1. If q3 = 1, then (HPP , m, ∆q) is a braided Hopf algebra. The braiding is given, for all P, Q ∈ PP, by: cq4(P ⊗ Q) = qP Q 4 Q ⊗ P. In particular, if q4 = 1, it is a Hopf algebra; if q4 = 0, it is an infinitesimal Hopf algebra. 2. If q4 = 0, the compatibility becomes the following: ∆q(xy) =X q xy′′ q 3 (xy′ q) ⊗ y′′ qy x′ 3 x′ q ⊗ (x′′ q y) − qxy 3 x ⊗ y. In particular, if q3 = 1, then it is an infinitesimal Hopf algebra. q +X q 15 3. If q2 = q4 = 0, then for all x, y ∈ HPP: ∆q(xy) = (xy) ⊗ 1 + 1 ⊗ (xy) + ε(x)(∆q(y) − y ⊗ 1 − 1 ⊗ y) +ε(y)(∆q(x) − x ⊗ 1 − 1 ⊗ x) + ε(x)ε(y)1 ⊗ 1. In other terms, for all x, y ∈ HPP , such that ε(x) = ε(y) = 0, xy is primitive. 4 Self-duality results 4.1 A first pairing on Hq Notations. If P, Q ∈ PP, we denote by Bij(P, Q) the set of bijections from P to Q. Definition 24 Let P, Q be two double posets and let σ ∈ Bij(P, Q). We put: φ1(σ) = ♯{(x, y) ∈ P 2 x <h y and σ(x) <h σ(y)} +♯{(x, y) ∈ P 2 x <h y and σ(x) >h σ(y)} +♯{(x, y) ∈ P 2 x <h y and σ(x) <r σ(y)} +♯{(x, y) ∈ P 2 x <r y and σ(x) <h σ(y)}, φ2(σ) = ♯{(x, y) ∈ P 2 x <h y and σ(x) <h σ(y)} +♯{(x, y) ∈ P 2 x <h y and σ(x) >h σ(y)} +♯{(x, y) ∈ P 2 x <h y and σ(x) >r σ(y)} +♯{(x, y) ∈ P 2 x <r y and σ(x) >h σ(y)}, φ3(σ) = ♯{(x, y) ∈ P 2 x <r y and σ(x) <r σ(y)}, φ4(σ) = ♯{(x, y) ∈ P 2 x <r y and σ(x) >r σ(y)}.   Lemma 25 Let P, Q be two double posets and let σ ∈ Bij(P, Q). For all i ∈ {1, . . . , 4}, φi(σ−1) = φi(σ). Proof. Indeed, as σ is a bijection: φ1(σ−1) = ♯{(x, y) ∈ Q2 x <h y and σ−1(x) <h σ−1(y)} +♯{(x, y) ∈ P 2 x <h y and σ−1(x) >h σ−1(y)} +♯{(x, y) ∈ P 2 x <h y and σ−1(x) <r σ−1(y)} +♯{(x, y) ∈ P 2 x <r y and σ−1(x) <h σ−1(y)} = ♯{(x, y) ∈ P 2 x <h y and σ(x) <h σ(y)} +♯{(x, y) ∈ P 2 x >h y and σ(x) <h σ(y)} +♯{(x, y) ∈ P 2 x <h y and σ(x) <r σ(y)} +♯{(x, y) ∈ P 2 x <r y and σ(x) <h σ(y)} = φ1(σ). The other equalities are proved similarly. ✷ Lemma 26 Let P1, P2, Q be double posets. There is a bijection:   Bij(P1P2, Q) −→ [I⊆Q Bij(P1, R \ I) × R(P2, I) σ −→ (σP1, σP2), with I = σ(P2). 16 Let σ ∈ Bij(P1P2, Q) and let (σ1, σ2) be its image by this bijection. Then: φ1(σ) = φ1(σ1) + φ1(σ2) + hI Q\I , φ2(σ) = φ2(σ1) + φ2(σ2) + hQ\I I φ3(σ) = φ3(σ1) + φ3(σ2) + rI Q\I , φ4(σ) = φ4(σ1) + φ4(σ2) + rQ\I . , I   Proof. We put P = P1P2. if x ∈ P1 and y ∈ P2, then x <r y, so: φ3(σ) = ♯{(x, y) ∈ P 2 1 x <r y and σ(x) <r σ(y)} 2 x <r y and σ(x) <r σ(y)} +♯{(x, y) ∈ P 2 +♯{(x, y) ∈ P1 × P2 x <r y and σ(x) <r σ(y)} +♯{(x, y) ∈ P2 × P1 x <r y and σ(x) <r σ(y)} = φ3(σ) + φ3(σ2) + ♯{(x, y) ∈ P1 × P2 σ(x) <r σ(y)} + 0 = φ3(σ) + φ3(σ2) + ♯{(x, y) ∈ (Q \ I) × I x <r y} = φ3(σ1) + φ3(σ2) + rI Q\I . The other equalities are proved similarly. ✷ Theorem 27 Let q = (q1, q2, q3, q4) ∈ K 4. We define a pairing on HPP by: hP, Qiq = Xσ∈Bij(P,Q) qφ1(σ) 1 qφ2(σ) 2 qφ3(σ) 3 qφ4(σ) 3 . This pairing is symmetric and for all x, y, z ∈ HPP, hxy, ziq = hx ⊗ y, ∆q(z)iq. Proof. Let P, Q be two double posets. Then, by lemma 25: hP, Qiq = Xσ∈Bij(P,Q) = Xσ∈Bij(Q,P ) = Xσ∈Bij(Q,P ) = hQ, P iq. qφ1(σ) 1 qφ2(σ) 2 qφ3(σ) 3 qφ4(σ) 3 qφ1(σ−1) 1 qφ2(σ−1) 2 qφ3(σ−1) 3 qφ4(σ−1) 3 qφ1(σ) 1 q+φ2(σ) 2 qφ3(σ) 3 qφ4(σ) 3 So this pairing is symmetric. Let P1, P2, Q be three double posets. By lemma 26: hP1P2, Qiq = Xσ∈Bij(P1P2,Q) qφ1(σ) 1 qφ2(σ) 2 qφ3(σ) 3 qφ4(σ) 3 = XI⊆Q Xσ1∈Bij(P1,Q\I) Xσ2∈Bij(P2,I) φ1(σ1)+φ1(σ2)+hI 1 Q\I q ×q φ2(σ1)+φ2(σ2)+hQ\I 2 I R\I hI 1 hR\I I 2 q rI R\I 3 q q q = XI⊆Q = hP1 ⊗ P2, ∆q(Q)iq. φ3(σ1)+φ3(σ2)+rI q 3 rR\I I 4 hP1, Q \ IiqhP2, Iiq Q\I φ4(σ1)+φ4(σ2)+rQ\I 4 I q So h−, −iq is a Hopf pairing. ✷ 17 Remark. More generally, adapting this proof, it is possible to show that for all x, y, z ∈ HPP: hm(0,0,q,0)(x ⊗ y), ∆(z)i(q1,q2,q3,q4) = hx ⊗ y, ∆(qq1,qq2,qq3,qq4)(z)i(q1,q2,q3,q4). Examples. h q, qiq = 1. The pairing of plane posets of degree 2 is given by the following array: q q q q q q 2q1q2 q1 + q2 q q q1 + q2 q3 + q4 Proposition 28 For all double posets P, Q: hP, Qi(q1,0,q3,q4) = Xσ∈S(P,Q) qφ1(σ) 1 qφ3(σ) 3 qφ4(σ) 4 . Consequently, h−, −i(1,0,1,1) is the pairing h−, −i described in section 1. Proof. Let σ ∈ Bij(P, Q), such that the contribution of σ in the sum defining hP, Qi(q1,0,q3,q4) is non zero. As q2 = 0, necessarily φ2(σ) = 0. Hence, if x <h y in P , then σ(x) <r σ(y) in Q; if σ(x) <h σ(y) in Q, then x <r y in P . So σ ∈ S(P, Q). ✷ 4.2 Properties of the pairing Lemma 29 Let P1, P2, Q be double posets. There is a bijection: Bij(P1 P2, Q) −→ [I⊆Q Bij(P1, R \ I) × R(P2, I) σ −→ (σP1, σP2), with I = σ(P2). Let σ ∈ Bij(P1P2, Q) and let (σ1, σ2) be its image by this bijection. Then:     φ1(σ) = φ1(σ1) + φ1(σ2) + hI φ2(σ) = φ2(σ1) + φ2(σ2) + hI ψ2(σ) = ψ2(σ1) + ψ2(σ2) + hI I + rI Q\I , I + rQ\I I , Q\I + hQ\I Q\I + hQ\I Q\I + hQ\I I , φ3(σ) = φ3(σ1) + φ3(σ2), φ4(σ) = φ4(σ1) + φ4(σ2). Proof. We put P = P1 P2. Let x, y ∈ P . If x <h y, then x, y ∈ P1, or x, y ∈ P2, or x ∈ P1, y ∈ P2. If x <r y, then x, y ∈ P1, or x, y ∈ P2. Moreover, if x ∈ P1 and x ∈ P2, then x <h y. So: φ1(σ) = φ1(σ1) + φ1(σ2) +♯{(x, y) ∈ P1 × P2 (x <h y) and σ(x) <h σ(y))} +♯{(x, y) ∈ P1 × P2 (x <h y) and σ(x) >h σ(y))} +♯{(x, y) ∈ P1 × P2 (x <h y) and σ(x) <r σ(y))} = φ1(σ1) + φ1(σ2) +♯{(x, y) ∈ P1 × P2 σ(x) <h σ(y))} +♯{(x, y) ∈ P1 × P2 σ(x) >h σ(y))} +♯{(x, y) ∈ P1 × P2 σ(x) <r σ(y))} = φ1(σ1) + φ1(σ2) + hI Q\I + hQ\I I + rI Q\I . The other equalities are proved similarly. ✷ 18 Proposition 30 For all x, y, z ∈ HPP, hx y, ziq = hx ⊗ y, ∆(q1q2,q1q2,q1,q2)(z)iq. Proof. Let P1, P2, Q be three double posets. By lemma 29: hP1 P2, Qiq = Xσ∈Bij(P1 P2,Q) qφ1(σ) 1 qφ2(σ) 2 qφ3(σ) 3 qφ4(σ) 3 = XI⊆Q Xσ1∈Bij(P1,Q\I) Xσ2∈Bij(P2,I) φ2(σ1)+φ2(σ2)+hI 2 ×q Q\I +hQ\I I +rQ\I I (q1q2)hI R\I (q1q2)hR\I I rI R\I 1 q q = XI⊆Q = hP1 ⊗ P2, ∆(q1q2,q1q2,q1,q2)(Q)iq, φ1(σ1)+φ1(σ2)+hI 1 q Q\I +hQ\I I +rI Q\I qφ3(σ1)+φ3(σ2) 3 rR\I I 2 hP1, Q \ IiqhP2, Iiq qφ4(σ1)+φ4(σ2) 4 which is the announced formula. ✷ Remarks. 1. In particular, if q = (1, 0, 1, 1), for all P, Q, R ∈ PP: hP Q, Ri = hP ⊗ Q, ∆(0,0,1,0)(R)i = XR=R1R2 hP ⊗ Q, R1 ⊗ R2i. this formula is already proved in [2]. 2. More generally, it is possible to show that for all x, y, z ∈ HPP : hm(q,0,0,0)(x ⊗ y), ∆(z)i(q1,q2,q3,q4) = hx ⊗ y, ∆(qq1q2,qq1q2,qq1,qq2)(z)i(q1,q2,q3,q4). Proposition 31 For all x, y ∈ HPP: hx, β(y)i(q1,q2,q3,q4) = hx, yi(q1,q2,q4,q3), hα(x), α(y)i(q1,q2,q3,q4) = hx, yi(q2,q1,q3,q4), hα(x), γ(y)i(q1,q2,q3,q4) = hx, yi(q2,q1,q4,q3), hγ(x), γ(y)i(q1,q2,q3,q4) = hx, yi(q1,q2,q3,q4). hx, γ(y)i(q1,q2,q3,q4) = hx, yi(q2,q1,q4,q3), hα(x), β(y)i(q1 ,q2,q3,q4) = hx, yi(q1,q2,q4,q3), hβ(x), β(y)i(q1 ,q2,q3,q4) = hx, yi(q2,q1,q3,q4), Proof. Let P, Q be two double posets. We put P ′ = α(P ) and Q′ = α(Q). Note that Bij(P, Q) = Bij(P ′, Q′). Let σ ∈ Bij(P, Q). We denote it by σ′ is we consider it as an element of Bij(P ′, Q′). By definition of P ′ and Q′, it is clear that φ1(σ′) = φ2(σ), φ2(σ′) = φ2(σ), φ3(σ′) = φ3(σ), and φ4(σ′) = φ4(σ). Hence: hP ′, Q′iq1,q2,q3,q4) = Xσ′∈Bij(P ′,Q′) = Xσ∈Bij(P,Q) qφ1(σ′) 1 qφ2(σ′) 2 qφ3(σ′) 3 qφ4(σ′) 4 qφ2(σ) 1 qφ1(σ) 2 qφ3(σ) 3 qφ4(σ) 4 = hP, Qi(q2,q1,q3,q4). The other assertions are proved in the same way. ✷ Remark. In general, the pairing defined by x ⊗ y −→ hx, α(y)i(q1,q2,q3,q4) is not a pairing h−, −iq. However, when q1 = q2, it is possible to prove that for all x, y ∈ HPP, hx, α(y)i(q1,q1,q3,q4) = hx, yi(q1,q1,q3,q4). Similarly, hβ(x), γ(y)i(q1,q1,q3,q4) = hx, yi(q1,q1,q4,q3). 19 4.3 Comparison of pairings with colinear parameters Proposition 32 Let q ∈ K. We define the following map: υq :(cid:26) HPP −→ HPP P ∈ PP −→ qh(P )P, where h(P ) = ♯{(x, y) ∈ P x <h y}. Then υq is an algebra and coalgebra morphism from (cid:0)HPP, m, ∆(qq1,qq2,q3,q4)(cid:1) to (cid:0)HPP, m, ∆(q1,q2,q3,q4)(cid:1). Moreover, for all x, y ∈ HPP: hυq(x), υq(y)i(q1,q2,q3,q4) = hx, yi(qq1,qq2,q3,q4). Proof. Let P1, P2 ∈ PP. Then h(P1P2) = h(P1) + h(P2), so υq(P1P2) = υq(P1)υq(P2), and υq is an algebra morphism. Let P ∈ PP. For any I ⊆ P , as P 2 = I 2 ⊔ (P \ I)2 ⊔ (I × (P \ I)) ⊔ ((P \ I) × I): h(P ) = h(I) + h(P \ I) + hI P \I + hP \I I . Consequently: ∆(q1,q2,q3,q4) ◦ υq(P ) = XI⊆P (qq1)hI P \I (qq2)hP \I I rI P \I 3 rP \I I 4 q q qh(P \I)(P \ I) ⊗ qh(I)I = (υq ⊗ υq) ◦ ∆(qq1,qq2,q3,q4)(P ). So υq is a coalgebra morphism. Let P, Q ∈ PP and let σ ∈ Bij(P, Q). We define: a1 = ♯{(x, y) ∈ P 2 x <h y, σ(x) <h σ(y)} a2 = ♯{(x, y) ∈ P 2 x <h y, σ(x) >h σ(y)} a3 = ♯{(x, y) ∈ P 2 x <h y, σ(x) <r σ(y)} a4 = ♯{(x, y) ∈ P 2 x <h y, σ(x) >r σ(y)} a5 = ♯{(x, y) ∈ P 2 x <r y, σ(x) <h σ(y)} a6 = ♯{(x, y) ∈ P 2 x <r y, σ(x) >h σ(y)} a7 = ♯{(x, y) ∈ P 2 x <r y, σ(x) <r σ(y)} a8 = ♯{(x, y) ∈ P 2 x <r y, σ(x) >r σ(y)}. In order to sum up the notations, we give a following array: ♯{(x, y) ∈ P 2 . . . x <h y, . . . x >h y, . . . x <r y, . . . x >r y, . . . σ(x) <h σ(y)} σ(x) >h σ(y)} σ(x) <r σ(y)} σ(x) >r σ(y)} a1 a2 a3 a4 a2 a1 a4 a3 a5 a6 a7 a8 a6 a5 a8 a7 So φ1(σ) = a1 + a2 + a3 + a5, φ2(σ) = a1 + a2 + a4 + a6, φ3(σ) = a7 and φ4(σ) = a8; h(P ) = a1 + a2 + a3 + a4 and h(Q) = a1 + a2 + a5 + a6. Then: qh(P )q(h(Q)qφ1(σ) 1 qφ2(σ) 2 qφ3(σ) 3 qφ4(σ) 4 Finally: = (q2q1q2)a1 (q2q1q2)a2(qq1)a3(qq2)a4 (qq1)a5 (qq2)a6qa7 = (qq1)φ1(σ)(qq2)φ2(σ)qφ3(σ) qφ4(σ) 4 3 . 3 qa8 4 hυq(P ), υq(Q)i(q1,q2,q3,q4) = Xσ∈Bij(P,Q) = Xσ∈Bij(P,Q) qh(P )q(h(Q)qφ1(σ) 1 qφ2(σ) 2 qφ3(σ) 3 qφ4(σ) 4 (qq1)φ1(σ)(qq2)φ2(σ)qφ3(σ) 3 qφ4(σ) 4 = hP, Qi(qq1,qq2,q3,q4). 20 So υq is an isometry. ✷ Remark. It is easy to prove that for P, Q ∈ PP, υq(P Q) = υq(P ) υq(Q). Similarly, one can prove: Proposition 33 Let q ∈ K. We define the following map: υ′ q :(cid:26) HPP −→ HPP P ∈ PP −→ qr(P )P, where r(P ) = ♯{(x, y) ∈ P x <r y}. Then υ′ q is an algebra and coalgebra morphism from (cid:0)HPP , , ∆(q1,q2,qq3,qq4)(cid:1) to (cid:0)HPP , , ∆(q1,q2,q3,q4)(cid:1). Moreover, for all x, y ∈ HPP: q(y)i(q1,q2,q3,q4) = hx, yi(q1,q2,qq3,qq4). hυ′ q(x), υ′ 4.4 Non-degeneracy of the pairing Lemma 34 For all P ∈ PP(n), S(P, ι(P )) is reduced to a single element and: hP, ι(P )i(q1,0,q3,q4) = q 1 2 . n(n−1) Proof. Clearly, IdP : P −→ P belongs to S(P, ι(P )). Let σ : P −→ P be a bijection. Then σ ∈ S(P, ι(P )) if, and only if, for all i, j ∈ P : • (i ≤h j in P ) =⇒ (σ(i) ≤h σ(j) in P ). • (σ(i) ≤r σ(j)) in P =⇒ (i ≤r j in P ). It is clear S(P, ι(P )) is a submonoid of SP . As the group SP is finite, S(P, ι(P )) is a subgroup of SP . So if σ ∈ S(P, ι(P )), then σ−1 ∈ S(P, ι(P )). So, if σ ∈ S(P, ι(P )): • (i ≤h j in P ) ⇐⇒ (σ(i) ≤h σ(j) in P ). • (i ≤r j in P ) ⇐⇒ (σ(i) ≤r σ(j) in P ). So σ is the unique increasing bijection from P to P , that is to say IdP . Moreover: φ1(IdP ) = ♯{(x, y) ∈ P 2 x <h y and x <r y} + ♯{(x, y) ∈ P 2 x <h y and x >r y} +♯{(x, y) ∈ P 2 x <h y and x <h y} + ♯{(x, y) ∈ P 2 x <r y and x <r y} = 0 + 0 + ♯{(x, y) ∈ P 2 x <h y} + ♯{(x, y) ∈ P 2 x <r y} = ♯{(x, y) ∈ P 2 x < y} = n(n − 1) 2 , φ3(IdP ) = ♯{(x, y) ∈ P 2 (x <r y) and (x <h y)} = 0, φ4(IdP ) = ♯{(x, y) ∈ P 2 (x >r y) and (x >h y)} = 0. So hP, ι(P )i(q1,0,q2,q3) = q 1 2 . n(n−1) For all n ∈ N, we give Sn the lexicographic order. For example, if n = 3: (123) ≤ (132) ≤ (213) ≤ (231) ≤ (312) ≤ (321). 21 ✷ We then define a total order ≪ on PP(n) by P ≪ Q if, and only if, Ψn(P ) ≤ Ψn(Q) in Sn. For example, if n = 3: q q q For all P ∈ PP(n), we put: ≪ q∨q q ≪ q∧q q ≪ q q q ≪ q q ≪ q q q. q m(P ) = min ≪ {Q ∈ PP(n) / S(P, Q) 6= ∅}. Lemma 35 For all P ∈ PP, m(P ) = ι(P ). Proof. By lemma 34, m(P ) ≪ ι(P ). Let Q ∈ PP, such that S(P, Q) 6= ∅. Let us prove that ι(P ) ≪ Q. We denote σ = Ψn(P ) and τ = Ψn(Q); we can suppose that P = Φn(σ) and Q = Φn(τ ). Moreover, it is not difficult to prove that Ψn(ι(P )) = (n · · · 1) ◦ σ. First step. Let us prove that τ (1) ≥ n − σ(1) + 1. In P = Ψn(σ), 1 ≤h j if, and only if, σ(1) ≤ σ(j). So there are exactly n − σ(1) + 1 elements of P which satisfy 1 ≤h j in P . Let α ∈ S(P, Q) (which is non-empty by hypothesis). Then if 1 ≤h j, α(1) ≤r α(j): there are at least n − σ(1) + 1 elements of Q which satisfy α(1) ≤r j. Let us put α−1(1) = i. Then α(i) = 1 ≤ α(1) in Q, that is to say 1 ≤h α(1) or 1 ≤r α(1) in Q. If 1 = α(i) ≤h α(1) in Q, then i ≤r 1 in P , so i = 1: in both cases, 1 ≤r α(1) in P . So there are at least n − σ(1) + 1 elements of Q which satisfy 1 ≤r j. In Q, 1 ≤r j if, and only if, τ (j) ≤ τ (1), so there are exactly τ (1) such elements. As a consequence, τ (1) ≥ n − σ(1) + 1. Moreover, if there is equality, necessarily 1 = α(1) for all α ∈ S(P, Q). Second step. We consider the assertion Hi: if τ (1) = n−σ(1)+1, · · · , τ (i−1) = n−σ(i−1)+1, then τ (i) ≥ n − σ(i) + 1 and, if there is equality, then α(1) = 1, · · · , α(i) = i for all α ∈ S(P, Q). We proved H0 in the first step. Let us prove Hi by induction on i. Let us assume Hi−1, 1 ≤ i ≤ n, and let us prove Hi. Let α ∈ S(P, Q) (non empty by hypothesis). By Hi−1, as the equality is satisfied, α(1) = 1, · · · , α(i − 1) = i − 1. In P , the number Ni of elements j such that i ≤h j is the cardinality of σ−1({σ(i), · · · , n}) ∩ {i, · · · , n}, so Ni = n − σ(i) + 1 − {k ≤ i / σ(k) > σ(i)}. Using α, there exists at least Ni elements j of Q such that α(i) ≤r j in Q. Let us put α−1(i) = j. As α(1) = 1, · · · , α(i − 1) = i − 1, j ≥ i and i ≤ α(i). If i ≤h α(i) in Q, then j ≤r i in P , so j = i. So we always have i ≤r α(i) in Q, so at least Ni elements k of Q satisfy i ≤r j in Q. Hence: n − σ(i) + 1 − {k ≤ i / σ(k) > σ(i)} ≤ τ (i) − {k ≤ i / τ (k) < τ (i)} Ni ≤ τ −1({1, · · · , τ (i)}) ∩ {i, · · · , n} ≤ τ (i) − {k ≤ i / n − σ(k) + 1 < n − σ(i) + 1} ≤ τ (i) − {k ≤ i / σ(k) > σ(i)}, so τ (i) ≥ n − σ(i) + 1. If there is equality, then necessarily i = α(i) for all α ∈ S(P, Q). Conclusion. The hypothesis Hi is true for all 0 ≤ i ≤ n. So τ ≥ (n · · · 1) ◦ σ in Sn, so ✷ ι(P ) ≪ Q in PP(n). As a conclusion, ι(P ) ≪ m(P ). Corollary 36 Let us assume that q2 = 0. Then the pairing h−, −iq is non-degenerate if, and only if, q1 6= 0. Proof. Let us fix an integer n ∈ N. We consider the basis PP(n) of H(n), totally ordered by ≪. In this basis, the matrix of h−, −iq restricted to H(n) has the following form, coming from 22 lemmas 34 and 35:   n!n(n−1) 0 ... ... 0 n(n−1) 2 q 1 · · · ... n(n−1) q 1 2 ∗ · · · ... ... ... · · · 0 n(n−1) q 1 2 ... · · · n(n−1) 2 q 1 ∗ ... ... ∗ ,   so its determinant is ±q 2 1 . Hence, h−, −iq is non-degenerate if, and only if, q1 6= 0. ✷ Remark. With the help of the isometry α (proposition 31), it is possible to prove that if q1 = 0, h−, −iq is non-degenerate if, and only if, q2 6= 0. Corollary 37 If q1, q2, q3, q4 are algebraically independent over Z, then the pairing h−, −iq is non-degenerate. Proof. For all n, let us consider the matrix of the pairing restricted to HPP(n) in the basis formed by the plane posets of degree n. Its determinant Dn is clearly an element of Z[q1, q2, q3, q4]. Moreover, Dn(1, 0, q3, q4) 6= 0 by corollary 36, so Dn is a non-zero polynomial. As a consequence, if q1, q2, q3, q4 are algebraically independent over Z, Dn(q1, q2, q3, q4) 6= 0. ✷ 5 Morphism to free quasi-symmetric functions 5.1 A second Hopf pairing on H(q1,0,q,1,q4) We here assume that q2 = 0 and q1 = q3. For all double poset P , ∆(q1,0,q1,q4)(P ) = XI h-ideal of P hI R\I +rI 1 R\I rP \I I 4 q q (P \ I) ⊗ I. For any h-ideal I of P , hP \I I = 0, so: hI R\I + rI R\I = ♯{(x, y) ∈ (P \ I) × I x < y}, rP \I I = rP \I = ♯{(x, y) ∈ (P \ I) × I x > y}. I + hP \I I Hence: ∆(q1,0,q1,q4)(P ) = XI h-ideal of P q♯{(x,y)∈(P \I)×Ix<y} 1 q♯{(x,y)∈(P \I)×Ix<y} 4 (P \ I) ⊗ I. Notations. Let P, Q ∈ P(n) and let σ ∈ Bij(P, Q). As totally ordered sets, P = Q = {1, . . . , n}, so σ can be seen as an element of the symmetric group Sn. Its length ℓ(σ) is then its length in the Coxeter group Sn [5]. Theorem 38 We define a pairing h−, −i′ q : Hq ⊗ Hq −→ K by: hP, Qi′ q = Xσ∈S′(P,Q) n(n−1) 2 −ℓ(σ) q 1 qℓ(σ) 4 , for all P, Q ∈ PP, with n = deg(P ). Then h−, −i′ on the braided Hopf algebra Hq = (H, m, ∆q). q is an homogeneous symmetric Hopf pairing 23 Proof. This pairing is clearly homogeneous. For any double posets P and Q, the map from S′(P, Q) to S′(Q, P ) sending σ to its inverse is a bijection and conserves the length: this implies that the hP, Qi′ q = hQ, P i′ q. Let P, Q, R be three double posets. There is a bijection: Bij(P Q, R) −→ [I⊆R σ −→ (σP , σQ), Bij(P, R \ I) × Bij(Q, I) Θ :  with I = σ(Q). Let us first prove that σ ∈ S′(P Q, R) if, and only if, I is a h-ideal of R and (σ1, σ2) ∈ S′(P, R \ I) × S′(Q, I). =⇒. Let x′ ∈ I and let y′ ∈ R such that x′ ≤h y′. We put x′ = σ(x) and y′ = σ(y). As σ ∈ S′(P Q, R), x ≤ y. As x ∈ Q, y ∈ Q, so y′ ∈ I: I is a h-ideal of R. By restriction, σP ∈ S′(P, R \ I) and σQ ∈ S′(Q, I). ⇐=. Let us assume that x ≤h y in P Q. Then x, y ∈ P or x, y ∈ Q. As σP ∈ S′(P, R \ I) and σQ ∈ S′(Q, I), σ(x) ≤ σ(Q) in R. Let us assume that σ(x) ≤h σ(y) in R. As I is a h-ideal, there are two possibilities: • σ(x), σ(y) ∈ R \ I or σ(x), σ(y) ∈ I. Hence, x, y ∈ P or x, y ∈ Q. As σP ∈ S′(P, R \ I) and σQ ∈ S′(Q, I), x ≤ y in P Q. • σ(x) ∈ R \ I and σ(y) ∈ I. Hence, x ∈ P and y ∈ Q, so x ≤ y in P Q. Finally, we obtain a bijection: S′(P Q, R) −→ [I h-ideal of R σ −→ (σP , σQ), S′(P, R \ I) × S′(Q, I) Θ :  Let σ ∈ S′(P Q, R) and we put Θ(σ) = (σ1, σ2). Let R \ I = {i1, . . . , ik} and I = {j1, . . . , jl}. Let ζ ∈ Sk+l defined by ζ −1(1) = i1 . . . , ζ −1(k) = ik, ζ −1(k + 1) = j1, . . . , ζ −1(k + l) = jl. Then ζ is a (k, l)-shuffle and σ = ζ −1 ◦ (σ1 ⊗ σ2), so ℓ(σ) = ℓ(ζ) + ℓ(σ1) + ℓ(σ2). Moreover: ℓ(ζ) = ♯{(p, q) jq < ip} = ♯{(i, j) ∈ (R \ I) × I j < i} = rR\I I + hR\I I = rR\I I , as I is a h-ideal of R. Finally, ℓ(σ) = ℓ(σ1) + ℓ(σ2) + rR\I lemma 13: I . Moreover, if R = n and I = k, by hI R\I + rI R\I + rR\I I = hI R\I + rI R\I + hR\I I + rR\I I = k(n − k), so: k(k − 1) 2 k(k − 1) + (n − k)(n − k − 1) 2 (n − k)(n − k − 1) 2 − rR\I I hI R\I + rI R\I + = k(n − k) + = n(n − 1) 2 2 − rR\I . I + 24 Finally: hP, Qi′ n(n−1) 2 −ℓ(σ) q 1 qℓ(σ) 4 σ1∈S′(P,R\I),σ2∈S′(Q,I) q = Xσ∈S′(P Q,R) = XI h-ideal of R = XI h-ideal of R = XI h-ideal of R = hP ⊗ Q, ∆q(R)i′ q. X X n(n−1) 2 −ℓ(σ1)−ℓ(σ2)−rR\I I q 1 ℓ(σ1)+ℓ(σ2)+rR\I 4 I q R\I + k(k−1) 2 + (n−k)(n−k−1) 2 hI R\I +rI 1 q ℓ(σ1)+ℓ(σ2)+rR\I 4 I q σ1∈S′(P,R\I),σ2∈S′(Q,I) hI R\I +rI 1 q R\I rR\I I 4 q hP, R \ Ii′ qhQ, Ii′ q So this pairing is a Hopf pairing on Hq. ✷ Remark. In particular, h−, −i′ (1,0,1,1) is the pairing h−, −i′ of the section 1. Examples. Here are the matrices of the pairing h−, −i′ q restricted to Hq(n), for n = 1, 2, 3. q 1 q q q q1 q1 q q q q q q q1 q1 + q4 q∨qq q3 1 q2 1(q1 + q4) q∧q q q3 1 q3 1 q q q q3 1 q2 1(q1 + q4) q q q q3 1 q3 1 q2 1(q1 + q4) q3 1 q2 1(q1 + q4) q2 1(q1 + q4) q2 1(q1 + q4) q3 1 q2 1(q1 + q4) 1 + q4 q1(q2 1) 1 + q1q4 + q2 4) q1(q2 q2 1(q1 + q4) 1 + q4 q1(q2 1) q2 1(q1 + q4) 1 + q1q4 + q2 q1(q2 4) q3 1 q3 1 q2 1(q1 + q4) q q q q3 1 q3 1 q3 1 q3 1 q3 1 q3 1 q q q q q∨q q∧q q q q q q q q q q q q q q q3 1 q2 1(q1 + q4) q2 1(q1 + q4) 1 + q1q4 + q2 q1(q2 4) 1 + q1q4 + q2 q1(q2 4) (q1 + q4)(q2 1 + q1q4 + q2 4) Proposition 39 For all x, y, z ∈ Hq, hx y, zi′ q = hx ⊗ y, ∆(q1,0,q1,0)(z)i′ q. Proof. Let P, Q, R ∈ PP. We consider the following map: Bij(P Q, R) −→ [I⊆R σ −→ (σP , σQ), Bij(P, R \ I) × Bij(Q, I) Θ :  with I = σ(Q). Let us first prove that σ ∈ S′(P Q, R) if, and only if, I is a biideal of R and (σ1, σ2) ∈ S′(P, R \ I) × S′(Q, I). =⇒. Obviously, (σ1, σ2) ∈ S′(P, R \ I) × S′(Q, I). Let x′ ∈ I, y′ ∈ R, such that x′ ≤ y′. We put x′ = σ(x) and y′ = σ(y). If y /∈ Q, then y ∈ P and x ∈ Q, so y <h x. As σ ∈ S′(P Q, R), y′ < x′: contradiction. So y ∈ Q and y′ ∈ I. ⇐=. Let x, y ∈ P Q, such that x ≤h y. Two cases are possible. • If x, y ∈ P or x, y ∈ Q, as (σ1, σ2) ∈ S′(P, R \ I) × S′(Q, I), σ(x) ≤ σ(y) in R. • If x ∈ P and y ∈ Q, as I = σ(Q) is a biideal, it is not possible to have y ≤ x, so x ≤ y. 25 Let x, y ∈ P Q, such that σ(x) ≤h σ(y) in R. As I = σ(Q) is a biideal, two cases are possible. • If x, y ∈ P or x, y ∈ Q, as (σ1, σ2) ∈ S′(P, R \ I) × S′(Q, I), x ≤ y in P Q. • If x ∈ P and y ∈ Q, then x ≤ y in P Q. Moreover, if I = σ(Q) is a biideal of R, then I = {k + 1, . . . , k + l} ⊆ R, where k = P and l = Q. Then σ = σ1 ⊗ σ2, so ℓ(σ) = ℓ(σ1) + ℓ(σ2). So, with n = R: hP Q, Ri′ q 1 q = XI biideal of R = XI biideal of R = XI biideal of R = XI biideal of R = XI biideal of R = hP ⊗ Q, ∆(q1,0,q1,0)(R)i′ q, n(n−1) 2 −ℓ(σ1)−ℓ(σ2) q 1 qℓ(σ1)+ℓ(σ2) 4 hP ⊗ Q, (R \ I) ⊗ Ii′ q X σ1∈S′(P,R\I),σ2∈S′(Q,I) n(n−1) 2 − k(k−1) 2 − (n−k)(n−k−1) 2 q(n−k)k 1 hP ⊗ Q, (R \ I) ⊗ Ii′ q qR\I.I 1 hP ⊗ Q, (R \ I) ⊗ Ii′ q hI R\I +rI 1 R\I q hP ⊗ Q, (R \ I) ⊗ Ii′ q with the observation that, as I is a biideal, hR\I I = rR\I I = 0, so R \ I.I = hI R\I + rI R\I . ✷ Remark. In particular, if q = (1, 0, 1, 1), for all P, Q, R ∈ PP: hP Q, Ri′ = hP ⊗ Q, ∆(1,0,1,0)(R)i′ = XI biideal of R hP ⊗ Q, (R \ I) ⊗ Ii′. This formula was already proved in [3]. 5.2 Quantization of the Hopf algebra of free quasi-symmetric functions Theorem 40 1. We define a coproduct on FQSym in the following way: for all σ ∈ Sn, ∆q(σ) = n Xk=0 k(n−k)−ℓ(σ)+ℓ(σ(1) 1 k )+ℓ(σ(2) k ) ℓ(σ)−ℓ(σ(1) 4 k )−ℓ(σ(2) k ) q q k ⊗ σ(2) σ(1) k . For all q ∈ K, FQSymq = (FQSym, m, ∆q) is a graded braided Hopf algebra. 2. One defines a Hopf pairing on FQSymq by hσ, τ i′ = q n(n−1) 2 −ℓ(σ) 1 qℓ(σ) 4 δσ,τ −1. 3. The following map is an isomorphism of braided Hopf algebras: Hq −→ FQSymq P ∈ PP −→ Xσ∈SP σ. Θ :  Moreover, for all x, y ∈ Hq, hΘ(x), Θ(y)i′ q = hx, yi′ q. Remarks. 26 1. Let σ ∈ Sn. If 0 ≤ k ≤ n, we put σ({1, . . . , k}) = {i1, . . . , ik} and σ({k + 1, . . . , n}) = {j1, . . . , jl}. Let ζ be the (k, l)-shuffle defined by ζ −1(1) = i1 . . . , ζ −1(k) = ik, ζ −1(k + 1) = k ) + ℓ(σ(2) j1, . . . , ζ −1(k + l) = jl. Then σ = ζ −1 ◦ (σ(1) k ). Moreover, 0 ≤ ℓ(ζ) ≤ k(n − k), so: k ), and ℓ(σ) = ℓ(ζ) + ℓ(σ(1) k ⊗ σ(2) 0 ≤ ℓ(σ) − ℓ(σ(1) k ) − ℓ(σ(2) k ) ≤ k(n − k). As a consequence, the coproduct on FQSymq is well-defined, even if q1 = 0 or q4 = 0. 2. It is clear that the pairing h−, −i′ q defined on FQSymq is non degenerate if, and only if, q1 6= 0 and q2 6= 0. Proof. We already know that Θ is an algebra isomorphism. Moreover, it is also proved there that S′(P, Q) = S(P ) ∩ S(Q)−1 for any P, Q ∈ P P , so: hΘ(P ), Θ(Q)i′ q = = = hσ, τ i′ q Xσ∈S(P ),τ ∈S(Q) Xσ∈S(P ),τ ∈S(Q) Xσ∈S(P )∩S(Q)−1 n(n−1) 2 −ℓ(σ) q 1 qℓ(σ) 4 δσ,τ −1 n(n−1) 2 −ℓ(σ) q 1 qℓ(σ) 4 n(n−1) 2 −ℓ(σ) q 1 qℓ(σ) 4 = Xσ∈S′(P,Q) = hP, Qi′ q. So Θ is an isometry. As a consequence, for all x ∈ Hq, y, z ∈ FQSymq: h∆ ◦ Θ(x), y ⊗ zi′ q = hΘ(x), yzi′ q = hx, Θ−1(yz)i′ q = hx, Θ−1(y)Θ−1(z)i′ q = h∆q(x), Θ−1(y) ⊗ Θ−1(z)i′ q = h(Θ ⊗ Θ) ◦ ∆q(x), y ⊗ zi′ q. Let us now assume that q1, q4 6= 0. Then the pairing h−, −i′ q on FQSymq is non-degenerate, so ∆ ◦ Θ = (Θ ⊗ Θ) ◦ ∆. Hence, Θ is an isometric isomorphism of braided Hopf algebras. This proves the three points immediately. Up to an extension, we can assume that K is infinite. It is not difficult to show that the set A of elements (q1, q4) ∈ K 2 such that the three points are satisfied is given by polynomial equations with coefficients in Z, As it contains (K − {0})2 from the preceding observations, which is dense in K 2, it is is equal to K 2, so the result also holds for any (q1, q4). ✷ Corollary 41 The pairing h−, −i′ q is non-degenerate if, and only if, q1 6= 0 and q4 6= 0. Proof. As the isometric pairing h−, −i′ q on FQSymq is non-degenerate if, and only if, q1 6= 0 ✷ and q4 6= 0. 27 References [1] Gérard Duchamp, Florent Hivert, and Jean-Yves Thibon, Noncommutative symmetric func- tions. VI. Free quasi-symmetric functions and related algebras, Internat. J. Algebra Comput. 12 (2002), no. 5, 671 -- 717, arXiv:math/0105065. [2] Loïc Foissy, Algebraic structures on double and plane posets, to be published in Journal of Algebraic Combinatorics, arXiv:1101.5231, 2011. [3] , Plane posets, special posets, and permutations, arXiv:1109.1101, 2011. [4] Loïc Foissy and Jeremie Unterberger, Ordered forests, permutations and iterated integrals, Int. Math. Res. Notices (2012), doi: 10.1093/imrn/rnr273 F, arXiv:1004.5208. [5] James E. Humphreys, Reflection groups and Coxeter groups, Cambridge Studies in Advanced Mathematics, vol. 29, Cambridge University Press, Cambridge, 1990. [6] Jean-Louis Loday and María Ronco, On the structure of cofree Hopf algebras, J. Reine Angew. Math. 592 (2006), 123 -- 155, arXiv:math/0405330. [7] Clauda Malvenuto and Christophe Reutenauer, Duality between quasi-symmetric functions and the Solomon descent algebra, J. Algebra 177 (1995), no. 3, 967 -- 982. [8] Claudia Malvenuto and Christophe Reutenauer, A self-dual Hopf algebra on double partially ordered sets, arXiv:0905.3508. 28
1612.05303	2	1612	2016-12-25T01:01:15	A Dichotomy for GK dimensions of simple modules over simple differential rings	[ "math.RA" ]	The Gelfand--Kirillov dimension has gained importance since its introduction as an tool in the study of non-commutative infinite dimensional algebras and their modules. In this paper we show a dichotomy for the Gelfand--Kirillov dimension of simple modules over certain simple rings of differential operators.	math.RA	math	A dichotomy for the Gelfand–Kirillov dimensions of simple modules over simple differential rings Ashish Gupta∗and Arnab Dey Sarkar† Abstract The Gelfand–Kirillov dimension has gained importance since its intro- duction as an tool in the study of non-commutative infinite dimensional algebras and their modules. In this paper we show a dichotomy for the Gelfand–Kirillov dimension of simple modules over certain simple rings of differential operators. We thus answer a question of J. C. McConnell in Representations of solvable Lie algebras V. On the Gelfand-Kirillov di- mension of simple modules. J. Algebra 76 (1982), no. 2, 489–493 concern- ing this question for a class of algebras that arise as simple homomorphic images of solvable lie algebras. We also determine the Gelfand–Kirillov dimension of an induced module. 1 Introduction The usefulness of the Gelfand-Kirllov dimension (GK dimension) as a tool in the investigation of non-commutative algebras is now well established. This is due to the fact that it is relatively easier to compute and work with. It is an exact dimension function for the finitely generated modules over almost commutative algebras. Moreover, for a finitely generated module over an almost commutative algebra the GK dimension is given by the degree of the Hilbert- Samuel polynomial which has coefficients from Q. Thus the GK dimension of such a module is a non-negative integer. It is an easy consequence of the Hilbert Nullstellensatz that each simple module over an affine commutative algebra has GK dimension equal to zero (e.g., [8, Lemma 1.17]). It was once believed that such a stability holds true for the GK dimension of simple modules over the Weyl algebras An(K) over a field K of characteristic zero. But in 1985 certain counterexamples were constructed by J. T. Stafford and it is now known that An(C) has a simple module Sj with GK dimension j for each n ≤ j ≤ 2n − 1 ([2]). Our aim in this paper is to show that for certain simple rings a weak stability holds in that the GK dimensions of simple modules always lie in a two element set containing the two extreme values. Note that simple rings are not amenable to any methods that are based on the study of ideals. Thus studying the simple modules of such rings may be helpful in learning about their structure. For example, it was shown by J. C. McConnell in [7] that for a simple algebra Rn+1 that occurs as an image of the universal enveloping algebra of a ∗Corresponding author †Graduate student. 1 solvable lie algebra there exist simple modules having GK dimension as low as one and as high as n where n+1 denotes the GK dimension of the algebra Rn+1. McConnell's example is the ring defined as follows. Let k be an extension field of Q such that dimQ k = ∞ and let λi, i ∈ {1, · · · , n} be linearly independent over Q. Let Gn be the free abelian group of rank n having the basis x1, x2, · · · , xn and set Kn = k[x±1 1 , x±1 2 , · · · , x±1 n ]. It is the group algebra kGn of Gn over k. Then Rn+1 is the Ore extension Kn[x, δ], where δ is the derivation of Kn defined by δ(xj ) = λjxj . The assump- tion that the scalars λi are linearly independent over Q means that Rn+1 is a simple ring [7]. By [7, Theorem 2.1] it has GK dimension n + 1. The question was asked whether there exist simple Rn+1-modules having GK dimension α where 1 < α < n. In this article we answer this question in the negative and show the following. Theorem A. Let Kn be the laurent polynomial algebra over a field k and let Kn := k[x±1 1 , · · · , x±1 n ] Rn+1 := Kn[x, δ], where δ is a k-derivation of Kn of the form δ = n X i=1 gi ∂ ∂xi . for gi ∈ kxi + k such that Rn+1 is a simple ring. Let M be a simple Rn+1- module. Then G K -dim(M ) ∈ {1, n}. Remark 1.1. Clearly, J. C. McConnell's example is obtained as a special case by taking gi = λixi. For a finitely generated module Q over an affine algebra A, the following inequalities hold with regard to the GK dimension 0 ≤ G K -dim(Q) ≤ G K -dim(A). (1) We refer to [3, Proposition 5.1] for a proof of this fact. In the case of the algebra Rn+1 in Theorem A the possibility G K -dim(M ) = G K -dim(Rn+1) = n + 1 is easily ruled out (e.g., [7, Proposition 2.3] for a simple module M . In the paper [7], it was also shown that G K -dim(M ) = 0 is not possible. In this case M would necessarily be finite dimensional over k implying that dimk(Endk(M )) < ∞. Rn+1 is a simple domain and so M is a faithful simple module and so the last inequality says that Rn+1 is finite dimensional over k which is clearly false. It follows that 1 ≤ G K -dim(M ) ≤ n. 2 Other examples of simple rings where such a dichotomy holds for the GK dimensions of simple modules are known. An example is the skew-Laurent extension Sn+1 over Kn defined as Sn+1 = Kn[x, σ] where σ is the automorphism of Kn defined by σ(xi) = λixi, where λj ∈ k∗ := k \ {0} are assumed to generate a subgroup of k∗ of the maximal possible rank n. The following theorem was proved in [6]. Theorem 1.2. Let M be a simple Sn+1-module. Then G K -dim(M ) ∈ {1, n}. Moreover for each j ∈ {1, n} there exists a simple Sn+1-module M with G K -dim(M ) = j. Our approach as exposed in this article is applicable to a wider class of rings than considered in this paper and can be easily adapted to more general Ore extensions Kn[x, σ, δ] that are simple rings with suitable conditions on the derivation or automorphism involved. In this paper all modules will be right modules. 2 GK dimension, Ore localization and critical modules In this section we review some basic facts on the GK dimension of finitely generated modules over affine algebras and also the basic tools that we will be employing in our proofs. The GK dimension of an affine k-algebra A is defined as follows. Let a generating set {a1, a2, · · · , as} be given for A. For convenience we may assume that 1 ∈ {a1, · · · , as}. There is an increasing sequence of k-subspaces of A which is given by n A0 := k, A1 := X j=1 kaj and, An := An 1 . This defines the standard finite dimensional filtration of A with respect to the given choice of generators. The GK dimension of A measures the asymptotic growth rate of the corresponding sequence of dimensions (measured over k). G K -dim(A) := lim n→∞ sup logn(dimk An). It turns out that this definition is independent of the choice of a generating set for A. This idea can be extended to a finitely generated A-module M as follows. Pick a finite dimensional generating subspace M0 of M and set Mn = M0Am. We define G K -dim(M ) := lim n→∞ sup logn dimk Mn (2) Again the GK dimension of M doesn't depend on the choice of a generating set of A or the generating subspace of M . We refer the interested reader to [3] for the proofs of these facts and an excellent treatment of the GK dimension. The GK dimension is rather well-behaved on the finitely generated modules over affine algebras. 3 Proposition 2.1. Let R be an affine algebra with a finitely generated module R is a finitely generated R′- MR. submodule of M then If R′ is an affine subalgebra of R and M ′ G K -dim(M ′ R′ ) ≤ G K -dim(MR) Proof. See Lemma 1.13(ii) of [8]. Proposition 2.2. Let A be an affine commutative k-algebra. Let 0 → N → M → L → 0 be a sequence of A-modules. Then G K -dim(M ) = max{G K -dim(N ), G K -dim(M/N )}. The last proposition is actually true for almost commutative algebras [3]. Proof. See [3, Proposition 5.1(ii)]. Recall that a multiplicative subset S of a ring R is called a right Ore subset if for any s ∈ S and r ∈ R the intersection sR ∩ rS is nonempty. A right Ore set in a domain R is always a right denominator set and R has a right (Ore) localization with respect to S denoted as RS−1. We refer the interested reader to [4, Chapter 8] for the proofs of the above statements further details. Let M be a right R-module and X a right Ore subset in R. The set of all X-torsion elements, namely, TX (M ) := {m ∈ M mx = 0, x ∈ X} is a submodule of M called the X-torsion submodule of M and is denoted as tX (M ). It arises as the kernel of the natural map M → M S−1 where M S−1 denotes the corresponding localization of M at S. Our approach to the determination of the GK dimensions of simple modules will involve the use of localization. To this end we note the following fact. Lemma 2.3. In the notation of Theorem A, the subsets of Rn+1 of the form X := kH \ {0} where H ≤ Gn are Ore subsets in Rn. Proof. This follows from [4, Exercise 10R] noting that X is trivially an Ore subset in the commutative domain Kn = kGn. In [1] C. J. B. Brookes and J. R. J. Groves gave an interesting characteri- zation of the GK dimension of finitely generated modules over crossed products of a free abelian group over a division ring. Since the ring Kn is of this type and we shall be making use of this property which we state below. Proposition 2.4 (Brookes and Groves). Let N be a finitely generated Kn = k[x±1 n ]-module. Then G K -dim(N ) equals the maximal integer t so that N is not torsion as k[x±1 1 , · · · , x±1 ]-module. 1 , · · · , x±1 t Proof. This follows from [1, Lemma 2.6] where this number is shown to be equal to the dimension of N in the sense of [1, Definition 2.2]. It is also shown in the paragraph following Proposition 4.2 of [1] that this dimension coincides with the GK dimension. 4 We also recall the definition of a GK-critical module. We shall need this notion for modules over Kn. But the definition can be made for an arbitrary affine algebra. Definition 2.5. Let A be an affine k-algebra and let M be non-zero finitely generated A-module. Then M is said to be GK-critical if G K -dim(M/N ) < G K -dim(M ) holds for each submodule N < M with N 6= 0. In particular each simple module is GK-critical. We shall also refer to a GK-critical module as simply critical. The usefulness of working with critical modules owes to the following fact. Proposition 2.6. Each nonzero Kn = kGn-module contains a critical submod- ule. Proof. This follows from the more general result which holds for twisted group algebras (c.f. [5, Proposition 4.3]). Proposition 2.7. Let M be a critical Kn-module. Then each nonzero submod- ule N of M is also critical. Proof. This assertion follows easily noting Proposition 2.2. Lemma 2.8. Let M be a Kn-module and let s be maximal with respect to the property that M is not torsion as K0 := k[x±1 s ]-module. Then M cannot embed a free K0-module of infinite rank. 1 . · · · , x±1 Proof. The assertion in the lemma is a special case of [1, Lemma 2.3]. 3 Proof of the main theorem Theorem A. Let Kn be the Laurent polynomial algebra over a field k and let Kn := k[x±1 1 , · · · , x±1 n ] Rn+1 := Kn[x, δ], where δ is a k derivation of Kn of the form δ = n X i=1 gi ∂ ∂xi . for gi ∈ kxi + k. Suppose that M is a simple Rn-module. Then G K -dim(M ) ∈ {1, n}. Proof. Set K := Kn and R := Rn. Suppose that a simple R-module M has GK dimension m such that 2 ≤ m ≤ n − 2. We show below that a contradiction necessarily results. Using Propositions 2.6 and 2.7, let N be a cyclic and critical K-submodule of M . If N R ∼= N ⊗K R then by lemma 3.1 below, G K -dim(M ) = G K -dim(N R) = G K -dim(N ) + 1. 5 Note that in this case N is a simple K-module since R is a flat K-module. Hence G K -dim(N ) = 0 as each simple K-module is finite dimensional (Hilbert Nullstellensatz) and so by the above equation we get G K -dim(M ) = 1. We now suppose that M 6∼= N ⊗K R. For d ≥ 1, pick a subset I := {i1, · · · , id} of {1, · · · , n} that is maximal with respect to the property that N is not S-torsion, where S is the set of nonzero elements of the subalgebra K ′ := k[x±1 i1 , · · · , x±1 id ] of Kn generated by the xi and their inverses. Without loss of generality we may assume I = {1, · · · , d}. If d = n then N and so M embeds a copy of K. By a well-known fact G K -dim(K) = n. It now follows from Proposition 2.1 that G K -dim(M ) ≥ n. But on the other hand G K -dim(M ) ≤ G K -dim(R) − 1 in view of Proposition 5.1(e) of [3] and so G K -dim(M ) ≤ n. This means that G K -dim(M ) = n. In this case also we are done. Hence we may suppose that d < n. Set S := K ′ \ {0} By Lemma 3.3, we know M does not embed an infinite direct sum of copies of K ′. As noted in Lemma 2.3 S is a right Ore subset in R and we may localize R at S. The resulting ring RS −1 which we will denote by T has the form: T := RS −1 = K[x, δ]S −1 = Qd[xd+1, · · · , xn][x, δ] where Qd denotes the fraction field of K ′. Note that the derivation δ of K restricts to a derivation on K ′ that extends uniquely to a derivation of Qd in accordance with the quotient rule for ordinary derivatives. As M is not S-torsion the T -module V := T S −1 6= 0. We claim that V is finite dimensional as Qd-space. Indeed, if this were not true M would necessarily embed a free K ′-module of infinite rank. But this is contrary to the assertion of Lemma 3.3. Lemma 3.1 (GK dimension of induced modules). Let N be a cyclic Kn sub- module of an Rn+1-module M . Then G K -dim(N ⊗Kn Rn+1) = G K -dim(N ) + 1. Proof. We write R for Rn+1 and K for Kn. Also let M := N ⊗K R. By (2) , G K -dim(M ) := lim n→∞ sup logn dimk N0Rn, where N0 denotes a finite dimensional generating subspace of N (and so also a finite dimensional generating subspace of M ). Let {Rm}m∈N denote the standard finite dimensional filtration of R with respect to the generating set {x±1 n , x}. We can write 1 , · · · , x±1 Rm = (⊕m−1 i=0 Kixm−i) ⊕ Km ⊕ (⊕m−1 i=0 Kixi−m) 6 where {Km}m∈N denotes the standard finite dimensional filtration of K with respect to the generating set {x±1 1 . · · · , x±1 n }. Now, dimk N0Rm = dimk[(⊕m−1 i=0 Kixm−i) ⊕ Km ⊕ (⊕m−1 i=0 Kixi−m)] = m−1 X i=0 dimk Ni + dimk Nm, (3) (4) where {Ni}i∈N stands for the standard filtration of the K-module N with respect to the generating subspace N0 and the filtration {Km}m∈N of K. Here we are using our assumption that the derivation δ has the form given in the theorem. This ensures that any word in x±1 n , x of length k may be expressed as a linear combination of words of length at most k in which the only power is x is at the end. 2 , · · · , x±1 1 , x±1 It is well known that for finitely generated module N over an affine com- mutative algebra there exists a polynomial HN , namely, the Hilbert polynomial such that for a standard finite dimensional filtration (Nm)∞ m=0 dimk(Nm) = HN (m) for almost all natural numbers m. In this case it is easily seen that G K -dim(M ) equals the degree d of HN . In view of (3) we have, dimk Nm = 2 m X t=0 HN (t) − HN (m). (5) The sum in the right side of the last equation is a polynomial in m of degree d + 1 by the well-known Faulhaber's formula. It is now clear that G K -dim(N ⊗K R) = d + 1. Remark 3.2. The foregoing lemma may easily be adapted to the case when the base ring is a polynomial ring k[x1, x2, · · · , xn]. In this case Rm = ⊕m i=0Kixi−m and the sum corresponding to equation (5) will be dimk Nm = m X t=0 HN (t) which gives G K -dim(N ⊗K R) = G K -dim(N ) + 1 by Faulhaber's theorem. Lemma 3.3. Suppose that M is a cyclic Rn-module with a cyclic and critical Kn-module N such that M = N Rn. Let t be maximal with respect to the property ]. If N Rn 6∼= N ⊗Kn Rn then M that N embeds a copy of K ′ := k[x±1 1 , · · · , x±1 cannot embed a free K ′-module of infinite rank. t 7 Proof. Set K := Kn. We try to mimic the proof of [1][Lemma 2.3] which is for the case of a crossed product of a free ablelian group over a division ring. ∞ Since the sum X i=0 K-module N xi is not direct, we can find a t such that for r > t the r X i=0 N xi/ r−1 X i=0 N xi is isomorphic to a proper quotient of N via the obvious map n 7→ nxr. The first part of the hypothesis on N means that G K -dim(M ) = t. This follows from Proposition 2.4. But N is critical and so each of these quotients is K ′-torsion. Indeed if this were not true then such a quotient would embed a copy of K ′ and so by (2.1) it would have GK-dimension at least t since G K -dim(K ′) = t. But this contradicts the definition of a critical module. Also the sum t X njΛY i cannot embed a free Λ′-submodule of infinite rank by Lemma i=0 2.8. This proves our claim. References [1] C. J. B. Brookes and J. R. J. Groves, Modules over crossed products of division ring with an abelian group I, J. Algebra, 229, 25–54, 2000. [2] S. C. Coutinho, A primer of algebraic D-modules, London Mathematical Society Student Texts, No. 33, 1995. [3] G. R. Krause, T. H. Lenagan, Growth of Algebras and Gelfand-Kirillov Dimension, American Mathematical Soc., 2000. [4] K. R. Goodearl and R. B. Jr. Warfield, An introduction to noncommu- tative Noetherian rings, London Mathematical Society Student Texts, 16. Cambridge University Press, Cambridge, 1989. [5] A. Gupta, Modules over quantum Laurent polynomials, J. Aust. Math. Soc. 91 (2011), 323-341. [6] A. Gupta, GK dimensions of simple modules over K[X ±1, σ], Comm. Al- gebra 41 (2013), no. 7, 2593-2597. [7] J. C. McConnell, Representations of solvable Lie algebras. V. On the Gelfand-Kirillov dimension of simple modules. J. Algebra 76 (1982), no. 2, 489-493. [8] J. C. McConnell and J. C. Robson, Noncommutative Noetherian rings, Graduate Studies in Mathematics, 30, American Mathematical Society, Providence, RI, 2001. 8
1701.04192	1	1701	2017-01-16T07:00:06	Pivotal decomposition schemes inducing clones of operations	[ "math.RA", "math.LO" ]	We study pivotal decomposition schemes and investigate classes of pivotally decomposable operations. We provide sufficient conditions on pivotal operations that guarantee that the corresponding classes of pivotally decomposable operations are clones, and show that under certain assumptions these conditions are also necessary. In the latter case, the pivotal operation together with the constant operations generate the corresponding clone.	math.RA	math	PIVOTAL DECOMPOSITION SCHEMES INDUCING CLONES OF OPERATIONS MIGUEL COUCEIRO AND BRUNO TEHEUX Abstract. We study pivotal decomposition schemes and investigate classes of pivotally decomposable operations. We provide sufficient conditions on pivotal operations that guarantee that the corresponding classes of pivotally decomposable operations are clones, and show that under certain assumptions these conditions are also necessary. In the latter case, the pivotal operation together with the constant operations generate the corresponding clone. 1. Introduction and Motivation Several classes of operations have the remarkable feature that each member f : An → A is decomposable into simpler operations that are then combined by a single operation, in order to retrieve the values of the original operation f . A noteworthy example is the class of Boolean functions f : {0, 1}n → {0, 1} that can be decomposed into expressions of the form f (x) = xkf (x1 k) + (1 − xk)f (x0 (1) for x = (x1, . . . , xn) ∈ {0, 1}n and k ∈ [n], and where xc k denotes the n-tuple obtained from x by substituting its k-th component by c ∈ {0, 1}. Such decom- position scheme is referred to as Shannon decomposition (or Shannon expansion) [18], or pivotal decomposition [1]. Boolean functions are similarly decomposable into expressions in the language of Boolean lattices k), (2) where xk = 1 − xk. f (x) = (xk ∧ f (x1 k)) ∨ (xk ∧ f (x0 k)) More recent examples include the class of polynomial operations over a distribu- tive lattice (essentially, combinations of variables and constants using the lattice operations ∧ and ∨) that were shown in [13] to be decomposable into expressions of the form (3) k), xk, f (x1 where med is the ternary lattice polynomial given by f (x) = med(f (x0 k)), med(x1, x2, x3) = (x1 ∧ x2) ∨ (x1 ∧ x3) ∨ (x2 ∧ x3) = (x1 ∨ x2) ∧ (x1 ∨ x3) ∧ (x2 ∨ x3). The latter decomposition scheme is referred to as median decomposition in [5] and [13]. We refer the reader to [15, 19, 4] for applications of the median decomposition formula to obtain median representations of Boolean functions. Note that decomposition schemes (1), (2) and (3) share the same general form, namely, Indeed, f (x) = Π(xk, f (x1 k), f (x0 k)). • in (1) we have Π(x, y, z) = xy + (1 − x)z, Date: Wednesday 14th July, 2021,07:19. 1 2 MIGUEL COUCEIRO AND BRUNO TEHEUX • in (2) we have Π(x, y, z) = (x ∧ y) ∨ (x ∧ z), and • in (3) we have Π(x, y, z) = med(x, y, z). These facts were observed in [14] where such pivotal decomposition schemes were investigated. These preliminary efforts were then further pursued under the obser- vation that certain classes of pivotal operations fulfill certain closure requirements, notably, closure under functional composition. This led to the study [6] of those classes of pivotally decomposable operations that constitute clones. In particular, we presented conditions on pivotal operations to ensure that the corresponding classes of pivotally decomposable operations constitute clones. However, several questions were stated without being answered. In this paper we settle many of these questions and provide new insights in this line of research. The paper is organised as follows. In Section 2 we recall basic notions and terminology that will be used throughout the paper (Subsection 2.1). We also in- troduce the concepts of pivotal operation and that of pivotally decomposable class (Subsection 2.2) and discuss normal form representations that arise from such piv- otal decompositions (Subsection 2.3). Moreover, we investigate certain symmetry properties that are common to pivotal operations (Subsection 2.4). In Section 3 we consider the problem of describing classes of pivotally decomposable operations that are clones. A general solution to this problem still eludes us, but we provide several sufficient conditions on pivotal operations that ensure the latter (Subsection 3.1). In fact, we show that under certain assumptions many of these conditions are also necessary (Subsection 3.2). The question of determining sets of generators for clones of pivotally decomposable operations is also addressed and partially an- swered. Taking this framework further into the realm of clone theory, many natural questions emerge. For instance, we construct an example of a pivotal operation Π for which the class of Π-decomposable operations is a clone that does not contain Π (Subsection 3.3); such an example is shown not to exist in the case of Boolean functions. Further questions that remain open are then discussed in Section 4. 2. Basic notions and notation In this section we recall basic terminology used throughout the paper. In partic- ular, we introduce the concepts of pivotal operation and of pivotally decomposable class, and we observe that, under certain conditions, pivotal decompositions lead to normal form representations that use a unique non trivial connective, namely, the pivotal operation. In the last subsection we investigate symmetric properties of pivotal operations and present some characterizations. 2.1. Preliminaries: Clones of operations. For any positive integer n, we denote by [n] the set {1, . . . , n}. For a nonempty set A, a function f : An → A is called an n-ary operation on A. We denote by O(n) A the set of n-ary operations on A and by OA = Sn≥1 O(n) A , S ⊆ [n] and a ∈ An we define the S-section f a S of f as the S-ary operation on A defined by f a S(x) = f (ax S is the n-tuple whose i-th coordinate is xi, if i ∈ S, and ai, otherwise. For k ∈ [n], we say that the k-th argument of f ∈ O(n) A is essential if there is a tuple b ∈ An such that f b k is non-constant. Otherwise, we way that it is inessential. A the set of operations on A. For any f ∈ O(n) S), where ax A clone on A is a set C ⊆ OA of operations on A that (1) contains all projections on A, i.e., operations pn i : An → A given by pn i (x1, . . . , xn) = xi, for i ∈ [n], and PIVOTAL DECOMPOSITION SCHEMES INDUCING CLONES OF OPERATIONS 3 (2) is closed under taking functional compositions, i.e., if f ∈ C ∩ O(n) A , then their composition f (g1, . . . , gn) ∈ O(m) g1, . . . , gn ∈ C ∩ O(m) defined by A and A that is f (g1, . . . , gn)(x) = f (g1(x), . . . , gn(x)) (x ∈ Am) also belongs to C. In the case when A is finite, the set of all clones on A forms an algebraic lattice, where the lattice operations are the following: meet is the intersection, join is the smallest clone that contains the union. The greatest element is the clone OA of all operations on A; the least element is the clone JA of all projections on A. For sets A of cardinality at least 3, this lattice is uncountable, and its structure remains a topic of current research; see, e.g., [8, 10]. In the case when A = 2, the lattice of clones on A is countably infinite, and it was completely described by E. Post [17]. In particular, it follows that each Boolean clone can be generated by a finite set of Boolean functions. For instance, • the clone O{0,1} of all Boolean functions can be generated by {¬, ∧} or, equivalently, by {0, ¬, med}; • the clone M of all monotone Boolean functions, i.e., verifying x ≤ y =⇒ f (x) ≤ f (y), can be generated by {0, 1, ∧, ∨} or, equivalently, by {0, 1, med}; • the clone SM of all self-dual monotone Boolean functions, i.e., monotone operations verifying f (¬x) = ¬f (x), is generated by {med}. For further background see, e.g., [8, 10]. 2.2. Pivotal operations and pivotally decomposable classes. In what fol- lows, A denotes an arbitrary fixed nonempty set, and 0 and 1 are two fixed elements of A. In the setting of operations, the notion of pivotal operation Π and that of Π-decomposable operation can be defined as follows. Definition 2.1 (Definition 2.1 in [14]). A pivotal operation on A is a ternary operation Π on A that satisfies the equation (4) If Π is a pivotal operation, then f ∈ O(n) Π(x, y, y) = y. A is Π-decomposable if (5) f (x) = Π(xi, f (x1 i ), f (x0 i )), x ∈ An, i ∈ [n]. Also, we denote by ΛΠ the class of Π-decomposable operations on A. Note that condition (4) ensures that Π-decomposability of an operation does not depend on its inessential arguments. Indeed, if the ith argument of f is inessential, i ) = f (x0 then f (x) = f (x1 i ) for every x ∈ An. It follows from (4) that f (x) = Π(xi, f (x1 i ), f (x0 i )) for any x ∈ An. In particular, we can state the following result. Lemma 2.2. If Π is a pivotal operation, then every constant operation on A is Π-decomposable. 2.3. Normal form representations induced by pivotal decompositions. Note that if an operation f is Π-decomposable, then we arrive at a representation of f by an expression built from the pivotal operation Π and applied to variables and constants, by iterating its Π-decomposition expression (5). This fact motivates the following notion of Π-normal form. Definition 2.3. Let Π ∈ O(3) inductively on k ≥ 0 by the following rules. A . We define the classes of k-ary Π-normal forms N k Π (1) N 0 Π = O(0) A . 4 MIGUEL COUCEIRO AND BRUNO TEHEUX (2) For any k ≥ 0, the class N k+1 Π is defined by We denote by NΠ the class Sk≥0 N k N k+1 Π = {Π(xk+1, g, g′) g, g′ ∈ N k Π}. Π of the Π-normal forms. Observe that N k Π ⊆ O(k) get the following result. A for every k ≥ 0. By repeated applications of (5), we Proposition 2.4. If Π is a pivotal operation, then ΛΠ ⊆ NΠ. 2.4. Symmetric pivotal operations. As we will see later in the paper, a pivotal operation Π is not necessarily Π-decomposable. However, when it is, then it verifies certain symmetry properties. Consider the following equations: (6) (7) (8) Π(x, y, z) = Π(z, x, y), Π(x, y, z) = Π(z, y, x), Π(x, 1, 0) = x, Clearly, Π is symmetric if and only if it satisfies (6) and (7). The following result states that if Π ∈ ΛΠ and satisfies (6) and (8), then it is symmetric. Proposition 2.5. If Π is a Π-decomposable pivotal operation that satisfies (8) and (6), then it satisfies (7). In particular, Π is a symmetric operation. Proof. We obtain successively (9) (10) (11) (12) (13) Π(x, y, z) = Π(x, Π(1, y, z), Π(0, y, z)) = Π(x, Π(y, Π(1, 1, z), Π(1, 0, z)), Π(z, Π(0, y, 1), Π(0, y, 0))) = Π(x, Π(y, 1, z), Π(z, y, 0)) = Π(x, Π(z, y, 1), Π(z, y, 0)) = Π(z, y, x). where (9), (10) and (13) were obtained by Π-decomposability of Π, and where (11) and (12) were obtained by (4), (8) and (7). (cid:3) Under the assumption of Π-decomposability of Π and (8), symmetry of a pivotal operation Π can be characterized in the following way. Theorem 2.6. Le Π be a Π-decomposable pivotal operation that satisfies (8) . The following conditions are equivalent. (i) Π is symmetric. (ii) Π satisfies the equations Π(0, 1, 0) = Π(0, 0, 1) and Π(1, 1, 0) = Π(1, 0, 1). Proof. It is clear that (i) implies (ii). Let us prove that (ii) implies (i). It suffices to prove that Π satisfies (14) (15) Π(x, y, z) = Π(x, z, y), Π(x, y, z) = Π(y, x, z). First, note that for every x ∈ A we obtain successively Π(x, 1, 0) = Π(x, Π(1, 1, 0), Π(0, 1, 0)) = Π(x, Π(1, 0, 1), Π(0, 0, 1)) (16) = Π(x, 0, 1), where the first identity is obtained by (7), the second by contition (ii) and the last one by Π-decomposability of Π. Then, for every y ∈ A we have (17) (18) Π(x, y, 0) = Π(y, Π(x, 1, 0), Π(x, 0, 0)) = Π(y, Π(x, 0, 1), Π(x, 0, 0)) = Π(x, 0, y), PIVOTAL DECOMPOSITION SCHEMES INDUCING CLONES OF OPERATIONS 5 where the first and last identities are obtained by decomposability of Π, and the second by (16). Using a similar argument, we obtain (19) Π(x, y, 1) = Π(x, 1, y). Finally, we obtain for any z ∈ A (20) Π(x, y, z) = Π(z, Π(x, y, 1), Π(x, y, 0)) = Π(z, Π(x, 1, y), Π(x, 0, y)) = Π(x, z, y), where the first and last identities are obtained by decomposability (5) of Π, and the second by (18) and (19). This proves that (14) holds. Now, using (8) and (4) in the first identity in (17) we obtain that (21) Π(x, y, 0) = Π(y, x, 0). Similarly, we have Π(x, y, 1) = Π(y, Π(x, 1, 1), Π(x, 0, 1)) = Π(y, 1, x), (22) = Π(y, x, 1), where the first identity is obtained by decomposability of Π, the second by (4), (16) and (8), and the last one by (19). Using (21) and (22) in the first identity of (20), we obtain (15) by Π-decomposability of Π. (cid:3) 3. Clones of pivotally decomposable operations In Subsection 3.1, we provide sufficient conditions on a pivotal operation Π for ΛΠ to be a clone, and in Subsection 3.2, we prove that these conditions are also necessary under the assumption that Π belongs to ΛΠ, and satisfies two additional equations (30) and (31) that involves only Π, and the elements 0 and 1. Certain natural questions are also discussed and answered negatively by counter-examples that are constructed in Subsection 3.3. 3.1. Sufficient conditions for ΛΠ to be a clone. Let us consider the following equations: (23) Π(Π(x, y, z), t, u) = Π(x, Π(y, t, u), Π(z, t, u)). The relevance of property (23) is made apparent by the following lemma. Lemma 3.1. Let Π be a pivotal operation that satisfies (23). If f : An → A and g1, . . . , gn : Am → A are Π-decomposable, then so is f (g1, . . . , gn). Proof. For every i ∈ [n] let g′ where xi = (x(i−1)m+1, . . . , xim). We prove that f (g′ For x ∈ Anm, set i : Anm → A be the operation defined by g′ 1, . . . , g′ i(x) = gi(xi) n) is Π-decomposable. 2(x), . . . , g′ n(x)), c0 x = f (0, g′ x = g′ a0 1(0, x2, . . . , xnm), 2(x), . . . , g′ n(x)), c1 x = f (1, g′ x = g′ a1 1(1, x2, . . . , xnm). We obtain by Π-decomposability of f that f (g′ 1(x), . . . , g′ n(x)) = Π(g′ 1(x), c1 x, c0 x). By iterating the pivotal decomposition expression (to each argument), we get the following equalities Π(g′ 1(x), c1 x, c0 x) = Π(Π(x1, a1 = Π(x1, Π(a1 = Π(x1, f (g′ x), c1 x, a0 x, c0 x, c1 1, . . . , g′ x, c0 x), x), Π(a0 n)(x1 x, c1 1), f (g′ x, c0 x)), 1, . . . , g′ n)(x0 1)), 6 MIGUEL COUCEIRO AND BRUNO TEHEUX where the first equality is obtained by Π-decomposability of g′ 1, the second one by equation (23) and the last one by Π-decomposability of f . Thus, we have proved that condition (5) holds for f (g′ n) and i = 1. We can proceed in a similar way to obtain f (g′ n(x)) = Π(xℓ, f (g′ 1(x), . . . , g′ 1, . . . , g′ 1, . . . , g′ n)(x1 ℓ ), f (g′ 1, . . . , g′ (24) n)(x0 ℓ )), for every ℓ ∈ [nm]. The decomposability of Π(g1, . . . , gn) follows from (24) by identifying all arguments in {xi, xm+i, . . . , x(n−1)m+i} for every i ∈ [m]. (cid:3) Similarly, if the pivotal operation satisfies equation (8), then ΛΠ must contain all projections. Lemma 3.2. Let Π be a pivotal operation. The following conditions are equivalent. (i) Π satisfies equation (8). (ii) ΛΠ contains all projections on A. (iii) ΛΠ contains the unary projection p1 1. Proof. (i) =⇒ (ii): Let n ≥ 1 and k ∈ [n]. For every i ∈ [n] such that i 6= k and for every x ∈ An, Π(xi, pn k (x1 i ), pn k (x0 i )) = Π(xi, xk, xk) = xk where the last equality is obtained by (4). If i = k, then Π(xi, pn k (x1 i ), pn k (x0 i )) = Π(xk, 1, 0) = xk where the last equality is obtained by (8). We conclude that pn k ∈ ΛΠ. (ii) =⇒ (iii): Trivial. (iii) =⇒ (i): If ΛΠ contains the unary projection p1 1, then for every x ∈ A we have x = p1 1(x) = Π(x, 1, 0). Thus Π satisfies equation (8), and the proof of the lemma is now complete. (cid:3) By combining Lemmas 2.2, 3.1, and 3.2, we obtain the following result. Proposition 3.3. Suppose that Π is a pivotal operation that satisfies equation (23). Then ΛΠ is a clone if and only if Π satisfies equation (8). In the latter case, ΛΠ is a clone that contains all constant operations. We illustrate the previous results by analyzing the particular case of Boolean functions. Example 3.4. Let Π be a Boolean pivotal operation such that ΛΠ is a clone. According to Proposition 3.2, the operation Π satisfies equation (8). Hence, the unary sections Π(x, 0, 0) and Π(x, 1, 1) are determined by (4) while the value of the section Π(x, 1, 0) is determined by (8): (25) Π(x, 0, 0) = 0, Π(x, 1, 1) = 1, Π(x, 1, 0) = x. Moreover, it is not difficult to check that the four possibilities for the unary section Π(x, 0, 1), namely, Π0(x, 0, 1) = x, Π2(x, 0, 1) = 0, Π1(x, 0, 1) = x, Π3(x, 0, 1) = 1, give rise to operations Π0, . . . , Π3 that satisfy equation (23). Simple computations then show that we must have Π0(x, y, z) = (x ∧ y) ∨ (x ∧ z) ∨ (y ∧ z), Π2(x, y, z) = y ∧ (x ∨ z), Π1(x, y, z) = (x ∧ y) ∨ (x ∧ z), Π3(x, y, z) = z ∨ (x ∧ y). Hence, the clones ΛΠ0 , . . . , ΛΠ3 are as follows: PIVOTAL DECOMPOSITION SCHEMES INDUCING CLONES OF OPERATIONS 7 (a) ΛΠ0 is the clone M of all monotone Boolean functions, since Π0 = med; (b) ΛΠ1 is the clone O{0,1} of all Boolean functions, since Π1 is the pivotal operation used in Shannon decomposition; (c) ΛΠ2 is the clone M of all monotone Boolean functions, since Π2(x, y, 0) = y ∧ x and Π2(x, 1, z) = x ∨ z, and every composition of Π2 with projections or constants is monotone; (d) ΛΠ3 is the clone M of all monotone Boolean functions (by a similar argument to that used for ΛΠ2). The situation can be summarized by the following result. Proposition 3.5. If C is Boolean clone, then there is a Boolean pivotal operation Π such that C = ΛΠ if and only if C is the clone of all monotone Boolean functions or the clone of all Boolean functions. 3.2. The case of a pivotally decomposable Π. In the section, we derive results about clones of Π-decomposable operations under the additional assumption that the operation Π itself is Π-decomposable, i.e., that Π ∈ ΛΠ. The next result states that under this assumption, the pivotal operation together with constant maps suffice to construct expressions representing each member of ΛΠ. Proposition 3.6. Let Π be a pivotal operation such that ΛΠ is a clone that contains Π. Then ΛΠ is the clone generated by Π and the constant maps. In particular, ΛΠ = NΠ. Proof. Let C be the clone generated by Π and the constant operations. We have to prove that ΛΠ = C. The right to left inclusion is trivial since ΛΠ is a clone and contains each of the mentioned generators of C by assumption and Lemma 2.2. We derive the converse inclusion and the last part of the statement from the following sequence of inclusions, ΛΠ ⊆ NΠ ⊆ C ⊆ ΛΠ, where the first inclusion is obtained by Proposition 2.4, the second inclusion follows from the definitions of NΠ and C, and the third inclusion is a consequence of the first part of this proof. (cid:3) In the presence of a Π-decomposable operation Π, equation (8) has interesting consequences on the equational theory of the the algebra hA, Π, 0, 1i, where Π is a pivotal operation. Lemma 3.7. If Π is a pivotal operation on A that satisfies (8), then it satisfies the following equations: (26) (27) (28) (29) Π(0, 1, z) = z, Π(1, 1, z) = 1, Π(0, y, 0) = 0, Π(1, y, 0) = y. Proof. The proof follows from straightforward applications of equations (5) and (8). For instance, we obtain successively Π(0, 1, z) = Π(z, Π(0, 1, 1), Π(0, 1, 0)) = Π(z, 1, 0) = z, where the first equality is obtained by (5) and the two last ones by (8). (cid:3) According to Proposition 3.3, if Π is a pivotal operation that satisfies equations (8) and (23), then ΛΠ is a clone. In the next theorem, we prove that the converse 8 MIGUEL COUCEIRO AND BRUNO TEHEUX statement also holds, under the assumption that Π ∈ ΛΠ and that (30) (31) Π(Π(1, 0, 1), 0, 1) = Π(1, Π(0, 0, 1), Π(1, 0, 1)), Π(Π(0, 0, 1), 0, 1) = Π(0, Π(0, 0, 1), Π(1, 0, 1)). Theorem 3.8. Let Π be a Π-decomposable pivotal operation that satisfies (30) and (31). The following conditions are equivalent: (i) ΛΠ is a clone, (ii) Π satisfies equations (8) and (23). In this case, ΛΠ is the clone generated by Π and the constant maps. Proof. Proposition 3.3 states that (ii) =⇒ (i). Conversely, assume that Π is a pivotal operation such that ΛΠ is a clone that contains Π. By Lemma 3.2, it follows that Π satisfies equation (8). We prove that (23) also holds. In what follows, we use without further warning the fact that ΛΠ is a clone that contains Π and every constant operation to apply (5) to operations which are compositions of Π and constant ones. Hence, if L(x, y, z) and R(x, y, z) denote the operations given by the left-hand side and right-hand side of equation (23), respectively, we have L(x, y, z) = Π(cid:0)x, Π(cid:0)Π(1, y, z), t, u(cid:1), Π(cid:0)Π(0, y, z), t, u(cid:1)(cid:1), R(x, y, z) = Π(cid:0)x, Π(cid:0)1, Π(y, t, u), Π(z, t, u)(cid:1), Π(cid:0)0, Π(y, t, u), Π(z, t, u)(cid:1)(cid:1). To prove that L(x, y, z) = R(x, y, z) it suffices to prove that the two following equations hold: (32) (33) Π(cid:0)Π(1, y, z), t, u(cid:1) = Π(cid:0)1, Π(y, t, u), Π(z, t, u)(cid:1), Π(cid:0)Π(0, y, z), t, u(cid:1) = Π(cid:0)0, Π(y, t, u), Π(z, t, u)(cid:1). We prove that (32) holds (with the help of (30)). Equation (33) can be obtained in a similar way (with the help of (31)). By decomposing with respect to y, we obtain that the right-hand side of (32) is equal to Π(cid:0)y, Π(cid:0)Π(1, 1, z), t, u(cid:1), Π(cid:0)Π(1, 0, z), t, u(cid:1)(cid:1), while the left-hand side of (32) is equal to Π(cid:0)y, Π(cid:0)1, Π(1, t, u), Π(z, t, u)(cid:1), Π(cid:0)1, Π(0, t, u), Π(z, t, u)(cid:1)(cid:1). Hence, to prove that equation (32) holds, we first observe that Π(cid:0)Π(1, 1, z), t, u(cid:1) = Π(1, t, u) by (27), = Π(cid:0)z, Π(1, t, u), Π(1, t, u)(cid:1) = Π(cid:0)z, Π(cid:0)1, Π(1, t, u), Π(1, t, u)(cid:1), Π(cid:0)1, Π(0, t, u), Π(1, t, u)(cid:1)(cid:1), by (5), (4), = Π(cid:0)1, Π(1, t, u), Π(z, t, u)(cid:1) by (5). by (4), It remains to prove that (34) Π(cid:0)Π(1, 0, z), t, u(cid:1) = Π(cid:0)1, Π(0, t, u), Π(z, t, u)(cid:1). By decomposing with regard to z we obtain Π(cid:0)Π(1, 0, z), t, u(cid:1) = Π(cid:0)z, Π(cid:0)Π(1, 0, 1), t, u(cid:1), Π(0, t, u)(cid:1), Π(cid:0)1, Π(0, t, u), Π(z, t, u)(cid:1) = Π(cid:0)z, Π(cid:0)1, Π(0, t, u), Π(1, t, u)(cid:1), Π(0, t, u)(cid:1), and it suffices to prove that (35) Π(cid:0)Π(1, 0, 1), t, u(cid:1) = Π(cid:0)1, Π(0, t, u), Π(1, t, u)(cid:1). PIVOTAL DECOMPOSITION SCHEMES INDUCING CLONES OF OPERATIONS 9 Observe that by decomposing with respect to u, Π(cid:0)Π(1, 0, 1), t, u(cid:1) = Π(cid:0)u, Π(cid:0)Π(1, 0, 1), t, 1(cid:1), Π(cid:0)Π(1, 0, 1), t, 0(cid:1)(cid:1), Π(cid:0)1, Π(0, t, u), Π(1, t, u)(cid:1) = Π(cid:0)u, Π(cid:0)1, Π(0, t, 1), Π(1, t, 1)(cid:1), Π(cid:0)1, 0, Π(1, t, 0)(cid:1)(cid:1), where we have applied (28) to obtain the second identity. Hence, to prove (35) it suffices to prove that (36) (37) Π(cid:0)Π(1, 0, 1), t, 1(cid:1) = Π(cid:0)1, Π(0, t, 1), Π(1, t, 1)(cid:1), Π(cid:0)Π(1, 0, 1), t, 0(cid:1) = Π(cid:0)1, 0, Π(1, t, 0)(cid:1). By (5) we obtain Π(cid:0)Π(1, 0, 1), t, 0(cid:1) = Π(cid:0)t, Π(1, 0, 1), 0(cid:1) = Π(cid:0)1, 0, Π(1, t, 0)(cid:1), which proves (37). Next, we observe that Π(cid:0)Π(1, 0, 1), t, 1(cid:1) = Π(cid:0)t, 1, Π(cid:0)Π(1, 0, 1), 0, 1(cid:1)(cid:1) Π(cid:0)1, Π(0, t, 1), Π(1, t, 1)(cid:1) = Π(cid:0)t, 1, Π(cid:0)1, Π(0, 0, 1), Π(1, 0, 1)(cid:1)(cid:1). We conclude that (36) is satisfied by applying (31), which holds by assumption. (cid:3) Since (30) and (31) are instances of (23), Theorem 3.8 can be restated as follows. Corollary 3.9. Let Π be a Π-decomposable pivotal operation. The following con- ditions are equivalent: (i) ΛΠ is a clone and Π satisfies (30) and (31), (ii) Π satisfies (8) and (23). By noting that equations (30) and (31) are satisfied by a symmetric pivotal operation that satisfies (8), we obtain the following corollary. Corollary 3.10. Let Π be a symmetric Π-decomposable pivotal operation that sat- isfies (8). The following conditions are equivalent: (i) ΛΠ is a clone, (ii) Π satisfies (23). 3.3. Further issues and counter-examples. In view of Theorem 3.8 and Corol- lary 3.9, a natural question arises: can the Π-decomposability of Π be deduced from equations (8) and (23)? Example 3.4 shows that the answer is positive if Π is a Boolean pivotal operation. Now, we prove that it is not true in general. We set A∆ = {(x, y) ∈ A2 x 6= y & (x, y) 6= (1, 0)}. Proposition 3.11. Let Π be a pivotal operation that satisfies (8). If there exits a function f : A∆ → A such that Π(x, y, z) = f (y, z) for every (y, z) ∈ A∆, then Π also satisfies (23). Proof. Note first that Π is well defined by the conditions in the statement. More- over, equations (4) and (8) ensure that (23) is satisfied when t = u or (t, u) = (1, 0), respectively. Now, if (t, u) ∈ A∆, then Π(Π(x, y, z), t, u) = f (t, u) = Π(x, f (t, u), f (t, u)) = Π(x, Π(y, t, u), Π(z, t, u)). This shows that (23) does indeed hold for such a Π. (cid:3) Example 3.12. Assume that A has at least three elements 0, 1, 2. Let f : A∆ → A be any mapping that satisfies f (2, 0) = f (2, 1) 6= f (2, 2). Then the pivotal operation Π defined as in Proposition 3.11 is not Π-decomposable since Π(1, 2, 2) = f (2, 2) while Π(2, Π(1, 2, 1), Π(1, 2, 0)) = f (2, 0). 10 MIGUEL COUCEIRO AND BRUNO TEHEUX Theorem 3.8 gives a characterization of pivotal operations Π such that ΛΠ is a clone, under the assumption that Π ∈ ΛΠ. We now give an example of a pivotal operation Π such that ΛΠ is a clone that does not contain Π. Note however that Example 3.4 shows that such a Π does not exist in the case of Boolean functions. Example 3.13. Let A = {0, a, 1} and N : A → A be the map defined by N (0) = 1, N (a) = a, and N (1) = 0. Define Π : A3 → A as the map that satisfies (8), (4) and (38) (39) (40) (41) (42) Π(x, 0, 1) = N (x) Π(x, 1, a) = 1 Π(x, 0, a) = 1 Π(x, a, 1) = 0 Π(x, a, 0) = 0. First, observe that Π 6∈ ΛΠ. Indeed, for any x ∈ A we have on the one hand Π(x, a, a) = a while Π(a, Π(a, Π(x, 1, a), Π(x, 0, a)) = Π(a, 1, 1) = 1. According to Proposition 3.3, it suffices to prove that Π satisfies equation (23). If t = u or (t, u) ∈ {(a, 0), (a, 1), (1, a), (0, a)} then Π(x, t, u) is constant and (23) holds trivially. If (t, u) = (1, 0) then Π(x, t, u) is the first projection and (23) holds as well. It remains to consider the case (t, u) = (0, 1). We have to prove (43) Π(x, N (y), N (z)) = N (Π(x, y, z)). If y = z then or (y, z) ∈ {(a, 0), (a, 1), (1, a), (0, a)} then(43) clearly holds by (39) - (42). If (y, z) ∈ {(0, 1), (1, 0)} then (43) holds by (38). 4. Conclusions and Further Research In this paper, we studied pivotal decompositions of operations from a clone theory perspective, and presented a characterization of classes of Π-decomposable operations that are clones in the case when the pivotal operation Π is itself Π- decomposable and satisfies (30) and (31). However in Example 3.13 we showed that there exists a clone of Π-decomposable operations that does not contain Π, i.e., Π is not Π-decomposable. This leaves open a complete description of classes of pivotally decomposable operations that are clones. Moreover, once such a description is obtained, a structural analysis of the set of all pivotally decomposable clones is to be expected. Another topic that will deserve our attention is motivated by Theorem 2.6 that states that if a pivotal operation Π is a Π-decomposable and satisfies Π(x, 1, 0) = x, Π(0, 0, 1) = 0, and Π(1, 0, 1) = 1, then Π is symmetric and hence is a majority operation. Furthermore, if Π satisfies (23), then Π is a median operation (see [2] and the bibliography therein). These observations establish noteworthy connections between pivotally decomposable classes and median algebras, and should deserve a deeper study in future research. As a third line of research that emerges from this paper deals with normal form representations of operations arising from pivotal decomposition schemes. Propo- sition 2.4 provides normal form representations for the elements of a pivotally de- composable class. Here, determining canonical expressions for these representations based on the pivotal operation, as well as studying the complexity of such repre- sentations (e.g., with respect to classical normal form representations) constitute an interesting topic of research which is under current investigation. We envision a similar study to that of [3] where, in particular, it was shown that normal form representations of Boolean functions that use the ternary median as the only logical connective, produce asymptotically shorter representations than the classical DNF, CNF and polynomial representations. PIVOTAL DECOMPOSITION SCHEMES INDUCING CLONES OF OPERATIONS 11 Acknowledgment This work was supported by the internal research project F1R-MTH-PUL-15MRO3 of the University of Luxembourg. References [1] R.E. Barlow, F. Proschan. Importance of system components and fault tree events, Stochastic Process. Appl. 3 (1975) 153 -- 172. [2] H. J. Bandelt and J. Hedl´ıkov´a. Median algebras. Discrete mathematics, 45:1 -- 30, 1983. [3] M. Couceiro, S. Foldes, E. Lehtonen. Composition of Post classes and normal forms of Boolean functions, Discrete Math. 306 (2006) 3223 -- 3243. [4] M. Couceiro, E. Lehtonen, J.-L. Marichal, T. Waldhauser. An algorithm for producing median normal form representations for Boolean functions, in the proceedings of the Reed -- Muller Workshop 2011, 49 -- 54, 2011. [5] M. Couceiro, and J.-L. Marichal. Polynomial functions over bounded distributive lattices, J. Multiple-Valued Logic Soft Comput. 18 (2012) 247 -- 256. [6] M. Couceiro, B. Teheux. Clones of pivotally decomposable functions. 45th IEEE International Symposium on Multiple-Valued Logic (ISMVL 2015), IEEE Computer Society, 195 -- 198. [7] M. Couceiro, M. Pouzet. On a quasi-order on Boolean functions, Theoret. Comput. Sci. 396 (2008) 71 -- 87. [8] K. Denecke, S. L. Wismath. Universal Algebra and Applications in Theoretical Computer Science, Chapman & Hall/CRC, Boca Raton, 2002. [9] L. Hellerstein. On generalized constraints and certificates, Discrete Math. 226 (2001) 211 -- 232. [10] D. Lau. Function Algebras on Finite Sets, Springer-Verlag, Berlin, Heidelberg, 2006. [11] E. Lehtonen. Descending chains and antichains of the unary, linear, and monotone subfunction relations, Order 23 (2006) 129 -- 142. [12] E. Lehtonen, ´A. Szendrei. Equivalence of operations with respect to discriminator clones, Discrete Math. 309 (2009) 673 -- 685. [13] J.-L. Marichal. Weighted lattice polynomials, Discrete Math. 309 (2009) 814 -- 820. [14] J.-L. Marichal, B. Teheux. Pivotal decompositions of functions, Discrete Appl. Math., 174:102 -- 112, 2014. [15] F. Miyata. Realization of arbitrary logical functions using majority elements, IEEE Trans. on Electronic Computers EC-12 (1963) 183 -- 191. [16] N. Pippenger. Galois theory for minors of finite functions, Discrete Math. 254 (2002) 405 -- 419. [17] E. L. Post. The Two-Valued Iterative Systems of Mathematical Logic, Annals of Mathematical Studies, vol. 5, Princeton University Press, Princeton, 1941. [18] C.E. Shannon. A symbolic analysis of relay and switching circuits, Trans. Am. Inst. Electr. Eng. 57 (1938) 713 -- 723. [19] Y. Tohma. Decompositions of logical functions using majority decision elements, IEEE Trans. on Electronic Computers EC-13 (1964) 698 -- 705. LORIA, (CNRS - Inria Nancy Grand Est - Universit´e de Lorraine), BP239 - 54506 Vandoeuvre les Nancy, France E-mail address: miguel.couceiro@{loria,inria}.fr Mathematics Research Unit, FSTC, University of Luxembourg, 6, rue Coudenhove- Kalergi, L-1359 Luxembourg, Luxembourg E-mail address: [email protected]
0812.0609	3	0812	2011-12-23T05:24:55	Degenerate Sklyanin algebras and Generalized Twisted Homogeneous Coordinate rings	[ "math.RA", "math.AG" ]	In this work, we introduce the point parameter ring B, a generalized twisted homogeneous coordinate ring associated to a degenerate version of the three-dimensional Sklyanin algebra. The surprising geometry of these algebras yields an analogue to a result of Artin-Tate-van den Bergh, namely that B is generated in degree one and thus is a factor of the corresponding degenerate Sklyanin algebra.	math.RA	math	DEGENERATE SKLYANIN ALGEBRAS AND GENERALIZED TWISTED HOMOGENEOUS COORDINATE RINGS CHELSEA WALTON Department of Mathematics University of Michigan Ann Arbor, MI 48109. E-mail address: [email protected] Abstract. In this work, we introduce the point parameter ring B, a generalized twisted homogeneous coordinate ring associated to a degen- erate version of the three-dimensional Sklyanin algebra. The surprising geometry of these algebras yields an analogue to a result of Artin-Tate- van den Bergh, namely that B is generated in degree one and thus is a factor of the corresponding degenerate Sklyanin algebra. 1. Introduction Let k be an algebraically closed field of characteristic 0. We say a k- R0 = k. algebra R is connected graded (cg) when R = Li∈N Ri is N-graded with A vital development in the field of Noncommutative Projective Algebraic Geometry is the investigation of connected graded noncommutative rings with use of geometric data. In particular, a method was introduced by Artin-Tate-van den Bergh in [3] to construct corresponding well-behaved graded rings, namely twisted homogeneous coordinate rings (tcr) [2, 12, 18]. However, there exist noncommutative rings that do not have sufficient geometry to undergo this process [12]. The purpose of this paper is to explore a recipe suggested in [3] for building a generalized analogue of a tcr for any connected graded ring. As a result, we provide a geometric approach to examine all degenerations of the Sklyanin algebras studied in [3]. We begin with a few historical remarks. In the mid-1980s, Artin and Schelter [1] began the task of classifying noncommutative analogues of the polynomial ring in three variables, yet the rings of interest were not well Date: January 17, 2009. 2000 Mathematics Subject Classification. 14A22, 16S37, 16S38, 16W50. Key words and phrases. noncommutative algebraic geometry, degenerate sklyanin al- gebra, point module, twisted homogeneous coordinate ring. The author was partially supported by the NSF: grants DMS-0555750, 0502170. 1 2 CHELSEA WALTON understood. How close were these noncommutative rings to the commutative counterpart k[x, y, z]? Were they Noetherian? Domains? Global dimension 3? These questions were answered later in [3] and the toughest challenge was analyzing the following class of algebras. Definition 1.1. Let k{x, y, z} denote the free algebra on the noncommuting variables x, y, and z. The three-dimensional Sklyanin algebras are defined as S(a, b, c) = k{x, y, z} (1.1)   ayz + bzy + cx2, azx + bxz + cy2, axy + byx + cz2   for [a : b : c] ∈ P2 k \ D where D = {[0 : 0 : 1], [0 : 1 : 0], [1 : 0 : 0]} ∪ {[a : b : c] a3 = b3 = c3 = 1}. As algebraic techniques were exhausted, two seminal papers [3] and [4] arose introducing algebro-geometric methods to examine noncommutative analogues of the polynomial ring. In fact, a geometric framework was specifi- cally associated to the Sklyanin algebras S(a, b, c) via the following definition and result of [3]. Definition 1.2. A point module over a ring R is a cyclic graded left R- module M where dimk Mi = 1 for all i. Theorem 1.3. Point modules for S = S(a, b, c) with [a : b : c] /∈ D are parameterized by the points of a smooth cubic curve E = Ea,b,c : (a3 + b3 + c3)xyz − (abc)(x3 + y3 + z3) = 0 ⊂ P2. (1.2) The curve E is equipped with σ ∈ Aut(E) and the invertible sheaf i∗OP2(1) from which we form the corresponding twisted homogeneous coordinate ring B. There exists a regular normal element g ∈ S, homogeneous of degree 3, so that B ∼= S/gS as graded rings. The ring B is a Noetherian domain and thus so is S. Moreover for d ≥ 1, we get dimkBd = 3d. Hence S has the same Hilbert series as k[x, y, z], namely HS(t) = 1 (cid:3) (1−t)3 . In short, the tcr B associated to S(a, b, c) proved useful in determining the Sklyanin algebras' behavior. Due to the importance of the Sklyanin algebras, it is natural to understand their degenerations to the set D. Definition 1.4. The rings S(a, b, c) from (1.1) with [a : b : c] ∈ D are called the degenerate three-dimensional Sklyanin algebras. Such a ring is denoted by S(a, b, c) or Sdeg for short. DEGENERATE SKLYANIN ALGEBRAS 3 In section 2, we study the basic properties of degenerate Sklyanin algebras resulting in the following proposition. Proposition 1.5. The degenerate three-dimensional Sklyanin algebras have Hilbert series HSdeg (t) = 1+t 1−2t , they have infinite Gelfand Kirillov dimension, and are not left or right Noetherian, nor are they domains. Furthermore, the algebras Sdeg are Koszul and have infinite global dimension. The remaining two sections construct a generalized twisted homogeneous coordinate ring B = B(Sdeg) for the degenerate Sklyanin algebras. We are specifically interested in point modules over Sdeg (Definition 1.2). Unlike their nondegenerate counterparts, the point modules over Sdeg are not pa- rameterized by a projective scheme so care is required. Nevertheless, the degenerate Sklyanin algebras do have geometric data which is described by the following definition and theorem. Definition 1.6. A truncated point module of length d over a ring R is a cyclic graded left R-module M where dimkMi = 1 for 0 ≤ i ≤ d and dimkMi = 0 for i > d. The dth truncated point scheme Vd parameterizes isomorphism classes of length d truncated point modules. Theorem 1.7. For d ≥ 2, the truncated point schemes Vd ⊂ (P2)×d corre- sponding to Sdeg are isomorphic to a union of ( three copies of (P1)× d−1 2 and three copies of (P1)× d+1 2 , six copies of (P1)× d 2 , for d odd; for d even. The precise description of Vd as a subset of (P2)×d is provided in Proposition 3.13. Furthermore, this scheme is not a disjoint union and Remark 4.2 describes the singularity locus of Vd. In the language of [16], observe that the point scheme data of degenerate Sklyanin algebras does not stabilize to produce a projective scheme (of finite type) and as a consequence we cannot construct a tcr associated to Sdeg. Instead, we use the truncated point schemes Vd produced in Theorem 1.7 and a method from [3, page 19] to form the N-graded, associative ring B defined below. Definition 1.8. The point parameter ring B =Ld≥0 Bd is a ring associated to the sequence of subschemes Vd of (P2)×d (Definition 1.6). We have Bd = H 0(Vd,Ld) where Ld is the restriction of invertible sheaf pr∗ 1OP2(1) ⊗ . . . ⊗ pr∗ dOP2(1) ∼= O(P2)×d(1, . . . , 1) to Vd. The multiplication map Bi × Bj → Bi+j is defined by applying H 0 to the isomorphism pr1,...,i(Li) ⊗OVi+j pri+1,...,i+j(Lj) → Li+j. 4 CHELSEA WALTON Despite point parameter rings not being well understood in general, the final section of this paper verifies the following properties of B = B(Sdeg). Theorem 1.9. The point parameter ring B for a degenerate three-dimen- sional Sklyanin algebra Sdeg has Hilbert series HB(t) = (1+t2)(1+2t) (1−2t2)(1−t) and is generated in degree one. Hence we have a surjection of Sdeg onto B, which is akin to the result involving Sklyanin algebras and corresponding tcrs (Theorem 1.3). Corollary 1.10. The ring B = B(Sdeg) has exponential growth and there- fore infinite GK dimension. Moreover B is neither right Noetherian, Koszul, nor a domain. Furthermore B is a factor of the corresponding Sdeg by an ideal K where K has six generators of degree 4 (and possibly more of higher degree). Therefore the behavior of B(Sdeg) resembles that of Sdeg. It is natural to ask if other noncommutative algebras can be analyzed in a similar fashion, though we will not address this here. Acknowledgements. I sincerely thank my advisor Toby Stafford for in- troducing me to this field and for his encouraging advice on this project. I am also indebted to Karen Smith for supplying many insightful suggestions. I have benefited from conversations with Hester Graves, Brian Jurgelewicz, and Sue Sierra, and I thank them. 2. Structure of degenerate Sklyanin algebras In this section, we establish Proposition 1.5. We begin by considering the degenerate Sklyanin algebras S(a, b, c)deg with a3 = b3 = c3 = 1 (Definition 1.1) and the following definitions from [10]. Definition 2.1. Let α be an endomorphism of a ring R. An α-derivation on R is any additive map δ : R → R so that δ(rs) = α(r)δ(s) + δ(r)s for all r, s ∈ R. The set of α-derivations of R is denoted α-Der(R). We write S = R[z; α, δ] provided S is isomorphic to the polynomial ring R[z] as a left R-module but with multiplication given by zr = α(r)z + δ(r) for all r ∈ R. Such a ring S is called an Ore extension of R. By generalizing the work of [7] we see that most degenerate Sklyanin algebras are factors of Ore extensions of the free algebra on two variables. Proposition 2.2. In the case of a3 = b3 = c3 = 1, assume without loss of generality a = 1. Then for [1 : b : c] ∈ D we get the ring isomorphism (2.1) S(1, b, c) ∼= k{x, y}[z; α, δ] (Ω) DEGENERATE SKLYANIN ALGEBRAS 5 where α ∈End(k{x, y}) is defined by α(x) = −bx, α(y) = −b2y and the element δ ∈ α-Der(k{x, y}) is given by δ(x) = −cy2, δ(y) = −b2cx2. Here Ω = xy + byx + cz2 is a normal element of k{x, y}[z, α, δ]. Proof. By direct computation α and δ are indeed an endomorphism and α- derivation of k{x, y} respectively. Moreover x·Ω = Ω·bx, y·Ω = Ω·by, z·Ω = Ω· z so Ω is a normal element of the Ore extension. Thus both rings of (2.1) have the same generators and relations. (cid:3) Remark 2.3. Some properties of degenerate Sklyanin algebras are easy to verify without use of the Proposition 2.2. Namely one can find a basis of irreducible monomials via Bergman's Diamond lemma [6, Theorem 1.2] to imply dimk Sd = 2d−13 for d ≥ 1. Equivalently S(1, b, c) is free with a basis {1, z} as a left or right module over k{x, y}. Therefore, HSdeg (t) = 1+t 1−2t . Therefore due to Proposition 2.2 (for a3 = b3 = c3 = 1) or Remark 2.3 we have the following immediate consequence. Corollary 2.4. The degenerate Sklyanin algebras have exponential growth, infinite GK dimension, and are not right Noetherian. Furthermore Sdeg is not a domain. Proof. The growth conditions follow from Remark 2.3 and the non-Noetherian property holds by [20, Theorem 0.1]. Moreover if [a : b : c] ∈ {[1 : 0 : 0], [0 : 1 : 0], [0 : 0 : 1]}, then the monomial algebra S(a, b, c) is obviously not a domain. On the other hand if [a : b : c] satisfies a3 = b3 = c3 = 1, then assume without loss of generality that a = 1. As a result we have f1 + bf2 + cf3 = (x + by + bc2z)(cx + cy + b2z), where f1 = yz + bzy + cx2, f2 = zx + bxz + cy2, and f3 = xy + byx + cz2 are the relations of S(1, b, c). (cid:3) Now we verify homological properties of degenerate Sklyanin algebras. Definition 2.5. Let A be a cg algebra which is locally finite (dimkAi < ∞). When provided a minimal resolution of the left A-module A/Li≥1 Ai ∼= k determined by matrices Mi, we say A is Koszul if the entries of the Mi all belong to A1. Proposition 2.6. The degenerate Sklyanin algebras are Koszul with infinite global dimension. Proof. For S = S(a, b, c) with a3 = b3 = c3 = 1, consider the description of S in Proposition 2.2. Since k{x, y} is Koszul, the Ore extension k{x, y}[z, α, δ] is also Koszul [9, Definition 1.1, Theorem 10.2]. By Proposition 2.2, the 6 CHELSEA WALTON element Ω is normal and regular in k{x, y}[z; α, δ]. Hence the factor S is Koszul by [17, Theorem 1.2]. To conclude gl.dim(S) = ∞, note that the Koszul dual of S is S(1, b, c)! ∼= k{x, y, z}   z2 − cxy, zy − b2yz, zx − bxz, yz − c2x2, y2 − bcxz, yx − b2xy .   Taking the ordering x < y < z, we see that all possible ambiguities of S! are resolvable in the sense of [6]. Bergman's Diamond lemma [6, Theorem 1.2] implies that S! has a basis of irreducible monomials {xi, xjy, xkz}i,j,k∈N. Hence S! is not a finite dimensional k-vector space and by [13, Corollary 5], S has infinite global dimension. For S = S(a, b, c) with [a : b : c] ∈ {[1 : 0 : 0], [0 : 1 : 0], [0 : 0 : 1]}, note that S is Koszul as its ideal of relations is generated by quadratic monomials [14, Corollary 4.3]. Denote these monomials m1, m2, m3. The Koszul dual of S in this case is S! ∼= k{x, y, z} (the six monomials not equal to mi) . Since S! is again a monomial algebra, it contains no hidden relations and has a nice basis of irreducible monomials. In particular, S! contains Li≥0 kwi where wi is the length i word: xyzxyzx . . . , if [a : b : c] = [1 : 0 : 0] Therefore S! is not a finite dimensional k-vector space. By [13, Corollary 5], the three remaining degenerate Sklyanin algebras are of infinite global dimension. (cid:3) 3. Truncated point schemes of Sdeg The goal of this section is to construct the family of truncated point schemes {Vd ⊆ (P2)×d} associated to the degenerate three-dimensional Sklyanin al- gebras Sdeg (see Definition 1.4). These schemes will be used in §4 for the construction of a generalized twisted homogeneous coordinate ring, namely the point parameter ring (Definition 1.8). Nevertheless the family {Vd} has immediate importance for understanding point modules over S = Sdeg. wi = xzyxzyx . . . ,   xi, i {z {z i } } if [a : b : c] = [0 : 1 : 0] if [a : b : c] = [0 : 0 : 1]. DEGENERATE SKLYANIN ALGEBRAS 7 Definition 3.1. A graded left S-module M is called a point module if M 1−t . Moreover a graded left S-module M is called a truncated point module of length d if M is again cyclic and i=0 ti = 1 is cyclic and HM (t) = P∞ HM (t) =Pd−1 i=0 ti. Note that point modules share the same Hilbert series as a point in pro- jective space in Classical Algebraic Geometry. Now we proceed to construct schemes Vd that will parameterize length d truncated point modules. This yields information regarding point modules over S(a, b, c) for any [a : b : c] ∈ P2 due to the following result. Lemma 3.2. [3, Proposition 3.9, Corollary 3.13] Let S = S(a, b, c) for any [a : b : c] ∈ P2. Denote by Γ the set of isomorphism classes of point modules over S and Γd the set of isomorphism classes of truncated point modules of length d + 1. With respect to the truncation function ρd : Γd → Γd−1 given by M 7→ M/Md+1, we have that Γ is the projective limit of {Γd} as a set. The sets Γd can be understood by the schemes Vd defined below. Definition 3.3. [3, §3] The truncated point scheme of length d, Vd ⊆ (P2)×d, is the scheme defined by the multilinearizations of relations of S(a, b, c) from Definition 1.1. More precisely Vd = V(fi, gi, hi)0≤i≤d−2 where fi = ayi+1zi + bzi+1yi + cxi+1xi gi = azi+1xi + bxi+1zi + cyi+1yi hi = axi+1yi + byi+1xi + czi+1zi. (3.1) For example, V1 = V(0) ⊆ P2 so we have V1 = P2. Similarly, V2 = V(f0, g0, h0) ⊆ P2 × P2. Lemma 3.4. [3] The set Γd is parameterized by the scheme Vd. In short, to understand point modules over S(a, b, c) for any [a : b : c] ∈ P2, Lemmas 3.2 and 3.4 imply that we can now restrict our attention to truncated point schemes Vd. On the other hand, we point out another useful result pertaining to Vd associated to S(a, b, c) for any [a : b : c] ∈ P2. Lemma 3.5. The truncated point scheme Vd lies in d copies of E ⊆ P2 where E is the cubic curve E : (a3 + b3 + c3)xyz − (abc)(x3 + y3 + z3) = 0. Proof. Let pi denote the point [xi : yi : zi] ∈ P2 and Mabc,i := Mi := cxi bzi ayi azi cyi bxi byi axi czi ! ∈ Mat3(kxi ⊕ kyi ⊕ kzi). (3.2) 8 CHELSEA WALTON A d-tuple of points p = (p0, p1, . . . , pd−1) ∈ Vd ⊆ (P2)×d must satisfy the system fi = gi = hi = 0 for 0 ≤ i ≤ d − 2 by definition of Vd. In other words, one is given Mabc,j · (xj+1 yj+1 zj+1)T = 0 or equivalently (xj yj zj)· Mabc,j+1 = 0 for 0 ≤ j ≤ d− 2. Therefore for 0 ≤ j ≤ d− 1, det(Mabc,j) = 0. This implies pj ∈ E for each j. Thus p ∈ E×d. 3.1. On the truncated point schemes of some Sdeg. We will show that to study the truncated point schemes Vd of degenerate Sklyanin algebras, it suffices to understand the schemes of specific four degenerate Sklyanin algebras. Recall that Vd parameterizes length d truncated point modules (Lemma 3.4). Moreover note that according to [21], two graded algebras A and B have equivalent graded left module categories (A-Gr and B-Gr) if A is a Zhang twist of B. The following is a special case of [21, Theorem 1.2]. (cid:3) Theorem 3.6. Given a Z-graded k-algebra S =Ln∈Z Sn with graded auto- morphism σ of degree 0 on S, we form a Zhang twist Sσ of S by preserving the same additive structure on S and defining multiplication ∗ as follows: a ∗ b = abσn for a ∈ Sn. Furthermore if S and Sσ are cg and generated in degree one, then S-Gr and Sσ-Gr are equivalent categories. (cid:3) Realize D from Definition 1.1 as the union of three point sets Zi: (3.3) Z1 := {[1 : 1 : 1], Z2 := {[1 : 1 : ζ], Z3 := {[1 : ζ : ζ], Z0 := {[1 : 0 : 0], [1 : ζ : ζ 2], [1 : ζ : 1], [1 : 1 : ζ 2], [0 : 1 : 0], [1 : ζ 2 : ζ]}, [1 : ζ 2 : ζ 2]}, [1 : ζ 2 : 1]}, [0 : 0 : 1]}. where ζ = e2πi/3. Pick respective representatives [1 : 1 : 1], [1 : 1 : ζ], [1 : ζ : ζ], and [1 : 0 : 0] of Z1, Z2, Z3, and Z0. Lemma 3.7. Every degenerate Sklyanin algebra is a Zhang twist of one the following algebras: S(1, 1, 1), S(1, 1, ζ), S(1, ζ, ζ), and S(1, 0, 0). Proof. A routine computation shows that the following graded automor- phisms of degenerate S(a, b, c), σ : {x 7→ ζx, y 7→ ζ 2y, z 7→ z} and τ : {x 7→ y, y 7→ z, z 7→ x}, yield the Zhang twists: S(1, 1, 1)σ = S(1, ζ, ζ 2), S(1, 1, ζ)σ = S(1, ζ, 1), S(1, ζ, ζ)σ = S(1, ζ 2, 1), S(1, 0, 0)τ = S(0, 1, 0), S(1, 1, 1)σ−1 S(1, 1, ζ)σ−1 S(1, ζ, ζ)σ−1 S(1, 0, 0)τ −1 = S(1, ζ 2, ζ) = S(1, ζ 2, ζ 2) = S(1, 1, ζ 2) = S(0, 0, 1) for Z1; for Z2; for Z3; for Z0. (cid:3) Therefore it suffices to study a representative of each of the four classes of degenerate three-dimensional Sklyanin algebras due to Theorem 3.6. DEGENERATE SKLYANIN ALGEBRAS 9 3.2. Computation of Vd for S(1, 1, 1). We now compute the truncated point schemes of S(1, 1, 1) in detail. Calculations for the other three repre- sentative degenerate Sklyanin algebras, S(1, 1, ζ), S(1, ζ, ζ), S(1, 0, 0), will follow with similar reasoning. To begin we first discuss how to build a trun- cated point module M ′ of length d, when provided with a truncated point module M of length d − 1. Let us explore the correspondence between truncated point modules and truncated point schemes for a given d; say d ≥ 3. When given a truncated point module M = Ld−1 i=0 Mi ∈ Γd−1, multiplication from S = S(a, b, c) is determined by a point p = (p0, . . . , pd−2) ∈ Vd−1 (Definition 3.3, (3.2)) in the following manner. As M is cyclic, Mi has basis say {mi}. Furthermore for x, y, z ∈ S with pi = [xi : yi : zi] ∈ P2, we get the left S-action on mi determined by pi: x · mi = ximi+1, x · md−1 = 0; y · mi = yimi+1, y · md−1 = 0; z · md−1 = 0. z · mi = zimi+1, (3.4) Conversely given a point p = (p0, . . . , pd−2) ∈ Vd−1, one can build a module M ∈ Γd−1 unique up to isomorphism by reversing the above process. We summarize this discussion in the following remark. Remark 3.8. Refer to notation from Lemma 3.2. To construct M ′ ∈ Γd from M ∈ Γd−1 associated to p ∈ Vd−1, we require pd−1 ∈ P2 such that p′ = (p, pd−1) ∈ Vd. Now we begin to study the behavior of truncated point modules over Sdeg through the examination of truncated point schemes in the next two lemmas. Lemma 3.9. Let p = (p0, . . . , pd−2) ∈ Vd−1 with pd−2 6∈ Zi (refer to (3.3)). Then there exists a unique pd−1 ∈ Zi so that p′ := (p, pd−1) ∈ Vd. Proof of 3.9. For Z1, we study the representative algebra S(1, 1, 1). If such a pd−1 exists, then fd−2 = gd−2 = hd−2 = 0 so we would have M111,d−2 · (xd−1 yd−1 zd−1)T = 0 (Definition 3.3, Eq. (3.2)). Since rank(M111,d−2) = 2 when pd−2 6∈ D, the tuple (xd−1, yd−1, zd−1) is unique up to scalar multiple and thus the point pd−1 is unique. To verify the existence of pd−1, say pd−2 = [0 : yd−2 : zd−2]. We require pd−2 and pd−1 to satisfy the system of equations: d−2 + z3 y3 d−2 = x3 fd−2 = gd−2 = hd−2 = 0 d−1 = 0 d−1 + y3 d−1 + z3 (Eq. (3.1)) (pd−2, pd−1 ∈ E, Lemma 3.5).   ([1 : −(1 + zd−2) : zd−2], ([1 : −ζ(1 + ζzd−2) : zd−2], ([1 : −ζ(ζ + zd−2) : zd−2], [1 : 1 : 1]), [1 : ζ : ζ 2]), [1 : ζ 2 : ζ]) .   10 CHELSEA WALTON However basic algebraic operations imply yd−2 = zd−2 = 0, thus producing a contradiction. Therefore, without loss of generality pd−2 = [1 : yd−2 : zd−2]. With similar reasoning we must examine the system yd−1zd−2 + zd−1yd−2 + xd−1 = 0 zd−1 + xd−1zd−2 + yd−1yd−2 = 0 xd−1yd−2 + yd−1 + zd−1zd−2 = 0 (3.5) d−2 = 3yd−2zd−2 d−1 = 3xd−1yd−1zd−1. There are three solutions (pd−2, pd−1) ∈ (E \ Z1) × E to (3.5): 1 + y3 x3 d−1 + y3 d−2 + z3 d−1 + z3 Thus when pd−2 6∈ Z1, there exists an unique point pd−1 ∈ Z1 so that (p0, . . . , pd−2, pd−1) ∈ Vd. Now having studied S(1, 1, 1) with care, we leave it to the reader to verify the assertion for the algebras S(1, 1, ζ), S(1, ζ, ζ), and S(1, 0, 0) in a similar manner. (cid:3) The next result explores the case when pd−2 ∈ Zi. Lemma 3.10. Let p = (p0, . . . , pd−2) ∈ Vd−1 with pd−2 ∈ Zi. Then for any [yd−1 : zd−1] ∈ P1 there exists a function θ of two variables so that pd−1 = [θ(yd−1, zd−1) : yd−1 : zd−1] 6∈ Zi which satisfies (p0, . . . , pd−2, pd−1) ∈ Vd. Proof. The point p′ = (p, pd−1) ∈ Vd needs to satisfy fi = gi = hi = 0 for 0 ≤ i ≤ d − 2 (Definition 3.3). Since p ∈ Vd−1, we need only to consider the equations fd−2 = gd−2 = hd−2 = 0 with pd−2 ∈ Zi. We study S(1, 1, 1) for Z1 so the relevant system of equations is fd−2 : yd−1zd−2 + zd−1yd−2 + xd−1xd−2 = 0 gd−2 : zd−1xd−2 + xd−1zd−2 + yd−1yd−2 = 0 hd−2 : xd−1yd−2 + yd−1xd−2 + zd−1zd−2 = 0. If pd−2 = [1 : 1 : 1] ∈ Z1, then xd−1 = −(yd−1 + yd−1) is required. On the other hand, if pd−2 = [1 : ζ : ζ 2] or [1 : ζ 2 : ζ], we require xd−1 = −ζ(yd−1+ζzd−1) or xd−1 = −ζ(ζyd−1+zd−1) respectively. Thus our function θ is defined as θ(yd−1, zd−1) =  −(yd−1 + zd−1), −(ζyd−1 + ζ 2zd−1), −(ζ 2yd−1 + ζzd−1), if pd−2 = [1 : 1 : 1] if pd−2 = [1 : ζ : ζ 2] if pd−2 = [1 : ζ 2 : ζ]. DEGENERATE SKLYANIN ALGEBRAS 11 The arguments for S(1, 1, ζ), S(1, ζ, ζ), and S(1, 0, 0) proceed in a likewise fashion. (cid:3) Fix a pair (Sdeg, Zi(Sdeg)). We now know if pd−2 6∈ Zi, then from every truncated point module of length d over Sdeg we can produce a unique truncated point module of length d + 1. Otherwise if pd−2 ∈ Zi, we get a P1 worth of length d + 1 modules. We summarize this in the following statement which is made precise in Proposition 3.13. Proposition 3.11. The parameter space of Γd over Sdeg is isomorphic to the singular and nondisjoint union of ( three copies of (P1)× d−1 six copies of (P1)× d 2 , 2 and three copies of (P1)× d+1 2 , for d odd; for d even. The detailed statement and proof of this proposition will follow from the results below. We restrict our attention to S(1, 1, 1) for reasoning mentioned in the proofs of Lemmas 3.9 and 3.10. 3.2.1. Parameterization of Γ2. Recall that length 3 truncated point modules of Γ2 are in bijective correspondence to points on V2 ⊂ P2 × P2 (Lemma 3.4) and it is our goal to depict this truncated point scheme. By Lemma 3.5, we know that V2 ⊆ E × E. Furthermore note that with ζ = e2πi/3, the curve E = E111 is the union of three projective lines: B : x = −(ζy + ζ 2z), P1 A : x = −(y + z), P1 C : x = −(ζ 2y + ζz) (3.6) P1 P1 C P1 B [1:1:1] ✹✹✹✹✹✹✹✹✹✹✹✹✹✹✹✹✹✹✹✹✹✹✹✹✹✹✹✹✹✹ ✡✡✡✡✡✡✡✡✡✡✡✡✡✡✡✡✡✡✡✡✡✡✡✡✡✡✡✡✡✡ [1:ζ 2:ζ] [1:ζ:ζ 2] / P1 A Figure 1: The curve E = E111 ⊆ P2 : x3 + y3 + z3 − 3xyz = 0. Now to calculate V2, recall that Γ2 consists of length 3 truncated point modules M(3) := M0 ⊕ M1 ⊕ M2 where Mi is a 1-dimensional k-vector space say with basis mi. The module M(3) has action determined by (p0, p1) ∈ V2 (Eq. (3.4)). Moreover Lemmas 3.9 and 3.10 provide the precise conditions for (p0, p1) to lie in E × E. Namely, Y Y E E o o / 12 CHELSEA WALTON Lemma 3.12. Refer to (3.6) for notation. The set of length 3 truncated point modules Γ2 is parametrized by the scheme V2 = V(f0, g0, h0) which is the union of the six subsets: P1 A × [1 : 1 : 1]; B × [1 : ζ : ζ 2]; P1 P1 C × [1 : ζ 2 : ζ]; [1 : 1 : 1] × P1 A; [1 : ζ : ζ 2] × P1 B; [1 : ζ 2 : ζ] × P1 C. of E × E. Thus Γ2 is isomorphic to 6 copies of P1. 3.2.2. Parameterization of Γd for general d. To illustrate the parametriza- tion of Γd, we begin with a truncated point module M(d+1) of length d + 1 corresponding to (p0, p1, . . . , pd−1) ∈ Vd ⊆ (P2)×d. Due to Lemmas 3.5, 3.9, and 3.10, we know that (p0, p1, . . . , pd−1) belongs to either (cid:3) (E \ Z1) × Z1 × (E \ Z1) × Z1 × . . . } where Z1 is defined in (3.3). {z d or Z1 × (E \ Z1) × Z1 × (E \ Z1) × . . . } {z d By adapting the notation of Lemma 3.10, we get in the first case that the point (p0, p1, . . . , pd−1) is of the form ([θ(y0, z0) : y0 : z0], [1 : ω : ω2], [θ(y2, z2) : y2 : z2], [1 : ω : ω2], . . . ) ∈ (P2)×d where ω3 = 1 and θ(y, z) = −(ωy + ω2z). Thus in this case, the set of length d truncated point modules is parameterized by three copies of (P1)×⌈d/2⌉ with coordinates ([y0 : z0], [y2 : z2], . . . , [y2⌈d/2⌉−1 : z2⌈d/2⌉−1]). In the second case (p0, p1, . . . , pd−1) takes the form ([1 : ω : ω2], [θ(y1, z1) : y1 : z1], [1 : ω : ω2], [θ(y3, z3) : y3 : z3], . . . ) ∈ (P2)×d and the set of truncated point modules is parameterized with three copies of (P1)×⌊d/2⌋ with coordinates ([y1 : z1], [y3 : z3], . . . , [y2⌊d/2⌋−1 : z2⌊d/2⌋−1]). In other words, we have now proved the next result. Proposition 3.13. Refer to (3.6) for notation. For d ≥ 2 the truncated point scheme Vd for S(1, 1, 1) is equal to the union of the six subsetsS6 i=1 Wd,i of (P2)×d where Wd,1 = P1 Wd,2 = [1 : 1 : 1] × P1 Wd,3 = P1 Wd,4 = [1 : ζ : ζ 2] × P1 Wd,5 = P1 Wd,6 = [1 : ζ 2 : ζ] × P1 A × [1 : 1 : 1] × P1 B × [1 : ζ : ζ 2] × P1 C × [1 : ζ 2 : ζ] × P1 A × [1 : 1 : 1] × . . . , A × . . . , B × [1 : ζ : ζ 2] × . . . , C × . . . , C × [1 : ζ 2 : ζ] × . . . , C × . . . . A × [1 : 1 : 1] × P1 B × [1 : ζ : ζ 2] × P1 C × [1 : ζ 2 : ζ] × P1 (cid:3) As a consequence, we obtain the proof of Proposition 3.11 for S(1, 1, 1) and this assertion holds for the remaining degenerate Sklyanin algebras due to Lemma 3.7, and analogous proofs for Lemmas 3.9 and 3.10. (cid:3) DEGENERATE SKLYANIN ALGEBRAS 13 We thank Karen Smith for suggesting the following elegant way of inter- preting the point scheme of S(1, 1, 1). Remark 3.14. We can provide an alternate geometric description of the point scheme of the Γ of S(1, 1, 1). Let G := Z3 ⋊ Z2 =< ζ, σ > where ζ = e2πi/3 and σ2 = 1. We define a G-action on P2 × P2 as follows: ζ([x : y : z], [u : v : w]) = ([x : ζ 2y : ζz], [u : ζv : ζ 2w]) σ([x : y : z], [u : v : w]) = ([u : v : w], [x : y : z]) Note that G stabilizes E × E and acts transitively on the W2,i. We extend the action of G to (P2 × P2)×∞ diagonally. Now we interpret Γ as W2d,i = G · (P1 A × [1 : 1 : 1])×∞, Γ = lim ←− Vd = lim ←− as sets. V2d = lim ←−[i 4. Point parameter ring of S(1, 1, 1) We now construct a graded associative algebra B from truncated point schemes of the degenerate Sklyanin algebra S = S(1, 1, 1). The analogous result for the other degenerate Sklyanin algebras will follow in a similar fashion and we leave the details to the reader. As is true for the Sklyanin algebras themselves, it will be shown that this algebra B is a proper factor of S(1, 1, 1) and its properties closely reflect those of S(1, 1, 1). We will for example show that B is not right Noetherian, nor a domain. The definition of the algebra B initially appears in [3, §3]. Recall that we have projection maps pr1,...,d−1 and pr2,...,d from (P2)×d to (P2)×d−1. Restrictions of these maps to the truncated point schemes Vd ⊆ (P2)×d (Definition 3.3) yield pr1,...,d−1(Vd) ⊂ Vd−1 and pr2,...,d(Vd) ⊂ Vd−1 for all d. Definition 4.1. Given the above data, the point parameter ring B = B(S) is an associative N-graded ring defined as follows. First Bd = H 0(Vd,Ld) where Ld is the restriction of invertible sheaf 1OP2(1) ⊗ . . . ⊗ pr∗ pr∗ dOP2(1) ∼= O(P2)×d(1, . . . , 1) to Vd. The multiplication map µi,j : Bi × Bj → Bi+j is then defined by applying H 0 to the isomorphism pr∗ 1,...,i(Li) ⊗OVi+j pr∗ i+1,...,i+j(Lj) → Li+j. We declare B0 = k. We will later see in Theorem 4.6 that B is generated in degree one; thus S surjects onto B. 14 CHELSEA WALTON To begin the analysis of B for S(1, 1, 1), recall that V1 = P2 so B1 = H 0(V1, pr∗ 1OP2(1)) = kx ⊕ ky ⊕ kz where [x : y : z] are the coordinates of P2. For d ≥ 2 we will compute dimk Bd and then proceed to the more difficult task of identifying the multiplication maps µi,j : Bi × Bj → Bi+j. Before we get to specific calculations for d ≥ 2, let us recall that the schemes Vd are realized as the union of six subsets {Wd,i}6 (3.6). These subsets intersect nontrivially so that each Vd for d ≥ 2 is singular. More precisely, i=1 of (P2)×d described in Proposition 3.13 and Eq. Remark 4.2. A routine computation shows that the singular subset, Sing(Vd), consists of six points: vd,1 := ([1 : 1 : 1], [1 : ζ : ζ 2], [1 : 1 : 1], [1 : ζ : ζ 2], . . . ) ∈ Wd,2 ∩ Wd,3, vd,2 := ([1 : 1 : 1], [1 : ζ 2 : ζ], [1 : 1 : 1], [1 : ζ 2 : ζ], . . . ) ∈ Wd,2 ∩ Wd,5, vd,3 := ([1 : ζ : ζ 2], [1 : 1 : 1], [1 : ζ : ζ 2], [1 : 1 : 1], . . . ) ∈ Wd,1 ∩ Wd,4, vd,4 := ([1 : ζ : ζ 2], [1 : ζ : ζ 2], [1 : ζ : ζ 2], [1 : ζ : ζ 2], . . . ) ∈ Wd,3 ∩ Wd,4, vd,5 := ([1 : ζ 2 : ζ], [1 : 1 : 1], [1 : ζ 2 : ζ], [1 : 1 : 1], . . . ) ∈ Wd,1 ∩ Wd,6, vd,6 := ([1 : ζ 2 : ζ], [1 : ζ 2 : ζ], [1 : ζ 2 : ζ], [1 : ζ 2 : ζ], . . . ) ∈ Wd,5 ∩ Wd,6. where ζ = e2πi/3. 4.1. Computing the dimension of Bd. Our objective in this section is to prove Proposition 4.3. For d ≥ 1, dimk Bd = 3(cid:16)2⌊ d+1 2 ⌉(cid:17) − 6. For the rest of the section, let 1 denote a sequence of 1s of appropriate length. Now consider the normalization morphism π : V ′ d → Vd where V ′ d is the disjoint union of the six subsets {Wd,i}6 i=1 mentioned in Proposition 3.13. This map induces the following short exact sequence of sheaves on Vd: 2 ⌋ + 2⌈ d−1 0 → OVd(1) → (π∗OV ′ d )(1) → S(1) → 0, (4.1) where S is the skyscraper sheaf whose support is Sing(Vd), that is S = L6 k=1 O{vd,k}. Note that we have H 0(Vd, (π∗OV ′ d )(1)) ∼= k−v.s. H 0(V ′ d,OV ′ d (1)) (4.2) since the normalization morphism is a finite map, which in turn is an affine map [11, Exercises II.5.17(b), III.4.1]. To complete the proof of the propo- sition, we make the following assertion: Claim: H 1(Vd,OVd(1)) = 0. DEGENERATE SKLYANIN ALGEBRAS 15 Assuming that the claim holds, we get from (4.1) the following long exact sequence of cohomology: 0 → H 0(Vd,OVd (1)) → H 0(Vd, (π∗OV ′ d )(1)) → H 0(Vd,S(1)) → H 1(Vd,OVd(1)) = 0. Thus, with writing h0(X,L) = dimk H 0(X,L), (4.2) implies that )(1)) − h0(S(1)) dimk Bd = h0(OVd(1)) = h0((π∗OV ′ d d (1)) − h0(S(1)) i=1 h0(OWd,i(1)) − 6. = h0(OV ′ =P6 Therefore applying Proposition 3.11 and Kunneth's Formula [8, A.10.37] completes the proof of Proposition 4.3. It now remains to verify the claim. Proof of Claim: By the discussion above, it suffices to show that δd : H 0(V ′ d,OV ′ d {vd,k}, S(1)! (1)) → H 0 6 [k=1 is surjective. Referring to the notation of Proposition 3.13 and Remark 4.2, we choose vd,i ∈ Supp(S(1)) and Wd,ki containing vd,i. This Wd,ki contains precisely two points of Supp(S(1)) and say the other is vd,j for j 6= i. After choosing a basis {ti}6 i=1 for the six-dimensional vector space H 0(S(1)) where ti(vd,j) = δij, we construct a preimage of each ti. Since (1) is a very ample sheaf, it separates points. In other words there OWd,ki exists si ∈ H 0(OWd,ki (1)) such that si(vd,j) = δij. Extend this section si to si ∈ H 0(OV ′ (1)) by declaring si = si on Wd,ki and si = 0 elsewhere. Thus δd(si) = ti for all i and the map δd is surjective as desired. (cid:3) d This concludes the proof of Proposition 4.3. Corollary 4.4. We have limd→∞(dimk Bd)1/d = √2 > 1 so B has expo- nential growth hence infinite GK dimension. By [20, Theorem 0.1], B is not left or right Noetherian. (cid:3) On the other hand, we can also determine the Hilbert series of B. Proposition 4.5. HB(t) = (1 + t2)(1 + 2t) (1 − 2t2)(1 − t) . Proof. Recall from Proposition 4.3 that dimk Bd = 3(cid:16)2⌈ d−1 for d ≥ 1 and that dimk B0 = 1. Thus td 2⌈ d−1 2 ⌋td − 2Xd≥1  td 2⌊ d 2 ⌋td − 2tXd≥0  . HB(t) = 1 + 3 Xd≥1 = 1 + 3 tXd≥0 2 ⌉td +Xd≥1 2⌈ d 2 ⌉td + 2tXd≥0 2⌊ d+1 2 ⌉ + 2⌊ d+1 2 ⌋(cid:17) − 6 16 CHELSEA WALTON Consider generating functions a(t) = Pd≥0 adtd and b(t) = Pd≥0 bdtd for the respective sequences ad = 2⌈d/2⌉ and bd = 2⌊d/2⌋. Elementary operations result in a(t) = 1+2t 1−2t2 and b(t) = 1+t 1 − 2t2(cid:19) + 2t(cid:18) 1 + t 1−2t2 . Hence 1 − 2t2(cid:19) − 2t(cid:18) 1 HB(t) = 1 + 3(cid:20)t(cid:18) 1 + 2t 1 − t(cid:19)(cid:21) = (1 + t2)(1 + 2t) (1 − 2t2)(1 − t) . (cid:3) 4.2. The multiplication maps µij : Bi × Bj → Bi+j . In this section we examine the multiplication of the point parameter ring B of S(1, 1, 1). In particular, we show that the multiplication maps are surjective which results in the following theorem. Theorem 4.6. The point parameter ring B of S(1, 1, 1) is generated in degree one. With similar reasoning, B = B(Sdeg) is generated in degree one for all Sdeg. Proof. It suffices to prove that the multiplication maps µd,1 : Bd × B1 → Bd+1 are surjective for d ≥ 1. Recall from Definition 4.1 that µd,1 = H 0(md) where md is the isomorphism md : OVd×P2(1, . . . , 1, 0) ⊗OVd+1 O(P2)×d(0, . . . , 0, 1) → OVd+1(1, . . . , 1). To use the isomorphism md, we employ the following commutative diagram: OVd×P2(1, . . . , 1, 0) ⊗O(P2)×d+1 O(P2)×d+1(0, . . . , 0, 1) OVd×P2(1, . . . , 1, 0) ⊗OVd+1 O(P2)×d+1(0, . . . , 0, 1) / OVd+1(1, . . . , 1). td +❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲❲ md (4.3) (1) −→ OVd×P2(1) The source of td is isomorphic to OVd×P2(1, . . . , 1) and the map td is given by restriction to Vd+1. Hence we have the short exact sequence td−→ OVd+1(1) −→ 0, 0 −→ I Vd+1 Vd×P2 is the ideal sheaf of Vd+1 defined in Vd× P2. Since the Kunneth where I Vd+1 Vd×P2 formula and the claim from §4.1 implies that H 1(OVd×P2(1)) = 0, the cok- ernel of H 0(td) is H 1(cid:16)I Vd+1 Proposition 4.7. H 1(cid:16)I Vd+1 (1)(cid:17). Now we assert (1)(cid:17) = 0 for d ≥ 1. (4.4) Vd ×P2 Vd×P2 + / DEGENERATE SKLYANIN ALGEBRAS 17 By assuming that Proposition 4.7 holds, we get the surjectivity of H 0(td) for d ≥ 1. Now by applying the global section functor to Diagram (4.3), we have that H 0(md) = µd,1 is surjective for d ≥ 1. This concludes the proof of Theorem 4.6. (cid:3) := IV2 Proof of 4.7. Consider the case d = 1. We study the ideal sheaf I V2 by using the resolution of the ideal of defining relations (f0, g0, h0) for V2 (Eqs. (3.1)) in the N2-graded ring R = k[x0, y0, z0, x1, y1, z1]. Note that each of the defining equations have bidegree (1,1) in R and we get the following resolution: P2×P2 0 → OP2×P2(−3,−3) → OP2×P2(−2,−2)⊕3 → OP2×P2(−1,−1)⊕3 → IV2 → 0. Twisting the above sequence with OP2×P2(1, 1) we get 0 → OP2×P2(−2,−2) → OP2×P2(−1,−1)⊕3 → O⊕3 P2×P2 f → IV2 (1, 1) → 0. Let K = ker(f ). Then h0(IV2(1, 1)) = 3 − h0(K) + h1(K). On the other hand, H 1(OP2(j)) = H 2(OP2(j) = 0 for j = −1,−2. Thus the Kunneth formula applied the cohomology of the short exact sequence 0 → OP2×P2(−2,−2) → OP2×P2(−1,−1)⊕3 → K → 0 results in h0(K) = h1(K) = 0. Hence h0(IV2(1, 1)) = 3. exact sequence Now using the long exact sequence of cohomology arising from the short 0 → IV2(1, 1) → OP2×P2(1, 1) → OV2(1, 1) → 0, and the facts: h0(IV2(1, 1)) = 3, h0(OP2×P2(1, 1)) = 9 h0(OV2 (1, 1)) = dimk B2 = 6, h1(OP2×P2(1, 1)) = 0, For d ≥ 2 we will construct a commutative diagram to assist with the (1). Recall from (4.1) that we conclude that H 1(IV2(1, 1)) = 0. study of the cohomology of the ideal sheaf I Vd+1 Vd ×P2 we have the following normalization sequence for Vd: 0 −→ OVd −→ 6 Mi=1 OWd,i −→ 6 Mk=1 O{vd,k} −→ 0. (†d) Consider the sequence pr∗ 1,...,d(cid:16)(†d) ⊗ O(P2)×d(1)(cid:17) ⊗O(P2)×d+1 pr∗ d+1OP2(1) 18 CHELSEA WALTON and its induced sequence of restrictions to Vd+1, namely 0 → O Vd×P2 (1)(cid:12)(cid:12)Vd+1 → 6 Mi=1 O Wd,i×P2 (1)(cid:12)(cid:12)Vd+1 → 6 Mk=1 O {vd,k}×P2 (1)(cid:12)(cid:12)Vd+1 → 0. (4.5) Now Vd+1 ⊆ Vd × P2 and (Wd,i × P2) ∩ Vd+1 = Wd+1,i due to Proposition 3.13 and Remark 4.2. We also have that ({vd,k} × P2) ∩ Vd+1 = {vd+1,k} for all i,k. Therefore the sequence (4.5) is equal to (†d+1) ⊗ O(P2)×d+1(1). In other words, we are given the commutative diagram: 0 / OVd×P2(1) 6 Mi=1 OWd,i×P2(1) 6 Mk=1 O{vd,k}×P2(1) / 0 0 / OVd+1(1) O{vd+1,k}(1) / 0. 6 6 Mi=1 OWd+1,i(1) Mk=1 Diagram 1: Understanding I Vd+1 Vd×P2 (1, . . . , 1). where the vertical maps are given by restriction to Vd+1. Observe that the (1), kernels of the vertical maps (from left to right) are respectively I Vd+1 Vd×P2 Mi I Wd+1,i Wd,i×P2 By the claim in §4.1 and the Kunneth formula, we have that (1), and Mk (1), and the cokernels are all 0. I {vd+1,k } {vd,k }×P2 H 1(OVd×P2(1)) = H 1(OVd+1(1)) = 0. Hence the application of the global section functor to Diagram 1 yields Diagram 2 below. Now by the Snake Lemma, we get the following sequence: 6 6 Wd,i ×P2 . . . −→ (1)(cid:19) −→ H 0 I Wd+1,i H 1 I Wd+1,i −→ H 1(cid:18)I Vd+1 In Lemma 4.8, we will show that Mi Mi=1 Mi=1 Vd ×P2 Wd,i×P2 6 (1)! ψ Mk=1 −→ (1)! → . . . . H 1(cid:16)I Wd+1,i Wd,i×P2 Furthermore the surjectivity of the map ψ will follow from Lemma 4.9. This will complete the proof of Proposition 4.7. H 0 I {vd+1,k } {vd,k }×P2 (1)! (1)(cid:17) = 0 for d ≥ 2. / / / / / / / / / / / / 0 0 0 H 0(cid:18)Vd × P2,I Vd+1 Vd×P2 (1)(cid:19) H 0 Wd,i × P2,I Wd+1,i Wd,i×P2 (1)! ψ /❴❴❴❴ 6 Mi=1 α / H 0(cid:0)Vd × P2,OVd×P2(1)(cid:1) 6 Mi=1 H 0(cid:0)Wd,i × P2,OWd,i×P2(1)(cid:1) ν 6 Mk=1 6 Mk=1 H 0 {vd,k} × P2,I {vd+1,k } {vd,k }×P2 (1)! λ H 0(cid:0){vd,k} × P2,O{vd,k}×P2(1)(cid:1) β γ / H 0(cid:0)Vd+1,OVd+1(1)(cid:1) 6 Mi=1 H 0(cid:0)Wd+1,i,OWd+1,i(1)(cid:1) 6 Mk=1 H 0(cid:0){vd+1,k},O{vd+1,k}(1)(cid:1) 0 0 / 0 / 0 H 1(cid:18)Vd × P2,I Vd+1 Vd×P2 (1)(cid:19) H 1 Wd,i × P2,I Wd+1,i Wd,i×P2 (1)! 6 Mi=1 H 1 {vd,k} × P2,I {vd+1,k } {vd,k }×P2 (1)! 6 Mk=1 Diagram 2: Induced Cohomology from Diagram 1 D E G E N E R A T E S K L Y A N N I A L G E B R A S 1 9 / / / / / / / / / / / / / 20 CHELSEA WALTON Lemma 4.8. 6 Mi=1 H 1(cid:16)Wd,i × P2, I Wd+1,i Wd,i×P2 (1)(cid:17) = 0 for d ≥ 2. Proof. We consider the different parities of d and i separately. For d even and i odd, Wd,i×P2 ∼= OWd,i×P2(0, . . . , 0,−1) I Wd+1,i because Wd+1,i is defined in Wd,i × P2 by one equation of degree (0, . . . , 0, 1) (Proposition 3.13). Twisting by O(P2)×d+1(1, . . . , 1) results in H 1 I Wd+1,i Wd,i×P2 (1, . . . , 1)! ∼= H 1(cid:16)OWd,i×P2(1, . . . , 1, 0)(cid:17) . (4.6) Since Wd,i is the product of P1 and points lying in P2 and H 1(OP1(1)) = H 1(O{pt}(1)) = H 1(OP2) = 0, the Kunneth formula implies that the right hand side of (4.6) is equal to zero. Consider the case of d and i even. As pr1,...,d(Wd+1,i) = Wd,i and prd+1(Wd+1,i) = [1 : ω : ω2] for ω = ωd,i a third of unity, we have that Wd+1,i is defined in Wd,i × P2 by two equations of degree (0, . . . , 0, 1). The defining equations (in variables x, y, z) of [1 : ω : ω2] form a k[x, y, z]-regular sequence and so we have the Koszul resolution of I Wd+1,i Wd,i×P2 ⊗ O(P2)×d+1(1, . . . , 1): 0 → OWd,i×P2(1, . . . , 1,−1) → OWd,i×P2(1, . . . , 1, 0)⊕2 (1, . . . , 1) → 0. → I Wd+1,i Wd,i×P2 (4.7) Now apply the global section functor to sequence (4.7) and note that H j(OWd,i(1, . . . , 1)) = H j(OP2) = H j(OP2(−1)) = 0 for j = 1, 2. Hence the Kunneth formula yields H 1(cid:16)OWd,i×P2(1, . . . , 1, 0)(cid:17)⊕2 = H 2(cid:16)OWd,i×P2(1, . . . , 1,−1)(cid:17) = 0. odd, the same conclusion is drawn by swapping the arguments for the i even and i odd subcases. (cid:3) Lemma 4.9. The map ψ is surjective for d ≥ 2. Proof. Refer to the notation from Diagram 2. To show ψ is onto, here is our plan of attack. Therefore H 1(cid:16)I Wd+1,i Wd,i×P2 (1)(cid:17) = 0 for d and i even. Mi=1 6 We conclude that for d even, we know H 1(cid:16)I Wd+1,i Wd,i×P2 (1)(cid:17) = 0. For d DEGENERATE SKLYANIN ALGEBRAS 21 } }×P2 {vd,k0 Wd,i×P2 {vd,k}×P2 (1)(cid:17) so that each basis element t H 0(cid:16)I {vd+1,k} (1)(cid:17) for some k = k0. For such a basis element (1) Choose a basis of Mk lies in H 0(cid:16)I {vd+1,k0 t, identify its image under λ in Mk H 0(cid:0)O{vd,k}×P2(1)(cid:1). (2) Construct for λ(t) a suitable preimage s ∈ ν−1(λ(t)). (3) Prove s ∈ ker(β). As a consequence, s lies inMi H 0(cid:16)I Wd+1,i (1)(cid:17) and serves as a preimage to t under ψ. In other words, ψ is surjective. To begin, fix such a basis element t and integer k0. Step 1: Observe that pr1,...,d({vd+1,k0}) = {vd,k0} and prd+1({vd+1,k0}) = [1 : ω : ω2] for some ω, a third root of unity (Remark 4.2). Thus our basis element t ∈ Mk (1)(cid:17) is of the form H 0(cid:16)I {vd+1,k } t = a(ωxd − yd) + b(ω2xd − zd) (4.8) for some a, b ∈ k, with {ωxd − yd, ω2xd − zd} defining [1 : ω : ω2] in the (d + 1)st copy of P2. Note that λ is the inclusion map so we may refer to λ(t) as t. This concludes Step 1. (cid:3) Step 2: Next we construct a suitable preimage s ∈ ν−1(λ(t)). Referring to Remark 4.2, let us observe that for all k, there is an unique even integer := i′′ k and unique odd integer := i′ for all k = 1, . . . , 6. For instance with k0 = 1, we consider the membership vd,1 ∈ Wd,2 ∩ Wd,3; hence i′′ k so that vd,k ∈ Wd,i′′ k ∩ Wd,i′ {vd,k }×P2 1 = 2 and i′ 1 = 3. k As a consequence, λ(t) has preimages under ν in k0 k0 × P2,OWd,i′ k0 × P2,OWd,i′′ H 0(cid:18)Wd,i′′ ×P2(1)(cid:19) ⊕ H 0(cid:18)Wd,i′ For d even (respectively odd) we write ik0 := i′′ k0 ×P2(1)(cid:19) . Therefore we intend to construct s ∈ ν−1(t) belonging to H 0(cid:16)OWd,ik0 Let us define the global section s ∈ H 0(cid:16)OWd,ik0 (1) is a very ample sheaf, we have a global section sk0 separating the OWd,ik0 points vd,k0 and vd,j; say sk0(vd,k) = δk0,k. We then use (4.8) to define s by ×P2(1)(cid:17). will also contain another point vd,j for some j 6= k0. ×P2(1)(cid:17) as follows. Since (respectively ik0 := i′ k0 However this Wd,ik0 ). k0 where [1 : ω : ω2] = prd+1({vd+1,k0}). We now extend this section s to s = sk0 · [a(ωxd − yd) + b(ω2xd − zd)]. H 0(cid:0)OWd,i×P2(1)(cid:1) ∼= 6 Mi=1 H 0(cid:0)OWd,i (1)(cid:1)! ⊗ H 0(OP2(1)). s ∈ 6 Mi=1 22 CHELSEA WALTON This is achieved by setting s = s on Wd,ik0 × P2 and 0 elsewhere. To check that ν(s) = t, note s = 6 Mi=1 si where si ∈ H 0(cid:0)OWd,i×P2(1)(cid:1) , si =(s, 0, i = ik0 , i 6= ik0 ; (4.9) Therefore by the construction of s, we have ν(s) = t(cid:12)(cid:12){vd,k0 }×P2. Hence we have built our desired preimage s ∈ ν−1(t) and this concludes Step 2. Step 3: Recall the structure of s from (4.9). By definition of β, we have that β(s) = β(cid:16)L6 i=1 si(cid:17) is equal to L6 For i 6= ik0, we clearly get that si(cid:12)(cid:12)Wd+1,i = 0. On the other hand, the = Wd,ik0 × [1 : ǫ : ǫ2] for key point of our construction is that Wd+1,ik0 some ǫ3 = 1 as ik0 is chosen to be even (respectively odd) when d is even (respectively odd) (Proposition 3.13). Moreover vd+1,k0 ∈ Wd+1,ik0 i=1(cid:16)si(cid:12)(cid:12)Wd+1,i(cid:17). and (cid:3) prd+1(Wd+1,ik0 ) = prd+1({vd+1,k0}) = [1 : ω : ω2] where ω is defined by Step 1 and Remark 4.2. Thus ǫ = ω. Now we have = sk0 · [a(ωxd − yd) + b(ω2xd − zd)](cid:12)(cid:12)(cid:12)[1:ω:ω2] = 0 for all i = 1, . . . , 6. Hence β(s) = 0. = 0. (cid:3) sik0(cid:12)(cid:12)Wd+1,ik0 Therefore si(cid:12)(cid:12)Wd+1,i Hence Steps 1-3 are complete which concludes the proof of Lemma 4.9. (cid:3) Consequently, we have verified Proposition 4.7. (cid:3) One of the main results why twisted homogeneous coordinate rings are so useful for studying Sklyanin algebras is that tcrs are factors of their corre- sponding Sklyanin algebra (by some homogeneous element; refer to Theorem 1.3). The following corollaries to Theorem 4.6 illustrate an analogous result for Sdeg. Corollary 4.10. Let B be the point parameter ring of a degenerate Sklyanin algebra Sdeg. Then B ∼= Sdeg/K for some ideal K of Sdeg that has six generators of degree 4 and possibly higher degree generators. Proof. By Theorem 4.6, Sdeg surjects onto B say with kernel K. By Remark 2.3 we have that dimk S4 = 57, yet we know dimk B4 = 63 by Proposition 4.3. Hence dimk K4 = 6. The same results also imply that dimk Sd = dimk Bd for d ≤ 3. (cid:3) Corollary 4.11. The ring B = B(Sdeg) is neither a domain or Koszul. DEGENERATE SKLYANIN ALGEBRAS 23 Proof. By Corollary 2.4, there exist linear nonzero elements u, v ∈ S with uv = 0. The image of u and v are nonzero, hence B is not a domain due to Corollary 4.10. Since B has degree 4 relations, it does not possess the Koszul property. (cid:3) References [1] M. Artin and W. Schelter, Graded algebras of global dimension 3, Adv. in Math. 66 (1987), 171 -- 216. [2] M. Artin and J. T. Stafford, Noncommutative graded domains with quadratic growth, Invent. Math. 122 (1995), 231 -- 276. [3] M. Artin, J. Tate, and M. van den Bergh, Some algebras associated to automor- phisms of elliptic curves, The Grothendieck Festschrift, vol. 1, Birkhauser (1990), 33 -- 85. [4] M. Artin, J. Tate, and M. van den Bergh, Modules over regular algebras of dimen- sion 3, Invent. Math. 106 (1991), 335 -- 389. [5] M. Artin and J. J. Zhang, Abstract Hilbert Schemes, Alg. Rep. Theory 4 (2001), 305 -- 394. [6] G. M. Bergman, The Diamond Lemma for Ring Theory, Adv. Math. 29 (1978), 178 -- 218. [7] J. E. Bjork and J. T. Stafford, email correspondence with M. Artin, January 31, 2000. [8] E. Bombieri and W. Gubler, Heights in Diophantine Geometry, Cambridge Uni- versity Press, 2006. [9] T. Cassidy and B. Shelton, Generalizing the Notion of Koszul Algebra, math.RA/0704.3752v1. [10] K. R. Goodearl and R. B. Warfield, Jr., An Introduction to Noncommutative Noe- therian Rings, London. Math. Soc. Student Texts vol. 61, Cambridge University Press, 2004. [11] R. Hartshorne, Algebraic Geometry, Graduate Text in Mathematics 52, Springer- Verlag, New York, 1977. [12] D. S. Keeler, D. Rogalski, and J. T. Stafford, Naıve Noncommutative Blowing Up, Duke Math. J. 126 (3) (2005), 491 -- 546. [13] U. Krahmer, Notes on Koszul Algebras, www.impan.gov.pl/ ∼ kraehmer/ connected.pdf. [14] A. Polishchuk and L. Positselski, Quadratic Algebras, Amer. Math. Soc. University Lecture Series 37, 2005. [15] D. Rogalski and J. T. Stafford, A Class of Noncommutative Projective Surfaces, arXiv:math/0612657v1. [16] D. Rogalski and J. J. Zhang, Canonical maps to twisted rings, Math. Z. 259 (2) (2008), 433 -- 455. [17] B. Shelton and C. Tingey, On Koszul algebras and a new construction of Artin Schelter regular algebras, J. Alg. 241 (2001), 789-798. [18] S. P. Smith and J. T. Stafford, Regularity of the 4-dimensional Sklyanin algebra, Compositio Math. 83 (1992), 259 -- 289. [19] J. T. Stafford, Math 715: Noncommutative Projective Algebraic Geometry (course notes), University of Michigan, Winter 2007. 24 CHELSEA WALTON [20] D. R. Stephenson and J. Zhang, Growth of Graded Noetherian Rings, Proc. Amer. Math. Soc. 125 (1997), 1593 -- 1605. [21] J. J. Zhang, Twisted Graded Algebras and Equivalences of Graded Categories, Proc. London Math Soc. (3) 72 (1996) 281 -- 311.
1112.5753	3	1112	2013-04-13T20:16:35	A construction of integer-valued polynomials with prescribed sets of lengths of factorizations	[ "math.RA" ]	For an arbitrary finite set S of natural numbers greater 1, we construct an integer-valued polynomial f, whose set of lengths in Int(Z) is S. The set of lengths of f is the set of all natural numbers n, such that f has a factorization as a product of n irreducibles in Int(Z)={g in Q[x] \| g(Z) contained in Z}.	math.RA	math	To appear in Monatsh. Math. A construction of integer-valued polynomials with prescribed sets of lengths of factorizations Sophie Frisch Abstract. For an arbitrary finite set S of natural numbers greater 1, we construct f ∈ Int(Z) = {g ∈ Q[x] g(Z) ⊆ Z} whose set of lengths is S . The set of lengths of f is the set of all n such that f has a factorization as a product of n irreducibles in Int(Z). MSC 2000: primary 13A05, secondary 13B25, 13F20, 20M13, 11C08. 1. Introduction Non-unique factorization has long been studied in rings of integers of number fields, see the monograph of Geroldinger and Halter-Koch [5]. More recently, non-unique factorization in rings of polynomials has attracted attention, for instance in Zpn [x], cf. [4], and in the ring of integer-valued polynomials Int(Z) = {g ∈ Q[x] g(Z) ⊆ Z} (and its generalizations) [1, 3]. We show that every finite set of natural numbers greater 1 occurs as the set of lengths of factorizations of an element of Int(Z) (Theorem 9 in section 4). Our proof is constructive, and allows multiplicities of lengths of factorizations to be specified. For example, given the multiset {2, 2, 2, 5, 5}, we construct a polynomial that has three different factorizations into 2 irreducibles and two different factorizations into 5 irreducibles, and no other factorizations. Perhaps a quick review of the vocabulary of factorizations is in order: Notation and Conventions. R denotes a commutative ring with identity. An element r ∈ R is called irreducible in R if r is a non-zero non-unit such that r = ab with a, b ∈ R implies that a or b is a unit. A factorization of r in R is an expression r = s1 . . . sn of r as a product of irreducible elements in R. The number n of irreducible factors is called the length of the factorization. The set of lengths L(r) of r ∈ R is the set of all natural numbers n such that r has a factorization of length n in R. 3 1 0 2 r p A 3 1 ] . A R h t a m [ 3 v 3 5 7 5 . 2 1 1 1 : v i X r a R is called atomic if every non-zero non-unit of R has a factorization in R. If R is atomic, then for every non-zero non-unit r ∈ R the elasticity of r is defined as ρ(r) = sup{ m, n ∈ L(r)} m n 1 and the elasticity of R is ρ(R) = supr∈R′ (ρ(r)), where R′ is the set of non-zero non-units of R. An atomic domain R is called fully elastic if every rational number greater than 1 occurs as ρ(r) for some non-zero non-unit r ∈ R. Two elements r, s ∈ R are called associated in R if there exists a unit u ∈ R such that r = us. Two factorizations of the same element r = r1 · . . . · rm = s1 · . . . · sn are called essentially the same if m = n and, after re-indexing the si, rj is associated to sj for 1 ≤ j ≤ m. Otherwise, the factorizations are called essentially different. 2. Review of factorization of integer-valued polynomials In this section we recall some elementary properties of Int(Z) and the fixed divisor d(f ), to be found in [1], [2] and [3]. The reader familiar with integer-valued polynomials is encouraged to skip to section 3. Definition. For f ∈ Z[x], (i) the content c(f ) is the ideal of Z generated by the coefficients of f , (ii) the fixed divisor d(f ) is the ideal of Z generated by the image f (Z). By abuse of notation we will identify the principal ideals c(f ) and d(f ) with their non-negative generators. Thus, for f = Pn k=0 akxk ∈ Z[x], c(f ) = gcd (ak k = 0, . . . , n) and d(f ) = gcd (f (c) c ∈ Z). A polynomial f ∈ Z[x] is called primitive if c(f ) = 1. Recall that a primitive polynomial f ∈ Z[x] is irreducible in Z[x] if and only if it is irreducible in Q[x]. Similarly, f ∈ Z[x] with d(f ) = 1 is irreducible in Z[x] if and only if it is irreducible in Int(Z). We denote p-adic valuation by vp . Almost everything that we need to know about the fixed divisor follows immediately from the fact that vp(d(f )) = min c∈Z (vp(f (c))). In particular, it is easy to deduce that for any f, g ∈ Z[x], d(f )d(g)(cid:12)(cid:12) d(f g). Unlike c(f ), which satisfies c(f )c(g) = c(f g), d(f ) is not multiplicative: d(f )d(g) is in general a proper divisor of d(f g). 2 Remark 1. (i) Every non-zero polynomial f ∈ Q[x] can be written in a unique way as f (x) = ag(x) b with g ∈ Z[x], c(g) = 1, a, b ∈ N, gcd(a, b) = 1. (ii) When expressed as in (i), f is in Int(Z) if and only if b divides d(g). (iii) For non-constant f ∈ Int(Z) expressed as in (i) to be irreducible in Int(Z) it is necessary that a = 1 and b = d(g). Proof. (i) and (ii) are easy. Ad (iii). Note that the only units in Int(Z) are ±1. By (ii), b divides d(g). Let d(g) = bc. Then f factors as a · c · (g/bc), where (g/bc) is non-constant and ac is a unit only if a = c = 1. (cid:3) Remark 2. (i) Every non-zero polynomial f ∈ Q[x] can be written in a unique way (up to the sign of a and the signs and indexing of the gi) as f (x) = a b Yi∈I gi(x), with gi primitive and irreducible in Z[x] for i ∈ I (a finite set) and a ∈ Z, b ∈ N with gcd(a, b) = 1. (ii) A non-constant polynomial f ∈ Int(Z) expressed as in (i) is irreducible in Int(Z) if and only if a = ±1, b = d(Qi∈I gi), and there do not exist ∅ 6= J $ I and b1, b2 ∈ N with b1b2 = b and b1 = d(Qi∈J gi), b2 = d(Qi∈I\J gi). (iii) Int(Z) is atomic. (iv) Every non-zero non-unit f ∈ Int(Z) has only finitely many factorizations into irreducibles in Int(Z). Proof. Ad (ii). If f is irreducible, the conditions on f follow from Remark 1 (ii) and (iii). Conversely, if the conditions hold, what chance does f have to be reducible? By Remark 1 (ii), we cannot factor out a non-unit constant, because no proper multiple of b divides d(Qi∈I gi). Any non-constant irreducible factor would, by Remark 1 (iii), be of the kind (Qi∈J gi)/b1 with b1 = d(Qi∈J gi), and its co-factor would be (Qi∈I\J gi)/b2 with b1b2 = b and b2 a divisor of d(Qi∈I\J gi). Also, b2 could not be a proper divisor of d(Qi∈I\J gi), because otherwise b1b2 = b would be a proper divisor of Qi∈I gi. So, the existence of a non-constant irreducible factor would imply the existence of J and b1, b2 of the kind we have excluded. 3 Ad (iii). With f (x) = ag(x)/b, g =Qi∈I gi as in (i), d(g) = cb for some c ∈ N, and f (x) = acg(x)/d(g) with g(x)/d(g) ∈ Int(Z). We can factor ac into irreducibles in Z, which are also irreducible in Int(Z). Either g(x)/d(g) is irreducible, or (ii) gives an expression as a product of two non-constant factors of smaller degree. By iteration we arrive at a factorization of g(x)/d(g) into irreducibles. Ad (iv). Let f ∈ Int(Z) = (ag(x)/b) with g = Qi∈I gi as in (i). Then all factorizations of f are of the form, for some c ∈ N such that bc divides d(g), f = a1 . . . anc1 . . . cm k Yj=1 Qi∈Ij dj gi , where a = a1 . . . an and c = c1 . . . cm are factorizations into primes in Z, I = gi). I1 ∪ . . . ∪ Ik is a partition of I into non-empty sets, d1 . . . dk = bc, dj = d(Qi∈Ij There are only finitely many such expressions. (cid:3) Remark 3. (i) The binomial polynomials n(cid:19) = (cid:18)x x(x − 1) . . . (x − n + 1) n! for n ≥ 0 are a basis of Int(Z) as a free Z-module. (ii) n!f ∈ Z[x] for every f ∈ Int(Z) of degree at most n. (iii) Let f ∈ Z[x] primitive, deg f = n and p prime. Then vp(d(f )) ≤ Xk≥1 (cid:20) n pk(cid:21) = vp(n!). In particular, if p divides d(f ) then p ≤ deg f . Proof. Ad (i). The binomial polynomials are in Int(Z) and they form a Q-basis of Q[x]. If a polynomial in Int(Z) is written as a Q-linear combination of binomial polynomials then an easy induction shows that the coefficients must be integers. (ii) follows from (i). Ad (iii). Let g = f /d(f ). Then g ∈ Int(Z) and d(f )Z = (Z[x] :Z g). Since n! ∈ (Z[x] :Z g) by (ii), d(f ) divides n!. (cid:3) 4 3. Useful Lemmata Lemma 4. Let p be a prime, I 6= ∅ a finite set and for i ∈ I , fi ∈ Z[x] primitive and irreducible in Z[x] such that d(Qi∈I fi) = p. Let g(x) = Qi∈I fi p . Then every factorization of g in Int(Z) is essentially the same as one of the following: g(x) = Qj∈J fj · Yi∈I\J where J ⊆ I is minimal such that d(Qi∈J fj) = p. p fi, Proof. Follows from Remark 1 (iii) and the fact that d(f )d(h) divides d(f h) for all f, h ∈ Z[x]. (cid:3) The following two easy lemmata are constructive, since the Euclidean algorithm makes the Chinese Remainder Theorem in Z effective. Lemma 5. For every prime p ∈ Z, we can construct a complete system of residues mod p that does not contain a complete system of residues modulo any other prime. Proof. By the Chinese Remainder Theorem we solve, for each k = 1, . . . , p the system of congruences sk = k mod p and sk = 1 mod q for every prime q < p. (cid:3) Lemma 6. Given finitely many non-constant monic polynomials fi ∈ Z[x], i ∈ I , we can construct monic irreducible polynomials Fi ∈ Z[x], pairwise non-associated in Q[x], with deg Fi = deg fi, and with the following property: Whenever we replace some of the fi by the corresponding Fi, setting gi = Fi for i ∈ J (J an arbitrary subset of I ) and gi = fi for i ∈ I \ J , then for all K ⊆ I , d(Yi∈K gi) = d(Yi∈K fi). Proof. Let n = Pi∈I deg fi. Let p1, . . . , ps be all the primes with pi ≤ n, and set αi = vpi(n!). Let q > n be a prime. For each i ∈ I , we find by the Chinese Remainder Theorem the coefficients of a polynomial ϕi ∈(Qs k )Z[x] of smaller degree than fi, such that Fi = fi + ϕi satisfies Eisenstein's irreducibility criterion k=1 pαk 5 with respect to the prime q. Then, with respect to some linear ordering of I , if Fi happens to be associated in Q[x] to any Fj of smaller index, we add a suitable to Fi, to make Fi non-associated in Q[x] k=1 pαk non-zero integer divisible by q2Qs to all Fj of smaller index. k The statement about the fixed divisor follows, because for every c ∈ Z and every prime pi that could conceivably divide the fixed divisor, Yi∈K (gi(c)) ≡ Yi∈K (fi(c)) mod pαi i , where pαi i polynomial of degree at most n. (cid:3) is the highest power of pi that can divide the fixed divisor of any monic 4. Constructing polynomials with prescribed sets of lengths We precede the general construction by two illustrative examples of special cases, corresponding to previous results by Cahen, Chabert, Chapman and McClain. Example 7. For every n ≥ 0, we can construct H ∈ Int(Z) such that H has exactly two essentially different factorizations in Int(Z), one of length 2 and one of length n + 2. Proof. Let p > n + 1, p prime. By Lemma 5 we construct a complete set a1, . . . , ap of residues mod p in Z that does not contain a complete set of residues mod any prime q < p. Let f (x) = (x − a2)(x − a3) . . . (x − ap) and g(x) = (x − an+2)(x − an+3) . . . (x − ap). By Lemma 6, we construct monic irreducible polynomials F, G ∈ Z[x], not as- sociated in Q[x], with deg F = deg f , deg G = deg g, such that any product of a selection of polynomials from (x − a1), . . . , (x − an+1), f (x), g(x) has the same fixed divisor as the corresponding product with f replaced by F and g by G. Let H(x) = F (x)(x − a1) . . . (x − an+1)G(x) p . By Lemma 4, H factors into two irreducible polynomials in Int(Z) H(x) = F (x) · (x − a1) . . . (x − an+1)G(x) p or into n + 2 irreducible polynomials in Int(Z) H(x) = F (x)(x − a1) p · (x − a2)(x − a3) . . . (x − an+1)G(x). (cid:3) 6 Corollary. (Cahen and Chabert [1]). ρ (Int(Z)) = ∞. Example 8. For 1 ≤ m ≤ n, we can construct a polynomial H ∈ Int(Z) that has in Int(Z) a factorization into m + 1 irreducibles and an essentially different factorization into n+1 irreducibles, and no other essentially different factorization. Proof. Let p > mn be prime, s = p − mn. By Lemma 5 we construct a complete system of residues R mod p that does not contain a complete system of residues for any prime q < p. We index R as follows: R = {r(i, j) 1 ≤ i ≤ m, 1 ≤ j ≤ n} ∪ {b1, . . . , bs}. Let b(x) = Qs 1 ≤ j ≤ n let gj(x) = Qm k=1(x − bk). For 1 ≤ i ≤ m let fi(x) = Qn k=1(x − r(k, j)). k=1(x − r(i, k)) and for By Lemma 6, we construct monic irreducible polynomials Fi, Gj ∈ Z[x], pair- wise non-associated in Q[x], such that the product of any selection of the poly- nomials (x − b1), . . . , (x − bs), f1, . . . , fm, g1, . . . , gn has the same fixed divisor as the corresponding product in which fi has been replaced by Fi and gj by Gj for 1 ≤ i ≤ m and 1 ≤ j ≤ n. Let H(x) = 1 p b(x) m Yi=1 Fi(x) n Yj=1 Gj (x), then, by Lemma 4, H has a factorization into m + 1 irreducibles H(x) = F1(x) · . . . · Fm(x) · b(x)G1(x) · . . . · Gn(x) p and an essentially different factorization into n + 1 irreducibles H(x) = b(x)F1(x) · . . . · Fm(x) p · G1(x) · . . . · Gn(x) and no other essentially different factorization. (cid:3) Corollary. (Chapman and McClain [3]). Int(Z) is fully elastic. Theorem 9. Given natural numbers 1 ≤ m1 ≤ . . . ≤ mn , we can construct a polynomial H ∈ Int(Z) that has exactly n essentially different factorizations into irreducibles in Int(Z), the lengths of these factorizations being m1 + 1, . . . , mn + 1. 7 i=1 mi)2 −Pn Proof. Let N = (Pn i , and p > N prime, s = p − N . By Lemma 5, we construct a complete system of residues R mod p that does not contain a complete system of residues for any prime q < p. We partition R into disjoint sets R = R0 ∪ {t1, . . . , ts} with R0 = N . The elements of R0 are indexed as follows: i=1 m2 R0 = {r(k, h, i, j) 1 ≤ k ≤ n, 1 ≤ h ≤ mk, 1 ≤ i ≤ n, 1 ≤ j ≤ mi; i 6= k}, meaning we arrange the elements of R0 in an m×m matrix with m = m1 +. . .+mn , whose rows and columns are partitioned into n blocks of sizes m1, . . . , mn. Now r(k, h, i, j) designates the entry in the h-th row of the k-th block of rows and the j -th column of the i-th block of columns. Positions in the matrix whose row and column are each in block i are left empty: there are no elements r(k, h, i, j) with i = k. For 1 ≤ k ≤ n, 1 ≤ h ≤ mk , let Sk,h be the set of entries in the (k, h)-th row: Sk,h = {r(k, h, i, j) 1 ≤ i ≤ n, i 6= k, 1 ≤ j ≤ mi}. For 1 ≤ i ≤ n, 1 ≤ j ≤ mi, let Ti,j be the set of elements in the (i, j)-th column: Ti,j = {r(k, h, i, j) 1 ≤ k ≤ n, k 6= i, 1 ≤ h ≤ mk}. For 1 ≤ k ≤ n, 1 ≤ h ≤ mk , set f (k) h (x) = Yr∈Sk,h (x − r) · Yr∈Tk,h (x − r). Also, let b(x) = Qs i=1(x − ti). By Lemma 6, we construct monic irreducible polynomials F (k) associated in Q[x], with deg F (k) of polynomials from (x − t1), . . . , (x − ts) and f (k) the same fixed divisor as the corresponding product in which the f (k) replaced by the F (k) h , pairwise non- h , such that any product of a selection for 1 ≤ k ≤ n, 1 ≤ h ≤ mk has have been h = deg f (k) h h h . Let H(x) = 1 p b(x) n mk Yk=1 Yh=1 F (k) h (x). Then deg H = N + p; and for each i = 1, . . . , n, H has a factorization into mi + 1 irreducible polynomials in Int(Z): H(x) = F (i) 1 (x) · . . . · F (i) mi (x) · b(x)Qk6=iQmk h=1 F (k) p h (x) 8 These factorizations are essentially different, since the F (i) associated in Q[x] and hence in Int(Z). j are pairwise non- By Lemma 4, H has no further essentially different factorizations. This is so because a minimal subset with fixed divisor p of the polynomials (x − ti) for 1 ≤ i ≤ s and F (k) for 1 ≤ k ≤ n, 1 ≤ h ≤ mk must consist of all the linear factors (x − ti) together with a minimal selection of F (k) such that all r ∈ R0 occur as roots in the product of the corresponding f (k) h . For all linear factors (x − r) with r ∈ R0 to occur in a set of polynomials f (k) h , it must contain for all but one k all f (k) h , h = 1, . . . mk . If, for i 6= k, f (k) are missing, then r(k, h, i, j) and r(i, j, k, h) do not occur among the roots of the polynomials f (k) h . A set consisting of all f (k) for n − 1 different values of k, however, has the property that all linear factors (x − r) for r ∈ R0 occur. (cid:3) and f (i) h h j h h Corollary. Every finite subset of N\{1} occurs as the set of lengths of a polynomial f ∈ Int(Z). 5. No transfer homomorphism to a block-monoid For some monoids, results like the above Corollary have been shown by means of transfer-homomorphisms to block monoids. For instance, by Kainrath [6], in the case of a Krull monoid with infinite class group such that every divisor class contains a prime divisor. Int(Z), however, doesn't admit this method: We will show a property of the multiplicative monoid of Int(Z) \ {0} that excludes the existence of a transfer- homomorphism to a block monoid. Theorem 10. For every n ≥ 1 there exist irreducible elements H, G1, . . . , Gn+1 in Int(Z) such that xH(x) = G1(x) . . . Gn+1(x). Proof. Let p1 < p2 < . . . < pn be n distinct odd primes, P = {p1, p2, . . . , pn}, and Q the set of all primes q ≤ pn + n. By the Chinese remainder theorem construct a1, . . . , an with ai ≡ 0 mod pi and ai ≡ 1 mod q for all q ∈ Q with q 6= pi. Similarly, construct b1, . . . bpn such that, firstly, for all p ∈ P , bk ≡ k mod p if k ≤ p and bk ≡ 1 mod p if k > p and, secondly, bk ≡ 1 mod q for all q ∈ Q \ P . So, for each pi ∈ P , a complete set of residues mod pi is given by b1, . . . bpi, ai, while all remaining aj and bk are congruent to 1 mod pi. Also, all aj and bk are congruent to 1 for all primes in Q \ P . Set f (x) = (x − b1) . . . (x − bpn ) and let F (x) be a monic irreducible polynomial in Z[x] with deg F = deg f such that the fixed divisor of any product of a selection 9 of polynomials from f (x), (x − a1), . . . , (x − an) is the same as the fixed divisor of the corresponding set of polynomials in which f has been replaced by F . Such an F exists by Lemma 6. Let H(x) = F (x)(x − a1) . . . (x − an) p1 . . . pn . Then H(x) is irreducible in Int(Z), and xH(x) = xF (x) p1 . . . pn · (x − a1) · . . . · (x − an), where xF (x)/(p1 . . . pn) and, of course, (x − a1), . . ., (x − an), are irreducible in Int(Z). (cid:3) Thanks to Alfred Geroldinger for pointing this out: Theorem 10 Remark. implies that there does not exist a transfer-homomorphism from the multiplicative monoid (Int(Z) \ {0}, ·) to a block-monoid. (For the definition of block-monoid and transfer-homomorphism see [5] Def. 2.5.5 and Def. 3.2.1, respectively.) This is so because, in a block-monoid, the length of factorizations of elements of the form cd with c, d irreducible, c fixed, is bounded by a constant depending only on c, cf. [5], Lemma 6.4.4. More generally, applying [5], Lemma 3.2.2, one sees that every monoid that admits a transfer-homomorphism to a block-monoid has this property, in marked contrast to Theorem 10. References [1] P.-J. Cahen and J.-L. Chabert, Elasticity for integral-valued polynomials, J. Pure Appl. Algebra 103 (1995), 303–311. [2] P.-J. Cahen and J.-L. Chabert, Integer-valued polynomials, vol. 48 of Mathematical Surveys and Monographs, Amer. Math. Soc., 1997. [3] S. T. Chapman and B. A. McClain, Irreducible polynomials and full elasticity in rings of integer-valued polynomials, J. Algebra 293 (2005), 595–610. [4] Ch. Frei and S. Frisch, Non-unique factorization of polynomials over residue class rings of the integers, Comm. Algebra 39 (2011), 1482–1490. [5] A. Geroldinger and F. Halter-Koch, Non-unique factorizations, vol. 278 of Pure and Appl. Math., Chapman & Hall/CRC, Boca Raton, FL, 2006. [6] F. Kainrath, Factorization in Krull monoids with infinite class group, Colloq. Math. 80 (1999), 23–30. 10 Institut fur Mathematik A Technische Universitat Graz Steyrergasse 30 A-8010 Graz, Austria [email protected] 11
1711.09995	7	1711	2019-11-07T03:12:00	Quiver mutations and Boolean reflection monoids	[ "math.RA", "math.CO", "math.GR" ]	In 2010, Everitt and Fountain introduced the concept of reflection monoids. The Boolean reflection monoids form a family of reflection monoids (symmetric inverse semigroups are Boolean reflection monoids of type $A$). In this paper, we give a family of presentations of Boolean reflection monoids and show how these presentations are compatible with quiver mutations of orientations of Dynkin diagrams with frozen vertices. Our results recover the presentations of Boolean reflection monoids given by Everitt and Fountain and the presentations of symmetric inverse semigroups given by Popova respectively. Surprisingly, inner by diagram automorphisms of irreducible Weyl groups and Boolean reflection monoids can be constructed by sequences of mutations preserving the same underlying diagrams. Besides, we show that semigroup algebras of Boolean reflection monoids are cellular algebras.	math.RA	math	QUIVER MUTATIONS AND BOOLEAN REFLECTION MONOIDS BING DUAN, JIAN-RONG LI, AND YAN-FENG LUO Abstract. In 2010, Everitt and Fountain introduced the concept of reflection monoids. The Boolean reflection monoids form a family of reflection monoids (symmetric inverse semigroups are Boolean reflection monoids of type A). In this paper, we give a fam- ily of presentations of Boolean reflection monoids and show how these presentations are compatible with quiver mutations of orientations of Dynkin diagrams with frozen vertices. Our results recover the presentations of Boolean reflection monoids given by Everitt and Fountain and the presentations of symmetric inverse semigroups given by Popova respectively. Surprisingly, inner by diagram automorphisms of irreducible Weyl groups and Boolean reflection monoids can be constructed by sequences of mu- tations preserving the same underlying diagrams. Besides, we show that semigroup algebras of Boolean reflection monoids are cellular algebras. Key words: Boolean reflection monoids; presentations; mutations of quivers; inner by diagram automorphisms; cellular semigroups 2010 Mathematics Subject Classification: 13F60; 20M18 9 1 0 2 v o N 7 ] . A R h t a m [ 7 v 5 9 9 9 0 . 1 1 7 1 : v i X r a 1. Introduction Around 2000, Fomin and Zelevinsky introduced a new family of commutative algebras called cluster algebras, [19]. The theory of cluster algebras has connections and appli- cations to diverse areas of mathematics and physics, including quiver representations, Teichmuller theory, tropical geometry, integrable systems, and Poisson geometry. Barot and Marsh applied the idea in cluster algebras to give new presentations of finite irreducible crystallographic reflection groups, [2]. Quiver mutations play an important role in the theory of cluster algebras and the work of Barot and Marsh [2]. Finite type cluster algebras are classified by finite type Dynkin diagrams, [20]. The quivers appearing in the work of Barot and Marsh [2] are quivers of finite type (i.e., these quivers are mutation equivalent to orientations of finite type Dynkin diagrams). Let Γ be a quiver whose underlying graph is a Dynkin diagram. In [2], Barot and Marsh gave new presentations of the reflection group WΓ determined by Γ and showed that these presentations are compatible with mutations of Γ quivers. More precisely, Barot and Marsh introduced some additional relations (cycle relations) corresponding to chordless cycles arising in quivers of finite type. For any quiver Q of finite type, they defined an abstract group W (Q) by generators (corresponding to vertices of Q) and relations and then proved that W (Q) ∼= WΓ. 1 2 BING DUAN, JIAN-RONG LI, AND YAN-FENG LUO Motivated by Barot and Marsh's work, new presentations of affine Coxeter groups, braid groups, Artin groups, and Weyl groups of Kac-Moody algebras have been studied using quiver mutations in [16, 17, 24, 28, 37], respectively. One of the aims in this paper is to give new presentations of Boolean reflection monoids and show that these presentations are compatible with mutation of certain quivers. In [10], Everitt and Fountain introduced the concept of reflection monoids. The Boolean reflection monoids are reflection monoids. Symmetric inverse semigroups are Boolean reflection monoids of type A. In [11], Everitt and Fountain obtained presenta- tions of the Boolean reflection monoids of type A, B/C, D. Let Aε n−1 (respectively, Bε n) be the Dynkin diagram with n (respectively, n + 1, n + 1) vertices, among them the first n − 1 (respectively, n, n) vertices are mutable vertices and the last vertex ε is a frozen vertex, which is shown in the 4-th column of Table 1. We label an edge if its weight is greater or equal to 2. n, Dε n, Dε n−1, Bε Let ∆ ∈ {Aε n} and Q any quiver that is mutation equivalent to a quiver whose underlying graph is ∆. We define an inverse monoid M (Q) from Q, see Sec- tion 4.3, and show that M (Q) ∼= M (Φ,B), where M (Φ,B) is the Boolean reflection monoid defined in [10], see Theorem 4.7 and Theorem 4.9. This implies that Boolean reflection monoids can also be classified by ∆, see Table 1. In [2, 16, 24, 28], the dia- grams corresponding to presentations of irreducible Weyl groups, affine Coxeter groups, braid groups, Artin groups have no frozen vertices. In the present paper, the diagrams corresponding to presentations of Boolean reflection monoids have frozen vertices. Type of Φ Boolean reflection monoids Generators ∆ = Φε An−1(n ≥ 2) M (An−1,B) {s1, . . . , sn−1, sε} Bn(n ≥ 2) M (Bn,B) {s0, . . . , sn−1, sε} Dn(n ≥ 4) M (Dn,B) {s0, . . . , sn−1, sε} 2 2 1 2 1 0 1 0 n − 1 n − 1 n − 1 ε ε ε Table 1. Boolean reflection monoids and Dynkin diagrams Aε n−1, Bε n, Dε n. In Proposition 3.1 of [10], Everitt and Fountain proved that the symmetric inverse semigroup In is isomorphic to the Boolean reflection monoid of type An−1. We recover the presentation of the symmetric inverse semigroup In defined in [9, 32]. The presenta- tion corresponds exactly to the presentation determined by Aε n−1. Moreover, we recover Everitt and Fountain's presentations of Boolean reflection monoids defined in Section 3 of [11], see Theorem 4.9. These presentations can be obtained from any ∆ quiver by a finite sequence of mutations. We show in Theorem 4.10 that the inner automorphism group of the Boolean reflection monoid M (Φ,B) is naturally isomorphic to W (Φ)/Z(W (Φ)). Surprisingly, inner by QUIVER MUTATIONS AND BOOLEAN REFLECTION MONOIDS 3 diagram automorphisms of irreducible Weyl groups and Boolean reflection monoids can be constructed by a sequence of mutations preserving the same underlying diagrams, see Theorem 3.5 and Theorem 4.12 respectively. Finally we study the cellularity of the semigroup algebras of Boolean reflection monoids. It is well known that Hecke algebras of finite type, q-Schur algebras, the Brauer alge- bra, the Temperley-Lieb algebras and partition algebras are cellular, see [21 -- 23, 41, 42]. Recently, the cellularity of semigroup algebras is investigated by East [7], Wilox [40], Guo and Xi [25], and Ji and Luo [29] respectively. Applying Geck's and East's results, we show that semigroup algebras of Boolean reflection monoids are cellular algebras, see Proposition 4.13. Moreover, we describe their cellular bases using the presentations we obtained. Automorphisms of finite symmetric inverse semigroups are studied in [38]. Coxeter arrangement monoids are defined in [10, 11] and studied in [10, 11, 13, 14]. Braid inverse monoids are defined in [12] and studied in [8, 12]. It would be interesting to study automorphisms of Boolean reflection monoids, presentations of Coxeter arrangement monoids and braid inverse monoids using the method in this paper. We plan to study these in future publications. The paper is organized as follows. In Section 2, we recall some notions and back- ground knowledge that will be used later. In Section 3, we recall Barot and Marsh's work about presentations of irreducible Weyl groups and then study inner by diagram automorphisms of irreducible Weyl groups (Theorem 3.5). In Section 4, we state our main results, Theorem 4.7 and Theorem 4.9, which show that presentations of Boolean reflection monoids are compatible with quiver mutations. In Section 5, we describe mu- tations of ∆ quivers and the oriented cycles appearing in the mutations. In Section 6, we define the inverse monoid M (Q). In Section 7, we prove Theorem 4.7. 2. Preliminaries 2.1. Mutation of quivers. Let Q be a quiver with finitely many vertices and finitely many arrows that have no loops or oriented 2-cycles. Given a quiver Q, let I be the set of its vertices and Qop its opposite quiver with the same set of vertices but with the reversed orientation for all the arrows. We consider a quiver with weight associated to a skew-symmetrizable matrix, see [20]. For each mutable vertex k of Q, one can define a mutation of Q at k, [3, 20]. This produces a new quiver denoted by µk(Q) which can be obtained from Q in the following way [20]: (i) The orientations of all edges incident to k are reversed and their weights intact. (ii) For any vertices i and j which are connected in Q via a two-edge oriented path going through k, the quiver mutation µk affects the edge connecting i and j in the way shown in Figure 1, where the weights c and c′ are related by ±√c ± √c′ = √ab, where the sign before √c (resp., √c′) is "+" if i, j, k form an oriented cycle in Q (resp., in µk(Q)), and is "−" otherwise. Here either c or c′ may be equal to 0, which means no arrows between i and j. 4 BING DUAN, JIAN-RONG LI, AND YAN-FENG LUO a @✂✂✂✂✂✂✂✂ i k c b ❂❂❂❂❂❂❂❂ µk←→ j i a ✂✂✂✂✂✂✂✂ k c′ b ^❂❂❂❂❂❂❂❂ j Figure 1. Quiver mutation (iii) The rest of the edges and their weights in Q remain unchanged. Two quivers Q1 and Q2 are said to be mutation equivalent if there exists a finite sequence of mutations taking one to the other. We write Q1 ∼mut Q2 to indicate that Q1 is mutation equivalent to Q2 and U (Q) for the mutation class of a quiver Q. The underlying diagram of a quiver Q is a undirected diagram obtained from Q by forgetting the orientation of all the arrows. Dynkin quivers are quivers whose underlying diagrams are Dynkin diagrams of finite type. A quiver Q is called connected if its underlying diagram is connected (every node is "reachable"). It was shown in [20, Theorem 1.4] that there are only finitely many quivers in the mutation classes of Dynkin quivers. A cycle in the underlying diagram of a quiver is called a chordless cycle if no two vertices of the cycle are connected by an edge that does not belong to itself, see [2]. As shown in [20, Proposition 9.7] (or see [2, Proposition 2.1]), all chordless cycles are oriented in the mutation classes of Dynkin quivers. 2.2. Cellular algebras and cellular semigroups. Let us first recall the basic defini- tion of cellular algebras introduced by Graham and Lehrer [23]. Let R be a commutative ring with identity. Definition 2.1. An associative R-algebra A is called a cellular algebra with cell datum (Λ, M, C, i) if the following conditions are satisfied: M (λ) of indices and there exists an R-basis {C λ (C1) Λ is a finite partially ordered set. Associated with each λ ∈ Λ there is a finite set S,T λ ∈ Λ; S, T ∈ M (λ)} of A; (C2) i is an R-linear anti-automorphism of A with i2 = i, which sends C λ S,T to C λ T,S; (C3) For each λ ∈ Λ, S, T ∈ M (λ), and each a ∈ A, S ′,T (mod A(< λ)), aC λ ra(S′, S)C λ S,T ≡ XS ′∈M (λ) S ′′,T ′′ µ < λ, S′′, T ′′ ∈ M (µ)}. where ra(S′, S) ∈ R is independent of T and A(< λ) is the R-submodule of A generated by {C µ Cellular algebras provide a general framework for studying the representation theory of many important classes of algebras including Hecke algebras of finite type, q-Schur algebras, the Brauer algebra, the Temperley-Lieb algebras and partition algebras, see [21 -- 23, 41, 42]. Recently, the cellularity of semigroup algebras is investigated by East [7], Wilox [40], Guo and Xi [25], and Ji and Luo [29] respectively. In the following, we recall some basic notions and facts from the theory of semigroups. Let S be a semigroup. For any a, b ∈ S, define a R b ⇔ aS1 = bS1, a J b ⇔ S1aS1 = S1bS1, a L b ⇔ S1a = S1b, @ ^ QUIVER MUTATIONS AND BOOLEAN REFLECTION MONOIDS 5 H = L∩R and D = L∨R = L◦R = R◦L, where S1 is the monoid obtained from S by adding an identity if necessary. A semigroup S is said to be inverse if for each element s ∈ S, there exists a unique inverse s−1 ∈ S such that ss−1s = s and s−1ss−1 = s−1. If S is a finite inverse semigroup, then D = J . For a in a semigroup S, denote by Ha (respectively, La, Ra, Da, Ja) the H (respectively, L, R, D, J )-class of a. For any s, t ∈ S, define Ds ≤ Dt if and only if s ∈ S1tS1. Let S be an inverse semigroup with the set E(S) of idempotents. Suppose that e1, . . . , ek ∈ D ∩ E(S). Choose a1 = e1, . . . , ak ∈ Le1 such that aj R ej for each j. Then D = Re1 ∪ Ra2 ∪ ··· ∪ Rak . Put HD = He1 and by Green's Lemma, for each x ∈ D we have x = aiga−1 for unique 1 ≤ i, j ≤ k and g ∈ HD. Using East's symbol in [7], let [ei, ej, g]D be the element x. Then x−1 = [ej , ei, g−1]D. For more detailed knowledge about semigroups, the reader is referred to [7, 27]. j A semigroup S is said to be cellular if its semigroup algebra R[S] is a cellular algebra. In [7], East proved the following theorem: Theorem 2.2 ([7, Theorem 15]). Let S be a finite inverse semigroup with the set E(S) of idempotents. If S satisfies the following conditions: (1) For each D-class D, the subgroup HD is cellular with cell datum (ΛD, MD, CD, iD); (2) The map i : R[S] → R[S] sending [e, f, g]D to [f, e, iD(g)]D is an R-linear anti- homomorphism. Then S is a cellular semigroup with cell datum (Λ, M, C, i), where Λ = {(D, λ) D ∈ S/D, λ ∈ ΛD} with partial order defined by (D, λ) ≤ (D′, λ′) if D < D′ or D = D′ and λ ≤ λ′, M (D, λ) = {(e, s) e ∈ E(S) ∩ D, s ∈ MD(λ)} for (D, λ) ∈ Λ, and C = {C (D,λ) s,t]D (D, λ) ∈ Λ; (e, s), (f, t) ∈ M (D, λ)} for (D, λ) ∈ Λ and (e, s), (f, t) ∈ M (D, λ). (e,s),(f,t) = [e, f, C λ By the definition of cellular algebras, we have the following corollary. Corollary 2.3. Suppose that A is a cellular algebra over R with cell datum (Λ, M, C, i) S,T ) for any λ ∈ Λ, (S, T ) ∈ and ϕ is an R-linear automorphism of A. Let C M (λ) × M (λ), and i = ϕiϕ−1. Then (Λ, M, C, i) is a cellular datum of A. Proof. Since {C λ linear automorphism of A, we have the fact that {C 2 is also an R-basis of A. It follows from the definition of i that i S,T λ ∈ Λ, (S, T ) ∈ M (λ) × M (λ)} is an R-basis of A and ϕ is an R- S,T λ ∈ Λ, (S, T ) ∈ M (λ) × M (λ)} λ T,S. λ S,T = ϕ(C λ λ = i and i(C λ S,T ) = C For each λ ∈ Λ, S, T ∈ M (λ), and each a ∈ A, aC λ S,T ≡ XS ′∈M (λ) ra(S′, S)C λ S ′,T (mod A(< λ)), where ra(S′, S) ∈ R is independent of T and A(< λ) is the R-submodule of A generated by {C µ S ′′,T ′′ µ < λ, S′′, T ′′ ∈ M (µ)}. For each a ∈ A, there exists a b ∈ A such that 6 BING DUAN, JIAN-RONG LI, AND YAN-FENG LUO a = ϕ(b). Then aC λ S,T = ϕ(b)ϕ(C λ S,T ) = ϕ(bC λ S,T ) ≡ XS ′∈M (λ) ≡ XS ′∈M (λ) rb(S′, S)ϕ(C λ S ′,T ) (mod A(< λ)) rb(S′, S)C λ S ′,T (mod A(< λ)), where rb(S′, S) ∈ R is independent of T . Therefore (Λ, M, C, i) is a cellular basis of A, as required. (cid:3) 3. Some new results of irreducible Weyl groups In this section, we recall Barot and Marsh's work about presentations of irreducible Weyl groups and then study inner by diagram automorphisms of irreducible Weyl groups. We refer the reader to [4, 18, 26, 33] for more information about Weyl groups, root sys- tems, and reflection groups. 3.1. Barot and Marsh's results. It is well known that finite irreducible crystallo- graphic reflection groups or irreducible Weyl groups are classified by Dynkin diagrams, see [26]. Let Γ be a Dynkin diagram and I the set of its vertices. Let WΓ be the finite irreducible Weyl group determined by Γ. A quiver is called a Γ quiver if its underlying diagram is Γ. In [2], Barot and Marsh gave presentations of WΓ. Their construction works as follows. Let Q be a quiver that is mutation equivalent to a Γ quiver. For two vertices i, j of Q, one defines 2 i and j are not connected, 3 i and j are connected by an edge with weight 1, 4 i and j are connected by an edge with weight 2, 6 i and j are connected by an edge with weight 3. (3.1) mij =  Definition 3.1. Let W (Q) be the group with generators si, i ∈ I, subjecting to the following relations: i = e for all i; (1) s2 (2) (sisj)mij = e for all i 6= j; (3) For any chordless cycle C in Q: i0 ω1−→ i1 ω2−→ ··· ωd−1−→ id−1 ω0−→ i0, where either all of the weights are 1, or ω0 = 2, we have: (si0si1 ··· sid−2sid−1sid−2 ··· si1)2 = e; where e is the identity element of W (Q). One of Barot and Marsh's main results in [2] is stated as follows. Theorem 3.2 ([2, Theorem A]). The group W (Q) does not depend on the choice of a quiver in U (Q). In particular, W (Q) ∼= WΓ for each quiver Q mutation equivalent to a Γ quiver. QUIVER MUTATIONS AND BOOLEAN REFLECTION MONOIDS 7 3.2. Inner by diagram automorphisms of irreducible Weyl groups. Let W be a Coxeter group defined by a set S of generators and relations (1) and (2) of Definition 3.1. The pair (W, S) is called a Coxeter system of W . In what follows, given any two Coxeter systems (W, S1) and (W, S2), when we say that there exists an automorphism of W , we mean that there is an automorphism α ∈ Aut(W ) such that α(S1) = S2. If such an automorphism can always be chosen from Inn(W ), the group of inner automorphisms of W , then W is called strongly rigid. In case W is strongly rigid, the group Aut(W ) has a very simple structure (see Corollary 3.2 of [6]): Aut(W ) = Inn(W ) × Diag(W ), where Diag(W ) consists of diagram automorphisms of the unique Coxeter diagram cor- responding to W . The following lemma is well known. Lemma 3.3. Let W be a finite group generated by a finite set S of simple reflections. Then the set of all reflections in W is {wsw−1 w ∈ W, s ∈ S}. In [1, Table I], Bannai computed the center Z(W (Φ)) of an irreducible Weyl group W (Φ). The longest element w0 in W (Φ) is a central element of W (Φ) except for Φ = An(n ≥ 2), D2k+1(k ≥ 2), E6. The following important notion was introduced by Franzsen in [15]. Definition 3.4 ([15, Definition 1.36]). An inner by diagram automorphism is an auto- morphism generated by some inner and diagram automorphisms in Aut(W ). The sub- group of inner automorphisms is a normal subgroup of Aut(W ), therefore any inner by diagram automorphism can be written as the product of an inner and a diagram auto- morphism. By Proposition 1.44 of [15], we know that all automorphisms of irreducible Weyl groups that preserve reflections are inner by diagram automorphisms. The following theorem reveals a connection between inner by diagram automorphisms of irreducible Weyl groups and quiver mutations. Theorem 3.5. Let Q be a Γ quiver and W (Q) the corresponding Weyl group generated by a set S of simple reflections. Then α is an inner by diagram automorphism of W (Q) if and only if there exists a sequence of mutations preserving the underlying diagram Γ such that α(S) can be obtained from Q by mutations. In particular, all reflections in W (Q) can be obtained from Q by mutations. Proof. By observation, every variable obtained by mutations must be some reflection of the corresponding Weyl group. The sufficiency follows from the fact that all automor- phisms of Weyl groups that preserve reflections are inner by diagram automorphisms. To prove necessity, assume without loss of generality that the vertex set of Q is {1, 2, . . . , n} and α is an inner by diagram automorphism of W (Q). Note that diagram automorphisms of any Dynkin diagram keep the underlying Dynkin diagram. Then relabelling the vertices of Q if necessary, there exists an inner automorphism α of W (Q) such that α(S) = α(S). It is sufficient to prove that α(S) can be obtained from Q by mutations and the sequence of mutations preserves the underlying diagram Γ. 8 BING DUAN, JIAN-RONG LI, AND YAN-FENG LUO Let g = si1si2 ··· sir ∈ W (Q) be a reduced expression for g, where sik ∈ S, 1 ≤ k ≤ r. We assume that α(S) = gSg−1 = {gsg−1 s ∈ S}. In the following we use induction to construct a sequence of mutations preserving the underlying diagram Γ such that gSg−1 is obtained from Q by mutations. Step 1. We mutate firstly Q at the vertex i1 twice. Then we get a quiver Q1, which has the same underlying diagram with Q. Moreover, the set S becomes si1Ssi1 = {si1s1si1, si1s2si1, . . . , si1sn−1si1, si1snsi1}. We keep vertices of Q1 having the same label as vertices of Q. Step 2. We then mutate Qt−1, t = 2, 3,··· , r at the vertex it twice. Note that the variable corresponding to the vertex it of Qt−1 is si1 . . . sit−1sitsit−1 . . . si1. Then we get a quiver Qt, which has the same underlying diagram with Q, Q1, . . . , Qt−1. Moreover, the set si1 . . . sit−1Ssit−1 . . . si1 of generators in Qt−1 becomes si1 ··· sit−1sitsit−1 ··· si1(si1 ··· sit−1Ssit−1 ··· si1)si1 ··· sit−1sitsit−1 ··· si1 = si1 ··· sit−1sitSsitsit−1 ··· si1. We keep vertices of Qt having the same label as vertices of Qt−1. Step 3. We repeat Step 2 until we get the quiver Qr. By induction, it is not difficult to show that the set of generators in Qr is si1si2 ··· sir Ssir ··· si2si1 = gSg−1 = α(S). Finally, every reflection in W (Q) is conjugate to a simple reflection by Lemma 3.3. So we assume that every reflection is of the form gsig−1, where g = si1si2 ··· sik is a reduced expression for g in W (Q). By the same arguments as before, we mutate the sequence i1, i1, i2, i2, . . . , ik, ik starting from Q, we obtain the reflection gsig−1. (cid:3) Remark 3.6. (1) In types An, Bn(n > 2), D2k+1, E6, E7, and E8, all inner by dia- gram automorphisms of the corresponding Weyl groups are inner automorphisms. W (Φ) is strongly rigid for Φ = An(n 6= 5), D2k+1, E6, E7. α of W , we have that (W, α(S)) is also a Coxeter system of W . (2) If (W, S) is a Coxeter system of W , then for any inner by diagram automorphism (3) Suppose that Q is a quiver that is mutation equivalent to a Γ quiver. Let W (Q) be the corresponding Weyl group with a set S of generators. Then Theorem 3.5 holds for Q. 4. Main results on Boolean reflection monoids Let Aε n−1 (respectively, Bε n) be the Dynkin diagram with n (respectively, n + 1, n + 1) vertices, among them the first n − 1 (respectively, n, n) vertices are mutable vertices and the last vertex ε is a frozen vertex, as is shown in Table 1. The label is left on an edge only if its weight is greater or equal to 2, otherwise left unlabelled. n, Dε In this section, we assume that ∆ is one of Aε n. A quiver is said to be a ∆ quiver if the underlying diagram of such quiver is ∆. It follows from the connectivity and finiteness of ∆ that any two ∆ quivers are mutation equivalent. Let U (∆) be the mutation class of any ∆ quiver. n and Dε n−1, Bε QUIVER MUTATIONS AND BOOLEAN REFLECTION MONOIDS 9 4.1. Boolean reflection monoids. In 2010, Everitt and Fountain introduced reflection monoids, see [10]. The Boolean reflection monoids are a family of reflection monoids (symmetric inverse semigroups are Boolean reflection monoids of type A). Let V be a Euclidean space with standard orthonormal basis {v1, v2, . . . , vn} and let Φ ⊆ V be a root system and W (Φ) the associated Weyl group of type Φ. A partial linear isomorphism of V is an isomorphism between two subspaces of V . Any partial linear isomorphism of V can be realized by restricting an isomorphism of V to some subspace. Let g, h be two isomorphisms of V and Y, Z be two subspaces of V . We denote by gY the partial isomorphism whose domain is Y and whose range is the restriction of g to Y . We denote by M (V ) (respectively, GL(V )) the general linear monoid (respectively, general linear group) on V consisting of partial linear isomorphisms (respectively, linear In M (V ), we have gY = hZ if and only if Y = Z and gh−1 is isomorphisms) of V . in the isotropy group GY = {g ∈ GL(V ) gv = v for all v ∈ Y }. Moreover, gY hZ = (gh)Z∩h−1Y and (gY )−1 = (g−1)gY . Let us recall the notion "system" in V for a group G ⊆ GL(V ) introduced in [10]. Definition 4.1 ([10, Definition 2.1]). Let V be a real vector space and G ⊆ GL(V ) a group. A collection S of subspaces of V is called a system in V for G if and only if (1) V ∈ S, (2) GS = S, that is, gX ∈ S for any g ∈ G and X ∈ S, (3) if X, Y ∈ S, then X ∩ Y ∈ S. For J ⊆ X = {1, 2, . . . , n}, let X(J) = Lj∈J Rvj ⊆ V , and B = {X(J) : J ⊆ X} with X(∅) = 0. Then B is a Boolean system in V for W (Φ), where Φ = An−1, Bn/Cn, or Dn. For example, the Weyl group W (An−1)-action on the subspaces X(J) ∈ B is just g(π)X(J) = X(πJ), where π 7→ g(π) induces an isomorphism from the symmetric group Sn on the set X to W (An−1), πJ = {π(j) j ∈ J}. Note that B is not a system for any of the exceptional W (Φ). Definition 4.2 ([10, Definition 2.2]). Let G ⊆ GL(V ) be a group and S a system in V for G. The monoid of partial linear isomorphisms given by G and S is the submonoid of M (V ) defined by M (G,S) := {gX : g ∈ G, X ∈ S}. If G is a reflection group, then M (G,S) is called a reflection monoid. Let G be the reflection group W (Φ) for Φ = An−1, Bn/Cn, or Dn, and S the Boolean system in V for G, then M (G,S) is called a Boolean reflection monoid. In general, we write M (Φ,B) instead of M (W (Φ),B), and call M (Φ,B) the Boolean reflection monoid of type Φ. We recall Everitt and Fountain's presentations in Section 4 of [11]: 10 BING DUAN, JIAN-RONG LI, AND YAN-FENG LUO Lemma 4.3. Everitt and Fountain's presentations of Boolean reflection monoids are as follows: M (An−1,B) =hs1, s2, . . . , sn−1, sε (sisj)mij = e for 1 ≤ i, j ≤ n − 1, s2 ε = sε, sisε = sεsi for i 6= 1, sεs1sεs1 = s1sεs1sε = sεs1sεi , M (Bn,B) =hs0, s1, . . . , sn−1, sε (sisj)mij = e for 0 ≤ i, j ≤ n − 1, s2 ε = sε, s0s1sεs1 = s1sεs1s0, s0sε = sε, sisε = sεsi for i 6= 1, s1sεs1sε = sεs1sεs1 = sεs1sεi, M (Dn,B) =hs0, s1, . . . , sn−1, sε (sisj)mij = e for 0 ≤ i, j ≤ n − 1, s2 ε = sε, sisε = sεsi for i > 1, sεs1sεs1 = s1sεs1sε = sεs1sε, s0sεs0 = s1sεs1, s0s2s1sεs1s2 = s2s1sεs1s2s0i , where mij is defined in (3.1). In [10], Everitt and Fountain proved that the Boolean reflection monoid M (An−1,B) (respectively, M (Bn,B), M (Dn,B)) is isomorphic to the symmetric inverse semigroup In (respectively, the monoid I±n of partial signed permutations, the monoid I e ±n of partial even signed permutations). 4.2. Mutation classes of ∆ quivers. By a result of Fomin and Zelevinsky [20], the mutation class of a Dynkin quiver is finite. Let U (Ak+1) be the mutation class of Dynkin quivers of type Ak+1. We use a description of U (Ak+1) given in [5]: (1) Each quiver has k + 1 vertices. (2) All nontrivial cycles are oriented and of length 3. (3) A vertex has at most four incident arrows. (4) If a vertex has four incident arrows, then two of them belong to one oriented 3-cycle and the other belong to another oriented 3-cycle. (5) If a vertex has three incident arrows, then two of them belong to one oriented 3-cycle and the third arrow does not belong to any oriented 3-cycles. Let φ(Aε k) be the class of quivers which satisfy the above conditions (1) -- (5) and (6) The frozen vertex ε has at most two incident arrows and if it has two incident arrows, then ε belongs to one oriented 3-cycle. Let U A (respectively, U Aε k)) for all k. For a quiver Q ∈ U A or Q ∈ U Aε , a vertex v in Q is called a connecting vertex if v has at most 2 neighbours and, moreover, if v has 2 neighbours, then v is a vertex in a 3-cycle in Q. ) be the union of all U (Ak) (respectively, φ(Aε Let φ(Bε n) be the class of quivers which belong to one of the following two types (here a is a connecting vertex for Q′ ∈ U Aε and b is a connecting vertex for Q′′ ∈ U A): QUIVER MUTATIONS AND BOOLEAN REFLECTION MONOIDS 11 0 2 2 0 2 a Q′ ε Q′′ b a Q′ ε Figure 2. φ(Bε n) As a generalization of mutation class of Dynkin quivers of type Dn in [39], let φ(Dε n) be the class of quivers which belong to one of the following four types: Type I: Q has two vertices a and b which have a neighbour and both a and b have n−2), and c is a an arrow to or from the same vertex c, and Q′ = Q\{a, b} is in φ(Aε connecting vertex for Q′, see Figure 3. a c ε Q′ b Figure 3. Type I Type II: Q has a full subquiver with four vertices which looks like the quiver drawn in the left hand side of Figure 4 and Q\{a, b, c → d} = Q′ ∪ Q′′ is disconnected with two components Q′ ∈ φ(Aε k) for some integer k and Q′′ ∈ U (Aℓ) for some integer ℓ and for which c and d are connecting vertices, see the right hand side of Figure 4. d a b c Q′′ d Figure 4. Type II a b c Q′ ε Type III: Q has a full subquiver which is an oriented 4-cycle drawn in the left hand side of Figure 5 and Q\{a, b} = Q′ ∪ Q′′ is disconnected with two components Q′ ∈ φ(Aε k) for some integer k and Q′′ ∈ U (Aℓ) for some integer ℓ and for which c and d are connecting vertices, see the right hand side of Figure 5. 12 BING DUAN, JIAN-RONG LI, AND YAN-FENG LUO a b d c Q′′ d Figure 5. Type III a b c Q′ ε Type IV: Q has a full subquiver which is an oriented d-cycle, where d ≥ 3. We will call this the central cycle. For each arrow α : a → b in the central cycle, there may (and may not) be a vertex cα which is not on the central cycle, such that there is an oriented 3-cycle traversing a → b → cα → a. Moreover, this 3-cycle is a full subquiver. Such a 3-cycle will be called a spike. There are no more arrows starting or ending in vertices on the central cycle. Now Q\{vertices in the central cycle and their incident arrows} = Q′ ∪ Q′′ ∪ Q′′′ ∪ . . . is a disconnected union of quivers, one for each spike, which Q′ ∈ φ(Aε k) for some integer k and the others are all in U A and for which the corresponding vertex c is a connecting vertex, see Figure 6. Q′′ c′′ Q′ c′ ε id id−1 i1 i2 i6 i5 c′′′ Q′′′ i3 i4 Figure 6. Type IV Let U (∆) be the mutation class of any ∆ quiver. Lemma 4.4. The set U (∆) consists of these diagrams described as before. Proof. The mutation class U (∆) is contained in the mutation class of the corresponding Dynkin quiver obtained by viewing the frozen vertex as a mutable vertex. The mutation classes of Dynkin quivers of types An, Bn, and Dn were given in [2, 5, 20, 39]. In particualr, Dynkin quivers of type Dn are also obtained by using Schiffler's geometric model of cluster categories of type Dn in [36]. QUIVER MUTATIONS AND BOOLEAN REFLECTION MONOIDS 13 k) (respectively, φ(Bε By Lemma 3.2 of [39], we deduce that these quivers in φ(Aε n), n)) belong to the corresponding mutation class U (∆) of ∆ quivers. It is not difficult n)) keeps invariant under the mutations. n) also keeps (cid:3) φ(Dε to show that the set φ(Aε By using the similar argument as Theorem 3.1 of [39], we show that the φ(Dε invariant under the mutations. k) (respectively, φ(Bε 4.3. Inverse monoids determined by quivers. Let I ∪ {ε} be the set of vertices of a quiver Q with a frozen vertex ε. For any i, j ∈ I and ε, define if i and j are not connected, if i and j are connected by an edge of weight 1, if i and j are connected by an edge of weight 2, if i and j are connected by an edge of weight 3. 2 3 1 2 4 2 if ε and j are not connected, if ε and j are connected by an edge of weight 1, if ε and j are connected by an edge of weight 2. if ε and j are not connected, if ε and j are connected by an edge of weight 1, if ε and j are connected by an edge of weight 2. 2 3 4 6 mij =  mεj =  mjε =  Setting, in addition, mii = 1 for any i ∈ I ∪ {ε}. Then (mij)i,j∈I is a Coxeter matrix. In order to define an inverse monoid M (Q) associated to Q, we introduce some nota- tions. Let (i1, i2, . . . , ε) denote the shortest path from i1 to ε in Q. We denote by e and P (si1, sε) the identity element and the element si1 . . . sε in M (Q) respectively. Denote by (aba . . .)m an alternating product of m terms. Definition 4.5. For any quiver Q ∈ U (∆), we define an inverse monoid M (Q) with generators si, i ∈ I ∪ {ε} and relations: (R1) s2 (R2) (sisj)mij = e for i, j ∈ I and (sεsjsε ··· )mεj = (sjsεsj ··· )mjε = (sεsjsε ··· )mεj +1 (R3) i = e for i ∈ I, s2 for any j ∈ I; (i) For every chordless oriented cycle C in Q: ε = sε; i0 w1−→ i1 w2−→ ··· → id−1 w0−→ i0, where ij ∈ I for j = 0, 1, . . . , d− 1, either all of the weights are 1, or w0 = 2, we have: (si0si1 ··· sid−2sid−1sid−2 ··· si1)2 = e. (ii) For every chordless oriented cycle C in Q: ε −→ i1 −→ ··· → id−1 −→ ε, where ij ∈ I for j = 1, . . . , d − 1, we have: sεsi1 ··· sid−2sid−1sid−2 ··· si1 = si1 ··· sid−2sid−1sid−2 ··· si1sε. 14 BING DUAN, JIAN-RONG LI, AND YAN-FENG LUO (iii) For every chordless oriented cycle C in Q: ε w1−→ i1 2−→ i2 w2−→ ε, where i1, i2 ∈ I, if w1 = 1 and w2 = 2, we have sεsi1si2si1 = si1si2si1sε; if w1 = 2 and w2 = 1, we have si1si2sεsi2 = si2sεsi2si1. (i) Without loss of generality, let (0, 1, . . . , d, ε) be the shortest path from 0 to (R4) ε in Q ∈ U (Bε n), we have P (s0, sε) = P (s1, sε), P (sε, s0) = P (sε, s1). (ii) For any Q ∈ U (Dε n) drawn in Figures 3 -- 5, we have P (sa, sε)P (sε, sa) = P (sb, sε)P (sε, sb). For any Q ∈ U (Dε si1P (sc′, sε)P (sε, sc′)si1 = si2 ··· sidP (sc′, sε)P (sε, sc′)sid ··· si2. n) drawn in Figure 6, we have Remark 4.6. (1) In (R2), if mjε = 2 then mεj = 2. In this case the equation (sεsjsε . . .)mεj = (sjsεsj . . .)mjε = (sεsjsε . . .)mεj +1 is simplified as sεsj = sjsε. (2) For relation (R3) (ii), though in this paper we only use the case d = 3, 4, the defined relation for arbitrary d is still meaningful, we will study it in future work. Now we are ready for our main results in this section. Theorem 4.7. Let ∆ ∈ {Aε M (Q) ∼= M (Q0). n−1, Bε n, Dε n} and Q0 be a ∆ quiver. If Q ∼mut Q0 then We will prove Theorem 4.7 in Section 7. Up to the above isomorphism, we denote by M (∆) the inverse monoid associated to any quiver appearing in U (∆). The following example is given to explain Theorem 4.7. Example 4.8. We start with a Aε Q1 = µ2(Q0) be the quiver obtained from Q0 by a mutation at vertex 2. 3 quiver Q0 which is shown in Figure 7 (a). Let (a) 1 ◦ 2 ◦ ε • µ2−→ (b) 1 ◦ 2 ◦ ε • Figure 7. (a) A Aε 3 quiver Q0; (b) The quiver Q1 = µ2(Q). It follows from Definition 4.5 that 2 = e, s2 1 = s2 ε = sε, s1s2s1 = s2s1s2, s1sε = sεs1, s2sεs2sε = sεs2sεs2 = sεs2sεi , M (Q0) =(cid:10)s1, s2, sε s2 M (Q1) =(cid:10)t1, t2, tε t2 Then an inverse monoid isomorphism ϕ : M (Q0) → M (Q1) is given by t2tεt2tε = tεt2tεt2 = tεt2tε, tεt2t1t2 = t2t1t2tεi . 2 = e, t2 1 = t2 ε = tε, t1t2t1 = t2t1t2, t1tεt1tε = tεt1tεt1 = tεt1tε, ϕ(si) =(t2tit2 ti if i = 1, otherwise. / / / / o o o o QUIVER MUTATIONS AND BOOLEAN REFLECTION MONOIDS 15 When we say that we mutate a sequence (n, n − 1,··· , 2, 1) of vertices of a quiver we mean that we first mutate the n-th vertex of the quiver, then we mutate the (n − 1)- th vertex, and so on until the first vertex. The following theorem shows that Everitt and Fountain's presentations of Boolean reflection monoids can be obtained from any ∆ quiver by mutations. Theorem 4.9. Let Φ = An−1, Bn or Dn. Then M (Φ,B) ∼= M (Φε). Proof. In view of Theorem 4.7, we only need to prove that there exists a quiver Q ∈ U (∆) such that M (Q) = M (Φ,B). n−1 quiver We mutate the sequence (n− 1, n− 2,··· , 2, 1) of vertices of the following Aε ◦1 / ◦2 / ◦3 /❴❴❴ ◦n−1 , / •ε •ε / ◦1 / ◦2 /❴❴❴ ◦n−2 . / ◦n−1 we obtain the quiver Q1: Then by Definition 4.5 M (Q1) =(cid:10)s1, s2, . . . , sn−1, sε (sisj)mij = e for 1 ≤ i, j ≤ n − 1, s2 sisε = sεsi for i 6= 1, sεs1sεs1 = s1sεs1sε = sεs1sεi , ε = sε, where mij =  1 if i = j, 3 if i − j = 1, 2 otherwise. By Lemma 4.3, we deduce that M (Q1) = M (An−1,B). After the sequence (n − 1, n − 2,··· , 1, 0) of mutations, the Bε n quiver ◦0 2 / ◦1 /❴❴❴ / ◦2 ◦n−1 / •ε becomes the following quiver (denoted by Q2) ε • 2 2 ☎☎☎☎☎☎☎☎☎ ◦0 \✿✿✿✿✿✿✿✿✿ / ◦1 . /❴❴❴ / ◦2 ◦n−2 / ◦n−1 Then by Definition 4.5 M (Q2) =hs0, s1, . . . , sn−1, sε (sisj)mij = e for 0 ≤ i, j ≤ n − 1, s2 ε = sε, s0s1sεs1 = s1sεs1s0, s0sε = sεs0 = sε, sisε = sεsi for i 6= 1, s1sεs1sε = sεs1sεs1 = sεs1sεi, / / / / / / / / / / / / / \ / / / 16 BING DUAN, JIAN-RONG LI, AND YAN-FENG LUO where mij =   1 2 3 4 if i = j, if i and j are not connected, if i and j are connected by an edge with weight 1, if i and j are connected by an edge with weight 2. By Lemma 4.3, we have M (Q2) = M (Bn,B). After the sequence (n − 1, n − 2,··· , 2, 0) of mutations, the Dε n quiver 3 ◦ /❴❴❴ n−1 ◦ ε • 0 ◦ 2 ◦ ◦1 becomes the following quiver (denoted by Q3) . /❴❴❴ 3 ◦ n−2 ◦ n−1 ◦ B✆✆✆✆✆✆✆✆ \✾✾✾✾✾✾✾✾✾ ε • 0 ◦ ◦1 ✾✾✾✾✾✾✾✾ ✆✆✆✆✆✆✆✆✆ 2 ◦ Then by Definition 4.5 M (Q3) =(cid:10)s0, s1, . . . , sn−1, sε (sisj)mij = e for 0 ≤ i, j ≤ n − 1, s2 ε = sε, sisε = sεsi for i > 1, sεsjsεsj = sjsεsjsε = sεsjsε for j = 0, 1, s0sεs0 = s1sεs1, s0s2s1sεs1s2 = s2s1sεs1s2s0i , where 1 2 3 if i = j, if i and j are not connected, if i and j are connected by an edge with weight 1. mij =  We claim that M (Q3) = M (Dn,B), which follows from Lemma 6.1 (3) and Lemma (cid:3) 4.3. For any quiver Q ∈ U (∆), by Theorem 4.7 and Theorem 4.9, M (Q) gives a presen- tation of the corresponding Boolean reflection monoid. In particular, our presentations recover the presentation of the symmetric inverse semigroup In defined in [9, 32]. 4.4. Inner by diagram automorphisms of Boolean reflection monoids. We first consider inner automorphisms of the Boolean reflection monoid M (Φ,B). An automorphism α of M (Φ,B) is called an inner automorphism if there exists a uniquely determined element g ∈ W (Φ) of the Weyl group W (Φ) such that α(t) = gtg−1 for all t ∈ M (Φ,B). Let Aut(G) (respectively, Inn(G)) be the automorphism / / / / / / / O O B / / / / / \ QUIVER MUTATIONS AND BOOLEAN REFLECTION MONOIDS 17 (respectively, inner automorphism) group of G and Z(G) be the center of G for any semigroup G. It was showed in [30, 38] that Aut(M (An−1,B)) = Inn(M (An−1,B)). In other words, As a generalization of the above result, we have the following theorem. Aut(M (An−1,B)) ∼= W (An−1)/Z(W (An−1)) for n ≥ 3 and Aut(M (A1,B)) ∼= W (A1). Theorem 4.10. Inn(M (Φ,B)) ∼= W (Φ)/Z(W (Φ)) for Φ = An−1(≥ 3), Bn(n ≥ 2), Dn(n ≥ 4). Proof. Let α ∈ Inn(M (Φ,B)). Since W (Φ) ⊆ M (Φ,B) is the unique unit group of M (Φ,B), we have the fact that αW (Φ) is an inner automorphism of W (Φ). We will prove the parts of Φ = Bn(n ≥ 2), Dn(≥ 4). Let Φε be one of Bε n and Dε n shown in Table 1. Suppose that Λ = {s0, s1, . . . , sn−1, sε} is a set of generators of M (Φ,B). For any element g ∈ W (Φ), we claim that the set gΛg−1 = {gs0g−1, gs1g−1, . . . , gsn−1g−1, gsεg−1} 1 = g2sεg−1 2 Finally, we show that g1sεg−1 is still a set of generators of M (Φ,B). Obviously, (gsig−1)2 = e, (gsεg−1)2 = gsεg−1, and gΛg−1 satisfies (R2) -- (R4) in Definition 4.5. for any g1, g2 ∈ Z(W (Φ)). It suffices to prove that sε = w0sεw−1 0 , where w0 is the longest element in W (Φ) and w0 is an involution. By Section 1.2 of [15], we have w0 = wnwn−1 ··· w1, where wi = si−1 ··· s1s0s1 ··· si−1 for i ≥ 1 in type Φ = Bn, and wi = si−1 ··· s3s2s1s0s2s3 ··· si−1 for i ≥ 3 and w1 = s0, w2 = s1 in type Φ = Dn. Then by (R1), (R2), and (R4) of Definition 4.5, w0sεw−1 0 = wnwn−1 ··· w1sεw1 ··· wn−1wn = wnsεwn in type Bn, in type Dn. (cid:3) =(sn−1 ··· s1(s0s1 ··· sn−1sεsn−1 ··· s1s0)s1 ··· sn−1 = sε sn−1 ··· s3s2s1(s0s2s3 ··· sn−1sεsn−1 ··· s3s2s0)s1s2s3 ··· sn−1 = sε The theorem is proved. n−1, Bε n, and Dε Let ∆ be one of Aε n shown in Table 1. Let Q be a ∆ quiver. Let I ∪{ε} be the set of vertices of Q and Q′ = µk(Q) the quiver obtained by a mutation of Q at a mutable vertex k. Following Barot and Marsh's work [2], one can define variables ti for i ∈ I, and tε in M (Q′) as follows: si ti =(sksisk tε =(sksεsk sε if there is an arrow i → k in Q (possibly weighted), otherwise, if there is an arrow ε → k in Q (possibly weighted), otherwise. (4.1) From Lemma 3.3 and Equation (4.1), it follows that new elements ti, i ∈ I, appearing in the procedure of mutations of quivers, must be some reflections in Weyl groups. We introduce the definition of inner by diagram automorphisms of M (Φ,B). Definition 4.11. An inner by diagram automorphism of M (Φ,B) is an automorphism generated by some inner automorphisms and diagram automorphisms in Aut(M (Φ,B)). 18 BING DUAN, JIAN-RONG LI, AND YAN-FENG LUO For simplicity, let W (Q\{ε}) be the unit group of M (Q). In the following theorem, we show that inner by diagram automorphisms of M (Φ,B) can be constructed by a sequence of mutations preserving the same underlying diagrams. n, Dε n. Theorem 4.12. Let Q be a ∆ quiver and M (Q) the corresponding Boolean reflection monoid with a set S of generators. Then α is an inner by diagram automorphism of M (Q) if and only if there exists a sequence of mutations preserving the underlying diagram ∆ such that α(S) can be obtained from Q by mutations. In particular, all reflections in W (Q\{ε}) and gsεg−1 for g ∈ W (Q\{ε}) can be obtained from Q by mutations. Proof. Suppose that S = {s1, s2, . . . , sn−1, sε} for ∆ = Aε for ∆ = Bε n−1 or S = {s0, s1, . . . , sn−1, sε} All automorphisms of M (An−1,B) are inner, see Theorem 4.10. A sequence µ of mutations preserving the underlying diagram of Q induces to an inner automorphism of M (An−1,B). The remainder proof of the sufficiency is to prove types Bn and Dn. Since all au- tomorphisms of irreducible Weyl groups that preserve reflections are inner by diagram automorphisms, we assume without loss of generality that M (µ(Q)) is generated by {t0, t1, . . . , tn−1, tε}, where ti = gsig−1 for 0 ≤ i ≤ n − 1, g ∈ W (Bn) or W (Dn). We claim that tε = gsεg−1. If tε = gsεg−1, then {t0, t1, . . . , tn−1, tε} is a set of generators of M (µ(Q)) and µ(Q) preserves the underlying diagram ∆. On the one hand, since tεti = titε for 0 ≤ i ≤ n − 2, we have tε ∈ Z(W (Bn−1)) or Z(W (Dn−1)), where {tεt0, tεt1, . . . , tεtn−2} is a set of generators of W (Bn−1) or W (Dn−1). On the other hand, the variable tε must be of the form g′sεg′−1 for some g′ ∈ W (Bn) or W (Dn)). So tε is not the longest element w0 in W (Bn−1) or W (Dn−1) and hence tε must be the unique identity element in W (Bn−1) or W (Dn−1). Therefore by the uniqueness tε = gsεg−1. Conversely, for each inner automorphism α of M (Q), by Theorem 4.10, there exists an element g ∈ W (Q\{ε}) such that α(t) = gtg−1 for all t ∈ M (Q). The remainder proof of the necessity is similar to the proof of the necessity of Theorem 3.5. Each reflection in W (Q\{ε}) is of the form gsig−1, where g = si1si2 ··· sik ∈ W (Q\{ε}) is a reduced expression for g. By the same arguments as Theorem 3.5, we mutate the sequence i1, i1, i2, i2, . . . , ik, ik starting from Q, we obtain gsig−1 and gsεg−1. (cid:3) 4.5. Cellularity of semigroup algebras of Boolean reflection monoids. In this section, we show that semigroup algebras of Boolean reflection monoids are cellular algebras. We use the presentations we obtained to interpret cellular bases of such cellular algebras. Let R be a commutative ring with identity. Recall that a semigroup S is said to be cellular if its semigroup algebra R[S] is a cellular algebra. Proposition 4.13. The Boolean reflection monoid M (Φ,B) for Φ = An−1, Bn, or Dn is a cellular semigroup. Proof. All maximal subgroups of the Boolean reflection monoid M (Φ,B) are finite reflec- tion groups. It has been shown in [22] that any finite reflection group W (Φ) is cellular with respect to which the anti-involution is inversion. Therefore, for each D-class D of QUIVER MUTATIONS AND BOOLEAN REFLECTION MONOIDS 19 M (Φ,B), the subgroup HD ⊆ M (Φ,B) is cellular with cell datum (ΛD, MD, CD, iD), where iD : HD → HD is given by iD(g) = g−1 for any g ∈ HD, which satisfies East's first assumption, see Theorem 15 in [7] or Theorem 2.2. We define a map i : R[M (Φ,B)] → R[M (Φ,B)] , rjg−1 j Xj rjgj 7→Xj where rj ∈ R, gj ∈ M (Φ,B). The map i is an R-linear anti-homomorphism and i([e, f, g]D) = [e, f, g]−1 D = [f, e, g−1]D = [f, e, iD(g)]D for any g ∈ HD, e D f in M (Φ,B). From Theorem 19 in [7] or Theorem 2.2, it follows that the Boolean reflection monoid (cid:3) M (Φ,B) is a cellular semigroup, as required. Remark 4.14. A finite inverse semigroup whose maximal subgroups are direct prod- ucts of symmetric groups has been considered by East, see Theorem 22 of [7]. Maximal subgroups of M (An−1,B) are isomorphic to symmetric groups. Maximal subgroups of M (Bn,B) (respectively, M (Dn,B)) are isomorphic to finite reflection groups of type Br (respectively, finite reflection groups of type Dr), where r ≤ n. n−1, Bε n, and Dε Let ∆ be one of Aε n shown in Table 1. For two quivers with the same underlying diagrams appearing in U (∆), we construct inner by diagram automorphisms of Boolean reflection monoids, see Theorem 4.12, and then we extend it to an R-linear automorphism of semigroup algebras of Boolean reflection monoids. Applying Corollary 2.3, we interpret cellular bases of semigroup algebras of Boolean reflection monoids in terms of the presentations and inner by diagram automorphisms we obtained. n 2 w1 w2 4.6. An example. In this section, we denote by In the symmetric inverse semigroup on [n] = {1, 2, . . . , n}. Let w be a partial permutation on a set A ⊆ [n] and denote the image of i ∈ A under the map w by wi and the image of i 6∈ A under the map w by wi = ∅. Then w is denoted by (cid:18) 1 ··· wn−1 wn(cid:19), see [31]. A partial ··· n − 1 permutation w is called an element of rank i if the number of non-empty entries wj for 1 ≤ j ≤ n is i. Example 4.15. Let Q0 be the quiver in Example 4.8 and by the results of preceding sections, M (Q0) ∼= I3 a Boolean reflection monoid. We have M (Q0)/D = {D0 < D1 < D2 < D3}, where each Di is the set of all elements of M (Q0) of rank i, and idempotents in each Di are all partial identity permutations of rank i. Let A be a subset of {1, 2, 3}. As shown in Example 23 of [7], the H-class containing the idempotent idA is the subgroup {x ∈ I3 im(x) = dom(x) = A} ∼= SA. It is well known that the group algebra R[Sn] has cellular bases with respect to which the anti- involution is inversion. Indeed the Khazdan-Luzstig bases and the Murphy basis both have this property (see, Example (1.2) of [23], Example (2.2) of [34] or Section 4 of [35]). Let s1 = (cid:18)1 2 3 2 1 3(cid:19), s2 = (cid:18)1 2 3 1 3 2(cid:19), and sε =(cid:18)1 2 3 1 2 ∅(cid:19). By mutating the quiver Q0, we obtain the following isomorphic quivers (Theorems 4.10 and 4.12). 20 BING DUAN, JIAN-RONG LI, AND YAN-FENG LUO (a) (b) (c) (d) (e) (f ) s1◦ s1◦ s2s1s2◦ s2◦ s2s1s2◦ s2◦ s2◦ s1s2s1◦ s2◦ s1s2s1◦ s1◦ s1◦ sε• s1s2sεs2s1 sε• s2sεs2• • s2sεs2• • s1s2sεs2s1 From Theorem 4.7 and Theorem 4.9, it follows that the inverse monoids associated to quivers (a) -- (f ) are isomorphic to the symmetric inverse semigroup I3 respectively. The presentation of I3 determined by the quiver (a) admits an initial cellular bases by Theorem 19 of [7] or Theorem 2.2. By using these presentations corresponding to quivers (a) -- (f ), we construct an R-linear automorphism of R[I3] and then applying Corollary 2.3, we give an alternative interpretation of cellular bases of R[I3]. 5. Mutations of quivers of finite type Throughout this section, let as before ∆ be one of Aε n in Table 1. We consider the way of mutations of ∆ quivers and the oriented cycles appearing in them, refer to [2, 20]. We first recall: n−1, Bε n, and Dε Proposition 5.1 ([20, Proposition 9.7]). Let Q be a quiver of finite type. Then a chordless cycle in the underlying diagram of Q is always cyclically oriented in Q. In addition, the chordless cycle must be one of those shown in Figure 8. 2 2 2 2 Figure 8. The chordless cycles in Q. We extend Corollary 2.3 in [2] to the case of ∆ quivers. Lemma 5.2. Let Q be a ∆ quiver and k a mutable vertex of Q. Suppose that k has two neighbouring vertices. Then the effect of the mutation of Q at k on the induced subdiagram must be as in Figure 9 (starting on one side or the other), up to switching the two neighbouring vertices. Proof. This follows from Proposition 5.1 and Corollary 2.3 in [2] by restricting quivers to Dynkin quivers of types An, Bn+1, and Dn+1 with a frozen vertex ε. (cid:3) / / / / / / / / / / / / / / / / / / / / / / / / QUIVER MUTATIONS AND BOOLEAN REFLECTION MONOIDS 21 (a) (c) (e) k ◦ B✆✆✆✆✆✆✆✆✆ \✾✾✾✾✾✾✾✾✾ ◦i µk←→ ◦j ◦i k ◦ ✾✾✾✾✾✾✾✾✾ ✆✆✆✆✆✆✆✆✆ ◦j µk←→ µk←→ k ◦ k ◦ B✆✆✆✆✆✆✆✆✆ B✆✆✆✆✆✆✆✆✆ ◦i ◦i \✾✾✾✾✾✾✾✾✾ 2 2 ✾✾✾✾✾✾✾✾✾ ◦j ◦j k ◦ k ◦ 2 ✆✆✆✆✆✆✆✆✆ ✆✆✆✆✆✆✆✆✆ ◦i ◦i 2 ✾✾✾✾✾✾✾✾✾ ◦j \✾✾✾✾✾✾✾✾✾ 2 / ◦j k ◦ B✆✆✆✆✆✆✆✆✆ \✾✾✾✾✾✾✾✾✾ µk←→ •ε ◦i k ◦ ✾✾✾✾✾✾✾✾✾ ✆✆✆✆✆✆✆✆✆ •ε (a') ◦i (c') (e') (g') ✆✆✆✆✆✆✆✆✆ 2 B✆✆✆✆✆✆✆✆✆ 2 ✆✆✆✆✆✆✆✆✆ ◦i ◦i ◦i k ◦ k ◦ k ◦ \✾✾✾✾✾✾✾✾✾ ✾✾✾✾✾✾✾✾✾ •ε •ε \✾✾✾✾✾✾✾✾✾ 2 / •ε (d') (f') µk←→ µk←→ µk←→ k ◦ k ◦ 2 k ◦ B✆✆✆✆✆✆✆✆✆ 2 ✆✆✆✆✆✆✆✆✆ 2 B✆✆✆✆✆✆✆✆✆ ◦i ◦i ◦i ✾✾✾✾✾✾✾✾✾ •ε \✾✾✾✾✾✾✾✾✾ / •ε 2 ✾✾✾✾✾✾✾✾✾ •ε (b) (d) k ◦ k ◦ B✆✆✆✆✆✆✆✆✆ 2 B✆✆✆✆✆✆✆✆✆ ✾✾✾✾✾✾✾✾✾ ✾✾✾✾✾✾✾✾✾ ◦j ◦j ◦i ◦i µk←→ µk←→ µk←→ k ◦ 2 B✆✆✆✆✆✆✆✆✆ 2 ✾✾✾✾✾✾✾✾✾ (f) ◦i (b') ◦j ◦i µk←→ µk←→ µk←→ k ◦ k ◦ B✆✆✆✆✆✆✆✆✆ 2 B✆✆✆✆✆✆✆✆✆ ✾✾✾✾✾✾✾✾✾ \✾✾✾✾✾✾✾✾✾ •ε •ε ◦i ◦i k ◦ \✾✾✾✾✾✾✾✾✾ 2 ✆✆✆✆✆✆✆✆✆ ◦i •ε ◦i k ◦ k ◦ ✆✆✆✆✆✆✆✆✆ 2 ✆✆✆✆✆✆✆✆✆ ◦i ◦i 2 k ◦ 2 ✆✆✆✆✆✆✆✆✆ \✾✾✾✾✾✾✾✾✾ / ◦j \✾✾✾✾✾✾✾✾✾ / ◦j \✾✾✾✾✾✾✾✾✾ 2 / ◦j k ◦ k ◦ \✾✾✾✾✾✾✾✾✾ / •ε ✾✾✾✾✾✾✾✾✾ •ε ✆✆✆✆✆✆✆✆✆ 2 ✆✆✆✆✆✆✆✆✆ ◦i ◦i 2 B✆✆✆✆✆✆✆✆✆ k ◦ 2 ✾✾✾✾✾✾✾✾✾ •ε Figure 9. Local pictures of mutations of Q. In a diagram, a vertex is said to be connected to another if there is an edge between them. Let Q be a quiver that is mutation equivalent to a Dynkin quiver. In Lemma 2.4 of [2], Barot and Marsh described the way that vertices in Q can be connected to a chordless cycle: A vertex is connected to at most two vertices of a chordless cycle, and if it is connected to two vertices, then the two vertices must be adjacent in the cycle. The following lemma is a generalization of Barot and Marsh's results [2, Lemma 2.5]. Lemma 5.3. Let Q be a ∆ quiver and Q′ = µk(Q) the mutation of Q at vertex k. We list various types of induced subquivers in Q (on the left) and corresponding cycles C ′ in Q′ (on the right) arising from the mutation of Q at k. The diagrams are drawn so that C ′ is always a clockwise cycle in Figures 10 and 11. Then every chordless cycle in Q′ arises in such a way. B \ B / \ B \ B / \ B / \ B o o / \ B \ B / \ \ B o o B \ B / \ \ B o o / \ B o o 22 BING DUAN, JIAN-RONG LI, AND YAN-FENG LUO Proof. Let C ′ be a chordless cycle in Q′. We will divide C ′ into two classes, depending on whether or not it contains the frozen vertex ε. (1) If C ′ does not contain the frozen vertex ε, then C ′ is one of (a) -- (g), which follows from Proposition 5.1 and Lemma 2.5 in [2]. (2) If C ′ contains the frozen vertex ε, then C ′ must be a cycle of length 3 or 4. Otherwise, C ′ is a cycle of length at least 5. By Proposition 5.1, all edges of C ′ have trivial weights and C ′ is mutation equivalent to the following subquiver • / ◦ ◦ . / ◦ / ◦ / ··· If ε has two incident arrows, then by Proposition 5.1, either both of them have trival weight or one of them has weigh 2 and the other has weight 1. This is because if ε has two incident arrows with weight 2, then C ′ is mutation equivalent to ◦ / • . (cid:3) / ◦ 2 (a) The vertex k does not connect to an oriented chordless cycle C in Q. Then C is the corresponding cycle in Q′. (b) The vertex k connects to one vertex of an oriented chordless cycle C in Q (via an edge of unspecified weight). Then C is the corresponding cycle in Q′. (c) The vertex k is one vertex of an oriented d-cycle without ε, d ≥ 4, in Q. Then the local picture of µk(Q) becomes a 3-cycle and a (d − 1)-cycle, and they share a common arrow, but the orientations of two cycles are opposite, or vice versa. (d) (f) k ◦ \✾✾✾✾✾✾✾✾✾ ✆✆✆✆✆✆✆✆✆ ◦i k ◦ \✾✾✾✾✾✾✾✾✾ 2 ✆✆✆✆✆✆✆✆✆ ◦i ◦j ◦i µk−→ ◦j µk−→ k ◦ C ′ B✆✆✆✆✆✆✆✆✆ ◦i ✾✾✾✾✾✾✾✾✾ ◦j k ◦ C ′ B✆✆✆✆✆✆✆✆✆ 2 2 ✾✾✾✾✾✾✾✾✾ ◦j (e) (g) ◦i ◦i µk−→ ◦j µk−→ k ◦ \✾✾✾✾✾✾✾✾✾ 2 ✆✆✆✆✆✆✆✆✆ k ◦ 2 ✆✆✆✆✆✆✆✆✆ \✾✾✾✾✾✾✾✾✾ 2 / ◦j k ◦ C ′ 2 B✆✆✆✆✆✆✆✆✆ 2 k ◦ C ′ 2 B✆✆✆✆✆✆✆✆✆ ✾✾✾✾✾✾✾✾✾ ◦j 2 ✾✾✾✾✾✾✾✾✾ ◦j ◦i ◦i Figure 10. Induced subquivers and the corresponding chordless cycles, part 1. 6. Cycle relations and path relations In this section, we find an efficient subset of the relations to define Boolean reflec- tion monoids, which generalizes Barot and Marsh's results, Lemmas 4.1, 4.2, 4.4 and Proposition 4.6 in [2]. / / / / / / \ B o o \ B o o \ B o o / \ B o o QUIVER MUTATIONS AND BOOLEAN REFLECTION MONOIDS 23 µk−→ •ε •ε ◦i ◦i µk−→ µk−→ k ◦ k ◦ k ◦ ✆✆✆✆✆✆✆✆✆ ✆✆✆✆✆✆✆✆✆ 2 ✆✆✆✆✆✆✆✆✆ \✾✾✾✾✾✾✾✾✾ \✾✾✾✾✾✾✾✾✾ 2 \✾✾✾✾✾✾✾✾✾ 2 k ◦ C ′ B✆✆✆✆✆✆✆✆✆ k ◦ C ′ B✆✆✆✆✆✆✆✆✆ 2 k ◦ C ′ 2 B✆✆✆✆✆✆✆✆✆ ✾✾✾✾✾✾✾✾✾ 2 ✾✾✾✾✾✾✾✾✾ 2 ✾✾✾✾✾✾✾✾✾ ◦i ◦i •ε (a') •ε •ε (c') (e') ◦i (g') j ◦ ◦i (i') j ◦ ◦k (k') k ◦ ◦j ◦i / •ε µk−→ j ◦ ◦ k •ε i ◦ ◦i µk−→ j ◦ / •ε ◦k $❏❏❏❏❏❏❏❏❏❏❏❏❏❏ d❏❏❏❏❏❏❏❏❏❏❏❏❏❏ :✉✉✉✉✉✉✉✉✉✉✉✉✉✉ i ◦ µk−→ k ◦ •ε ◦j C ′ C ′ C ′ k ◦ •ε i ◦ •ε i ◦ •ε (b') (d') (f') ✆✆✆✆✆✆✆✆✆ 2 ✆✆✆✆✆✆✆✆✆ ◦i ◦i 2 ✆✆✆✆✆✆✆✆✆ •ε (h') j ◦ ◦i (j') j ◦ ◦k (l') k ◦ ◦j k ◦ k ◦ k ◦ \✾✾✾✾✾✾✾✾✾ \✾✾✾✾✾✾✾✾✾ •ε •ε \✾✾✾✾✾✾✾✾✾ 2 / ◦i k ◦ •ε i ◦ •ε i ◦ •ε k ◦ C ′ B✆✆✆✆✆✆✆✆✆ k ◦ C ′ 2 B✆✆✆✆✆✆✆✆✆ 2 k ◦ C ′ 2 B✆✆✆✆✆✆✆✆✆ ✾✾✾✾✾✾✾✾✾ ✾✾✾✾✾✾✾✾✾ •ε •ε 2 ✾✾✾✾✾✾✾✾✾ ◦i $❏❏❏❏❏❏❏❏❏❏❏❏❏❏ C ′ d❏❏❏❏❏❏❏❏❏❏❏❏❏❏ C ′ i ◦ / •ε µk−→ µk−→ µk−→ ◦i ◦i •ε µk−→ j ◦ ◦i µk−→ j ◦ ◦k µk−→ k ◦ :✉✉✉✉✉✉✉✉✉✉✉✉✉✉ C ′ ◦j k ◦ •ε i ◦ •ε Figure 11. Induced subquivers and the corresponding chordless cycles, part 2. Let ∆ be one of Aε n in Table 1. Suppose that Q is any quiver that is mutation equivalent to a ∆ quiver. The following lemma gives an efficient subset of relations (R3) and (R4) in Definition 4.5. n, and Dε n−1, Bε Lemma 6.1. Let M (Q) be an inverse monoid with generators subject to relations (R1) and (R2) in Definition 4.5. (1) If Q contains a chordless cycle C ′ 3, see Figure 12, then the following statements are equivalent: (a) sεs1s2s1 = s1s2s1sε; (b) s1s2sεs2 = s2sεs2s1. Furthermore, if one of the above holds, then the following statements are equiv- alent: (c) sεs1sε = sεs1sεs1 = s1sεs1sε; (d) sεs2sε = sεs2sεs2 = s2sεs2sε. \ B o o \ B o o \ B o o \ B o o / \ B o o / \ B o o $ o o O O o o O O / / O O o o / / O O o o $ o o O O o o O O / / / d / / O O o o / / O O o o / / / d o o : o o / / O O o o / / O O o o o o : o o 24 BING DUAN, JIAN-RONG LI, AND YAN-FENG LUO (2) If Q contains a chordless cycle C ′′ (3) If Q contains a chordless cycle C ′ 3 , see Figure 12, then s1s2sεs2 = s2sεs2s1. 4, see Figure 12, then the following statement holds: (a) sεs1s2s3s2s1 = s1s2s3s2s1sε; (b) s1s2s3sεs3s2 = s2s3sεs3s2s1. Furthermore, if one of the above holds, then the following statements are equiv- alent: (c) sεs1sε = sεs1sεs1 = s1sεs1sε; (d) sεs3sε = sεs3sεs3 = s3sεs3sε. (4) If Q contains a subquiver C ′ d, see Figure 6, then the following statements are equivalent: (a) si1P (sc′, sε)P (sε, sc′)si1 = si2 ··· sidP (sc′, sε)P (sε, sc′)sid ··· si2; (b) sia−1 ··· si1P (sc′, sε)P (sε, sc′)si1 ··· sia−1 = sia ··· sidP (sc′, sε)P (sε, sc′)sid ··· sia for any a = 3,··· , d. 2 ◦ C ′ 3 B✆✆✆✆✆✆✆✆✆ ◦1 ✾✾✾✾✾✾✾✾✾ •ε ◦1 2 ◦ ′′ C 3 2 B✆✆✆✆✆✆✆✆✆ ✾✾✾✾✾✾✾✾✾ 2 2 ◦ ◦1 •ε C ′ 4 3 ◦ •ε Figure 12. A chordless 3-cycle C ′ less 4-cycle C ′ 4. (see Lemma 6.1) 3, a chordless 3-cycle C ′′ 3 , and a chord- Proof. For (1), the equivalence of (a) and (b) follows from: s1s2sεs2 = s2(s1s2s1sε)s2, s2sεs2s1 = s2(sεs1s2s1)s2, using (R2). Suppose that (a) and (b) hold. Then by (R1), (R2), (a), and (b), the equivalence of (c) and (d) follows from: s1s2s1(sεs1sε)s1s2s1 = sεs1s2s1s1s1s2s1sε = sεs2sε, s1s2s1(sεs1sεs1)s1s2s1 = sε(s1s2sεs2)s1 = sεs2sεs2, s1s2s1(s1sεs1sε)s1s2s1 = s1(s2sεs2s1)sε = s2sεs2sε. For (3), the equivalence of (a) and (b) follows from: s1s2s3sεs3s2 = s2s3(s1s2s3s2s1sε)s3s2, s2s3sεs3s2s1 = s2s3(sεs1s2s3s2s1)s3s2, using first s1s2s3s2s1 = s3s2s1s2s3 and then (R1). Suppose that (a) and (b) hold. Using first s2sε = sεs2 and then (a), we have: s1s2s3s2s1s2(sεs1sε)s2s1s2s3s2s1 = (s1s2s3s2s1sε)s2s1s2(sεs1s2s3s2s1) = (sεs1s2s3s2s1)s2s1s2(s1s2s3s2s1sε) = sεs1s2s3s2s3s2s1sε = sεs3sε, (by (R2)) B o o B o o / / O O o o QUIVER MUTATIONS AND BOOLEAN REFLECTION MONOIDS 25 where in the last equation we used that s1 and s3 commute. Using (R1), (R2), (a), and (b), by a similar argument, we have s1s2s3s2s1s2(s1sεs1sε)s2s1s2s3s2s1 = s3sεs3sε. s1s2s3s2s1s2(sεs1sεs1)s2s1s2s3s2s1 = sεs3sεs3. Therefore (c) and (d) are equivalent. For (4), using (R1), it is obvious. (cid:3) At an end of this section, we show that M (Q) could be defined using only the under- lying unoriented weighted diagram of Q, by taking relations (R1) -- (R4) corresponding to both Q and Qop as the defining relations. Our result can be viewed as a generalization of Proposition 4.6 of [2]. Proposition 6.2. Let M (Φ,B) be a Boolean reflection monoid with generators si, i ∈ I ∪ {ε}. Then the generators satisfy (R1) -- (R4) with respect to Q if and only if they satisfy (R1) -- (R4) with respect to Qop. Proof. We assume that generators si, i ∈ I ∪{ε} satisfy relations (R1) -- (R4) with respect to Q, and show that these generators satisfy relations (R1) -- (R4) with respect to Qop. The converse follows by replacing Q with Qop. Since (R1) and (R2) do not depend on the orientation of Q, generators si, i ∈ I ∪{ε} satisfy relation (R1) and (R2) with respect to Qop. The cases of chordless cycles appearing in quivers of finite type have been proved in Proposition 4.6 of [2]. The remaining needed to check the cases are C ′ 4 shown in Figure 12 and C ′ 3, C ′′ 3 , C ′ d shown in Figure 6. Case 1. In C ′ 3, we have sεs2s1s2 = sεs1s2s1 = s1s2s1sε = s2s1s2sε, s2s1sεs1 = s1s2(s1s2sεs2)s2s1 = s1s2(s2sεs2s1)s2s1 = s1sεs1s2. Case 2. In C ′′ 3 , we have sεs2s1s2 = s2s1(s1s2sεs2)s1s2 = s2s1(s2sεs2s1)s1s2 = s2s1s2sε. Case 3. In C ′ 4, note that s1s2s3s2s1 = s3s2s1s2s3. We have sεs3s2s1s2s3 = sεs1s2s3s2s1 = s1s2s3s2s1sε = s3s2s1s2s3sε, s3s2s1sεs1s2 = s2s1s3s2(s1s2s3sεs3s2)s2s3s1s2 = s2s1s3s2(s2s3sεs3s2s1)s2s3s1s2 = s2s1sεs1s2s3. Case 4. In C ′ d, it follows from Lemma 6.1 (4) that (R4) does not depend on the orientation of chordless cycles in C ′ d. Since every chordless cylce in Qop corresponds to a chordless cycle in Q, the result (cid:3) holds. 7. The proof of Theorem 4.7 In this section, we give the proof of Theorem 4.7. Let ∆ be one of Aε n in Table 1. We fix a ∆ quiver Q. Let Q′ = µk(Q) be the mutation of Q at vertex k, where k ∈ I. Throughout the section, we use si and n, and Dε n−1, Bε 26 BING DUAN, JIAN-RONG LI, AND YAN-FENG LUO ri for i ∈ I ∪ {ε} to denote generators of M (Q) and M (Q′) respectively. Similar to [2], we define elements ti, i ∈ I, and tε in M (Q) as follows: if there is an arrow i → k in Q (possibly weighted), otherwise, if there is an arrow ε → k in Q (possibly weighted), otherwise. (7.1) si ti =(sksisk tε =(sksεsk i = e for i ∈ I and t2 sε Then t2 proposition, which we will prove it in Section 7.2. ε = tε. In order to prove Theorem 4.7, we need the following Proposition 7.1. For each i ∈ I ∪ {ε}, the map Φ : M (Q′) −→ M (Q) ri 7−→ ti, is an inverse monoid homomorphism. 7.1. Proof of Theorem 4.7. For each vertex i ∈ I ∪ {ε} of Q define the elements t′ M (Q′) as follows: i in t′ t′ ri i =(rkrirk ε =(rkrεrk rε if there is an arrow k → i in Q′ (possibly weighted), otherwise, if there is an arrow k → ε in Q′ (possibly weighted), otherwise. We claim that these elements t′ i, for each vertex i ∈ I ∪ {ε}, satisfy the relations (R1) -- (R4) defining M (Q). This follows from Proposition 7.1 by interchanging Q and Q′ and using the fact that the definition of M (Q) is unchanged under reversing the orientation of all the arrows in Q (see Proposition 6.2). Therefore there is an inverse monoid homomorphism Θ : M (Q) → M (Q′) such that Θ(si) = t′ If there is no arrow i → k in Q, then there is also no arrow k → i in Q′ and consequently Θ ◦ Φ(ri) = Θ(si) = ri. If there is an arrow i → k in Q, then there is an arrow k → i in Q′ and therefore Θ ◦ Φ(ri) = Θ(sksisk) = Θ(sk)Θ(si)Θ(sk) = rk(rkrirk)rk = ri. So Θ ◦ Φ = idM (Q′), and, similarly, Φ ◦ Θ = idM (Q), and hence Θ and Φ are isomorphisms. 7.2. The proof of Proposition 7.1. We will prove Proposition 7.1 by showing that the elements ti, i ∈ I ∪ {ε} satisfy the (R1) -- (R4) relations in M (Q′). We denote by m′ ij the value of mij for Q′. (R1) is obvious. In the sequel, the proof that the elements ti, i ∈ I ∪ {ε}, satisfy (R2) in M (Q′) follows from Lemma 7.2 and the rest of proof is completed case by case. i for each i. Lemma 7.2. The elements ti, for i a vertex of Q, satisfy the following relations. (1) If i = k or j = k and i, j 6= ε, then (titj)m′ (2) If at most one of i, j is connected to k in Q and i, j 6= ε, then (titj)m′ ij = e. ij = e. QUIVER MUTATIONS AND BOOLEAN REFLECTION MONOIDS 27 (3) Let i be in I. Then titε = tεti tεtitε = tεtitεti = titεtitε tε = titε = tεti if ε, i are not connected in Q′, if ε, i are connected by an edge with weight 1 in Q′, if ε, i are connected by an edge with weight 2 in Q′.   Proof. In Lemma 5.1 of [2], Barot and Marsh proved the parts (1) and (2). We only need to prove the part (3). Suppose without loss of generality that i = k. The only nontrivial case is when there is an arrow ε → k = i with a weight q in Q. If q = 1, then tεtktεtk = (sksεsk)sk(sksεsk)sk =(sksεsksε = sεsksεsk = sk(sksεsk)sk(sksεsk) = tktεtktε, sεsksε = sksεsksεsk = (sksεsk)sk(sksεsk) = tεtktε. If q = 2, note that sεsk = sksε = sε, then tktε = sk(sksεsk) = sεsk = sksε = sε = tεtk = tε. Suppose that i 6= k. We divide this proof into three cases. Case 1. There are no arrows from i, ε to k, then ti = si, tε = sε hold (3). Case 2. There are arrows from one of i, ε to k and there are no arrows from the other of i, ε to k in Q, then we assume that there are arrows from ε to k and there are no arrows from i to k in Q. If ε, i are not connected in Q, then titε = si(sksεsk) = (sksεsk)si = tεti. If ε, i are connected by an edge with weight 1 in Q, then tεtitε = (sksεsk)si(sksεsk) = sksεsisεsk =(sksεsisεsksi = (sksεsk)si(sksεsk)si = tεtitεti, sisksεsisεsk = si(sksεsk)si(sksεsk) = titεtitε. That ε, i are connected by an edge with weight 2 and there are no arrows from i to k in Q is impossible. Case 3. There are arrows from i, ε to k. The possibilities for the subquivers induced by i, ε, and k are enumerated in (a') -- (g') of Figure 9. We show that ti and tε satisfy (3) by checking each case. Within each case, subcase (i) is when the subquiver of Q is the diagram on the left, and subcase (ii) is when the subquiver of Q is the diagram on the right. (a′)(i) We have titε = (sksisk)(sksεsk) = sksisεsk = sksεsisk = (sksεsk)(sksisk) = tεti. (a′)(ii) We have titε = sisε = sεsi = tεti. (b′) (i) We have tεtitε = sε(sksisk)sε = sεsisksisε = si(sεsksε)si, =(sisεsksεsksi = sεsisksisεsisksi = sε(sksisk)sε(sksisk) = tεtitεti, sisksεsksεsi = sisksisεsisksisε = (sksisk)sε(sksisk)sε = titεtitε. (b′) (ii) We have titε = si(sksεsk) = sk(sksisksε)sk = sk(sεsksisk)sk = (sksεsk)si = tεti. 28 BING DUAN, JIAN-RONG LI, AND YAN-FENG LUO (c′) (i) We have tεtitε = (sksεsk)si(sksεsk) = sksεsisksisεsk = sksisεsksεsisk, =(sksisεsksεsksisk = sksεsisksisεsksi = (sksεsk)si(sksεsk)si = tεtitεti, sksisksεsksεsisk = sisksεsisksisεsk = si(sksεsk)si(sksεsk) = titεtitε. (c′) (ii) We have titε = (sksisk)sε = sisksisε = sεsisksi = sεsksisk = tεti. (d′) (i) We have titε = (sksisk)(sksεsk) = sksisεsk = sksεsisk = (sksεsk)(sksisk) = tεti. (d′)(ii) We have titε = sisε = sεsi = tεti. (e′)(i) Note that sisksε = sksε and sεsksi = sεsk. We have titε = (sksisk)sε = sk(sksε) = sε = tε, tεti = sε(sksisk) = (sεsk)sk = sε = tε. (e′)(ii) We have titε = si(sksεsk) = sisk(sεsksisk)sksi = sisk(sksisksε)sksi = sksεsksi = tεti. (f ′)(i) Note that sisksε = sksε and sεsksi = sεsk. We have titε = si(sksεsk) = sksεsk = tε, tεti = (sksεsk)si = sksεsk = tε. (f ′)(ii) We have titε = (sksisk)sε = sk(sisksεsk)sk = sk(sksεsksi)sk = sεsksisk = tεti. (g′)(i) Note that sksε = sεsk = sε. We have tεtitε = (sksεsk)si(sksεsk) = sεsisε =(sisεsisε = titεtitε, sεsisεsi = tεtitεti. (g′)(ii) Note that sksε = sεsk = sε. We have tεtitε = sε(sksisk)sε =(sεsisεsk = sεsisεsisk = sε(sksisk)sε(sksisk) = tεtitεti, sksεsisε = sksisεsisε = (sksisk)sε(sksisk)sε = titεtitε. (cid:3) The possibilities for chordless cycles in U (∆) are enumerated in Lemma 5.3. For (R3), (a) is trivial and (b) follows from the commutative property of tk and ti, where i is not incident to k. Barot and Marsh proved in [2] that (R3) (i) holds for (a) -- (g). It is enough to show that (R3) (ii) and (R3) (iii) hold by checking (a′) -- (l′). In each case, we need to check that the corresponding cycle relations hold. In the sequel, we frequently use (R1) and (R2) without comment. (a′) We have tεtktitk = sεsk(sksisk)sk = sεsi = sisε = sk(sksisk)sksε = tktitktε. (b′) We have tεtitkti = (sksεsk)sisksi = sksεsisk = sksisεsk = sisksi(sksεsk) = titktitε. (c′) We have tεtktitk = sεsk(sksisk)sk = sεsi = sisε = sk(sksisk)sksε = tktitktε. (d′) We have titktεtk = sisk(sksεsk)sk = sisε = sεsi = sk(sksεsk)sksi = tktεtkti. (e′) Note that sksε = sεsk = sε and sksisεsi = sisεsisk. We have tεtitkti = (sksεsk)sisksi = sεsisksi = sisksisε = sisksi(sksεsk) = titktitε. QUIVER MUTATIONS AND BOOLEAN REFLECTION MONOIDS 29 (f ′) Note that sksε = sεsk = sε and sεsisksi = sisksisε. We have tktitεti = sk(sksisk)sε(sksisk) = sisεsisk = sksisεsi = (sksisk)sε(sksisk)sk = titεtitk. (g′) Note that sksi = sisk and sεsisjsi = sisjsisε.We have tεtitjtktjti = (sksεsk)sisjsksjsi = sksεsisksjsksjsi = sk(sεsisjsi)sk = sk(sisjsisε)sk = sisjsksjsi(sksεsk) = titjtktjtitε. (h′) Note that sjsε = sεsj and sεsisjsksjsi = sisjsksjsisε. We have tεtktjtk = sεsk(sksjsk)sk = sεsj = sjsε = sk(sksjsk)sksε = tktjtktε, tεtitjti = sεsi(sksjsk)si = sεsisjsksjsi = sisjsksjsisε = si(sksjsk)sisε = titjtitε. (i′) Note that sksi = sisk and sεsjsisj = sjsisjsε. We have tεtktjtitjtk = sεsk(sksjsk)si(sksjsk)sk = sεsjsksisksj = sεsjsisj = sjsisjsε = sk(sksjsk)si(sksjsk)sksε = tktjtitjtktε. (j′) Note that sjsε = sεsj and sεsksjsisjsk = sksjsisjsksε. We have tεtjtktj = (sksεsk)sjsksj = sksεsjsk = sksjsεsk = sjsksj(sksεsk) = tjtktjtε, tεtjtitj = (sksεsk)sjsisj = sjsisjsksεsk = tjtitjtε. (k′) Note that sεsjsisj = sjsisjsε. We have tεtjtktitktj = sεsjsk(sksisk)sksj = sεsjsisj = sjsisjsε = sjsk(sksisk)sksjsε = tjtktitktjtε. (l′) Note that sisj = sjsi, sεsk = sksε, and sεsjsksisksj = sjsksisksjsε. We have titktjtk = sisk(sksjsk)sk = sisj = sjsi = sk(sksjsk)sksi = tktjtkti, tεtjtitj = sε(sksjsk)si(sksjsk) = sk(sεsjsksisksj)sk = sk(sjsksisksjsε)sk = (sksjsk)si(sksjsk)sε = tjtitjtε. In order to prove the relation (R4), we need the following lemma. Lemma 7.3. Let Q ∈ U (Aε ℓ) for some ℓ ≥ 1 and c be a fixed vertex in Q. Let k be a mutable vertex. Suppose that (c, . . . , ε) (respectively, (c′ = c, . . . , ε′ = ε)) is the shortest path from c (respectively, c′) to ε (respectively, ε′) in Q (respectively, µk(Q)). Then P (tc′, tε′) = P (sc, sε) or P (sc, sε)sk or skP (sc, sε) or skP (sc, sε)sk. and P (tε′, tc′) = P (sε, sc) or skP (sε, sc) or P (sε, sc)sk or skP (sε, sc)sk. In particular, P (tc′, tε′)P (tε′, tc′) = P (sc, sε)P (sε, sc) or skP (sc, sε)P (sε, sc)sk. Proof. Case 1. The vertex k does not connect to any vertex in {c, . . . , ε}. Then the shortest path relations keep unchanged. Case 2. The vertex k connects to a vertex i ∈ {c, . . . , ε}. Either P (tc′, tε′) = P (sc, sε) or by the commutative property of sk and sj, where j goes over all vertices which are not incident to k, P (tc′, tε′) = skP (sc, sε)sk. Case 3. The vertex k connects to two vertices i, j ∈ {c, . . . , ε}. By Proposition 5.1 and the definition of φ(Aε k) in Section 4.2, either vertices i and j are adjacent or k ∈ {c, . . . , ε} and vertices i and j are neighbours of k. In the first case, P (tc′, tε′) = P (sc, sε)sk or 30 BING DUAN, JIAN-RONG LI, AND YAN-FENG LUO skP (sc, sε). P (tc′, tε′) = P (sc, sε)sk or skP (sc, sε) or skP (sc, sε)sk. In the later case, either the shortest path relation keeps unchanged or Case 4. By the definition of φ(Aε ℓ ) in Section 4.2, the vertex k is impossible to connect to three or more vertices in {c, . . . , ε}. We reverse the order of P (tc′, tε′), we get P (tε′, tc′). (cid:3) In type Bε n, without loss of generality, we assume that (0, 1, . . . , d, ε) is the shortest path from 0 to ε in Q, where the weight between vertex 0 and vertex 1 is 2. Without loss of generality, we assume that (0, 1′, . . . , d′, ε) is the shortest path from 0 to ε in µk(Q), where the weight between vertex 0 and vertex 1′ is 2. In Figure 2, if k = 0 or k = 1 or k, 0, 1, is an oriented 3-cycle, it is easy to check that P (t0, tε) = P (t1′, tε) and P (tε, t0) = P (tε, t1′). Otherwise, by Lemma 7.3 and the commutative property of sk and s0, we have P (t0, tε) = P (t1′, tε) and P (tε, t0) = P (tε, t1′). In type Dε n, Lemma 7.3 allows us to reduce the proof of (R4) to the proof of the following cases: For any quiver in Figure 3, we check that k = a, b, c, for any quiver in Figures 4 -- 5, we check that k = a, b, c, d, and for any quiver in Figure 6, we check k = i1, i2, . . . , id, c′, c′′, c′′′, . . .. In Figures 3 -- 5, if k = a, then either the shortest path relations keep unchanged or P (ta, tε)P (tε, ta) = saP (sa, sε)P (sε, sa)sa, P (tb, tε)P (tε, tb) = saP (sb, sε)P (sε, sb)sa by the commutative relation sasb = sbsa, so P (ta, tε)P (tε, ta) = P (tb, tε)P (tε, tb). We prove that k = b by interchanging a and b. If k = c or k = d, then we prove the following case, other possibilities are similar: µ2←→ B✆✆✆✆✆✆✆✆ ✾✾✾✾✾✾✾✾✾ 2 ◦ 0 ◦ ◦1 ε • /❴❴❴ d ◦ 4 ◦ B✆✆✆✆✆✆✆✆ ✾✾✾✾✾✾✾✾✾ 3 ◦ 0 ◦ ◦1 ✾✾✾✾✾✾✾✾ B✆✆✆✆✆✆✆✆✆ 2 ◦ ε • /❴❴❴ d ◦ 4 ◦ 3 ◦ (i) We have P (t0, tε)P (tε, t0) = t0t3P (t4, tε)P (tε, t4)t3t0 = s0(s2s3s2)P (s4, sε)P (sε, s4)(s2s3s2)s0 = P (s0, sε)P (sε, s0), P (t1, tε)P (tε, t1) = t1t3P (t4, tε)P (tε, t4)t3t1 = s1(s2s3s2)P (s4, sε)P (sε, s4)(s2s3s2)s1 = P (s1, sε)P (sε, s1). (ii) We have P (t0, tε)P (tε, t0) = t0t2P (t3, tε)P (tε, t3)t2t0 = (s2s0s2)s2P (s3, sε)P (sε, s3)s2(s2s0s2) = s2P (s0, sε)P (sε, s0)s2, P (t1, tε)P (tε, t1) = t1t2P (t3, tε)P (tε, t3)t2t1 = (s2s1s2)s2P (s3, sε)P (sε, s3)s2(s2s1s2) = s2P (s1, sε)P (sε, s1)s2. / / / / / / / B / / / / / B o o B QUIVER MUTATIONS AND BOOLEAN REFLECTION MONOIDS 31 Finally, in Figure 6, we prove the following two cases, other possibilities are similar: /❴❴❴ ε • 0 ◦ 1 ◦ 2 ◦ ) ◦d 3 ◦ ❍ ❉ µ1←→ ✽ ✝ ✱ ✒ ③ ✈ ◦d−1 /❴❴❴ ε • 0 ◦ 1 ◦ ) ◦d 2 ◦ ❍ ❉ ③ ✈ ◦d−1 ✽ ✝ ✱ ✒ (i) We have t3 ··· tdP (t0, tε)P (tε, t0)td ··· t3 = s3 ··· sds1P (s0, sε)P (sε, s0)s1sd ··· s3, and t2t1P (t0, tε)P (tε, t0)t1t2 = s1s2s1P (s0, sε)P (sε, s0)s1s2s1 = s2s1P (s0, sε)P (sε, s0) s1s2. (ii) We have tdt1P (t0, tε)P (tε, t0)t1td = s1sdP (s0, sε)P (sε, s0)sds1 and td−1 ··· t1P (t0, tε)P (tε, t0)t1 ··· td−1 = sd−1 ··· s1P (s0, sε)P (sε, s0)s1 ··· sd−1 = sd−1 ··· s2s1s2P (s0, sε)P (sε, s0)s2s1s2 ··· sd−1 = s1(sd−1 ··· s2s1P (s0, sε)P (sε, s0)s1s2 ··· sd−1)s1. /❴❴❴ ε • 0 ◦ 1 ◦ ◦d 2 ◦ /❴❴❴ k−1 ◦ µk←→ k ◦ /❴❴❴ ε • 0 ◦ ◦d−1 o❴ ❴ ❴ ◦k+1 1 ◦ ◦d 2 ◦ /❴❴❴ k−1 ◦ ◦d−1 o❴ ❴ ❴ ◦k+1 k ◦ (i) We have t1P (t0, tε)P (tε, t0)t1 = s1P (s0, sε)P (sε, s0)s1 and t2 ··· tk−1tk+1 ··· tdP (t0, tε)P (tε, t0)td ··· tk+1tk−1 ··· t2 = s2 ··· (sksk−1sk)sk+1 ··· sdP (s0, sε)P (sε, s0)sd ··· sk+1(sksk−1sk)··· s2 = s2 ··· (sk−1sksk−1)sk+1 ··· sdP (s0, sε)P (sε, s0)sd ··· sk+1(sk−1sksk−1)··· s2 = s2 ··· sk−1sksk+1 ··· sdP (s0, sε)P (sε, s0)sd ··· sk+1sksk−1 ··· s2. (ii) We have t1P (t0, tε)P (tε, t0)t1 = s1P (s0, sε)P (sε, s0)s1 and t2 ··· tk−1tktk+1 ··· tdP (t0, tε)P (tε, t0)td ··· tk+1tktk−1 ··· t2 = s2 ··· sk−1sk(sksk+1sk)··· sdP (s0, sε)P (sε, s0)sd ··· (sksk+1sk)sksk−1 ··· s2 = s2 ··· sk−1sk+1sk ··· sdP (s0, sε)P (sε, s0)sd ··· sksk+1sk−1 ··· s2 = s2 ··· sk−1sk+1 ··· sdP (s0, sε)P (sε, s0)sd ··· sk+1sk−1 ··· s2. Proposition 7.1 is proved. / / O O { { ✤ / / / ) o o / / O O { { ✤ / ) o o / / O O / / ) ) o o v v o / / O O / / ) ) h h o o K K o 32 BING DUAN, JIAN-RONG LI, AND YAN-FENG LUO Acknowledgements The authors would like to express their gratitude to B. Everitt, W. N. Franzsen, R. Schiffler, and C. C. Xi for helpful discussions. The authors are very grateful to the anonymous reviewer for the comments and valuable suggestions, especially, the reviewer pointed out an error in the former version. B. Duan was supported by China Schol- arship Council to visit Department of Mathematics at University of Connecticut and he would like to thank R. Schiffler for hospitality during his visit. This work was par- tially supported by the National Natural Science Foundation of China (no. 11771191, 11371177, 11501267, 11401275). The research of J.-R. Li on this project is supported by the Minerva foundation with funding from the Federal German Ministry for Education and Research, and by the Austrian Science Fund (FWF): M 2633 Meitner Program. References [1] E. Bannai, Automorphisms of irreducible Weyl groups, J. Fac. Sci. Univ. Tokyo Sect. I 16 (1969), 273 -- 286. [2] M. Barot and R. J. Marsh, Reflection group presentations arising from cluster algebras, Trans. Amer. Math. Soc. 367 (2015), no. 3, 1945 -- 1967. [3] I. N. Bernstein, I. M. Gel'fand, and V. A. Ponomarev, Coxeter functors, and Gabriel's theorem, Uspehi Mat. Nauk 28 (1973), no. 2, 19 -- 33. [4] N. Bourbaki, Lie groups and Lie algebras, Chapters 4 -- 6, Springer-Verlag, Berlin, 2002. [5] A. B. Buan and D. F. Vatne, Derived equivalence classification for cluster-tilted algebras of type An, J. Algebra 319 (2008), no. 7, 2723 -- 2738. [6] R. Charney and M. Davis, When is a Coxeter system determined by its Coxeter group?, J. London Math. Soc. (2) 61 (2000), no. 2, 441 -- 461. [7] J. East, Cellular algebras and inverse semigroups, J. Algebra 296 (2006), no. 2, 505 -- 519. [8] [9] , Braids and partial permutations, Adv. Math. 213 (2007), no. 1, 440 -- 461. , Generators and relations for partition monoids and algebras, J. Algebra 339 (2011), 1 -- 26. [10] B. Everitt and J. Fountain, Partial symmetry, reflection monoids and Coxeter groups, Adv. Math. 223 (2010), no. 5, 1782 -- 1814. [11] , Partial mirror symmetry, lattice presentations and algebraic monoids, Proc. Lond. Math. Soc. (3) 107 (2013), no. 2, 414 -- 450. [12] D. Easdown and T. G. Lavers, The inverse braid monoid, Adv. Math. 186 (2004), no. 2, 438 -- 455. [13] D. G. FitzGerald, A presentation for the monoid of uniform block permutations, Bull. Austral. Math. Soc. 68 (2003), no. 2, 317 -- 324. [14] D. G. FitzGerald and J. Leech, Dual symmetric inverse monoids and representation theory, J. Austral. Math. Soc. Ser. A 64 (1998), no. 3, 345 -- 367. [15] W. N. Franzsen, Automorphisms of Coxeter Groups, PhD Thesis, University of Sydney, Australia (2001), 1 -- 92. [16] A. Felikson and P. Tumarkin, Coxeter groups and their quotients arising from cluster algebras, Int. Math. Res. Not. IMRN 2016, no. 17, 5135 -- 5186. [17] , Coxeter groups, quiver mutations and geometric manifolds, J. Lond. Math. Soc. (2) 94 (2016), no. 1, 38 -- 60. [18] S. Fomin and N. Reading, Root systems and generalized associahedra, Geometric combinatorics, IAS/Park City Math. Ser., vol. 13, Amer. Math. Soc., Providence, RI, 2007, pp. 63 -- 131. [19] S. Fomin and A. Zelevinsky, Cluster algebras I: Foundations, J. Amer. Math. Soc. 15 (2002), no. 2, 497 -- 529. , Cluster algebras. II. Finite type classification, Invent. Math. 154 (2003), no. 1, 63 -- 121. [20] [21] M. Geck, Relative Kazhdan-Lusztig cells, Represent Theory 10 (2006), 481 -- 524. [22] [23] J. J. Graham and G. I. Lehrer, Cellular algebras, Invent. Math. 123 (1996), no. 1, 1 -- 34. , Hecke algebras of finite type are cellular, Invent. Math. 169 (2007), no. 3, 501 -- 517. QUIVER MUTATIONS AND BOOLEAN REFLECTION MONOIDS 33 [24] J. Grant and R. J. Marsh, Braid groups and quiver mutation, Pacific J. Math. 290 (2017), no. 1, 77 -- 116. [25] X. J. Guo and C. C. Xi, Cellularity of twisted semigroup algebras, J. Pure Appl. Algebra 213 (2009), no. 1, 71 -- 86. [26] J. E. Humphreys, Reflection groups and Coxeter groups, Cambridge Studies in Advanced Mathe- matics, 29, Cambridge University Press, Cambridge, 1990. [27] J. M. Howie, Fundamental of Semigroup Theory, Oxford University Press, New York, 1995. [28] J. Haley, D. Hemminger, A. Landesman, and H. Peck, Artin group presentations arising from cluster algebras, Algebr. Represent. Theory 20 (2017), no. 3, 629 -- 653. [29] Y. D. Ji and Y. F. Luo, Cellularity of Some Semigroup Algebras, Bull. Malays. Math. Sci. Soc. 40 (2017), no. 1, 215 -- 235. [30] A. E. Liber, On symmetric generalized groups, (Russian) Mat. Sbornik N.S. 33 (1953), no. 75, 531 -- 544. [31] S. Lipscomb, Symmetric inverse semigroups, Mathematical Surveys and Monographs, 46, American Mathematical Society, Providence, RI, 1996. [32] L. M. Popova, Defining relations in some semigroups of partial transformations of a finite set, Uchenye Zap. Leningrad Gos. Ped. Inst. 218 (1961), 191 -- 212. [33] R. J. Marsh, Lecture notes on cluster algebras, Zurich Lectures in Advanced Mathematics, European Mathematical Society (EMS), Zurich, 2013. [34] A. Mathas, Iwahori-Hecke algebras and Schur algebras of the symmetric group, University Lecture Series, 15, American Mathematical Society, Providence, RI, 1999. [35] G. E. Murphy, The representations of Hecke algebras of type An, J. Algebra 173 (1995), no. 1, 97 -- 121. [36] R. Schiffler, A geometric model for cluster categories of type Dn, J. Algebraic Combin. 27 (2008), no. 1, 1 -- 21. [37] A. I. Seven, Reflection group relations arising from cluster algebras, Proc. Amer. Math. Soc. 144 (2016), no. 11, 4641 -- 4650. [38] B. M. Schein and B. Teclezghi, Endomorphisms of finite symmetric inverse semigroups, J. Algebra 198 (1997), no. 1, 300 -- 310. [39] D. F. Vatne, The mutation class of Dn quivers, Comm. Algebra 38 (2010), no. 3, 1137 -- 1146. [40] S. Wilcox, Cellularity of diagram algebras as twisted semigroup algebras, J. Algebra 309 (2007), no. 1, 10 -- 31. [41] C. C. Xi, Partition algebras are cellular, Compositio Math. 119 (1999), no. 1, 99 -- 109. [42] , Cellular algebras, available at https://webusers.imj-prg.fr/~bernhard.keller/ictp2006/lecturenotes/xi.pdf. Bing Duan: School of Mathematics and Statistics, Lanzhou University, Lanzhou 730000, P. R. China. E-mail address: [email protected] Jian-Rong Li: Dept. of Mathematics, The Weizmann Institute of Science, Rehovot 7610001, Israel; school of Mathematics and Statistics, Lanzhou University, Lanzhou 730000, P. R. China. E-mail address: [email protected] Yan-Feng Luo: School of Mathematics and Statistics, Lanzhou University, Lanzhou 730000, P. R. China. E-mail address: [email protected]
1607.05958	1	1607	2016-07-20T13:56:05	Restricted Poisson Algebras	[ "math.RA" ]	We re-formulate Bezrukavnikov-Kaledin's definition of a restricted Poisson algebra, provide some natural and interesting examples, and discuss connections with other research topics.	math.RA	math	RESTRICTED POISSON ALGEBRAS Y.-H. BAO, Y. YE AND J.J. ZHANG Abstract. We re-formulate Bezrukavnikov-Kaledin's definition of a restricted Poisson algebra, provide some natural and interesting examples, and discuss connections with other research topics. 0 2 ] . A R h t a m [ 1 v 8 5 9 5 0 . 7 0 6 1 : v i X r a 0. Introduction The Poisson bracket was introduced by Poisson as a tool for classical dynamics in 1809 [Po]. Poisson geometry has become an active research field during the past 50 years. The study of Poisson algebras over R or a field of characteristic zero [L-GPV] also has a long history, and is closely related to noncommutative algebra, differential geometry, deformation quantization, number theory, and other areas. The notion of a restricted Poisson algebra was introduced about ten years ago in an important paper of Bezrukavnikov-Kaledin [BK] in the study of deformation quantization in positive characteristic. The project in [BK] is a natural extension of the classical deformation quantization of symplectic (or Poisson) manifolds. Our first goal is to better understand Bezrukavnikov-Kaledin's definition via a Lie algebraic approach. We re-interpret their definition in the following way. Throughout the paper let k be a base field of characteristic p ≥ 3. All vector spaces and algebras are over k. Definition 0.1. Let (A, {−, −}) be a Poisson algebra over k. (1) We call A a weakly restricted Poisson algebra if there is a p-map operation x 7→ x{p} such that (A, {−, −}, (−){p}) is a restricted Lie algebra. (2) We call A a restricted Poisson algebra if A is a weakly restricted Poisson algebra and the p-map (−){p} satisfies (E0.1.1) for all x ∈ A. {p} (x2) = 2xpx{p} The formulation in (E0.1.1) is slightly simpler than the original definition. We will show that Definition 0.1(2) is equivalent to [BK, Definition 1.8] in Lemma 3.7. Generally it is not easy to prove basic properties for restricted Poisson algebras. For example, it is not straightforward to show that the tensor product preserves the restricted Poisson structure. Different formulations are helpful in understanding and proving some elementary properties. 2010 Mathematics Subject Classification. 17B63, 17B50. Key words and phrases. Restricted Poisson algebras, deformation quantization, restricted Lie algebra, restricted Lie-Rinehart algebra, restricted Poisson Hopf algebra. 1 2 Y.-H. BAO, Y. YE AND J.J. ZHANG Since there are several structures on a restricted Poisson algebra, it is delicate to verify all compatibility conditions. There are not many examples given in the literature. Our second goal is to provide several canonical examples from different research subjects. Restricted Poisson algebras can be viewed as a Poisson version of restricted Lie algebras, so the first few examples come from restricted (or modular) Lie theory. Let L be a restricted Lie algebra over k. Then the trivial extension algebra k ⊕ L (with L2 = 0) is a restricted Poisson algebra. More naturally we have the following. Theorem 0.2 (Theorem 6.5). Let L be a restricted Lie algebra over k and let s(L) be the p-truncated symmetric algebra. Then s(L) admits a natural restricted Poisson structure induced by the restricted Lie structure of L. To use ideas from Poisson geometry, it is a good idea to extend the restricted Poisson structure to the symmetric algebra of a restricted Lie algebra [Example 6.2]. The following result is slightly more general and useful in other setting. Theorem 0.3 (Theorem 6.1). Let T be an index set and A = k[xi i ∈ T ] be a polynomial Poisson algebra. If, for each i ∈ T , there exists γ(xi) ∈ A such that xi = adγ(xi), then A admits a restricted Poisson structure (−){p} : A → A such adp that xi {p} = γ(xi) for all i ∈ T . The next example comes from deformation theory, which is also considered in [BK]. See (E7.0.1) for the definition of M p n(f ). Proposition 0.4 (Proposition 7.1). Let (A, ·, {−, −}) be a Poisson algebra over k and let (A[[t]], ∗) be a deformation quantization of A. If M p n(f ) = 0 for 1 ≤ n ≤ p−2 and f p is central in A[[t]] for all f ∈ A, then A admits a restricted Poisson structure. A Lie-Rinehart algebra is an algebraic counterpart of a Lie algebroid, and appears naturally in the study of Gerstenhaber algebras, Batalin-Vilkovisky algebras and Maurer-Cartan algebras [Hu1, Hu2]. In this paper, we also study the relationship between restricted Poisson algebras and restricted Lie-Rinehart algebras. Theorem 0.5 (Theorem 8.2). Let (A, ·, {−, −}, (−){p}) be a restricted Poisson algebra. If the Kahler differential ΩA/k is free over A, then (A, ΩA/k, (−)[p]) is a restricted Lie-Rinehart algebra, where the p-map of ΩA/k is determined by (xdu)[p] = xpdu{p} + (xdu)p−1(x)du, for all xdu ∈ ΩA/k. The category of restricted Poisson algebras is a symmetric monoidal category. In particular, the tensor product of two restricted Poisson algebras is again a restricted Poisson algebra [Proposition 9.2]. Advances of algebra are tremendously benefited from geometric viewpoint and methods and vice versa. Restricted Poisson algebras are, to some extent, the algebraic counterpart of symplectic differential geometry in positive characteristic. Following this idea, restricted Poisson-Lie groups should correspond to restricted Poisson Hopf algebras which connects both Poisson geom- etry in positive characteristic and quantum groups at the root of unity. Hence, it is meaningful to introduce the notion of a restricted Poisson Hopf algebra, see Definition 9.3. One natural example of such an algebra is given in Example 9.4. RESTRICTED POISSON ALGEBRAS 3 The paper is organized as follows. Sections 1 and 2 contain basic definitions about restricted Lie algebras and Poisson algebras. In Section 3, we re-introduce the notion of a restricted Poisson algebra. In Sections 4 to 7, we give several natural examples. In Section 8, we prove Theorem 0.5. The notion of a restricted Poisson Hopf algebra is introduced in Section 9. The Appendix contains a combinatorial proof of (E7.2.1) which is needed for Example 7.2. 1. Restricted Lie algebras We give a short review about restricted Lie algebras. Lie algebras over a field of positive characteristic often admit an additional struc- ture involving a so-called p-map. The Lie algebra together with a p-map is called a restricted Lie algebra, which was first introduced and systematically studied by Jacobson [J1, J2]. Let L := (L, [−, −]) be a Lie algebra over k. For convenience, for each x ∈ L, we denote by adx : L → L the adjoint representation given by adx(y) = [x, y] for all y ∈ L. We recall the definition of a restricted Lie algebra from [J1, Section 1]. As always, we assume that k is of positive characteristic p ≥ 3. Definition 1.1. [J1] A restricted Lie algebra (L, (−)[p]) over k is a Lie algebra L over k together with a p-map (−)[p] : x 7→ x[p] such that the following conditions hold: (1) adp (2) (λx)[p] = λpx[p] for all λ ∈ k, x ∈ L; x = adx[p] for all x ∈ L; (3) (x + y)[p] = x[p] + y[p] + Λp(x, y), where Λp(x, y) = and si(x, y) is the coefficient of ti−1 in the formal expression adp−1 tx+y(x). si(x,y) i for all x, y ∈ L p−1Pi=1 For simplicity of notation, we write all multiple Lie brackets with the notation (E1.1.1) for x1, · · · , xn ∈ L. Clearly, adi [x1, [x2, · · · , [xn−1, xn] · · · ]] =: [x1, x2, · · · , xn−1, xn], x(y) = [x, · · · , x , y] for every i. Under this notation, we have (E1.1.2) and, hence (E1.1.3) i copies {z } [x1, · · · , xp−2, y, x], 1 #(x) [x1, · · · , xp−1, xp]. #{kxk=x}=i−1 si(x, y) = Xxk =x or y Λp(x, y) = Xxk =x or y xp−1=y,xp =x Note that Λp(x, y) is denoted by L(x, y) in [BK] and denoted by σ(x, y) in [Ho2]. Another way of understanding Λp(x, y) is to use the universal enveloping algebra U(L) of the Lie algebra L. By [Ho2, Condition (3) on p. 559], (E1.1.4) for all x, y ∈ L ⊂ U(L), where (−)p is the multiplicative p-th power in U(L). Λp(x, y) = (x + y)p − xp − yp We give a well-known example which will be used later. 4 Y.-H. BAO, Y. YE AND J.J. ZHANG Example 1.2. Let A be an associative algebra over k. We denote by AL the induced Lie algebra with the bracket given by [x, y] := xy − yx, for all x, y ∈ A. Then (AL, (−)p) is a restricted Lie algebra, where (−)p is the Frobenius map given by x 7→ xp. In [J2, Theorem 11], Jacobson gives a necessary and sufficient condition in which an ordinary Lie algebra over k is restricted. Lemma 1.3. [J2, Theorem 11] Let L be a Lie algebra with a k-basis {xi}i∈I for some index set I. Suppose that there exists an element γ(xi) ∈ L for each i ∈ I such that adp xi = adγ(xi). Then there exists a unique restricted structure on L such that x[p] i ∈ I. i = γ(xi) for all 2. Poisson algebras and their enveloping algebras In this section we recall some definitions. We refer to [L-GPV] for some basics concerning Poisson algebras. Definition 2.1. [L-GPV, Definition 1.1] Let A be a commutative algebra over k. A Poisson structure on A is a Lie bracket {−, −} : A ⊗ A → A such that the following Leibniz rule holds (E2.1.1) {xy, z} = x{y, z} + y{x, z}, ∀ x, y, z ∈ A. The algebra A together with a Poisson structure is called a Poisson algebra. The Lie bracket {−, −} (which replaces [−, −] in the previous section) is called the Poisson bracket, and the associative multiplication of A is sometimes denoted by ·. In this paper all Poisson algebras are commutative as an associative algebra. Recall that the Kahler differentials, denoted by ΩA/k, of a commutative algebra A over k is an A-module generated by elements (or symbols) dx for all x ∈ A, and subject to the relations d(x + y) = dx + dy, d(xy) = xdy + ydx, dλ = 0, where x, y ∈ A, λ ∈ k ⊆ A. When (A, {−, −}) is a Poisson algebra, the Kahler differentials ΩA/k admits a Lie algebra structure with Lie bracket given by [xdu, ydv] = x{u, y}dv + y{x, v}du + xyd{u, v} for all xdu, ydv ∈ ΩA/k. Moreover, A is also a Lie module over ΩA/k with the action given by (xdu).a = x{u, a} for all xdu ∈ ΩA/k, a ∈ A. In fact, the pair (A, ΩA/k) is a Lie-Rinehart algebra in the following sense. Definition 2.2. [Do, Definition 1.5] A Lie-Rinehart algebra over A is a pair (A, L), where A is a commutative associative algebra over k, L is a Lie algebra equipped with the structure of an A-module together with a map called anchor α : L → Derk(A) which is both an A-module and a Lie algebra homomorphism such that (E2.2.1) [X, aY ] = a[X, Y ] + α(X)(a)Y RESTRICTED POISSON ALGEBRAS 5 for all a ∈ A and X, Y ∈ L. Note that, in the situation of Poisson algebra, the anchor map α : ΩA/k → Der(A) is given by (E2.2.2) α(xdu)(z) = x{u, z} for all xdu ∈ ΩA/k and z ∈ A. Let (A, L) be a Lie-Rinehart algebra. In [Ri], Rinehart introduced the notion of universal enveloping algebra U(A, L) of (A, L), which is an associative k-algebra satisfying the appropriate universal property, see [Hu1] for more details. We recall the definition next. Denote by A ⋊ L the semi-direct product of the Lie algebra L and the L-module A. More precisely, A ⋊ L is the direct sum of A and L as a vector space, and the Lie bracket is given by [(a, X), (b, Y )] = (X(b) − Y (a), [X, Y ]) for all (a, X), (b, Y ) ∈ A⋊L. Let (U(A⋊L), ι) be the universal enveloping algebra of the Lie algebra A ⋊ L, where ι : A ⋊ L → U(A ⋊ L) is the canonical embedding. We consider the subalgebra U +(A ⋊ L) (without unit) generated by A ⋊ L. Moreover, A⋊L has the structure of an A-module via a(a′, X) = (aa′, aX) for all a, a′ ∈ A and X ∈ L. The (universal) enveloping algebra U(A, L) associated to the Lie-Rinehart algebra (A, L) is defined to be the quotient U(A, L) = U +(A ⋊ L) (ι((a, 0))ι((a′, X)) − ι(a(a′, X))) . Note that (1A, 0) becomes the algebra identity of U(A, L). There are two canonical maps ι1 : A → U(A, L), a 7→ (a, 0) and ι2 : L → U(A, L), X 7→ (0, X). Observe that ι1 is an algebra homomorphism and ι2 is a Lie algebra homomorphism. Moreover, we have the following relations ι1(a)ι2(X) = ι2(aX), and [ι2(X), ι1(a)] = ι1(X(a)) for all a ∈ A and X ∈ L. As a consequence of [Ri, Theorem 3.1], we have the following. Lemma 2.3. Let (A, L) be a Lie-Rinehart algebra and U(A, L) the enveloping algebra of (A, L). If L is a projective A-module, then the Lie algebra homomorphism ι2 : L → U(A, L) is injective. It is worth spending half page to re-state the above construction for Poisson algebras since it is needed later. Denote by A ⋊ ΩA/k the semidirect product of A and ΩA/k with the Lie bracket given by [(a, xdu), (b, ydv)] = (x{u, b} − y{v, a}, x{u, y}dv + y{x, v}du + xyd{u, v}) for (a, xdu), (b, ydv) ∈ A ⋊ ΩA/k. The Poisson enveloping algebra of A, denoted by P(A) (which is a new notation), is defined to be the enveloping algebra of the Lie-Rinehart algebra (A, ΩA/k), which can be realized as an associated algebra P(A) := U(A, ΩA/k) = U +(A ⋊ ΩA/k)/J, 6 Y.-H. BAO, Y. YE AND J.J. ZHANG where U(A ⋊ ΩA/k) is the universal enveloping algebra of the Lie algebra A ⋊ ΩA/k, and J is the ideal generated by (E2.3.1) (a, 0)(b, xdu) − (ab, axdu) for all a, b ∈ A, xdu ∈ ΩA/k [MM, Ri]. Here we have two maps ι1 : A → A ⋊ ΩA/k → P(A), ι1(a) = (a, 0) and ι2 : ΩA/k → A ⋊ ΩA/k → P(A), ι2(xdu) = (0, xdu). Then ι1 and ι2 are homomorphisms of associative algebras and Lie algebras, re- spectively. Moreover, we have (E2.3.2) (E2.3.3) for all x, y ∈ A. ι1({x, y}) = [ι2(dx), ι1(y)], ι2(d(xy)) = ι1(x)ι2(dy) + ι1(y)ι2(dx) If ΩA/k is a projective A-module, then the canonical map ι2 : ΩA/k → P(A) is injective [Lemma 2.3]. It follows that ΩA/k can be seen as a Lie subalgebra of P(A). We now recall the definition of a free Poisson algebra, see [Sh, Section 3]. Let V be k-vector space. Let Lie(V ) be the free Lie algebra generated by V . The free Poisson algebra generated by V , denoted by F P (V ), is the symmetric algebra over Lie(V ), namely (E2.3.4) F P (V ) = k[Lie(V )]. The following universal property is well-known [Sh, Lemma 1, p. 312]. Lemma 2.4. Let A be a Poisson algebra and V be a vector space. Every k-linear map g : V → A extends uniquely to a Poisson algebra morphism G : F P (V ) → A such that g factors through G. In [Sh, Section 3], the notion of a free Poisson algebra is defined by the universal property stated in Lemma 2.4, and then Shestakov proved that the free Poisson algebra can be constructed by using (E2.3.4) [Sh, Lemma 1, p. 312]. In [Sh], Shestakov also considered the super (or Z2-graded) version of Poisson algebras. For each associative commutative algebra A, let Ap denote the subalgebra gen- erated by {f p f ∈ A}. The free Poisson algebras have the following special property. Lemma 2.5. Let A be a free Poisson algebra F P (V ). (1) ΩA/k is a free module over A. As a consequence, the Lie algebra map ι2 : ΩA/k → P(A) is injective. (2) The kernel of d : A → ΩA/k is Ap. Proof. (1) Since A is a commutative polynomial ring, ΩA/k is free over A. (The proof is omitted). The consequence follows from Lemma 2.3. (2) Check directly. (cid:3) RESTRICTED POISSON ALGEBRAS 7 Let V be a k-vector space. There are two gradings that can naturally be assigned to F P (V ). The first one is determined by deg1(x) = 1, ∀ 0 6= x ∈ Lie(V ). Since F P (V ) is the symmetric algebra associated to Lie(V ), the above extends to an N-grading on F P (V ). Since the Lie bracket {−, −} has degree −1, the Poisson bracket on F P (V ) has degree −1. Note that the multiplication on F P (V ) is homogeneous with respect to deg1. For the second grading, we assume that deg2(x) = 1, ∀ 0 6= x ∈ V and make the free Lie algebra Lie(V ) N-graded (namely, [−, −] is homogeneous of degree zero). Then we extend the N-grading to F P (V ) so that both the Poisson bracket and the multiplication are homogeneous of degree zero. Let {vi}i∈I be a k-basis of V and {xj}j∈J a k-basis of Lie(V ). Let A be the free Poisson algebra F P (V ) and let Ac be the Ap-submodule of A generated by monomials xi1 n , for x1, · · · , xn ∈ Lie(V ), which are not in Ap. 1 · · · xin Recall that (E2.5.1) {f1, f2, · · · , fn} := {f1, {f2, · · · , {fn−1, fn}} for all fi ∈ A. Lemma 2.6. Let A be a free Poisson algebra F P (V ). (1) Let f1, · · · , fn be polynomials in vi (not xi). If p does not divide n − 1, then {f1, f2, · · · , fn} ∈ Ac. (2) Let f, g be polynomials in vi. Then Λp(f, g) ∈ Ac. (3) The following elements are in Ac for any polynomials in f, g, h in vi: (a) Λp(f, g), Λp(f 2, g2), Λp(f 2 + g2, 2f g). (b) Λp(f g, h), Λp((f g)2, h2), Λp((f g)2 + h2, 2f gh). (c) Λp(f g, f h). Proof. (1) By linearity, we may assume that all fs are monomials in {vi} ⊆ V . Then deg1fs = deg2fs for s = 1, · · · , n. Let F := {f1, f2, · · · , fn}. Then deg1F = −n + 1 + deg2F. Since p does not divide n − 1, p can not divide both deg1F and deg2F . This implies that F ∈ Ac. (2) Note that Λp(f, g) is a linear combination of terms of the form (E2.5.1) when n = p and fi = f or g. By part (1), Λp(f, g) ∈ Ac. (3) This is a special case of part (2) for different choices of f, g. (cid:3) 3. Restricted Poisson algebras, Definition In this section we present a formulation of a restricted Poisson algebra that is equivalent to [BK, Definition 1.8]. Inspired by the notion of a restricted Lie algebra, we first introduce the definition of a weakly restricted Poisson structure over a field k of characteristic p ≥ 3. 8 Y.-H. BAO, Y. YE AND J.J. ZHANG If A admits a p-map Definition 3.1. Let (A, ·, {−, −}) be a Poisson algebra. (−){p} : A → A such that (A, {−, −}, (−){p}) is a restricted Lie algebra, then A is called a weakly restricted Poisson algebra. This definition requires no compatibility condition between the p-map (−){p} and the multiplication ·. We will see that an additional requirement is very natural from a Lie algebraic point of view. Lemma 3.2. Let (A, ·, {−, −}) be a Poisson algebra and let x, y ∈ A. (1) If there exists ex and ey in A such that adp xy = adxp ey+yp ex+Φp(x,y), adp where x = adex and adp y = adey, then (E3.2.1) Φp(x, y) = (xp + yp)Λp(x, y) − 1 2 (Λp(x2, y2) + Λp(x2 + y2, 2xy)). In particular, adp x2 = ad2xp ex. (2) If (A, ·, {−, −}) is a weakly restricted Poisson algebra, then (E3.2.2) ad(xy){p} = adxpy{p}+ypx{p}+Φp(x,y). In particular, (E3.2.3) ad(x2){p} = ad2xpx{p}. Proof. (1) We first prove the assertion when x = y. By the Leibniz rule, we have ad(f g) = f adg + gadf for any f, g ∈ A. Clearly, adp x2 = (2xadx)p = (2x)p(adx)p = 2xpadp x = 2xpadex = ad2xp ex. In the general case, considering the universal enveloping algebra of the Lie algebra (A, {−, −}) and using (E1.1.4), we get adΛp(f,g) = adp g for any f, g ∈ A. Therefore, f +g − adp f − adp adxp ey+yp ex+Φ(x,y) =adxp ey+yp ex+(xp+yp)Λp(x,y)− 1 2 (Λp(x2,y2)+Λp(x2+y2,2xy)) =xpadp y + ypadp + 1 2(cid:16)adp =xpadp y + ypadp x+y − adp x − adp y) x2 + adp x + (xp + yp)(adp 2xy − adp x + (xp + yp)(adp y2 + adp (x+y)2(cid:17) x+y − adp xy − (x + y)padp x+y x − adp y) + xpadp x + ypadp y + adp =adp xy, which completes the proof. (2) It is an immediate consequence of (1). (cid:3) Concerning the notation Φp in (E3.2.1), we also have the following characteriza- tion by considering the Poisson enveloping algebra. Proposition 3.3. Let A be a Poisson algebra and P(A) the Poisson enveloping algebra of A. Then, for all x, y ∈ A, we have (E3.3.1) ι2(dΦp(x, y)) = (ι2(d(xy)))p − ι1(xp)(ι2(dy))p − ι1(yp)(ι2(dx))p RESTRICTED POISSON ALGEBRAS 9 Proof. By the definition of P(A), we have (0, dx2)p = (0, 2xdx)p = ((2x, 0)(0, dx))p = (2x, 0)p(0, dx)p = 2(xp, 0)(0, dx)p and hence (E3.3.2) (ι2(dx2))p = 2ι1(xp)(ι2(dx))p for any x ∈ A. It follows that the equation (E3.3.1) holds when x = y. Considering the Frobenius map of P(A), we have (ι2(d(x + y)))p =(0, d(x + y))p = ((0, dx) + (0, dy))p =(0, dx)p + (0, dy)p + Λp((0, dx), (0, dy)) =(ι2(dx))p + (ι2(dy))p + ι2(dΛp(x, y)) since ι2 is a homomorphism of Lie algebras. By the above computation and (E3.3.2), we have (ι2(d(x + y)2))p = 2ι1((x + y)p)(ι2(d(x + y)))p = 2ι1(xp + yp)((ι2(dx))p + (ι2(dy))p + ι2(dΛp(x, y))). By a direct calculation and (E3.3.2), (ι2(d(x + y)2))p =(ι2(dx2 + dy2 + 2d(xy)))p =(ι2(dx2 + dy2))p + (ι2(2d(xy)))p + ι2(dΛp(x2 + y2, 2xy)) =(ι2(dx2))p + (ι2(dy2))p + ι2(dΛp(x2, y2)) + 2(ι2(d(xy)))p + ι2(dΛp(x2 + y2, 2xy)) =2ι1(xp)(ι2(dx))p + 2ι1(yp)(ι2(dy))p + ι2(dΛp(x2, y2)) + 2(ι2(d(xy)))p + ι2(dΛp(x2 + y2, 2xy)) Comparing the above two equations, we get (ι2(d(xy)))p+ (ι2(d(Λp(x2, y2) + Λp(x2 + y2, 2xy)))) 1 2 =ι1(xp)(ι2(dy))p + ι1(yp)(ι2(dx))p + ι1(xp + yp)ι2(dΛp(x, y)) =ι1(xp)(ι2(dy))p + ι1(yp)(ι2(dx))p + ι2(d((xp + yp)Λp(x, y))). Therefore, ι2(dΦp(x, y)) = ι2(d((xp + yp)Λp(x, y) − 1 2 = (ι2(d(xy)))p − ι1(xp)(ι2(dy))p − ι1(yp)(ι2(dx))p. (Λp(x2, y2) + Λp(x2 + y2, 2xy)))) This finishes the proof. (cid:3) For a weakly restricted Poisson algebra, it is desired to consider some compati- bility between the p-map and the associative multiplication. By removing ad from (E3.2.3) (which can be done in some cases), we obtain (E3.4.1) below. Similarly, if we remove ad from (E3.2.2), we obtain (E3.5.1) below. Both Lemma 3.2 and Proposition 3.3 suggest the following definition. Following Lemma 3.2(2), condi- tion (E3.4.1) is forced. 10 Y.-H. BAO, Y. YE AND J.J. ZHANG Definition 3.4. Let (A, ·, {−, −}, (−){p}) be a weakly restricted Poisson algebra over k. We call A a restricted Poisson algebra, if, for every x ∈ A, (E3.4.1) {p} (x2) = 2xpx{p}. In this case, the p-map (−){p} is a restricted Poisson structure on A. Next we give another description of condition (E3.4.1) which is convenient for some computation. Proposition 3.5. Let A be a weakly restricted Poisson algebra. (1) Suppose (E3.4.1) holds. Then (λ1A){p} = 0, for all λ ∈ k. (2) Equation (E3.4.1) holds for all x ∈ A if and only if every pair of elements (x, y) in A satisfies (E3.5.1) (xy){p} = xpy{p} + ypx{p} + Φp(x, y). As a consequence, A is a restricted Poisson algebra if and only if (E3.5.1) holds. (3) Suppose (E3.5.1) holds. Then (E3.5.2) (xn){p} = nx(n−1)px{p} for all n. As a consequence, (xp){p} = 0 for all x ∈ A. (4) If (1A){p} = 0, then (E3.5.1) holds for pairs (x, λ1A) and (λ1A, x) for all x ∈ A and all λ ∈ k. Proof. (1) Clearly, 1A (λ1A){p} = λp1A {p} = 0. {p} = 2 · 1p A1A {p} and hence 1A {p} = 0. For every λ ∈ k, (2) The " if " part is trivial since Φp(x, x) = 0 for any x ∈ A. Next, we show the " only if " part. By (E3.4.1) and Definition 1.1(3), we have ((x + y)2) {p} = 2(x + y)p(x + y){p} = 2(xp + yp)(x{p} + y{p} + Λp(x, y)) Since (A, {−, −}, (−){p}) is a restricted Lie algebra, 1.1(2,3) that it follows from Definition ((x + y)2) {p} = (x2 + y2 + 2xy) {p} = (x2 + y2) {p} + 2p(xy){p} + Λp(x2 + y2, 2xy) {p} {p} + (y2) + Λp(x2, y2) + 2p(xy){p} + Λp(x2 + y2, 2xy) = (x2) = 2xpx{p} + 2ypy{p} + Λp(x2, y2) + 2(xy){p} + Λp(x2 + y2, 2xy) Comparing the above two equations and using 2 6= 0, we obtain equation(E3.5.1). (3) This follows by induction. (4) First of all, (λ1A){p} = λp1A the fact Φp(λ1A, x) = Φp(x, λ1A) = 0. {p} = 0 for all λ ∈ k. The assertion follows by (cid:3) Remark 3.6. Several remarks are collected below. RESTRICTED POISSON ALGEBRAS 11 (1) As in the paper [BK], we assume that p ≥ 3. So the polynomial Φp(x, y) in (E3.2.1) is well-defined. When p = 3, we have Φ3(x, y) = x2y{y, y, x} + xy2{x, x, y} + xy{x, y}2. For any p > 3, it is too long to write out all terms like above. (2) Considering Φp(x, y) as an element in F P (V ), where V = kx ⊕ ky, it is homogeneous of degree p + 1 with respect to deg2 and homogeneous of degree 2p with respect to deg1. (3) In [BK, Definition 1.8], Bezrukavnikov-Kaledin defines a restricted Poisson algebra as a weakly restricted Poisson algebra (A, {−, −}, (−){p}) such that the p-map satisfies (E3.6.1) (xy){p} = xpy{p} + ypx{p} + P (x, y) for all x, y ∈ A. Here P (x, y) is a canonical quantized polynomial deter- mined by [BK, (1.3)]. We will show that Equation (E3.6.1) is equivalent to (E3.5.1). (4) The polynomial P (x, y) is defined implicitly, but it follows from [BK, (1.3)] that P (x, x) = 0. Therefore a restricted Poisson algebra in the sense of [BK, Definition 1.8] is a restricted Poisson algebra in the sense of Definition 3.4. (5) There are other interpretations of Φp(x, y). By using the equation xy = 1 4 [(x + y)2 − (x − y)2] we obtain that (E3.6.2) (xy){p} = xpy{p} + ypx{p} + Φ′ p(x, y) where (E3.6.3) Φ′ p(x, y) = 1 4 Λp((x + y)2, −(x − y)2) + 1 2 One can show that Φp(x, y) = Φ′ p(x, y) in the free Poisson algebra generated by x and y. ((xp + yp)Λp(x, y) − (xp − yp)Λp(x, −y)). (6) The following is clear by definition. (a) Λp(x, y) = Λp(y, x) for all x, y ∈ A. (b) If {x, y} = 0, then Λp(x, y) = 0. (c) Φp(x, y) = Φp(y, x) for all x, y ∈ A. (d) If {x, y} = 0, then Φp(x, y) = 0. Lemma 3.7. Definitions of restricted Poisson algebras in Definition 3.4 and [BK, Definition 1.8] are equivalent. Proof. Let P (x, y) be the polynomial defined in [BK, (1.3)]. By Proposition 3.5(2), it remains to show that P (x, y) = Φp(x, y). Let Lie(V ) be the free Lie algebra over a vector space V and consider the tensor (free) algebra T (V ) as a universal enveloping algebra over Lie(V ). Then we have a Poincar´e-Birkhoff-Witt filtration on T (V ). The free quantized algebra Q•(V ) is the Rees algebra associated to this filtration. By definition, for each n, Fn := FnT (V ) = k ⊕ L•(V ) ⊕ (L•(V ))2 ⊕ · · · ⊕ (L•(V ))n. 12 Y.-H. BAO, Y. YE AND J.J. ZHANG We are omitting the symbol h which represents the natural embedding h : F• → F•+1 in the Rees ring. Taking V = kx ⊕ ky, we have the following computation inside the Rees ring (xp + yp)2+(Λp(x, y))2 + Λp(x, y)(xp + yp) + (xp + yp)Λp(x, y) =(xp + yp + Λp(x, y))2 =(x + y)2p =(x2 + y2 + xy + yx)p =(x2 + y2)p + (xy + yx)p + Λp(x2 + y2, xy + yx) =x2p + y2p + Λp(x2, y2) + (xy)p + (yx)p + Λp(xy, yx) + Λp(x2 + y2, xy + yx), and hence (xy)p + (yx)p − xpyp − ypxp =Λp(x, y)(xp + yp) + (xp + yp)Λp(x, y) + (Λp(x, y))2 − Λp(x2, y2) − Λp(xy, yx) − Λp(x2 + y2, xy + yx). On the other hand, [x, y]p = (xy − yx)p = (xy)p − (yx)p + Λp(xy, −yx). So we have 2P (x, y) = 2((xy)p − xpyp) =Λp(x, y)(xp + yp) + (xp + yp)Λp(x, y) − Λp(x2, y2) − Λp(x2 + y2, xy + yx) + (Λp(x, y))2 − Λp(xy, yx) − Λp(xy, −yx) + [x, y]p − [xp, yp]. In fact, it is easily seen that (Λp(x, y))2 ∈ F2, [x, y]p ∈ Fp. On the other hand, [xp, yp] = adp x(yp)) = −adp−1 x (adp y(x)) ∈ F1, where adx(y) = [x, y]. By the equation (E1.1.3), we have Λp(xy, yx) = Xxk=xy or yx 1 #(xy) adx1 · · · adxp−2([yx, xy]). Since [yx, xy] = [yx, yx + [x, y]] = [yx, [x, y]] ∈ F2, we have Λp(xy, yx) ∈ Fp. Similarly, Λp(xy, −yx) ∈ Fp. By definition [BK, (1.3)], P (x, y) is homogeneous of degree p + 1. Therefore, after removing lower degree components, 2P (x, y) = Λp(x, y)(xp + yp) + (xp + yp)Λp(x, y) − Λp(x2, y2) − Λp(x2 + y2, xy + yx). Since the multiplication is commutative in a Poisson algebra, we have P (x, y) = (xp + yp)Λp(x, y) − 1 2 (Λp(x2, y2) + Λp(x2 + y2, 2xy)) = Φp(x, y). (cid:3) RESTRICTED POISSON ALGEBRAS 13 4. Elementary properties and examples We start with something obvious. Definition 4.1. Let (A, ·, {−, −}, (−){p}) be a restricted Poisson algebra. A Pois- son ideal I of A is said to be restricted, if x{p} ∈ I for any x ∈ I. The proofs of the following three assertions are easy and omitted. Lemma 4.2. Let A be a restricted Poisson algebra. Suppose that I is a Poisson ideal of A that is generated by {xi i ∈ S} as an ideal of the commutative ring A. If xi {p} ∈ I for any i ∈ S, then I is a restricted Poisson ideal. Proposition 4.3. Let A be a restricted Poisson algebra and I a restricted Poisson ideal of A. Then the quotient Poisson algebra A/I is a restricted Poisson algebra. As a consequence, we have. Corollary 4.4. Let f : A → A′ be a homomorphism of restricted Poisson algebras. Then Kerf is a restricted Poisson ideal of A. Let Ap be the subalgebra of A generated by {f p f ∈ A} – the image of the Frobenius map. Lemma 4.5. Let A be a Poisson algebra and f, g, h ∈ A. Then the following hold: (1) f pΦp(g, h) − Φp(f g, h) + Φp(f, gh) − hpΦp(f, g) = 0. (2) If f is in the Poisson center of A, then f pΦp(g, h) = Φp(f g, h) = Φp(g, f h). (3) Φp(f, g + h) − Φp(f, g) − Φp(f, h) = Λp(f g, f h) − f pΛp(g, h). Proof. It is clear that (2) is a consequence of (1). It suffices to show assertions (1) and (3) for the free Poisson algebra F P (A) since there is a surjective Poisson algebra map F P (A) → A [Lemma 2.4]. So the hypothesis becomes that f, g, h are in a k-space V sitting inside a free Poisson algebra F P (V ). When A is a free Poisson algebra F P (V ), by Lemma 2.5(1), ι2 is injective. It follows from Lemma 2.5(2) that (a) the kernel of the map is Ap. A d−→ ΩA/k ι2−→ P(A) Let {vi}i∈S be a basis of the V . Let Ac be the Ap-submodule of A = F P (V ) defined before Lemma 2.6. Then (b) Ac ∩ Ap = {0} and Λp(x, y) ∈ Ac for all x, y ∈ k[V ] by Lemma 2.6(2). Now we prove (1) and (3) under conditions (a) and (b). (1) For all f, u ∈ A, d(f pu) = f pdu and ι2(d(f pu)) = (f p, 0)(0, du) ∈ P(A). By Proposition 3.3, we have ι2(d(f pΦp(g, h))) = (f p, 0)(0, d(gh))p − (f pgp, 0)(0, dh)p − (f php, 0)(0, dg)p, 14 Y.-H. BAO, Y. YE AND J.J. ZHANG ι2(dΦp(f g, h)) = (0, d(f gh))p − ((f g)p, 0)(0, dh)p − (hp, 0)(0, d(f g))p, ι2(dΦp(f, gh)) = (0, d(f gh))p − (f p, 0)(0, d(gh))p − ((gh)p, 0)(0, df )p, ι2(d(hpΦp(f, g)) = (hp, 0)(0, d(f g))p − (hpf p, 0)(0, dg)p − (hpgp, 0)(0, df )p. for all f, g, h ∈ V . It follows that ι2(d(f pΦp(g, h) − Φp(f g, h) + Φp(f, gh) − Φp(f, g)hp)) = 0. By condition (a), we get X := f pΦp(g, h)− Φp(f g, h)+ Φp(f, gh)− hpΦp(f, g) ∈ Ap. By definition, X is in the Ap-submodule generated by Λp(x, y) for all x, y ∈ A, or in Ac as given in condition (b). But Ap ∩ Ac = {0} by condition (b), we obtain that X = 0 and that the desired identity holds. (3) The proof of part (3) is similar to the proof of (1) and is omitted. (cid:3) Proposition 4.6. Let A be a weakly restricted Poisson algebra. (1) If (x, y) satisfies (E3.5.1), then so do (x, λy) and (λx, y) for all λ ∈ k. (2) Let f, g, h ∈ A. Suppose that (f, g) and (g, h) satisfy (E3.5.1). Then (f g, h) satisfies (E3.5.1) if and only if (f, gh) does. (3) If (f, g) and (f, h) satisfies (E3.5.1), then so does (f, g + h). (3') If (g, f ) and (h, f ) satisfies (E3.5.1), then so does (g + h, f ). (4) Fix an x ∈ A and let Rx be the set of y ∈ A such that (x, y) satisfies (E3.5.1). Then Rx is a k-subspace of A. (4') Fix an x ∈ A and let Lx be the set of y ∈ A such that (y, x) satisfies (E3.5.1). Then Lx is a k-subspace of A. Proof. (1) Assuming (E3.5.1) for (x, y), we have (xλy){p} = λ(xy){p} = λp(xy){p} = λp(xpy{p} + ypx{p} + Φp(x, y)) = xp(λy){p} + (λy)px{p} + λpΦp(x, y)) = xp(λy){p} + (λy)px{p} + Φp(x, λy)), where the last equation is Lemma 4.5(2). So (x, λy) satisfies (E3.5.1). Similarly for (λx, y). (2) By symmetry, we only prove one implication and assume that (f g, h) satisfies (E3.5.1). We show next that (f, gh) satisfies (E3.5.1): (f (gh)){p} = ((f g)h){p} = (f g)ph{p} + hp(f g){p} + Φp(f g, h) = (f g)ph{p} + hp(f pg{p} + gpf {p} + Φp(f, g)) + Φp(f g, h) = f pgph{p} + f phpg{p} + gphpf {p} + Φp(f g, h) + hpΦp(f, g) = f pgph{p} + f phpg{p} + gphpf {p} + f pΦp(g, h) + Φp(f g, h) by Lemma 4.5(1) = f p(gph{p} + hpg{p} + Φp(g, h)) + (gh)pf {p} + Φp(f, gh) = f p(gh){p} + (gh)pf {p} + Φp(f, gh). RESTRICTED POISSON ALGEBRAS 15 (3) Assume (f, g) and (f, h) satisfies (E3.5.1). Then (f (g + h)){p} =(f g + f h){p} =(f g){p} + (f h){p} + Λp(f g, f h) =f pg{p} + gpx{p} + Φp(f, g) + xph{p} + hpf {p} + Φp(f, h) + Λp(f g, f h) =f p(g{p} + h{p} + Λp(g, h)) + (g + h)pf {p} + Φp(f, g + h) =f p(g + h){p} + (g + h)pf {p} + Φp(f, g + h), where the second last equality is deduced from Lemma 4.5(3). So (f, g + h) satisfies (E3.5.1). (3') is equivalent to (3). (4) Let Rx = {y ∈ A (E3.5.1) holds for the pair (x, y)}. By Proposition 4.6(1), we have (i) if y ∈ Rx, then so is λy for all λ ∈ k. By Proposition 4.6(3), (ii) if g, h ∈ Rx, then so is g + h. By (i) and (ii) above, Rx is a k-subspace of A. (4') This is true because Lx = Rx. The following result will be used several times. (cid:3) Theorem 4.7. Let A be a weakly restricted Poisson algebra. Let b := {bi}i∈S be a k-basis of A. If (E3.5.1) holds for every pair (x, y) ⊆ b, then A is a restricted Poisson algebra. Proof. We need to show that (E3.5.1) holds for all x, y ∈ A. First we fix any x ∈ b and let Rx = {y ∈ A (E3.5.1) holds for the pair (x, y)}. By Proposition 4.6(4), Rx is a k-subspace of A. By hypothesis, we see that b ⊆ Rx. Since b is a basis of A, Rx = A. Next we fix y ∈ A and consider Ly = {x ∈ A (E3.5.1) holds for the pairs (x, y)}. Similarly, by Proposition 4.6(4'), Ly is a k-subspace. It contains b because Rx = A for all x ∈ b (see the first paragraph). Hence, Ly = A. This means that (E3.5.1) holds for all pairs (x, y) in A. Therefore A is a restricted Poisson algebra. (cid:3) One of the main goals of this paper is to provide some interesting examples of restricted Poisson algebras. In the rest of this section we give some elementary (but nontrivial) examples. We would like to give a gentle warning before the examples. We have checked that all p-maps given below satisfy (E3.5.1), however our proofs 16 Y.-H. BAO, Y. YE AND J.J. ZHANG are tedious computations and therefore omitted. On the other hand, since the p- maps are explicitly expressed by partial derivatives, one can verify the assertions with enough patient. More sophisticated examples are given in later sections. Example 4.8. Let A = k[x, y] be a polynomial algebra in two variables x, y, where the (classical) Poisson bracket is given by (E4.8.1) {f, g} = fxgy − fygx. for all f, g ∈ A, and fx, fy are the partial derivative of f with respect to the variables x and y, respectively. (The bracket defined in (E4.8.1) was the original Poisson bracket studied by many people including Poisson [Po] when k = R.) (1) Let k be a base field of characteristic 3. For every f ∈ A, we define (E4.8.2) f {3} = f 2 x fyy + f 2 y fxx + fxfyfxy, where fxx, fyy and fxy are the second order partial derivatives of f . Then (A, ·, {−, −}, (−){3}) is a restricted Poisson algebra. (2) Let k be a base field of characteristic 5. For every f ∈ A, define f {5} =f 4 1 f2f1222 + f 2 1 f 2 2 f1122 + f1f 3 2 f1112 + f 4 2 f1111 (E4.8.3) 1 f2222 + f 3 + f12(f 3 − f1f22(f 2 − f2f11(f 2 + 2(f 2 1 f222 − f 2 1 f2f122 − f1f 2 1 f122 − 2f1f2f112 + f 2 2 f112 − 2f2f1f122 + f 2 2 f112 + f 3 2 f111) 1 f222) 2 f111) 12 − f11f22)(f 2 1 f22 − 2f1f2f12 + f 2 2 f11), where fi1i2···ik denotes the k-th order partial derivative of f with respect to the variables xi1 , xi2 , · · · , xik . Then (A, ·, {−, −}, (−){5}) is a restricted Poisson algebra. See Example 7.2 for general p. It would be interesting to understand the meaning of (E4.8.2) and (E4.8.3) and to find its connection with other subjects. The next two are slight generalizations of the previous example. Example 4.9. Suppose char k = 3 and let A = k[x, y] be a polynomial Poisson algebra in two variables x, y, where the Poisson bracket is given by {f, g} = ϕ(fxgy − fygx), and ϕ = λx + µy + ν, λ, µ, ν ∈ k. For every f ∈ A, we define (E4.9.1) f {3} = λϕfxf 2 xfy + ϕ2(f 2 x fyy + f 2 y + µϕf 2 y fxx + fxfyfxy) + λ2yf 3 y + µ2xf 3 x . Then (A, ·, {−, −}, (−){3}) is a restricted Poisson algebra. Example 4.10. Suppose char k = 3 and let A = k[x1, x2, · · · , xn] be a Poisson algebra, where the Lie bracket is given by {xi, xj } = 2cij ∈ k with cij + cji = 0 for 1 ≤ i, j ≤ n. Clearly, {f, g} =P1≤i,j≤n cij(figj − fjgi) for f, g ∈ A, where fi denotes the partial derivative of f with respect to the variable xi for i = 1, 2, · · · , n. Then A is a restricted Poisson algebra with the p-map given by f {3} = X1≤i,j,k,l≤n cijcklfifkfjl RESTRICTED POISSON ALGEBRAS 17 for any f ∈ A, where fjl is the second partial derivation of f with respect to the variables xj and xl. 5. Existence and uniqueness of restricted structures By Lemma 3.2(2), a weakly restricted Poisson structure on a Poisson algebra is very close to a restricted Poisson structure (up to a factor in the Poisson center). In this section, we study the existence and uniqueness of (weakly) restricted Poisson structure. First we consider the trivial extension. Lemma 5.1. Let A be a Poisson algebra and A = k1A ⊕ m as a Lie algebra decomposition. (1) If x 7→ x{p} is a restriction p-map of the Lie algebra m, then it can naturally {p} = 0. As a consequence, A is a weakly be extended to A by defining 1A restricted Poisson algebra. (2) If, further, the p-map on m satisfies (E3.4.1), then so does the extended p-map on A. In this case, A is a restricted Poisson algebra. Proof. (1) This follows from Lemma 1.3. For all λ ∈ k and x ∈ m, the p-map is defined by (λ1A + x){p} = x{p}. (2) We check (E3.4.1) for elements in A as follows: ((λ1A + x)2) {p} = (λ21A + 2λx + x2) {p} {p} {p} = (2λx){p} + (x2) = (2λx + x2) = 2λpx{p} + 2xpx{p} = 2(λ1A + x)px{p} = 2(λ1A + x)p(λ1A + x){p}. Therefore A is a restricted Poisson algebra. (cid:3) The following example is immediate. Example 5.2. (1) Let L be a restricted Lie algebra and let A = k1A ⊕L where associate product L2 = 0. Then A is a Poisson algebra in the obvious way. Both sides of (E3.4.1) are zero for elements in L (since L2 = 0). By Lemma 5.1(2), A is a restricted Poisson algebra. (2) Considering a special case when L = kx + ky is a solvable Lie algebra with [x, y] = x. For f = λ1x + λ2y ∈ L, we define the p-map by f {p} = λp−1 2 (λ1x + λ2y). It is straightforward to check that (L, (−){p}) is a restricted Lie algebra. Let A = k1A ⊕ L. Then, by part (1), A is a restricted Poisson algebra. As a commutative algebra, A = k[x,y] (x2,xy,y2) with k-linear basis {1, x, y}. The Poisson bracket is given by {x, y} = x. Let L be a restricted Lie algebra. It is well known that the p-map of L is unique up to a semilinear map from L to Z(L), where Z(L) is the center of L. Recall that a semilinear map γ : L → Z(L) means that for any x, y ∈ A, λ ∈ k, γ(x + y) = γ(x) + γ(y), 18 Y.-H. BAO, Y. YE AND J.J. ZHANG γ(λx) = λpγ(x). The following lemma is well-known and easy to prove. Lemma 5.3. Let (L, (−)[p]) be a restricted Lie algebra. (1) Let (−){p} be another restricted Lie structure on L. Then there is a maps γ : L → Z(L) such that (−){p} = (−)[p] + γ. (2) Let γ be a map from L to Z(L). Then (−)[p] +γ is a restricted Lie structure on L if and only if γ is a semilinear map from L to Z(L). Let A be a Poisson algebra over k and Z(A) the center of A. Observe that Z(A) is a left A-module with the action given by A × Z(A) → Z(A), (a, z) 7→ apz. A semilinear map ψ : A → Z(A) is called a Frobenius derivation of A with the value in Z(A) provided that ψ(ab) = apψ(b) + bpψ(a) for any a, b ∈ A. For example, if ψ0 : A → A is a derivation, then ψ : A → Z(A), defined by ψ(a) = (ψ0(a))p for all a ∈ A, is a Frobenius derivation of A with the value in Z(A). By Lemma 5.3(1), any two restricted Poisson structures on A differ by a semi- linear map γ which appears in the next proposition. Proposition 5.4. Let (A, ·, {−, −}, (−){p}) be a restricted Poisson algebra and γ a map from A to itself. Then the map (−){p} + γ is a restricted Poisson structure if and only if γ is a Frobenius derivation of A with value in Z(A). Proof. Let (−){p}1 : A → A be another p-map such that (A, ·, {−, −}, (−){p}1) is also a restricted Poisson algebra. Since (−){p}1 and (−){p} are restricted structures on Lie algebra (A, {−, −}), γ = (−){p}1 −(−){p} is a semilinear map from A to Z(A) by Lemma 5.3. Moreover, for any x, y ∈ A, (xy){p}1 = xpy{p}1 + ypx{p}1 + Φp(x, y), and γ(xy) =(xy){p}1 − (xy){p} =xp(y{p}1 − y{p}) + yp(x{p}1 − x{p}) =xpγ(y) + ypγ(x) It follows that γ is a Frobenius derivation of A with values in Z(A). Conversely, it follows from Lemma 5.3 that the map (−){p} +γ is also a restricted Lie structure on (A, {−, −}), since γ is a semilinear map from A to Z(A) and (−){p} is a p-map of Lie algebra (A, {−, −}). Moreover, for any x, y ∈ A, (xy){p} + γ(xy) = xp(y{p} + γ(y)) + yp(x{p} + γ(x)) + Φp(x, y). It follows that the Poisson algebra A together with the map (−){p} +γ is a restricted structure. (cid:3) By Proposition 5.4, the p-map of a restricted Poisson algebra is unique up to Frobenius derivations. RESTRICTED POISSON ALGEBRAS 19 Remark 5.5. Let (A, ·, {−, −}, (−){p}) be a restricted Poisson algebra and let γ : A → Z(A) be a semilinear map. Suppose that γ is not a Frobenius derivation (which is possible for many A) and defines a new p-map (−)′{p} = (−){p} + γ. Then by Proposition 5.4, (A, ·, {−, −}, (−)′{p}) is not a restricted Poisson algebra, but it is still a weakly restricted Poisson algebra by Lemma 5.3(2). 6. Restricted Poisson algebras from restricted Lie algebras We start with a general result. Theorem 6.1. Let A = k[xi i ∈ T ] be a polynomial Poisson algebra with an index set T . If for each i ∈ T , there exists γ(xi) ∈ A such that adp xi = adγ(xi), then A admits a restricted Poisson structure (−){p} such that xi {p} = γ(xi) for all i ∈ T . Proof. First we show that A has a weakly restricted Poisson structure, and then verify that the weakly restricted Poisson structure satisfies (E3.5.1). For the sake of simplicity, we assume that T = {1, 2, · · · , n}. To apply Lemma 1.3, we choose a canonical monomial k-basis of A, which is We define (xi1 1 xi2 2 · · · xin n ) inductively on the degree i1 + i2 + · · · + in such that 1 xi2 2 · · · xin n i1, i2, · · · , in ≥ 0}. {xi1 {p} adp (xi1 1 xi2 2 ···xin n ) = ad (xi1 1 xi2 2 ···xin n ) {p}, and therefore get the restricted Lie structure on (A, {−, −}) by Lemma 1.3. For convenience, we denote xI = xi1 n and I = i1 +· · ·+in for I = (i1, · · · , in). If I = 0, then xI = 1, we define 1{p} = 0 and if I = 1, then xI = xi for {p} = γ(xi) for each 1 ≤ i ≤ n. By hypothesis, 2 · · · xin 1 xi2 some 1 ≤ i ≤ n. We define xi adp xI = ad(xI ){p} for any I with I = 0, 1. Proceeding by induction and assuming that (xI ) has been defined such that adp xI = ad(xI ){p} for any xI with I ≤ m. For each monomial xI of degree m + 1, we assume that k is the smallest subscript such that ik ≥ 1 in I, i.e. I = (0, · · · , 0, ik, · · · , in) and define {p} (E6.1.1) {p} (xI ) =xp + (xik −1 k xik+1 k+1 · · · xin n )pxk {p} {p} k k(xik −1 + Φp(xk, xi1−1 xik+1 k+1 · · · xin n ) xik+1 k+1 · · · xin n ). xik+1 k+1 · · · xin k By Lemma 3.2(1) for (x, y) = (xk, xik−1 have adp Lemma 1.3, A has a weakly restricted Poisson structure. n ) and the above definition, we xI = ad(xI ){p} for any I = m + 1, which completes the induction. By k Now let b be the set of all monomials, which is a k-basis of A. We prove that (E3.5.1) holds for any pair of elements (x, y) in b by induction on degx + degy. If x or y is 1, then (E3.5.1) holds trivially, which also takes care of the case when m := degx + degy ≤ 1. Suppose that the assertion holds for m and now assume that degx + degy = m + 1. Let xy = xik n where ik > 0. By (E6.1.1), the pair (xk, xik−1 n ) satisfies (E3.5.1). By symmetry, we may assume that x = xkg. Then the above says that the pair (xk, gy) satisfies (E3.5.1). By xik+1 k+1 · · · xin k+1 · · · xin k xik+1 k 20 Y.-H. BAO, Y. YE AND J.J. ZHANG induction hypothesis, the pairs (xk, g) and (g, y) satisfy (E3.5.1). By Proposition 4.6(2), (x, y) = (xkg, y) satisfies (E3.5.1). By induction, (E3.5.1) holds for any two elements in b. Finally the main statement follows from Theorem 4.7. (cid:3) As a consequence, we have the following. Example 6.2. Let L be a restricted Lie algebra. We claim that the polynomial Poisson algebra A := k[L] (also denoted by S(L)) is a restricted Poisson algebra. Let {xi}i∈I be a basis of L. Then, for each i, there is an γ(xi) := x[p] i ∈ L such that adp xi = adγ(xi) when restricted to L. Since A is a polynomial ring over L, both adp xi = adγ(xi) holds when applying to A. The claim follows from Theorem 6.1 and there is a unique restricted structure (−){p} on A such that xi and adγ(xi) extends uniquely to derivations of A. Thus adp x{p} = x[p], ∀ x ∈ L. Let V be a vector space. Then the free restricted Lie algebra RLie(V ) can be defined by using the universal property or by taking the restricted Lie subalgebra of the free associative algebra generated by V with the p-map being the p-powering map. Now we can define the free restricted Poisson algebra generated by V . Definition 6.3. Let V be a k-space. The free restricted Poisson algebra generated by V is defined to be F RP (V ) = k[RLie(V )]. The following universal property is standard [Sh, Lemma 1, p. 312]. Lemma 6.4. Let A be a restricted Poisson algebra and V be a vector space. Every k-linear map g : V → A extends uniquely to a restricted Poisson algebra morphism G : F RP (V ) → A such that g factors through G. Continuing Example 6.2, when L is a restricted Lie algebra over k and S(L) := k[L] the symmetric algebra on L, then S(L) admits an induced restricted Poisson structure. One natural setting in positive characteristic is to replace the symmetric algebra S(L) by the truncated (or small) symmetric algebra s(L). By definition, when L has a k-basis {xi}i∈I , (E6.4.1) s(L) = k[xi i ∈ I]/(xp i , ∀ i ∈ I). It is easily seen that s(L) admits a Poisson structure with the bracket {f, g} = Pi,j ( ∂f ∂xi ∂g ∂xj − ∂f ∂xj ∂g ∂xi ){xi, xj } for any f, g ∈ s(L). Next we show that s(L) has a natural restricted Poisson structure. Theorem 6.5. Let L be a restricted Lie algebra over k of characteristic p and let s(L) be the Poisson algebra with the bracket induced by L. Then s(L) admits a natural restricted Poisson structure induced by the p-map of L. Proof. By Example 6.2, S(L) has an induced restricted Poisson algebra structure. By (E6.4.1), s(L) = S(L)/J where J is the Poisson ideal generated by xp (xp i ){p} = 0. By Lemma 4.2, J is a restricted Poisson ideal as desired. i for all i ∈ I. By Proposition 3.5(3), (cid:3) RESTRICTED POISSON ALGEBRAS 21 7. Restricted Poisson algebras from deformation quantization We will produce more examples in this section. Let A be a commutative (associative) algebra. Let k[[t]] be the formal power series ring in one variable t. A formal deformation of A means an associative algebra A[[t]] over k[[t]] with multiplication, denoted by mt, satisfying mt(a ⊗ b) = a ∗ b = ab + m1(a, b)t + · · · + mn(a, b)tn + · · · , for all a, b ∈ A ⊂ A[[t]]. We should view A[[t]] as the power series ring in one variable t with coefficients in A where the associative multiplication mt (or the star product ∗) is induced by a family of k-bilinear maps {mi : A ⊗ A → A}i≥0 with m0(a, b) = ab. Define a bilinear map {−, −} : A⊗ A → A by setting {a, b} = m1(a, b)− m1(b, a). It is easy to check that A together with the bracket {−, −} is a Poisson algebra. Then (A, {−, −}) is called the classical limit of (A[[t]], mt), and (A[[t]], mt) is called a deformation quantization of the Poisson algebra (A, {−, −}). For every f ∈ A, we write the p-power of f as (E7.0.1) f ∗p = M p n(f )tn = f p + M p 1 (f )t + M p 2 (f )t2 + · · · ∈ A[[t]] ∞Xn=0 where M p i (f ) ∈ A for all i = 0, 1, 2, · · · . Proposition 7.1. Let (A, ·, {−, −}) be a Poisson algebra over k and let (A[[t]], ∗) n(f ) = 0 for 1 ≤ n ≤ p − 2 and f p is be a deformation quantization of A. If M p central in A[[t]] for all f ∈ A, then A admits a restricted Poisson structure. Proof. Recall that f ∗ g = mn(f, g)tn ∈ A[[t]] for all f, g ∈ A, where mn(f, g) ∈ A for all n. By the definition of the deformation quantization, ∞Pn=0 {f, g} = lim t→0 f ∗ g − g ∗ f t = m1(f, g) − m1(g, f ). for all f, g ∈ A. Considering the Frobenius map f 7→ f ∗p in A[[t]], we get (E7.1.1) for all f, g ∈ A. , g]∗ [f ∗p, g]∗ = [f, · · · , f p copies {z } Since [f, g]∗ = {f, g}t (mod t2) and [−, −]∗ is k[[t]]-bilinear, we have , g]∗ ≡ {f, · · · , f , g}tp (mod tp+1). [f, · · · , f p copies {z } p copies {z } By assumption, M p fact that (E7.1.1) or adf ∗p(g) = (adf )p(g), it follows that n(f ) = 0 for 1 ≤ n ≤ p − 2 and f p is central in A[[t]]. Using the {M p p−1(f ), g}tp = {f, · · · , f p copies {z } , g}tp (mod tp+1) 22 or (E7.1.2) Y.-H. BAO, Y. YE AND J.J. ZHANG {M p p−1(f ), g} = m1(M p p−1(f ), g) − m1(g, M p p−1(f )) = {f, · · · , f p copies , g} {z } for all g ∈ A. We define f {p} = M p f 7→ M p p−1(f ) gives rise to a restricted Poisson structure on A. p−1(f ) for any f ∈ A, and prove that the map Note that Definition 1.1(1) follows from (E7.1.2). Definition 1.1(2) follows from the fact that (λf )∗p = λpf ∗p. For condition in Definition 1.1(3), we consider the Frobenius map of A[[t]], and get a restricted Lie structure of (A[[t]], [−, −]∗). It follows from Example 1.2 that Computing the coefficients of tp−1 of the above equation, we get (f + g)∗p − f ∗p − g∗p = Λ∗ p(f, g). (E7.1.3) as desired. (f + g){p} − f {p} − g{p} = Λp(f, g) Finally it remains to show (E3.4.1). By assumption, M p n(f ) = 0 for all 1 ≤ n ≤ p − 2. We compute the coefficient of tp−1 in the expression of f ∗,2p as follows: f ∗,2p = f ∗p ∗ f ∗p = (f p + tp−1M p ≡ f 2p + 2f pM p p−1(f ) + · · · ) ∗ (f p + tp−1M p p−1(f )tp−1 (mod tp) p−1(f ) + · · · ) Assume that f ∗ f = f 2 + tW , where W = m1(f, f ) + m2(f, f )t + · · · , and it follows that f ∗,2p = (f ∗2)∗p = (f 2 + tW )∗p = (f 2)∗p + (tW )∗p + Λ∗ ≡ f 2p + M p p(f 2, tW ) p−1(f 2)tp−1 (mod tp) Therefore, for all f ∈ A, f 2{p} = 2f pf {p}, which is (E3.4.1). We now give some explicit examples. (cid:3) Example 7.2. Let A = k[x, y] be a Poisson algebra over a field k of characteristic p ≥ 3 with the bracket given by {x, y} = 1. Let µ be the multiplication of the commutative algebra A[[t]]. By a direct cal- culation, the Poisson algebra A admits a deformation quantization (A[[t]], ∗) with the star product given by f ∗ g = µ(exp(t(∂1 ⊗ ∂2))(f ⊗ g)) for all f, g ∈ A, where ∂1 and ∂2 are the partial derivatives of f with respect to the variables x and y, respectively. To be precise, we have tn n! 1 f )(∂n 2 g). (∂n f ∗ g = X0≤n≤p−1 mn(f, g)tn = X0≤n≤p−1 Clearly, f p ∗ g = f pg = g ∗ f p for any f, g ∈ A and hence f p is central in A[[t]]. Moreover, for every f ∈ A, we claim that (E7.2.1) Mn(f ) = 0 for 1 ≤ n ≤ p − 2. RESTRICTED POISSON ALGEBRAS 23 The proof of the above is given in Appendix. By Proposition 7.1, A admits a restricted Poisson structure with the p-map f {p} = M p p−1(f ) for any f ∈ A. The p-map agrees with (E4.8.2) when p = 3 and (E4.8.3) when p = 5. The next is a generalization of the previous example. Example 7.3. Let A = k[x1, · · · , xm] be a Poisson algebra with the bracket given by {xi, xj} = cij ∈ k for 1 ≤ i < j ≤ n. By direct calculation, a deformation quantization (A[[t]], ∗) of the Poisson algebra A is given by f ∗ g = µexpt X1≤i<j≤m cij ∂i ⊗ ∂j (f ⊗ g) for all f, g ∈ A, where ∂i is the partial derivative of f with respect to the variable xi. This is well-defined by Remark 10.1(2). Clearly, f p ∈ A ⊂ A[[t]] is central for any f ∈ A. Being similar to the proof of Example 7.2 in Appendix, we have Mn(f ) = 0 for 1 ≤ n ≤ p − 2 and all f ∈ A. By Proposition 7.1, A admits a restricted Poisson structure with the p-map f {p} = M p p−1(f ) for any f ∈ A. When p = 3, the p-map is given in Example 4.10. Example 7.4. Let B2n = k[x1, · · · , x2n]/I be the p-truncated polynomial Poisson algebra in 2n variables over k, where the Poisson bracket is defined by {f, g} = (∂i(f )∂n+i(g) − ∂n+i(f )∂i(g)) nXi=1 for all f, g ∈ B2n, and I is generated by xp In [Sk], Skryabin introduced the notion of the normalized p-map on (B2n, {−, −}), say, 1{p} = 0 and f {p} ∈ m2 for all f ∈ m2, where m is the maximal ideal of B2n as an associative algebra. i , i = 1, · · · , 2n. We consider the Poisson algebra A = k[x1, · · · , x2n] in Example 7.3 with the bracket given by cij = δi+n,j for all 1 ≤ i < j ≤ 2n. Clearly, xp i is central and I is a i ){p} = 0 for all i ∈ I, and by Lemma Poisson ideal of A. By Proposition 3.5(3), (xp 4.2, I is a restricted Poisson ideal of A. Therefore, it follows from Proposition 4.3 that the Poisson algebra B2n admits a restricted Poisson structure. Clearly, this p-map is normalized. 8. Connection with restricted Lie-Rinehart Algebras Some definitions concerning Lie-Rinehart algebras were given in Section 2. Let A be a Poisson algebra and ΩA/k its Kahler differentials. Then the pair (A, ΩA/k) is a Lie-Rinehart algebra over k, where the anchor map α : ΩA/k → Der(A) is given in (E2.2.2). Dokas introduced the notion of a restricted Lie-Rinehart algebra and study its cohomology theory in [Do]. The goal of this section is to show that the Lie-Rinehart algebra (A, ΩA/k) admits a natural restricted structure if the Poisson algebra A is weakly restricted and ΩA/k is a free module over A. Let (L, (−)[p] and (L′, (−)[p]) be restricted Lie algebras. A map f : (L, (−)[p]) → (L′, (−)[p]) is called a restricted Lie homomorphism, if f is a Lie algebra homomor- phism and satisfies f (x[p]) = f (x)[p] for all x ∈ L. 24 Y.-H. BAO, Y. YE AND J.J. ZHANG The following definition was introduced by Dokas [Do]. Definition 8.1. [Do, Definition 1.7] A restricted Lie-Rinehart algebra (A, L, (−)[p]) over a commutative k-algebra A, is a Lie-Rinehart algebra over A such that (a) (L, (−)[p]) is a restricted Lie algebra over k, (b) the anchor map is a restricted Lie homomorphism, and (c) the following relation holds: (aX)[p] = apX [p] + (aX)p−1(a)X for all a ∈ A and X ∈ L. We now prove Theorem 0.5. Theorem 8.2. Let (A, ·, {−, −}, (−){p}) be a weakly restricted Poisson algebra. If the Kahler differential ΩA/k is a free, then the Lie-Rinehart algebra (A, ΩA/k, (−)[p]) is restricted, where the p-map of ΩA/k is defined by (xdu)[p] = xpdu{p} + (xdu)p−1(x)du, for all xdu ∈ ΩA/k. Proof. Since ΩA/k is a free A-module, ΩA/k can be embedded into the universal enveloping algebra U(A, ΩA/k) [Lemma 2.3]. By the proof of [Do, Proposition 2.2], it suffices to show that adp xdu(ydv) = [xpdu{p} + (xdu)p−1(x)du, ydv] for all xdu, ydv ∈ ΩA/k. By Hochschild's relation in [Ho1, Lemma 1], we get in U(A, L) the relation (ι2(xdu))p = ι1(xp)(ι2(du))p + ι2((xdu)p−1(x)du) for all xdu ∈ ΩA/k. Considering the Frobenius map of U(A, L), we have [(ι2(du))p, ι1(y)] = [ι2(du), · · · , ι2(du), ι1(y)] = ι1((adu)p(y)), and hence ι2(du)pι1(y) = ι1(y)ι2(du)p + ι1((adu)p(y)) for all du ∈ ΩA/k, y ∈ A. Moreover, for xdu, ydv ∈ ΩA/k ⊂ U(A, L), [ι1(xp)(ι2(du))p, ι2(ydv)] =ι1(xp)(ι2(du))pι1(y)ι2(dv) − ι1(y)ι2(dv)ι1(xp)(ι2(du))p =ι1(xp)(ι1(y)(ι2(du))p + ι1(adu)p(y))ι2(dv) − ι1(y)(ι1(xp)ι2(dv) + ι1({v, xp})(ι2(du))p =ι1(xpy)[(ι2(du))p, ι2(dv)] + ι2(xp(adu)p(y)dv) =ι1(xpy)ι2(adp =ι1(xpy)ι2(d(adp du(dv)) + ι2(xp(du)p(y)dv) u(v))) + ι2(xp(adu)p(y)dv), and therefore, ι2(adp xdu(ydv)) =[(ι2(xdu))p, ι2(ydv)] =[ι1(xp)(ι2(du))p + ι2((xadu)p−1(x)du), ι2(ydv)] =ι1(xpy)ι2(d(adp u(v))) + ι2(xp(adu)p(y)dv) + ι2([(xadu)p−1(x)du, ydv]) =ι2(xpyd(adp u(v))) + ι2(xp(adu)p(y)dv) RESTRICTED POISSON ALGEBRAS + ι2([(xadu)p−1(x)du, ydv]) =ι2([xpdu{p} + (xadu)p−1(x)du, ydv]), and hence adp xdu(ydv) = [xpdu{p} + (xadu)p−1(x)du, ydv] as desired. 25 (cid:3) For Poisson algebras A in Examples 4.8-4.10, 6.2, Theorem 6.5, Examples 7.2-7.4, it is automatic that ΩA/k is free over A. 9. Restricted Poisson Hopf algebras We first recall the definition of Poisson Hopf algebras. The notion of a Poisson Hopf algebra was probably first introduced by Drinfel'd [Dr1, Dr2] in 1980s, see also [DHL]. Definition 9.1. Let A be a Poisson algebra. We say that A is a Poisson Hopf algebra if (1) A is a Hopf algebra with usual operations ∆, ǫ, S. (2) ∆ : A → A⊗A and ǫ : A → k are Poisson algebra morphisms and S : A → A is a Poisson algebra anti-automorphism. To define restricted Poisson Hopf algebras, we need first show that tensor product of two restricted Poisson algebras is again a restricted Poisson algebra. Proposition 9.2. Let A and B be two restricted Poisson algebras. Then there is a unique restricted Poisson structure on A ⊗ B such that (E9.2.1) (a ⊗ b){p} = a{p} ⊗ bp + ap ⊗ b{p} for all a ∈ A and b ∈ B. Proof. First of all, it is well-known that A ⊗ B is a Poisson algebra with bracket defined by {a1 ⊗ b1, a2 ⊗ b2} = {a1, a2} ⊗ b1b2 + a1a2 ⊗ {b1, b2} for all a1, a2 ∈ A and b1, b2 ∈ B. Let {ai}i∈I (respectively, {bj}j∈J ) be a k-basis of A (respectively, B) and assume that 1A ∈ {ai}i∈I and 1B ∈ {bj}i∈J . Then {ai ⊗ bj}i∈I,j∈J is a k-basis of A ⊗ B. adp a⊗b(1 ⊗ d) b (d)) a⊗b is a derivation. For any c ⊗ d ∈ A ⊗ B, we have a⊗b(c ⊗ 1) + (c ⊗ 1)adp a(c) ⊗ bp) + (c ⊗ 1)(ap ⊗ adp For any a ∈ A and b ∈ B, adp a⊗b(c ⊗ d) = (1 ⊗ d)adp = (1 ⊗ d)(adp = (1 ⊗ d)(ada{p}(c) ⊗ bp) + (c ⊗ 1)(ap ⊗ adb{p} (d)) = (1 ⊗ d)(ada{p}⊗bp (c ⊗ 1)) + (c ⊗ 1)(adap⊗b{p}(1 ⊗ d)) = (1 ⊗ d)(ada{p}⊗bp (c ⊗ 1)) + (c ⊗ 1)(ada{p}⊗bp (1 ⊗ d)) + (1 ⊗ d)(adap⊗b{p}(c ⊗ 1)) + (c ⊗ 1)(adap⊗b{p}(1 ⊗ d)) = ada{p}⊗bp (c ⊗ d) + adap⊗b{p} (c ⊗ d) = ada{p}⊗bp+ap⊗b{p}(c ⊗ d). 26 Y.-H. BAO, Y. YE AND J.J. ZHANG In particular, adp ai⊗bj = ad(ai {p}⊗bp j +ap i ⊗bj {p}) for all i and j. Since {ai ⊗ bj}i∈I,j∈J is a k-basis of A ⊗ B, by Lemma 1.3, there is a unique weak restricted Poisson structure on A ⊗ B such that (E9.2.2) (ai ⊗ bj){p} = ai {p} ⊗ bp j + ap i ⊗ bj {p} for all i, j, which agrees with (E9.2.1). It remains to show that this weak restricted Poisson structure on A ⊗ B is indeed a restricted Poisson structure and (E9.2.1) holds. We first prove (E9.2.1). By (E9.2.2), (ai ⊗ 1){p} = ai {p} ⊗ 1. It follows from Definition 1.1 that (a ⊗ 1){p} = a{p} ⊗ 1 (E9.2.3) for all a ∈ A. By symmetry, (1 ⊗ b){p} = 1⊗b{p} for all b ∈ B. Since {ai⊗1, 1⊗bj} = 0, (E9.2.2) implies that the pair (ai ⊗ 1, 1 ⊗ bj) satisfies (E3.5.1). By Proposition 4.6(4), Rai⊗1 is a k-vector space; and by assumption, {bj} is a k-basis of B, we have that Rai⊗1 ⊇ B. Or, for any b ∈ B, the pair (ai ⊗ 1, 1 ⊗ b) satisfies (E3.5.1). By switching a and b and applying the same argument, one sees that any pair (a ⊗ 1, 1 ⊗ b) satisfies (E3.5.1). This means that (a ⊗ b){p} = (a ⊗ 1){p}(1 ⊗ b)p + (a ⊗ 1)p(1 ⊗ b){p} + Φp(a ⊗ 1, 1 ⊗ b) = (a ⊗ 1){p}(1 ⊗ b)p + (a ⊗ 1)p(1 ⊗ b){p} = a{p} ⊗ bp + ap ⊗ b{p}. So we proved (E9.2.1). For the rest, we claim that for any pair of elements (ai ⊗ bj, ak ⊗ bl), (E3.5.1) holds. By using (E9.2.3), (E3.5.1) holds for all pairs of the form (a ⊗ 1, a′ ⊗ 1). By symmetry, (E3.5.1) holds for all pairs of the form (1⊗b, 1⊗b′). By (E9.2.1), (E3.5.1) holds for pairs of the form (a⊗1, 1⊗b). Set f = a⊗1, g = a′⊗1 and h = 1⊗b for any a, a′ ∈ A and b ∈ B. Then (f, g), (g, h) and (f g, h) satisfy (E3.5.1). By Proposition 4.6(2), (f, gh) satisfies (E3.5.1). Or equivalently, (a⊗ 1, a′ ⊗ b) satisfies (E3.5.1). By symmetry, (1 ⊗ b, a ⊗ b′), (a ⊗ b, a′ ⊗ 1) and (a ⊗ b, 1 ⊗ b′) satisfy (E3.5.1). Recycle the letters and let f = a ⊗ b, g = a′ ⊗ 1 and h = 1 ⊗ b′. We have that (f, g), (g, h) and (f g, h) all satisfy (E3.5.1). By Proposition 4.6(2), (f, gh) satisfies (E3.5.1). By choosing special a, a′, b, b′ we have that (ai ⊗ bj, ak ⊗ bl) satisfies (E3.5.1) as desired. This says that every pair of elements from the k-basis {ai ⊗ bj}i∈I,j∈J satisfies (E3.5.1). By Theorem 4.7, the weak restricted Poisson structure on A ⊗ B is actually a restricted Poisson structure. The above proof shows that there is a unique restricted Poisson structure on A ⊗ B satisfying (E9.2.2). Since (E9.2.1) is a consequence of (E3.5.1), the assertion follows. (cid:3) Now it is reasonable to define a restricted Poisson Hopf algebra. Definition 9.3. A restricted Poisson algebra H is called a restricted Poisson Hopf algebra if there are restricted Poisson algebra maps ∆ : H → H ⊗ H, ǫ : H → k and restricted Poisson algebra anti-automorphism S : H → H such that H together with (∆, ǫ, S) becomes a Hopf algebra. RESTRICTED POISSON ALGEBRAS 27 One canonical example is the following. Example 9.4. Let L be a restricted Lie algebra. Then s(L) (given in Theorem 6.5) is a restricted Poisson Hopf algebra with the structure maps determined by ∆ : x → x ⊗ 1 + 1 ⊗ x, ǫ : x → 0, S : x → −x for all x ∈ L. It is straightforward to check that s(L) is a restricted Poisson Hopf algebra. Similarly, S(L) (given in Example 6.2) is a a restricted Poisson Hopf algebra with structure maps determined as above. 10. Appendix: The proof of (E7.2.1). Let A = k[x, y] be the Poisson algebra with the Poisson bracket determined by {x, y} = 1. Recall from Example 7.2 that the deformation quantization of A is isomorphic to an associative algebra (A[[t]], ∗) such that the star product of f, g ∈ A ⊂ A[[t]] is given by (E10.0.1) f ∗ g = µ(exp(t( ∂ ∂x ⊗ ∂ ∂y ))(f ⊗ g)) = µ ∞Xi=0 ti i! ∂if ∂xi ⊗ ∂ig ∂yi! , where µ : A ⊗ A → A is the multiplication operation of A. Define a sequence of Hasse-Schmidt derivations (or divided power derivations) ∂(i) 1 = 1 i!(cid:18) ∂ ∂x(cid:19)i and ∂(i) 2 = 1 i!(cid:18) ∂ ∂y(cid:19)i , ∀ i ≥ 0. Then all of them are k-linear operations from A to A. Using these we can re-write part of (E10.0.1) as (E10.0.2) exp(t( ∂ ∂x ⊗ ∂ ∂y ))(f ⊗ g) = ∞Xi=0 i!∂(i) 1 (f )∂(i) 2 (g) = i!∂(i) 1 (f )∂(i) 2 (g), p−1Xi=0 which is a sum of finitely many terms. Therefore (E10.0.1) is well-defined and the summation in (E10.0.1) is finite. Remark 10.1. Consider a generalization of (E10.0.1) in n variables. Let B = k[x1, · · · , xn] and ∂i = ∂ ∂xi for i = 1, · · · , n. (1) For each cij ∈ k, exp(t cij ∂i ⊗ ∂j)(f ⊗ g) is well-defined for all f, g ∈ B, and it is a sum of finitely many terms as in (E10.0.2). (2) For a set of {cij}1≤i,j≤n, (E10.1.1) exp(tXi,j cij ∂i ⊗ ∂j)(f ⊗ g) =Yi,j (exp(tcij ∂i ⊗ ∂j)) (f ⊗ g), which is well-defined for all f, g ∈ B and is a sum of finitely many terms in a similar fashion as (E10.0.2) (but more than p terms in general). 28 Y.-H. BAO, Y. YE AND J.J. ZHANG We now go back to the case of two variables. Clearly, ∂ ◦ µ = µ(∂ ⊗ id + id ⊗ ∂) for ∂ = ∂ ∂x or ∂ ∂y , and hence ( ∂ ∂x ⊗ ∂ ∂y )n(µ ⊗ id) = (µ ⊗ id)( ∂ ∂x ⊗ id ⊗ ∂ ∂y + id ⊗ ∂ ∂x ⊗ ∂ ∂y )n for all n ≥ 1. It follows that exp(t ∂ ∂x ⊗ ∂ ∂y and therefore, )(µ ⊗ id) = (µ ⊗ id) exp(t( ∂ ∂x ⊗ id ⊗ ∂ ∂y + id ⊗ ∂ ∂x ⊗ ∂ ∂y )), (f ∗ g) ∗ h = µ(exp(t ∂ ∂x ⊗ ∂ ∂y )((f ∗ g) ⊗ h) = µ(exp(t ∂ ∂x ⊗ ∂ ∂y = µ(µ ⊗ id)(exp(t( )((µ ⊗ id) exp(t ∂ ∂x ⊗ ∂ ∂x ⊗ id ⊗ ∂ ∂y + id ⊗ ∂ ∂y ∂ ∂x ⊗ id)(f ⊗ g ⊗ h))) ⊗ ∂ ∂y + ∂ ∂x ⊗ ∂ ∂y ⊗ id))(f ⊗ g ⊗ h)) In general, for k ≥ 2 and for f1, · · · , fk ∈ A ⊂ A[[t]], (E10.1.2) f1 ∗ · · · ∗ fk = µk(exp(t X1≤i<j≤k ∂i j(f1 ⊗ · · · ⊗ fk))). where ∂i j = id⊗i−1 ⊗ ∂ ∂x ∂ ∂y ⊗ id⊗j−i−1 ⊗ ⊗ id⊗k−j is a map from A⊗k to itself for all 1 ≤ i < j ≤ k, µ2(a ⊗ b) = (ab) for all a, b ∈ A (extended to a commutative multiplication on A[[t]]), and µk = µ2(µk−1 ⊗ id), k ≥ 3. Denote by M p n(f ) the coefficient of tn in f ∗p ∈ A[[t]], see (E7.0.1). For simplicity, we denote the map Φi1,··· ,in j1,··· ,jn = µp ◦ (∂i1 j1 ◦ · · · ◦ ∂in jn ) : A⊗p → A for 1 ≤ it < jt ≤ p, t = 1, · · · , n, n ≥ 1. It follows from the equation (E10.1.2) that (E10.1.3) M p n(f ) = for all 0 ≤ n ≤ p − 1. 1 n! X1≤ir <jr ≤p r=1,··· ,n Φi1,··· ,in j1,··· ,jn (f ⊗p) Claim 10.2. Retain the above notation. Then M p and all f ∈ A. n(f ) = 0 for all 1 ≤ n ≤ p − 2 This appendix is devoted to the proof of Claim 10.2. We need more notations. Recall that an oriented graph is a pair G = (V (G), E(G)), where V (G) is the set of vertices and E(G) is the set of edges. For α ∈ E(G), we denote by s(α) and t(α) the source and the target of α, respectively. Definition 10.3. An oriented graph G = (V (G), E(G)) is called a totally ordered graph (called tograph for short), if the set V (G) of vertices is totally ordered set with the ordering ≤ and for every α ∈ E(G), s(α) < t(α). RESTRICTED POISSON ALGEBRAS 29 Let G be a tograph (possibly with multiple edges). Being similar to usual ori- ented graphs, for each v ∈ V (G), we denote the indegree of v by the outdegree of v by the degree of v by d+ G(v) = #{α ∈ E(G) t(α) = v}, d− G(v) = #{α ∈ E(G) s(α) = v}, dG(v) = d+ G(v) + d− G(v). For u, v ∈ V (G), we denote by ν(u, v) the number of the edges with the source u and the target v, i.e. νG(u, v) = #{α ∈ E(G) s(α) = u, t(α) = v}. Let G and G′ be tographs. A bijection f : V (G) → V (G′) is called an iso- morphism, if f preserves the order of vertices and νG(u, v) = νG′ (f (u), f (v)) for all u, v ∈ V (G). Two tographs G and G′ are said to be isomorphic, denoted by G ∼= G′, provided that there exists an isomorphism between G and G′ . Clearly, the automorphism group of a tograph G is a trivial group since f preserves the order 1, · · · , G′ of vertices and V (G) is totally ordered. Suppose that G1, · · · , Gk and G′ m are the connected components of G and G′, respectively. Two tographs G and G′ are said to be equivalent, and denoted by G ∼ G′, if m = k and there exists a permutation σ ∈ Sm such that Gi and G′ σ(i) are isomorphic for each i = 1, · · · , m. Denote Γn = {(i1, · · · , in; j1, · · · , jn) 1 ≤ it < jt ≤ p, t = 1, · · · , n}. For each given (i1, · · · , in; j1, · · · , jn) ∈ Γn, we can assign a tograph, denoted by G( i1,··· ,in j1,··· ,jn ), where • the set of vertices V (G( i1,··· ,in j1,··· ,jn )) = {1, 2, · · · , p} with the usual ordering of natural numbers, and • the set of edges E(G( i1,··· ,in j1,··· ,jn )) = {(it, jt) t = 1, · · · , n}. We denote by Gn the set of tographs G( i1,··· ,in j1,··· ,jn ) for all (i1, · · · , in; j1, · · · , jn) ∈ Γn. We consider the lexicographical order on the set {(i, j) 1 ≤ i < j ≤ p}. To be precise, (i, j) < (i′, j′) if and only if i < i′ or i = i′, j < j′. n; j′ 1, · · · , j′ Let G and G′ be the tographs associated to elements (i1, · · · , in; j1, · · · , jn) and n) ∈ Γn, respectively. Clearly, G = G′ if and only if there 1, · · · , i′ (i′ exists a permutation σ ∈ Sn such that (i′ k) = (iσ(k), jσ(k)) for all k = 1, · · · , n. Therefore, for each (i1, · · · , in; j1, · · · , jn) ∈ Γn, there exists a permutation σ ∈ Sn such that G( i1,··· ,in j1,··· ,jn ) with (iσ(1), jσ(1)) ≤ · · · ≤ (iσ(n), jσ(n)). ) = G( iσ(1),··· ,iσ(n) jσ(1),··· ,jσ(n) k, j′ Lemma 10.4. Retain the above notation. (1) Let G be the tograph associated to (i1, · · · , in; j1, · · · , jn) ∈ Γn. Then (E10.4.1) Φi1,··· ,in j1,··· ,jn (f ⊗p) = ∂dG(1)f G(1)∂yd+ ∂xd− G(1) · · · ∂dG(p)f G(p)∂yd+ ∂xd− G(p) where d− of the vertex i ∈ V (G), respectively. G(i), d+ G(i) and dG(i) are the outdegree, the indegree and the degree 30 Y.-H. BAO, Y. YE AND J.J. ZHANG (2) Let G and G′ be the tographs associated to elements (i1, · · · , in; j1, · · · , jn) and (i′ 1, · · · , i′ n; j′ n) ∈ Γn, respectively. Then 1, · · · , j′ j1,··· ,jn (f ⊗p) = Φi′ Φi1,··· ,in 1,··· ,i′ j′ 1,··· ,j′ n n (f ⊗p) for all f ∈ A if and only if there exists a permutation σ ∈ Sn such that (d+ G′ (i), d− G′ (i)) = (d+ G(σ(i)), d− G(σ(i))) for all i = 1, · · · , p. Proof. (1) By the definition of Φi1,··· ,in (E10.4.1). j1,··· ,jn , we immediately get the desired equality (2) By (1), it is clear. For convenience, we denote G(i1,··· ,in j1,··· ,jn )(f ) = ∂xd− (f ⊗p) = G(i1,··· ,in j1,··· ,jn )(f ). ∂dG(1)f G(1)∂yd+ G(1) and hence Φi1,··· ,in j1,··· ,jn (cid:3) , · · · ∂dG(p)f G(p)∂yd+ ∂xd− G(p) Corollary 10.5. Let G and G′ be the tographs in Gn. If G is equivalent to G′, then G(f ) = G′(f ) for any f ∈ k[x, y]. Remark 10.6. The converse of Corollary 10.5 does not hold and a counter-example is G = G( 112 234 ) when p = 5 and n = 3. 234 ) and G′ = G( 113 Sketch Proof of Claim 10.2. For each (i1, · · · , in; j1, · · · , jn) ∈ Γn, we denote by N (i1,··· ,in j1,··· ,jn ) the number of the tographs which are equivalent to G( i1,··· ,in j1,··· ,jn We consider the decomposition ). G(i1,··· ,in j1,··· ,jn ) = G11 ∪ · · · ∪ G1k1 ∪ · · · ∪ Gr1 ∪ · · · ∪ Grkr , where Gis, for 1 ≤ s ≤ ki and 1 ≤ i ≤ r, are connected components of G with Gis ∼= Git for all 1 ≤ s, t ≤ ki, and Gis 6∼= Gjt for i 6= j. Denote V (Gis) = ni for each i = 1, · · · , r. By definition, a tograph G′ is equivalent to G(i1,··· ,in ), if and j1,··· ,jn only if for each i = 1, · · · , r, G′ admits ki connected components being isomorphic to Gi1. Therefore, by combinatorial counting, we have that N (i1,··· ,in j1,··· ,jn ) = p! (n1!)k1 · · · (nr!)kr k1! · · · kr! for each (i1, · · · , in; j1, · · · , jn) ∈ Γn. Clearly, if n ≤ p − 2, then the underly- ing graph of G( i1,··· ,in )) = n < p − 1 = j1,··· ,jn V (G( i1,··· ,in )) − 1. Therefore, r ≥ 2 and 1 ≤ kt, nt < p for each t. Therefore, by j1,··· ,jn (E10.1.3), ) is not connected since E(G( i1,··· ,in j1,··· ,jn M p n(f ) = 1 n! Φi1,··· ,in j1,··· ,jn (f ⊗p) G(f ) (i1,··· ,in;j1,··· ,jn)∈Γn X Q1≤u<v≤p ν(u, v)! n! = 1 n! XG∈Gn RESTRICTED POISSON ALGEBRAS 31 n! Q1≤u<v≤p ν(u, v)! p! (n1!)k1 · · · (nr!)kr k1! · · · kr! G(f ) 1 = n! X[G]∈Gn/∼ where the sum P[G]∈Gn/∼ (mod p) ≡0 means that one take one element in each equivalence class of Gn with respect to the relation ∼. (cid:3) Acknowledgments. Both Y.-H. Bao and Y. Ye were supported by NSFC (Grant No. 11401001, 11431010 and 11571329) and J.J. Zhang by the US National Science Foundation (grant No. DMS 1402863). References [BK] R. Bezrukavnikov and D. Kaledin, Fedosov quantization in positive characteristic, J. Amer. Math. Soc. 21 (2008), no. 2, 409–438. [Do] I. Dokas, Cohomology of restricted Lie-Rinehart algebras and the Brauer group, Adv. Math. 231 (2012) 2573–2592. [Dr1] V.G. Drinfel'd, Hopf algebras and the quantum Yang-Baxter equation (Russian), Dokl. Akad. Nauk SSSR 283 (1985), no. 5, 1060–1064. [Dr2] V.G. Drinfel'd, Quantum groups. In: International Congress of Mathematicians 1986, Amer. Math. Soc., Providence, RI: 1987, pp. 798–820. [DHL] H.-D. Doebner,J.D. Hennig and W. Lucke, Mathematical guide to quantum groups, Quan- tum groups (Clausthal, 1989), 29–63, Lecture Notes in Phys., 370, Springer, Berlin, 1990. [Ho1] G. Hochschild, Simple algebras with purely inseparable splitting fields of exponent 1, Trans. Amer. Math. Soc. 79 (1955) 477–489. [Ho2] G. Hochschild, Cohomology of restricted Lie algebras, Amer. J. Math. 76, (1954). 555–580. [Hu1] J. Huebschmann, Poisson cohomology and quantization, J. reine angew. Math. 408 (1990), 57–113. [Hu2] J. Huebschmann, Higher homotopies and Maurer-Cartan algebras: quasi-Lie-Rinehart, Ger- stenhaber, and Batalin-Vilkovisky algebras. The breadth of symplectic and Poisson geometry, 237–302, Progr. Math., 232, Birkhuser Boston, Boston, MA, 2005. [J1] N. Jacobson, Restricted Lie algebras of characteristic p, Trans. Amer. Math. Soc. 50 (1941) 15–25. [J2] N. Jacobson, Lie algebgras, John Wiley, New York, (1962). [L-GPV] C. Laurent-Gengoux, A. Pichereau, and P. Vanhaecke, Poisson Structures, Springer, Berlin Heidelberg, 2013. [MM] I. Moerdijk and J. Mrcun, On the universal enveloping algebra of a Lie algebroid, Proc. Amer. Math. Soc. 138 (2010), no. 9, 3135–3145. [Po] S.-D. Poisson, Sur la variation des constantes arbitraires dans les questions de m´ecanique, J. Ecole Polytechnique 8 Cah. 15 (1809), 266–344. [Ri] G.S. Rinehart, Differential forms on general commutative algebras, Trans. Amer. Math. Soc. 108 (1963), 195–222. [Sh] I.P. Shestakov, Quantization of Poisson superalgebras and speciality of Jordan Poisson su- peralgebras, Algebra Logika 32 (5) (1993) 571584; English translation: Algebra Logic 32 (5) (1993) 309317. [Sk] S. Skryabin, Invariant polynomial functions on the Poisson algebra in characteristic p, J. Algebra 256 (2002), 146–179. Bao: School of Mathematical Sciences, Anhui University, Hefei, 230601, China E-mail address: [email protected], [email protected] 32 Y.-H. BAO, Y. YE AND J.J. ZHANG Ye: School of Mathematical Sciences, University of Sciences and Technology of China, Hefei, 230026, China 1, Wu Wen-Tsun Key Laboratory of Mathematics, USTC, Chinese Academy of Sciences, Hefei, 230026, China 2 E-mail address: [email protected] Zhang: Department of Mathematics, Box 354350, University of Washington, Seattle, Washington 98195, USA E-mail address: [email protected]
1610.08156	3	1610	2016-11-30T04:18:30	On the number of generators of an algebra	[ "math.RA" ]	A classical theorem of Forster asserts that a finite module $M$ of rank $\leq n$ over a Noetherian ring of Krull dimension $d$ can be generated by $n + d$ elements. We prove a generalization of this result, with "module" replaced by "algebra". Here we allow arbitrary finite algebras, not necessarily unital, commutative or associative. Forster's theorem can be recovered as a special case by viewing a module as an algebra where the product of any two elements is $0$.	math.RA	math	ON THE NUMBER OF GENERATORS OF AN ALGEBRA URIYA A. FIRST AND ZINOVY REICHSTEIN Abstract. A classical theorem of Forster asserts that a finite module M of rank ≤ n over a Noetherian ring of Krull dimension d can be generated by n + d elements. We prove a generalization of this result, with "module" replaced by "algebra". Here we allow arbitrary finite algebras, not necessarily unital, commutative or associative. Forster's theorem can be recovered as a special case by viewing a module as an algebra where the product of any two elements is 0. 1. Introduction Throughout this paper R will denote a commutative Noetherian ring with 1. For p ∈ Spec R, R(p) will denote the fraction field of R/p. The starting point of this paper is the following classical theorem of Forster. Theorem 1.1 ([3]). Suppose R is Noetherian of Krull dimension d and let M be a finite R-module. If the R(p)-module M (p) := M ⊗R R(p) can be generated by n elements for every p ∈ Max R, then M can be generated by n + d elements. Swan [13] showed that Theorem 1.1 remains valid when the Krull dimension of R is replaced by the dimension of Max R. ∗ Further generalizations and refinements of Forster's Theorem can be found in [13, 2, 14, 5]. This note offers yet another generalization, replacing the finite R-module M by a finite R-algebra A, i.e., by a finitely generated R-module A with an R-bilinear multiplication map A × A → A. This bilinear map can be arbitrary; we do not require A to be commutative or associative, or to have a unit element. For p ∈ Spec R, write A(p) := A ⊗R R(p). Theorem 1.2. Assume dim Max R = d and let A be a finite R-algebra such that A(p) can be generated by n elements as a non-unital R(p)-algebra for every p ∈ Max R. Then A can be generated by n + d elements as a non-unital R-algebra. In the case where the multiplication map A × A → A is identically zero, we recover Forster's Theorem 1.1. Other applications of Theorem 1.2 can be found in Section 4. Before proceeding with the proof, we remark that our argument also proves the following variants of Theorem 1.2. (i) If A is a unital algebra, Theorem 1.2 remains valid if we replace "generated as a non-unital algebra" with "generated as a unital algebra". (ii) Both the original and the unital versions of Theorem 1.2 remain valid in the setting of [13], where A is equipped with a left Λ-module structure, Λ being an R-algebra, and generation means generation as an R-algebra carrying an additional Λ-module structure. Note that, unlike [13, The- orem 1], we do not require Λ to be finitely generated as an R-module. Department of Mathematics, University of British Columbia, Vancouver, CANADA E-mail addresses: [email protected], [email protected]. 2010 Mathematics Subject Classification. 17A01, 13C15, 13E15. The second author has been partially supported by an NSERC Discovery Grant. ∗Recall that the dimension of a topological space X is the maximal length d of a chain ∅ 6= X0 ( X1 ( · · · ( Xd ⊆ X of closed irreducible subsets (or −∞). The Krull dimension of R is the dimension of Spec R endowed with the Zariski topology. The maximal spectrum Max R is a subspace of Spec R, hence dim Max R 6 dim Spec R. 1 2 ON THE NUMBER OF GENERATORS OF AN ALGEBRA (iii) More generally, A can be taken to be a finite R-multialgebra, i.e., a finite right R-module equipped with an indexed family of homogeneous maps {fi : Ani → A}i∈I . Here we say that f : Ak → A is (m1, . . . , mk)-homogeneous, if f (a1r1, . . . , akrk) = f (a1, . . . , ak)rm1 for all a1, . . . , ak ∈ A and r1, . . . , rk ∈ R. Note that k = 0 is allowed; in this case f can be any map from A0 = 0 to A. The family is {fi}i∈I is clearly amenable to base change, hence A(p) carries the structure of an R(p)-multialgebra. A multisubalgebra of A is an R-submodule closed under each fi, and the multisubalgera generated by S ⊂ A is the smallest multisubalgbera containing S. Multialgebras can be used to encode many types of structures. For example, a (non-unital) R-algebra structure on A is a (1, 1)-homogeneous map A2 → A, a unit element can be specified by a map A0 → A, an involution by a (1)-homogeneous map A → A, a quadratic Jordan algebra structure by a (2, 1)-homogeneous map A2 → A, etc. Furthermore, if Λ is an associative R-algebra, then a left Λ-module structure, as in (ii), can be represented by the family of (1)-homogeneous maps {fλ : A → A}λ∈Λ, given by fλ(a) = λa. . . . rmk 1 k 2. Preliminary Lemmas Let A be a finite R-algebra. For p ∈ Spec R and a ∈ A, denote the image of a in A(p) by a(p). Lemma 2.1. Let a1, . . . , an ∈ A. Then a1, . . . , an generate A as an R-algebra if and only if for all p ∈ Max R, the elements a1(p), . . . , an(p) generate A(p) as an R(p)-algebra. Proof. Let B be the R-subalgebra generated by a1, . . . , an. The map B(p) → A(p) induced by the inclusion B ֒→ A is an isomorphism for all p. Since A is a finite R-algebra, Nakayama's Lemma implies that the map Bp → Ap is an isomorphism for all p ∈ Max R. It is well known that this implies B = A. (cid:3) Lemma 2.2. Suppose a1, . . . , an ∈ A and p ∈ Spec R. If a1(p), . . . , an(p) generate A(p) as an R(p)-algebra, then there exists an open neighborhood U of p in Spec R such that a1(q), . . . , an(q) generate A(q) for any q ∈ U . Proof. By our assumption, there exist (non-associative) monomials ω1, . . . , ωt on n letters such that A(p) is spanned by {ωi(a1(p), . . . , an(p))}t i=1 as an R(p)-module. Write bi = ωi(a1, . . . , an). By Nakayama's Lemma, Ap is spanned as an Rp-module by the images of b1, . . . , bt. Let B = Pi biR. Then (A/B)p = 0. Since A is finitely generated, there is s ∈ R \ p such that (A/B)s = 0. Thus, for any q ∈ Spec R not containing s, we have (A/B)q = 0. Hence, a1(q), . . . , an(q) generate A(q) as an R(q)-algebra. (cid:3) To state the next lemma, we need some additional notation. Let n ∈ N. For any commutative associative unital R-algebra S, let AS = A ⊗R S and write Vn(S) = {(a1, . . . , an) ∈ An S : a1, . . . , an generate AS as an S-algebra} . For all 0 ≤ i ≤ n, we further let Vn,i(S) =(cid:8)(a1, . . . , ai) ∈ Ai S : ∃ ai+1, . . . , an ∈ AS such that (a1, . . . , an) ∈ Vn(S)(cid:9) . Lemma 2.3. Let p ∈ Spec R, let a1, . . . , ai ∈ A, and assume that (a1(p), . . . , ai(p)) ∈ Vn,i(R(p)). Then there exists an open neighborhood U of p in Spec R such that (a1(q), . . . , ai(q)) ∈ Vn,i(R(q)) for all q ∈ U . Proof. There are bi+1, . . . , bn ∈ A(p) such that a1(p), . . . , ai(p), bi+1, . . . , bn generate A(p). After multiplying bi+1, . . . , bn by suitable invertible elements of R(p), we may assume that each bj is the image of some element aj ∈ A. By Lemma 2.2, there is an open neighborhood U of p such ON THE NUMBER OF GENERATORS OF AN ALGEBRA 3 that for all q ∈ U , the elements a1(q), . . . , an(q) generate A(q). In particular, (a1(q), . . . , ai(q)) ∈ Vn,i(R(q)). (cid:3) 3. Proof of Theorem 1.2 We claim that for every 0 ≤ j ≤ n + d, there exist elements a1, . . . , aj ∈ A and a partition of n with the following properties: X := Max R into locally closed subsets X = F (j) 1 ⊔ · · · ⊔ F (j) 0 ⊔ F (j) (1) For any i > 1 and any p ∈ F (j) i , there are t1, . . . , ti ∈ {1, . . . , j} such that (at1 (p), . . . , ati (p)) ∈ Vn,i(R(p)). (2) dim F (j) i ≤ dim X + i − j for every 0 ≤ i < n. For j = n + d, condition (2) implies that F (n+d) = ∅ for all 0 ≤ i < n, hence X = F (n+d) . Condition (1) then tells us that for every p ∈ X, there are t1, . . . , tn ∈ {1, . . . , n + d} such that (at1 (p), . . . , atn (p)) ∈ Vn,n(R(p)) = Vn(R(p)). In particular, a1(p), . . . , an+d(p) generate A(p) as an R(p)-algebra for every p ∈ X. Lemma 2.1 now implies that a1, . . . , an+d generate A as an R-algebra, proving the theorem. n i To prove the claim, we will construct the elements a1, . . . , aj ∈ A and the partition X = F (j) F (j) 1 ⊔· · ·⊔F (j) Condition (2) clearly holds and condition (1) is vacuous. n by induction on j. For the base case j = 0, set F (0) 1 = · · · = F (0) := X and F (0) 0 ⊔ := ∅. n 0 For the induction step, assume that elements a1, . . . , aj ∈ A and a partition X = F (j) 1 ⊔ · · · ⊔ F (j) satisfying conditions (1) and (2) have been constructed for some 0 6 j < n + d. We n shall choose an element aj+1 ∈ A as follows. For each 0 6 i < n, choose finitely many distinct points pi,1, . . . , pi,Ni ∈ F (j) (here we are using our standing assumption that R is Noetherian). By condition (1), for each point pi,s, there exist integers t1, . . . , ti ∈ {1, . . . , j} (depending on i and s) such that (at1 (pi,s), . . . , ati (pi,s)) ∈ Vn,i(R(pi,s)). Therefore, for each point pi,s there exists bi,s ∈ A(pi,s) such that (at1 (pi,s), . . . , ati (pi,s), bi,s) ∈ Vn,i+1(R(pi,s)). Since the sets F (j) n−1 are disjoint, the points pi,s are all distinct. By the Chinese Remainder Theorem, there exists aj+1 ∈ A such that aj+1(pi,s) = bi,s for every i = 1, . . . , n − 1, and every s = 1, . . . , Ni. i meeting all irreducible components of F (j) 1 , . . . , F (j) 0 ⊔ F (j) 0 , F (j) i Now, by Lemma 2.3, for each i and s as above, there is an open subset Ui,s of X containing pi,s such that (3.1) (at1(p), . . . , ati (p), aj+1(p)) ∈ Vi+1,n(R(p)) for all p ∈ Ui,s. Let Ui be the union of Ui,s, as s ranges from 1 to Ni, and set Un = ∅. Now set (3.2) F (j+1) i :=( (F (j) i−1 ∩ Ui−1) ∪ (F (j) 0 \ U0 F (j) i \ Ui) if i = 1, . . . , n, and if i = 0. It is easy to see that {F (j+1) Ui meets all irreducible components of F (j) , . . . , F (j+1) n 0 , hence } is a partition of X. Let 0 ≤ i < n. By our construction, (3.3) i dim(F (j) i \ Ui) 6 dim(F (j) i ) − 1 . Conditions (1), (2) for the elements a1, . . . , aj, aj+1 ∈ A and the partition X = Fi F (j+1) readily follow from (3.1), (3.2) and (3.3). This completes the proof of Theorem 1.2. i now (cid:3) 4 ON THE NUMBER OF GENERATORS OF AN ALGEBRA 4. Applications Let A and B be R-algebras. We say that A is a form of B (or equivalently, B is a form of A) if there is a faithfully flat commutative unital R-algebra S such that A ⊗R S ∼= B ⊗R S as S-algebras. For example, an Azumaya R-algebra of degree n is a form of the matrix algebra Mn(R), a finite étale R-algebra of rank n is a form of R × · · · × R (n times), a Cayley R-algebra is a form of the split octonion algebra OR (see [7, Corollary 4.11] or [9, Theorem 3.9]), and when 2 ∈ R×, an Albert R-algebra is a form of the split Albert algebra H3(OR), where H3 denotes the space of 3 × 3 Hermitian matrices (see [9, Theorem 6.9]). Proposition 4.1. Let A and B be finite-dimensional algebras over an infinite field F and assume A is a form of B. If B can be generated by n elements, then A can also be generated by n elements. Proof. Let r := dimF (A) = dimF (B). Choose an F -basis for A and use it to identify A with the F -points of the affine space Ar F )n such that for every field extension K/F , the K-points of U are the n-tuples (x1, . . . , xn) ∈ An K that generate AK as a K-algebra. Our goal is to show that U has an F -point. Since U is an open subscheme of an affine space and F is an infinite field, it suffices to check that U 6= ∅. F . It is easy to see that there exists an open subscheme U of (Ar Choose an F -field K such that AK ∼= BK. Since B is generated by n elements as an F -algebra, the same n elements will generate BK as a K-algebra. As AK ∼= BK, this implies that U has a K-point. Hence U 6= ∅, as claimed. (cid:3) Corollary 4.2. Assume the dimension of Max R is d. Then (a) every Azumaya R-algebra is generated by d + 2 elements, (b) every Cayley R-algebra is generated by d + 3 elements, (c) every Albert R-algebra is generated by d + 3 elements, provided 2 ∈ R×, † and (d) every finite étale R-algebra of rank n is generated by d + 1 elements, provided R/p is infinite for any p ∈ Max R. We remind the reader that in the statement of the corollary, "generated" means "generated as a non-unital algebra"; allowing use of the unit element in the proof does not improve the bounds. Note that, in (d), the assumption that R/p is infinite is automatic if R contains an infinite field. Proof. By Theorem 1.2, it is enough to prove the corollary when R is a field F , in which case d = 0. We let A denote an F -algebra that is Azumaya (resp. Cayley, Albert, étale of rank n). (a) First note that Mn(F ) is generated by the two matrices, E1,1 and E1,2 + · · · + En−1,n + En,1. Here Ei,j denotes the n × n matrix having 1 in the (i, j)-position and 0 elsewhere. Proposition 4.1 now tells us that when F is infinite, any form of Mn(F ) is also generated by two elements. When F is a finite field, the only form of Mn(F ) is Mn(F ) itself, by Wedderburn's theorem, so we are done. (b) By [11, §III.4], a Cayley F -algebra A is formed from a central simple F -algebra Q of degree 2 via the Cayley–Dickson process. In particular, A is generated by one element over Q. As we saw in the proof of part (a), Q is generated by two elements over F . Hence, A is generated by three elements over F . (c) A split Albert F -algebra is generated by three elements; see [8, p. 112]. By Proposition 4.1, this is also the case for any Albert F -algebra when F is infinite. Thus we may assume that F is finite. In this case Serre's Conjecture I (proved by Steinberg) implies that every Albert F -algebra is split. Indeed, isomorphism classes of Albert F -algebras are classified by the first Galois cohomology set H1(F, G), where G is the split simply-connected algebraic group of type F4 defined over F [6, †If 2 is not invertible in R, then an Albert R-algebra should be regarded as quadratic Jordan algebra [9, Section 4]; cf. Remark (iii) in the Introduction. ON THE NUMBER OF GENERATORS OF AN ALGEBRA 5 Proposition 37.11]. By Serre's Conjecture I, H1(F, G) = 0 whenever F has cohomological dimension 6 1; see [12, Theorem III.2.2.1]. On the other hand, finite fields are of cohomological dimension 6 1; see [4, Theorem 6.2.6, Proposition 6.2.3]. This shows that A is split, thus completing the proof of part (c). (d) We need to show that any étale F -algebra A of rank n over an infinite field F is generated by a single element. By Proposition 4.1, we may assume that A = F × · · · × F (n times). In this case, A is generated by any element (α1, . . . , αn) with distinct entries. (cid:3) Remark 4.3. In part (d), the assumption that R/p is infinite for any p ∈ Spec R cannot be removed in general. Indeed, when R is a field F with q elements and A = F × · · · × F , one needs at least ⌈logq(n + 1)⌉ generators, since xq = x for any x ∈ A. In fact, it can be shown that any étale F -algebra of rank n can be generated by ⌈logq(n + 1)⌉ elements (or ⌈logq n⌉ if use of the unity is allowed). Thus, if we drop the assumption that R/p is infinite in Corollary 4.2(d), we can still assert that A is generated by d + ⌈logq(n + 1)⌉ elements, where q = minp∈Max R R/p. Remark 4.4. Recall that a unital associative algebra A is called separable if A is projective relative to the left A⊗R Aop-module structure given by (a⊗bop)x = axb (a, b, x ∈ A). Examples of separable algebras include Azumaya and finite étale algebras; see [1] for further details. In the case where dim Max(R) = d and R has no finite homomorphic images, Corollary 4.2(a) can be generalized as follows: every finite separable R-algebra can be generated by d + 2 elements. Indeed, by Theorem 1.2 it suffices to show that every separable algebra B over an infinite field F is generated by two elements. Let K be an algebraic closure of F . By [1, Corollary 2.4], B ⊗F K is a product of matrix algebras Md1(K) × · · · × Mdr (K). By Proposition 4.1 we may assume B itself is a product of matrix algebras Md1(F ) × · · · × Mdr (F ). In this case a proof can be found in [10, Proposition 2.10]. Acknowledgement. We are grateful to Thomas Rüd and the anonymous referee for their help with the exposition. References [1] F. DeMeyer and E. Ingraham, Separable algebras over commutative rings, Lecture Notes in Mathematics, Vol. 181 (1971), Springer-Verlag. MR0280479 [2] D. Eisenbud and E. G. Evans, Jr., Generating modules efficiently: theorems from algebraic K-theory, J. Algebra 27 (1973), 278–305. MR0327742 [3] O. Forster, Über die Anzahl der Erzeugenden eines Ideals in einem Noetherschen Ring, Math. Z. 84 (1964), 80–87. MR0163932 [4] P. Gille, and T. Szamuely, Central simple algebras and Galois cohomology, Cambridge Studies in Advanced Mathematics, 101 (2006). MR2266528 [5] S. Kumar Upadhyay, S. D. Kumar and R. Sridharan, On the number of generators of a projective module, Proc. Indian Acad. Sci. Math. Sci. 123 (2013), no. 4, 469–478. MR3146601 [6] M.-A. Knus, A. Merkurjev, M. Rost, J.-P. Tignol, The book of involutions, With a preface in French by J. Tits., American Mathematical Society Colloquium Publications, 44 (1998). MR1632779 [7] O. Loos, H. P. Petersson and M. L. Racine, Inner derivations of alternative algebras over commutative rings, Algebra Number Theory 2 (2008), no. 8, 927–968. MR2457357 [8] K. McCrimmon, A taste of Jordan algebras, Universitext, Springer, New York (2004). MR2014924 [9] H. Petersson, Albert algebras, http://www.fernuni-hagen.de/petersson/download/alb-alg-ottawa-2012.pdf [10] Z. Reichstein, On automorphisms of matrix invariants, Trans. Amer. Math. Soc. 340 (1993), no. 1, 353–371. MR1124173 [11] R. D. Schafer, An introduction to nonassociative algebras, corrected reprint of the 1966 original, Dover, New York, 1995. MR1375235 [12] J.-P. Serre, Galois cohomology, corrected reprint of the 1997 English edition, Springer-Verlag, Berlin, 2002. MR1867431 [13] R. G. Swan, The number of generators of a module, Math. Z. 102 (1967), 318–322. MR0218347 6 ON THE NUMBER OF GENERATORS OF AN ALGEBRA [14] R. B. Warfield, Jr., The number of generators of a module over a fully bounded ring, J. Algebra 66 (1980), no. 2, 425–447. MR0593603
1704.08748	1	1704	2017-04-27T21:17:32	Conjugacy of Cartan subalgebras in EALAs with a non-fgc centreless core	[ "math.RA", "math.RT" ]	We establish the conjugacy of Cartan subalgebras for extended affine Lie algebras whose centreless core is "of type A", i.e., matrices over a quantum torus Q whose trace lies in the commutator space of Q. This settles the last outstanding part of the conjugacy problem for Extended Affine Lie Algebras that remained open.	math.RA	math	CONJUGACY OF CARTAN SUBALGEBRAS IN EALAS WITH A NON-FGC CENTRELESS CORE V. CHERNOUSOV, E. NEHER, AND A. PIANZOLA Abstract. We establish the conjugacy of Cartan subalgebras for extended affine Lie algebras whose centreless core is "of type A", i.e., ℓ × ℓ matrices over a quantum torus Q whose trace lies in the commutator space of Q. This settles the last outstanding part of the conjugacy problem for Extended Affine Lie Algebras that remained open. Dedicated to E. B. Vinberg on the occasion of his 80th birthday Introduction This work is the last of a series of papers [CGP, CNP, CNPY] devoted to proving the Conjugacy Theorem for Extended Affine Lie Algebras: Conjugacy Theorem. Let (E, H) and (E, H ′) be two extended affine Lie algebras, both defined on the same underlying Lie algebra E over an algebraically closed field of characteristic 0. Then there exists an automorphism f of E such that f (H) = H ′. Conjugacy has been established for all but one family of EALAs, and it is this remaining case that our paper settles. Below we give a brief historical account of the "Conjugacy problem". Let g be a finite-dimensional split simple Lie algebra over a field k of characteristic 0, and let G be the simply connected Chevalley-Demazure algebraic group associated to g. Chevalley's theorem ([Bo, VIII, §3.3, Cor. de la Prop. 10]) asserts that all split Cartan subalgebras h of g are conjugate under the adjoint action of G(k) on g. This is one of the central results of classical Lie theory. One of its immediate consequences is that the corresponding root system is an invariant of the Lie algebra (i.e., it does not depend on the choice of Cartan subalgebra). We now look at the analogous question in the infinite dimensional set up as it relates to extended affine Lie algebras (EALAs for short). Even if the field k is assumed to be algebraically closed, the reader should keep in mind that our results are more akin to the setting of Chevalley's theorem for general k than to conjugacy of Cartan subalgebras in finite-dimensional simple Lie algebras over algebraically closed fields. The role of (g, h) is now played by a pair (E, H) consisting of a Lie algebra E and a "Cartan subalgebra" H. There are other Cartan subalgebras in E, and the question is whether they are conjugate and, if so, under the action of which group. 2010 Mathematics Subject Classification. 17B67; (secondary) 16S36, 17B40. Key words and phrases. Extended affine Lie algebras, Lie torus, conjugacy, Cartan subalgebras, quantum torus, special linear Lie algebra. V. Chernousov was partially supported by the Canada Research Chairs Program and an NSERC research grant. E. Neher was partially supported by a Discovery grant from NSERC. A. Pianzola wishes to thank NSERC and CONICET for their continuous support. The second author wishes to thank the Department of Mathematical Sciences at the University of Alberta for hospitality during part of the work on this paper. 1 2 V. CHERNOUSOV, E. NEHER, AND A. PIANZOLA The first example is that of untwisted affine Kac-Moody Lie algebras. Let R = k[t±1]. Then (0.0.1) and (0.0.2) E = g ⊗k R ⊕ kc ⊕ kd H = h ⊗ 1 ⊕ kc ⊕ kd. The relevant information is as follows. The k-Lie algebra g⊗k R⊕kc is a central extension (in fact the universal central extension) of the k-Lie algebra g ⊗k R. The derivation d of g ⊗k R corresponds to the degree derivation td/dt acting on R. Finally h is a fixed Cartan subalgebra of g. The nature of H is that it is abelian, it acts k-diagonalizably on E, and it is maximal with respect to these properties. Correspondingly, these subalgebras are called MADs (Maximal Abelian Diagonalizable) subalgebras. A celebrated theorem of Peterson and Kac [PK] states that all MADs of E are conjugate (under the action of a group that they construct which is the analogue of the simply connected group in the finite-dimensional case). Similar results hold for the twisted affine Lie algebras. These algebras are of the form E = L ⊕ kc ⊕ kd. The Lie algebra L is a loop algebra L = L(g, σ) for some finite order automorphism σ of g (see [K] for details). If σ is the identity, we are in the untwisted case. The ring R can be recovered as the centroid of L. 1 , . . . , t±1 Extended affine Lie algebras can be thought of as multi-variable generalizations of finite- dimensional simple Lie algebras and affine Kac-Moody algebras. For example, taking R = k[t±1 n ] in (0.0.1) and increasing kc and kd correspondingly in the obvious way leads to toroidal algebras, an important class of examples of EALAs. But as is already the case for affine Kac-Moody algebras, there are many interesting examples of EALAs where g ⊗k R is replaced by a more general algebra, a so-called Lie torus (see 2.1). In the EALA set up, the Lie algebras g as above are the case of nullity n = 0, while In higher nullity n we have R = m ] for some m ≤ n, where again R is the centroid of the centreless core Ecc of the affine Lie algebras are the case of nullity n = 1. k[t±1 the given EALA. The theory of EALAs divides naturally into two cases: 1 , . . . , t±1 (a) m = n. In this case Ecc is a module of finite type over the centroid R. It is refereed to as the "fgc case" (short for finitely generated over the centroid). If k is algebraically closed,1 the R-Lie algebra Ecc is a multiloop algebra based on a (unique) g as above. In particular Ecc is twisted form of g ⊗k R. This fact allows the powerful methods of descent theory and reductive group schemes to be used. Conjugacy at the level of Ecc was established in [CGP]. The lift of this conjugacy theorem from Ecc to E is the main result of [CNPY]. (b) m < n. This is the so-called non-fgc case. Now Ecc is not a module of finite type over its centroid and Ecc is not a twisted form of g ⊗k R. The non-abelian Galois cohomology methods used in (a) are not available. Fortunately, in the non-fgc case the nature of Ecc is fully understood. Indeed Ecc = slℓ(Q) for some quantum torus Q and positive integer ℓ (see below for details). Conjugacy at the level of Ecc was established in [CNP] by means of a "specialization" trick of its own interest. The main result of the present paper is the lift of conjugacy for Ecc to E in the non-fgc case. This completes the proof that "Conjugacy of Cartan subalgebras" holds for all EALAs. The canonical procedure that associates to an EALA E its core Ec and centreless core Ecc = Ec/Z(Ec) can be reversed in the sense that one can re-construct E from its centreless core Ecc by a special type of a 2-fold extension (in this paper we generalize this to so-called 1See Remark 0.1 below 3 "interlaced extensions"). Moreover, going from E to Ecc is also a well-behaved procedure at the level of the Cartan subalgebras: Consider Hc = H ∩ Ec and let π : Ec → Ecc be the canonical map, then Hcc = π(Hc) and the analogously defined H ′ cc are special types of MADs in Ecc. Even more, every automorphism f of E leaves Ec and hence also Z(Ec) invariant and so gives rise to an automorphism fcc of Ecc. Thus, if our Main Theorem holds, then necessarily there exists some automorphism fcc ∈ Autk(Ecc) such that fcc(Hcc) = H ′ cc. From this perspective, our approach of proving conjugacy "upstairs" on the EALA level is the most natural one: we want to show that (A) there exists fcc ∈ Autk(Ecc) satisfying fcc(Hcc) = H ′ (B) the automorphism fcc of (A) can be "lifted" to an automorphism f of E such that cc, and f (H) = H ′. Problem (A) has been solved in the two papers [CGP] (the fgc case) and [CNP] (the non-fgc case). This leaves us with problem (B). Its difficulty lies in the fact that a lift f ∈ Aut(E) of fcc (if it exists at all) will not necessarily satisfy f (H) = H ′. However, for any EALA (E, H) and automorphism f of E it is easily seen that(cid:0)E, f (H)(cid:1) is an EALA which satisfies (cid:0)f (H))cc = fcc(Hcc). We can therefore split a solution of problem (B) into two steps: cc then there exists f ∈ Autk(E) such that f (H) = (B1) ([CNPY, Thm. 7.1]) If Hcc = H ′ H ′. (B2) The automorphism used to solve problem (A) can be lifted to an automorphism of E. We have solved Problem (B2) and thus established the Conjugacy Theorem for extended affine Lie algebras in the fgc case in [CNPY, Thm. 7.6]. Thus the Conjugacy Theorem for extended affine Lie algebras is reduced to proving (B2) in the non-fgc case. As explained in 2.2(d), in the non-fgc case Ecc ≃ slℓ(Q) for some ℓ ≥ 2, and Q a quantum torus which is not finitely generated over its centre. But as in [CNP] we will deal here with the Lie algebra slℓ(Q) for an arbitrary quantum torus Q.2 The conjugacy theorem of [CNP] for L = slℓ(Q), i.e., the solution of Problem (A) in the non-fgc case, uses an interior automorphism Int(g) for some g ∈ GLℓ(Q). The final step in the proof of the Conjugacy Theorem for EALAs is therefore that such g can be suitably chosen. More precisely. Main Theorem. Let L = slℓ(Q), ℓ ≥ 2 with Q a quantum torus, then Problem (A) can be solved with a g ∈ GLℓ(Q) such that Int(g) can be lifted to an automorphism of any extended affine Lie algebra E with Ecc = L. The somewhat curious formulation of our result refers to the fact that we are not claiming that all automorphisms Int(g) can be lifted to the EALA level. 0.1. Remark. A word on the nature of our base field k. The solution of Problem (A) in the fgc case ([CGP]) assumes k algebraically closed (and of course of characteristic 0). The reason for this assumption is the Realization Theorem of [ABFP]. More precisely, [CGP] holds as long as one knows that Ecc is a multiloop algebra, while [ABFP] shows that this holds in the fgc case under the assumption that k be algebraically closed. In the non-fgc case k there is no need not to assume that k be algebraically closed to solve problem (A) (see [CNP]). The lifting result (B1) works for any field of characteristic 0. In the remainder of this paper we will assume that our base field k has characteristic 0, but need not be algebraically closed. It is in this setting that we will prove our Main 2Assuming that Q is not-fgc would not simplify our arguments. The additional generality may be of future independent interest. 4 V. CHERNOUSOV, E. NEHER, AND A. PIANZOLA Theorem in the non-fgc case, namely the Conjugacy Theorem for EALAs with a non-fgc centreless core. Notation. For elements g, h of a group G we denote by [[g, h]] = ghg−1h−1 the commutator of g and h, and by D(G) = (G, G) the commutator subgroup of G. As usual Int(g)(h) = ghg−1. We use D < L to indicate that D is a subalgebra of the algebra L. For any (associative or Lie) algebra A we denote by Derk(A) the Lie algebra of k-linear derivations of A, and by Z(A) its centre. 1. Interlaced extensions In this section we introduce a general construction of Lie algebras, so-called interlaced extensions. We will see in §2 that extended affine Lie algebras are examples of interlaced extensions. In addition, one of the principal components of our proof of the Main Theorem can and will be done in the setting of interlaced extensions (Theorem 3.6). 1.1. Cocycles. Let L be a Lie algebra and let V be an L-module. A 2-cocycle with coeffi- cients in V is an alternating map σ : L × L → V satisfying for li ∈ L (1.1.1) l1 · σ(l2, l3) + l2 · σ(l3, l1) + l3 · σ(l1, l2) = σ([l1, l2], l3) + σ([l2, l3], l1) + σ([l3, l1], l2) Given such a 2-cocycle σ, the vector space L ⊕ V becomes a Lie algebra with respect to the product [l1 + v1, l2 + v2] = [l1, l2]L +(cid:0)l1 · v2 − l2 · v1 + σ(l1, l2)(cid:1). We will denote this Lie algebra by L ⊕σ V . Note that the projection onto the first factor prL : L ⊕σ V → L is an epimorphism of Lie algebras whose kernel is the abelian ideal V . We refer to such an extension as an abelian extension. A special case of this construction is the situation when V is a trivial L-module. In this case a 2-cocycle will be called a central 2-cocycle. Note that all terms on the left hand side of (1.1.1) vanish. For a central 2-cocycle, prL : L ⊕σ V → L is an epimorphism whose kernel V is the central ideal V of L ⊕σ V , i.e., prL is a central extension. A basic construction of a central 2-cocycle goes as follows. We assume that β is a bilinear form on L which is invariant in the sense that β([l1, l2], l3) = β(l1, [l2, l3]) holds for all li ∈ L. We denote by SDerk,β(L) (or simply SDerk(L) if β is fixed within our context) the subalgebra of Derk(L) consisting of skew derivations, i.e., those derivations d satisfying β(cid:0)d(l), l(cid:1) = 0 for all l ∈ L. We further suppose that D is a Lie algebra acting on L by skew derivations. It is well-known and easy to check that (1.1.2) σD,β : L × L → D∗ := Homk(D, k), is a central 2-cocycle. σD,β(cid:0)l1, l2) (d) = β(cid:0)d(l1), l2) 5 1.2. Interlaced extensions. As we explained in the introduction, one of the main prob- lems solved in this paper is lifting an automorphism from the centreless core slℓ(Q) of an EALA E to E. We will see that this can be done without additional work in a more gen- eral setting than extended affine Lie algebras. By working on this more general edifice not only do we strip the lifting process from unnecessary assumptions, but we also suggest the possibility of recasting EALA theory in a more general cadre. In this subsection we will introduce this general framework. It uses the following data: (i) a Lie algebra L equipped with an invariant bilinear form β; (ii) a Lie algebra D acting on L by skew derivations of (L, β); we write this action as d · l or sometimes d(l) for d ∈ D and l ∈ L; (iii) a subspace C ⊂ D∗ which is invariant under the co-adjoint action of D on D∗, defined by (d · c)(d′) = c([d′, d]), and satisfies (1.2.1) σD,β(l1, l2) ∈ C (li ∈ L) for σD,β as in (1.1.2); (iv) a 2-cocycle τ : D × D → C. Given these data, we define a product on the vector space (1.2.2) E = L ⊕ C ⊕ D by (li ∈ L, ci ∈ C and di ∈ D) (1.2.3) [l1 + c1 + d1, l2 + c2 + d2] =(cid:0)[l1, l2]L + d1 · l2 − d2 · l1(cid:1) ⊕(cid:0)σD,β(l1, l2) + d1 · c2 − d2 · c1 + τ (d1, d2)(cid:1) ⊕ [d1, d2]D. In this formula [., .]L and [., .]D denote the Lie algebra products of L and D respectively. We use ⊕ on the right hand side of (1.2.3) as a mnemonic device to indicate the components of the product with respect to the decomposition (1.2.2). To avoid any possible confusion we will sometimes indicate the product of E by [., .]E . We often abbreviate σ = σD,β. Our construction is a special case of [CNPY, 1.4]. Thus, by [CNPY, 1.5], the vector space E together with the product (1.2.3) is a Lie algebra. Since it is obtained by interlacing the central extension 0 → C → L ⊕ C → L → 0 (obvious maps) with the abelian extension 0 → C → C ⊕ D → D → 0 (again obvious maps) we call this Lie algebra the interlaced extension given by the data (L, β, D, C, τ ) and denote it IE(L, D, C) or IE(L, β, D, C, τ ) if more precision is helpful. Later on the bilinear form β on L will be unique, up to a scalar in k×. In general, we have for s ∈ k× (1.2.4) σD,sβ = sσD,β and IE(L, β, D, C, τ ) ≃ IE(L, sβ, D, C, sτ ) via the isomorphism l ⊕ c ⊕ d 7→ l ⊕ sc ⊕ d. 1.3. Lemma. 3 Let E = IE(L, D, C) = L ⊕ C ⊕ D be an interlaced extension, and let f : E → E be a linear map of the form (1.3.1) where l ∈ L, c ∈ C, d ∈ D and f (l ⊕ c ⊕ d) =(cid:0)fL(l) + η(d)(cid:1) ⊕(cid:0)ψ(l) + c + ϕ(d)(cid:1) ⊕ d (1.3.2) fL : L → L, η : D → L, ψ : L → C, ϕ : D → C are linear maps. Then f is an automorphism of the Lie algebra E if and only if the following conditions hold for all l, l1, l2 ∈ L and d, d1, d2 ∈ D. 3 This lemma holds in the more general setting of [CNPY, 1.4]. But we have no use for this generality. 6 V. CHERNOUSOV, E. NEHER, AND A. PIANZOLA (a) fL is an automorphism of the Lie algebra L, (c) fL(d · l) = [η(d), fL(l)]L + d · fL(l), (b) σ(cid:0)fL(l1), fL(l2)(cid:1) = ψ(cid:0)[l1, l2](cid:1) + σ(l1, l2) for σ = σD,β, (d) ψ(d · l) = σ(cid:0)η(d), fL(l)(cid:1) + d · ψ(l), (e) η(cid:0)[d1, d2]D(cid:1) = [η(d1), η(d2)]L + d1 · η(d2) − d2 · η(d1), (f) ϕ(cid:0)[d1, d2](cid:1) = σ(cid:0)η(d1), η(d2)(cid:1) + d1 · ϕ(d2) − d2 · ϕ(d1). Proof. The map f is bijective if and only if fL is so. Moreover, the definition of the product of E in (1.2.3) and the definition of f in (1.3.1) show that f is a homomorphism of the Lie algebra E if and only if it respects the products [l1, l2]E, [d, l]E and [d1, d2]E. This leads to the conditions (a) -- (f). (cid:3) We will call an automorphism of type (1.3.1) a special automorphism. Not all automor- phisms of E are special, but we have the following result. 1.4. Proposition. ([CNPY, Prop. 1.6]) Let E = IE(L, D, C) be an interlaced extension. Every elementary automorphism of L lifts to a special automorphism of E. We recall that an elementary automorphism of a Lie algebra M is a product of automor- phisms of type exp adM x with adM x ∈ Endk(M ) (locally) nilpotent. The reader can easily verify that for fL = exp adL x, the maps η, ψ and ϕ of (1.3.2) are given by 1 ψ(l) =Pn≥1 η(d) = −Pn≥1 ϕ(d) = −Pn≥2 1 n! (adL x)n−1(d · x), 1 n! σ(cid:0)x, (adL x)n−1(l)(cid:1) n! σ(cid:0)x, (adL x)n−2(d · x)(cid:1) for σ = σD,β, These formulas indicate that the maps ψ, η and ϕ are in general not zero. 1.5. Enlarging interlaced extensions. In the process of lifting an automorphism from L to an interlaced extension E, we will enlarge E to a bigger interlaced extension using the following construction. (i) E = IE(L, β, D, C, τ ) is an interlaced extension; (ii) L is a subalgebra of a Lie algebra L′ equipped with an invariant bilinear form β′ such that β′ L×L= β; (iii) the action of D on L extends to an action of D on L′ by skew derivations, and (iv) σD,β′(l′ 2) ∈ C for l′ 2 ∈ L.4 1, l′ 1, l′ The data (L′, β′, D, C, τ ) satisfy the assumptions (i) -- (iv) of 1.2, so that we can form the interlaced extension E′ = IE(L′, β′, D, C, τ ) = L′ ⊕ C ⊕ D Since for l1, l2 ∈ L we have σD,β′(l1, l2)(d) = β′(d · l1, l2) = β(d · l1, l2) = σD,β(l1, l2)(d), i.e., it is immediate that E is a subalgebra of E′. σD,β′(cid:12)(cid:12)L×L = σD,β, In this setting suppose that f ′ is a special automorphism of E′, thus given by the data L′ : L′ → L′, f ′ η′ : D → L′, ψ′ : L′ → C, ϕ′ : D → C 4 Note that because of assumption (iii) we necessarily have that σD,β ′ : L′ × L′ → C ⊂ D∗ coincides with the central 2-cocycle of (1.1.2) when restricted to L × L. as in (1.3.2), satisfying the conditions (a) -- (f) of Lemma 1.3. It is then immediate that 7 (1.5.1) In this case f ′ E is obviously an automorphism of E, in fact a special automorphism given by the data L′(L) = L and η′(D) ⊂ L. f ′(E) = E ⇐⇒ f ′ (1.5.2) fL = f ′ η = η′, ψ = ψ′, ϕ = ϕ′. L′(cid:12)(cid:12)L, 2. Review: Lie tori and extended affine Lie algebras (EALAs) In this section we review the theory of extended affine Lie algebras, in order to give the reader a perspective about the achievements of this paper. The structure of extended affine Lie algebra is intimately connected to Lie tori. We therefore start with a short summary of the pertinent facts from the theory of Lie tori. We then introduce EALAs and describe their construction as a special case of an interlaced extension (1.2) based on a Lie torus. 2.1. Lie tori. We use the term "root system" to mean a finite, not necessarily reduced root system ∆ in the usual sense, except that we will assume 0 ∈ ∆, as for example in [AABGP]. We denote by ∆ind = {0} ∪ {α ∈ ∆ : 1 2 α 6∈ ∆} the subsystem of indivisible roots and by Q(∆) = spanZ(∆) the root lattice of ∆. To avoid some degeneracies we will always assume that ∆ 6= {0}. Let ∆ be a finite irreducible root system, and let Λ be free abelian group of finite type.5 A Lie torus of type (∆, Λ) is a Lie algebra L satisfying the following conditions (LT1) -- (LT4). (LT1) (a) L is graded by Q(∆) ⊕ Λ. We write this grading as L = Lα∈Q(∆),λ∈Λ Lλ α+β. It is convenient to define thus have [Lλ β] ⊂ Lλ+µ α, Lµ α and (2.1.1) (b) We further assume that suppQ(∆) L = {α ∈ Q(∆); Lα 6= 0} = ∆, so that Lα =Lλ∈Λ Lλ α and Lλ =Lα∈Q(∆) Lλ α. L =Lα∈∆ Lα. (LT2) (a) If Lλ α 6= 0 and α 6= 0, then there exist eλ α and f λ α ∈ L−λ −α such that Lλ α = keλ α, L−λ α ∈ Lλ −α = kf λ α , and [[eλ for all β ∈ ∆ and xβ ∈ Lβ.6 (b) L0 α 6= 0 for all 0 6= α ∈ ∆ind. α, f λ α ], xβ] = hβ, α∨ixβ (LT3) As a Lie algebra, L is generated by S06=α∈∆ Lα. (LT4) As an abelian group, Λ is generated by suppΛ L = {λ ∈ Λ : Lλ 6= 0}. We define the nullity of a Lie torus L of type (∆, Λ) as the rank of Λ. We will say that L is a Lie torus (without qualifiers) if L is a Lie torus of type (∆, Λ) for some pair (∆, Λ). A Lie torus is called centreless if its centre Z(L) = {0}. If L is an arbitrary Lie torus, its centre Z(L) is contained in L0 from which it easily follows that L/Z(L) is in a natural way a centreless Lie torus of the same type as L and nullity (see [Yo2, Lemma 1.4]). The structure of Lie tori is known, see [Al] for a recent survey. Some more background on Lie tori is given in the papers [ABFP, Ne3, Ne4]. Lie tori can of course be defined for 5 Thus Λ ∼= Zn for some n ∈ N. But it is not helpful to assume Λ = Zn. 6 Here and elsewhere α∨ denotes the coroot corresponding to α in the sense of [Bo]. 8 V. CHERNOUSOV, E. NEHER, AND A. PIANZOLA any abelian group Λ (see for example [Yo2]), but only the case of a free abelian group of finite rank is of interest for EALAs. An obvious example of a Lie torus of type (∆, Zn) is the Lie k-algebra g ⊗ R where g is a finite-dimensional split simple Lie algebra of type ∆ and R = k[t±1 n ] is the Laurent polynomial ring in n-variables with coefficients in k equipped with the natural Zn- grading. Another important example, first studied in [BGK], is the Lie algebra slℓ(Q) for Q a quantum torus (see 3.7 and 3.8). 1 , . . . , t±1 2.2. Some known properties of centreless Lie tori. We review some of the properties of Lie tori needed in the following. We assume that L is a centreless Lie torus of type (∆, Λ) and nullity n. (a) For eλ α as in (LT2) we put hλ 0 and observe that (eλ α = [eλ α, f λ α ] ∈ L0 α and f λ an sl2-triple. Then α, hλ α, f λ α ) is (2.2.1) is a toral 7 subalgebra of L whose root spaces are the Lα, α ∈ ∆. h = spank{hλ α} = L0 0 (b) Up to scalars, L has a unique nondegenerate symmetric bilinear form (··) which is Λ-graded in the sense that (Lλ Lµ) = 0 if λ + µ 6= 0, [NPPS, Yo2]. Since the subspaces Lα are the root spaces of the toral subalgebra h we also know (Lα Lτ ) = 0 if α + τ 6= 0. (c) Let Ctdk(L) = {χ ∈ Endk(L) : χ([l1, l2]) = [l1, χ(l2)] ∀ l1, l2 ∈ L} be the centroid of L (see for example [BN] for general facts about centroids). Since L is perfect, Ctdk(L) is a commutative associative unital subalgebra of Endk(L). It is graded with respect to the Λ-grading (2.1.1) of L: Ctd(L) =Lλ∈Λ Ctd(L)λ where Ctd(L)λ consists of those centroidal transformations χ satisfying χ(Lµ) ⊂ Lλ+µ for all µ ∈ Λ. One knows that Ctdk(L) is graded-isomorphic to the group ring k[Ξ] for a subgroup Ξ of Λ, the so-called central grading group. Hence Ctdk(L) is a Laurent polynomial ring in ν variables, 0 ≤ ν ≤ n, ([Ne1, 7], [BN, Prop. 3.13]). (All possibilities for ν do in fact occur, for example for L = slℓ(Q), see 3.7 and 3.8.) (d) The space L is naturally a Ctdk(L)-module via χ · a = χ(a). As a Ctdk(L)-module, L is free. If L is fgc, i.e., namely finitely generated as a module over its centroid, then L is a multiloop algebra [ABFP]. If L is not fgc, equivalently ν < n, one knows ([Ne1, Th. 7]) that L has root-grading type A. Lie tori with this root-grading type are classified in [BGK, BGKN, Yo1]. It follows from this classification together with [NY, 4.9] that L ≃ sll(Q) for Q a quantum torus in n variables and structure matrix q = (qij) an n × n quantum matrix with at least one qij not a root of unity (3.7). (e) Any θ ∈ HomZ(Λ, k) induces a so-called degree derivation ∂θ of L defined by ∂θ(lλ) = θ(λ)lλ for lλ ∈ Lλ. We put D = {∂θ : θ ∈ HomZ(Λ, k)} and note that θ 7→ ∂θ is a vector space isomorphism from HomZ(Λ, k) to D, whence D ≃ kn. We define evλ ∈ D∗ by evλ(∂θ) = θ(λ). One knows ([Ne1, 8]) that D induces the Λ-grading of L in the sense that Lλ = {l ∈ L : ∂θ(l) = evλ(∂θ)l for all θ ∈ HomZ(Λ, k)} holds for all λ ∈ Λ. (f) If χ ∈ Ctdk(L) then χd ∈ Derk(L) for any derivation d ∈ Derk(L). We call (2.2.2) CDerk(L) := Ctdk(L)D =Lξ∈Ξ Ctd(L)ξD 7A subalgebra T of a Lie algebra L is toral, sometimes also called ad-diagonalizable, if L = Lα∈T ∗ Lα(T ) for Lα(T ) = {l ∈ L : [t, l] = α(t)l for all t ∈ T }. In this case {ad t : t ∈ T } is a commuting family of ad-diagonalizable endomorphisms. Conversely, if {ad t : t ∈ T } is a commuting family of ad-diagonalizable endomorphisms and T is a finite-dimensional subalgebra, then T is a toral. 9 the centroidal derivations of L. It is easily seen that CDer(L) is a Ξ-graded subalgebra of Derk(L), a generalized Witt algebra. Note that D is a toral subalgebra of CDerk(L) whose root spaces are the Ctd(L)ξ D = {d ∈ CDer(L) : [t, d] = evξ(t)d for all t ∈ D}. One also knows ([Ne1, 9]) that (2.2.3) Derk(L) = IDer(L) ⋊ CDerk(L) (semidirect product), where IDer(L) is the ideal of inner derivations of L. (g) For the construction of EALAs, the Ξ-graded subalgebra SCDerk(L) of skew-centroidal derivations is important: SCDerk(L) = {d ∈ CDerk(L) : (d · l l) = 0 for all l ∈ L} SCDerk(L)ξ = Ctd(L)ξ{∂θ : θ(ξ) = 0}. =Lξ∈Ξ SCDerk(L)ξ, 2.3. Extended affine Lie algebras (EALAs). An extended affine Lie algebra or EALA for short, is a pair (E, H) consisting of a Lie algebra E over k and a subalgebra H of E satisfying the axioms (EA0) -- (EA5) below. (EA0) E has an invariant nondegenerate symmetric bilinear form (··). (EA1) H is a nontrivial finite-dimensional toral and self-centralizing subalgebra of E. Thus E = Lα∈H ∗ Eα for Eα = {e ∈ E : [h, e] = α(h)e for all h ∈ H} and E0 = H. We denote by Ψ = {α ∈ H ∗ : Eα 6= 0} the set of roots of (E, H) -- note that 0 ∈ Ψ! Because the restriction of (··) to H × H is nondegenerate, one can in the usual way transfer this bilinear form to H ∗ and then introduce anisotropic roots Ψan = {α ∈ Ψ : (α α) 6= 0} and isotropic (= null) roots Ψ0 = {α ∈ Ψ : (α α) = 0}. The core of (cid:0)E, H, (··)(cid:1) is by definition the subalgebra generated by Sα∈Ψan Eα. It will be henceforth denoted by Ec. (EA2) For every α ∈ Ψan and xα ∈ Eα, the operator ad xα is locally nilpotent on E. (EA3) Ψan is connected in the sense that for any decomposition Ψan = Ψ1 ∪Ψ2 with Ψ1 6= ∅ and Ψ2 6= ∅ we have (Ψ1 Ψ2) 6= 0. (EA4) The centralizer of the core Ec of E is contained in Ec, i.e., {e ∈ E : [e, Ec] = 0} ⊂ Ec. (EA5) The subgroup Λ = spanZ(Ψ0) ⊂ H ∗ generated by Ψ0 in (H ∗, +) is a free abelian group of finite rank. The attentive reader will have noticed that the choice of invariant nondegenerate sym- metric bilinear form in (EA0) is part of the structural data defining an EALA. However, one can show that another choice of an invariant nondegenerate symmetric bilinear form leads to the same set of anisotropic and isotropic roots Ψan and Ψ0, and thus also to the same core Ec and centreless core Ecc = Ec/Z(Ec), see [CNPY, Rem. 2.4 and Cor. 3.3]. The core Ec of an EALA E is always an ideal of E. Some references for EALAs are [AABGP, BGK, Ne2, Ne3, Ne4]. It is immediate that any finite-dimensional split simple Lie algebra g is an EALA of nullity 0 and g = gc = gcc. The converse is also true, [Ne4, Prop. 5.3.24]. It is also known that any affine Kac-Moody algebra over C is an EALA -- in fact, by [ABGP], the affine Kac-Moody algebras g are precisely the EALAs over C of nullity 1. For those, gc = [g, g] and gcc is an (twisted or untwisted) loop algebra. 10 V. CHERNOUSOV, E. NEHER, AND A. PIANZOLA 2.4. The roots of an EALA. The set Ψ of roots of an EALA (E, H) is an extended affine root system in the sense of [AABGP, Ch. I] (see also the surveys [Ne3, §2, §3] and [Ne4, §5.3]). Thus, there exists an irreducible finite (but possibly non-reduced) root system ∆ ⊂ H ∗, an embedding ∆ind ⊂ Ψ and a family (Λα : α ∈ ∆) of subsets Λα ⊂ Λ such that (2.4.1) spank(Ψ) = spank(∆) ⊕ spank(Λ) and Ψ =Sα∈∆(α + Λα). Using this (non-unique) decomposition of Ψ, we write any ψ ∈ Ψ as ψ = α + λ with α ∈ ∆ and λ ∈ Λα ⊂ Λ and define (Ec)λ α is a Lie torus of type (∆, Λ), and the centreless core Ecc = Ec/Z(Ec) is a centreless Lie torus. α = Ec ∩ Eψ. Then the core Ec =Lα∈∆,λ∈Λ(Ec)λ 2.5. Construction of EALAs. To construct an EALA one reverses the process described in 2.4. We will use data (L, σD, τ ) described below. Some more background material can be found in [Ne3, §6] and [Ne4, §5.5]: • L is a centreless Lie torus of type (∆, Λ). We fix a Λ-graded invariant nondegenerate symmetric bilinear form (··) (see 2.2(b)) and let Ξ be the central grading group of L (see 2.2(c)). • D = Lξ∈Ξ Dξ is a graded subalgebra of SCDerk(L) (see 2.2(g)) such that the evaluation map evD0 : Λ → D0 ∗, λ → evλ D0, defined in 2.2(e), is injective. Since (Lλ Lµ) = 0 if λ + µ 6= 0 and since Dξ(Lλ) ⊂ Lξ+λ it follows that the central cocycle σD of (1.1.2) has values in the graded dual Dgr∗ =: C of D. Recall C = Lξ∈Ξ C ξ with C ξ = (D−ξ)∗ ⊂ D∗. The contragredient action of D on D∗ leaves C invariant. • τ : D × D → C is an affine cocycle defined to be a 2-cocycle satisfying τ (d0, d) = 0 and τ (d1, d2)(d3) = τ (d2, d3)(d1) for all d, di ∈ D and d0 ∈ D0. It is important to point out that there do exist non-trivial affine cocycles, see [BGK, Rem. 3.71]. The data (L, σD, τ ) with β the unique invariant bilinear form (··) of 2.2(b) satisfy all the axioms of our general construction 1.2. Thus the interlaced extension (2.5.1) E = L ⊕ C ⊕ D is a Lie algebra with respect to the product (1.2.3). Note that E contains the toral subal- gebra H = h ⊕ C 0 ⊕ D0 for h as in (2.2.1). The symmetric bilinear form (··) on E, defined by is nondegenerate and invariant, thus fulfilling the axiom (E0). (cid:0)l1 + c1 + d1 l2 + c2 + d2(cid:1) = (l1 l2)L + c1(d2) + c2(d1), Examples: (a) In case L = g is a finite-dimensional split simple Lie algebra, Ctd(g) = 0, Ξ = {0}, and so also SCDer(g) = 0. The construction above therefore yields E = g. (b) In case L is a twisted or untwisted loop algebra based on g as in (a) over C, the centroid Ctd(L) is isomorphic to a Laurent polynomial ring R, CDer(L) is a free R-module of rank 1, but SCDer(g) is 1-dimensional over C. The only non-trivial choice is therefore D ≃ C. In this case necessarily τ = 0. Thus the construction of affine Kac-Moody algebras is a special case of our construction above. 11 affine Lie algebra, denoted EA(L, D, τ ). Its core is L ⊕ Dgr ∗ and its centreless core is L. 2.6. Theorem ([Ne2, Th. 6]). (a) The triple (cid:0)E, H, (··)(cid:1) constructed above is an extended (b) Conversely, let (cid:0)E, H, (··)(cid:1) be an extended affine Lie algebra, and let L = Ec/Z(Ec) τ satisfying the conditions in 2.5 such that (cid:0)E, H, (··)(cid:1) ≃ EA(L, (··)L, D, τ ) for some Λ- be its centreless core. Then there exists a subalgebra D ⊂ SCDerk(L) and an affine cocycle graded invariant nondegenerate bilinear form (··)L on L. 3. Lifting automorphisms from slℓ(A) to IE(slℓ(A), D, C). 3.1. The Lie algebras glℓ(A) and slℓ(A). We assume throughout that ℓ ≥ 2. The letter A will always denote a unital associative k -- algebra. It becomes a Lie algebra Lie(A) with respect to the commutator. We denote by [A, A] the commutator subalgebra of Lie(A), [A, A] = spanZ{ab − ba : a, b ∈ A} and by Z(A) = {z ∈ A : [z, A] = 0} the centre of A, which is also the centre of Lie(A). We denote by Mℓ(A) the unital associative algebra of ℓ × ℓ matrices with coefficients in A, and by glℓ(A) its associated Lie algebra: glℓ(A) = Lie(cid:0)Mℓ(A)(cid:1). The derived algebra of glℓ(A) is the special linear Lie algebra slℓ(A) with coefficients from A: (3.1.1) slℓ(A) = [glℓ(A), glℓ(A)]. We let Tr be the trace of a matrix in Mℓ(A). The reader should be warned that Tr(xy) 6= Tr(yx) in general, rather we have the well-known fact (3.1.2) slℓ(A) = {x ∈ glℓ(A) : Tr(x) ∈ [A, A]}.8 Moreover, we will need (3.1.3) where C denotes the centralizer and Eℓ the ℓ × ℓ identity matrix. Cglℓ(A)(cid:0)slℓ(A)(cid:1) = Z(A) Eℓ = Z(cid:0)glℓ(A)(cid:1), Any d ∈ Derk(A) stabilizes Z(A) and [A, A], and induces a derivation of the associative algebra Mℓ(A) by (3.1.4) d · x =(cid:0)d(xij)(cid:1) for x = (xij) ∈ Mℓ(A). It is then also a derivation of glℓ(A), stabilizing Z(cid:0)glℓ(A)(cid:1) = Z(A)Eℓ and slℓ(A). In the following, a subalgebra D < Derk(A) will be a standard feature of our work. We will always use the action of Derk(A) and hence of D described in (3.1.4) without further explanation. Also, we will sometimes write dx or d(x) for d · x. 3.2. The groups GLℓ(A) and ELℓ(A). We denote by GLℓ(A) the group of invertible ℓ × ℓ matrices with coefficients from the unital associative k-algebra A. Every g ∈ GLℓ(A) gives rise to an automorphism Int(g) of the associative algebra Mℓ(A), defined by Int(g)(a) = gag−1. A fortiori, Int(g) is an automorphism of glℓ(A). It stabilizes slℓ(A), whence is by restriction an automorphism of slℓ(A), again denoted Int(g). Moreover, Int(g) induces the The elementary linear group ELℓ(A) is the subgroup of GLℓ(A) generated by the matrices Eℓ +aEij for arbitrary a ∈ A and i 6= j. Since (aEij)2 = 0 in Mℓ(A) the derivation identity on Z(cid:0)glℓ(A)(cid:1) as can be seen for the last equality of (3.1.3). ad aEij ∈ Der(cid:0)slℓ(A)(cid:1) is nilpotent, in fact (ad aEij)3 = 0, and the inner automorphism Int(Eℓ + aEij) ∈ Autk(cid:0)slℓ(A)(cid:1) is elementary in the sense of 1.4: Int(Eℓ +aEij) = exp(ad aEij). 8 Of course if A is commutative, then [A, A] = 0 and we recover the "usual" definition of slℓ(A). 12 V. CHERNOUSOV, E. NEHER, AND A. PIANZOLA It follows that (3.2.1) Int(g) ∈ EAutk(cid:0)slℓ(A)(cid:1) for every g ∈ ELℓ(A), where EAut(cid:0)slℓ(A)(cid:1) is the group of elementary automorphisms of slℓ(A). Moreover, the commutator relation [[Eℓ +aEij, Eℓ +Ejℓ]] = Eℓ + aEiℓ (i, j, ℓ 6=) shows that (3.2.2) 3.3. Lemma. Let A be a unital associative k -- algebra satisfying ELℓ(A) ⊂ D(cid:0)GLℓ(A)(cid:1). (3.3.1) Then (3.3.2) (3.3.3) A = Z(A) ⊕ [A, A]. glℓ(A) = Z(A) Eℓ ⊕slℓ(A), and (dg)g−1 ∈ slℓ(A) ⇐⇒ g−1dg ∈ slℓ(A) for any d ∈ Derk(A) and g ∈ GLℓ(A). Moreover, for D ⊂ Derk(A) the set H = HD = {g ∈ GLℓ(A) : (dg)g−1 ∈ slℓ(A) for all d ∈ D} is a normal subgroup of GLℓ(A) containing the commutator subgroup D(cid:0)GLℓ(A)(cid:1) of GLℓ(A). Proof. Our assumption (3.3.1) implies A Eℓ = Z(A) Eℓ ⊕[A, A] Eℓ. Since [A, A] Eℓ = A Eℓ ∩slℓ(A) by (3.1.2), the equation (3.3.2) follows from the decomposition with Tr(x − 1 ℓ Tr(x) Eℓ) + (x − 1 ℓ Tr(x) Eℓ) = 0 for arbitrary x ∈ glℓ(A). x = ( 1 ℓ Tr(x) Eℓ) The equivalence (3.3.3) is a consequence of g−1dg = (dg)g−1 + [g−1, dg] and [g−1, dg] ∈ slℓ(A). This shows that g ∈ H =⇒ g−1 ∈ H since (3.3.4) 0 = d(g−1g) = (dg−1)g + g−1(dg) Given that Eℓ ∈ H, for H to be a subgroup it suffices to show that g1, g2 ∈ H =⇒ g1g2 ∈ H. But this follows from (cid:0)d(g1g2)(cid:1)(g1g2)−1 = (dg1)g2g−1 2 g−1 1 + g1(dg2)g−1 2 g−1 1 = (dg1)g−1 1 + Int(g1)(cid:0)(dg2)g−1 2 (cid:1) since Int(g1) stabilizes slℓ(A). Thus H is a subgroup, and it will be a normal subgroup as soon as we have shown that H contains any commutator [[g1, g2]] where g1, g2 ∈ GLℓ(A). We have = (dg1) (g2g−1 (cid:0)d[[g1, g2]](cid:1)[[g1, g2]]−1 = d(cid:0)g1g2g−1 2 g2g1(cid:1)g−1 + g1g2(cid:0)d(g−1)g−1 2 g2g1g−1 = (dg1)g−1 1 g−1 2 g−1 1 g−1 2 g−1 1 1 g−1 2 (cid:1) g2g1g−1 1 ) + g1(cid:0)(dg2)g−1 1 (cid:0)d(g−1 2 ) + Int(g1g2)(cid:0)(dg−1 1 + g1g2g−1 2 g−1 2 g2g1g−1 1 2 g−1 1 2 (cid:1)g−1 2 g2)(cid:1)g1g−1 1 )g1(cid:1) + Int(g1g2g−1 1 + Int(g1)((dg2)g−1 To proceed, we use (3.3.2), thus uniquely writing any x ∈ glℓ(A) as x = xz + xs with 1 )(cid:0)(dg−1 2 )g2(cid:1) xz ∈ Z(cid:0)glℓ(A)(cid:1) and xs ∈ slℓ(A). Decomposing y ∈ glℓ(A) in the same way, we have (xy)z = (yx)z (3.3.5) since xy = yx + [x, y] with [x, y] ∈ slℓ(A). Because Int(g) stabilizes slℓ(A) and satisfies Int(g)(zEℓ) = zEℓ for z ∈ Z(A) we now get (cid:0)(dg1)g−1 1 (cid:1)z +(cid:16)Int(g1g2)(cid:0)(dg−1 1 )g1(cid:1)(cid:17)z =(cid:0)(dg1)g−1 1 (cid:1)z + Int(g1g2)(cid:0)(dg−1 1 )g1(cid:1)z =(cid:0)(dg1)g−1 1 (cid:1)z +(cid:0)(dg−1 thus proving that (dg1)g−1 1 )g1(cid:1)z =(cid:0)(dg1)g−1 1 +(cid:0)Int(g1g2)(cid:0)(dg−1 2 ) + Int(g1g2g−1 1 )(cid:1)z =(cid:0)d(g1g−1 1 (cid:1)z +(cid:0)g1(dg−1 1 )g1(cid:1) ∈ slℓ(A). Similarly 1 )(cid:0)(dg−1 2 )g2(cid:1) ∈ slℓ(A). Int(g1)((dg2)g−1 1 )(cid:1)z = 0, Hence [[g1, g2]] ∈ H, and therefore D(cid:0)GLℓ(A)(cid:1) ⊂ H. 3.4. Interlaced extensions based on slℓ(A). We specialize the setting of 1.2 to L = slℓ(A) with the aim of constructing a suitable interlaced extension that will allow us to lift the automorphisms used in conjugacy. Being an interlaced extension, we need to specify data (β, D, C, τ ). 13 (cid:3) (i) We fix a linear form (3.4.1) ε : A → k, ε([A, A]) = 0 and define β = βε : L × L → k by (3.4.2) for x = (xij and y = (yij). Then β is an invariant bilinear form on L, and every invariant bilinear form on L is of the type βε for a unique linear form ε satisfying (3.4.1) ([Ne3, 7.10]). βε(x, y) = ε(cid:0) Tr(xy)(cid:1) =Pℓ i,j=1 ε(xijyji) (ii) We let D be a subalgebra of derivations of A, which are skew with respect to the bilinear form (a, b) 7→ ε(ab), and let D act on L as in (3.1.4). Then D acts on L by skew derivations with respect to β. (iii) We choose C ⊂ D∗ and τ as in (iii) and (iv) of 1.2. D < SDerk(A), Using these data we form the interlaced extension IE(L, βε, D, C, τ ) = E = L ⊕ C ⊕ D. 3.5. Enlarging interlaced extensions. To suitably enlarge an interlaced extension E = IE(L, D, C) with L = slℓ(A) as in 3.4, we embed L into L′ = slℓ+m(A), m ∈ N arbitrary, via (3.5.1) slℓ(A) → slℓ+m(A), 0 0(cid:19) . ε as defined in (3.4.2): β′(x′, y′) = ε(cid:0) Tr(x′y′)(cid:1) = Pℓ+m Following the outline of 1.5 we next need an invariant bilinear form β′ on L′. We take β′ = β′ ij) and ij) ∈ slℓ+m(A). Then the condition (ii) of 1.5 is fulfilled: β′(l1, l2) = β(l1, l2) for y′ = (y′ l1, l2 ∈ L. l 7→(cid:18)l ji) for x′ = (x′ i,j=1 ε(x′ ijy′ 0 We also have condition (iii) of 1.5, i.e., D acts on L′ by skew derivations extending the action of D on L. Finally, 1.5(iv) also holds. Indeed, for x′, y′ ∈ L′ as before and d ∈ D, we have σD,β′(x′, y′)(d) = β′(d · x′, y′) =Pℓ+m =Pℓ+m i,j=1 ε(cid:0)(d · x′ 21E21) =(cid:16)Pℓ+m i,j=1 β(d · (x′ ijE12), y′ ji) i,j=1 σD,β(x′ ij)y′ ijE12, y′ jiE12)(cid:17)(d) which shows σD,β′(x′, y′) ∈ C. In sum, we have shown that for any m ∈ N the interlaced extension E = IE(L, βε, D, C, τ ) is a subalgebra of E′ = IE(slℓ+m(A), β′ ε, D, C, τ ). We are now ready to prove the main result of this section. 14 V. CHERNOUSOV, E. NEHER, AND A. PIANZOLA 3.6. Theorem. Let A be a unital associative k-algebra satisfying A = Z(A)⊕[A, A], and let E = IE(L, D, C) be an interlaced extension based on L = slℓ(A) as specified in 3.4. Assume that g ∈ GLℓ(A) is stably elementary in the sense that there exists m ∈ N such that 0 g′ =(cid:18)g 0 Em(cid:19) ∈ ELℓ+m(A). Then the automorphism Int(g) of slℓ(A) lifts to an automorphism of E. Proof. We embed L = slℓ(A) into L′ = slℓ+m(A) as in (3.5.1). We then know that E can be enlarged to an interlaced extension E′ = IE(L′, D, C). Moreover, by (3.2.1) and Proposition 1.4 the elementary automorphisms Int(g′) of L′ lifts to a special automorphism f ′ of E′, determined by maps Int(g′) = fL′ ∈ Autk(L′) and linear maps η′ : D → L′, ψ′ : L′ → C and ϕ′ : D → C as in Lemma 1.3. It will be sufficient to show f ′(E) = E. Since f ′ L′(L) = Int(g)(L) = L, it is in view of (1.5.1) enough to prove By 1.3(c) we have η′(d) ∈ L for all d ∈ D. d · fL′(l′) − fL′(d · l′) = [fL′(l′), η′(d)] (3.6.1) for all d ∈ D and l′ ∈ L′. For l ∈ L we know fL′(l) = Int(g)(l) ∈ L and also d · fL′(l) − fL′(d · l) ∈ L. It thus follows from (3.6.1) for l′ = l ∈ L that η′(d) normalizes L. One easily calculates that then η′(d) has the form η′(d) =(cid:18)α 0 0 β(cid:19) , α ∈ glℓ(A), β ∈ glm(A) (we have suppressed in our notation that α and β depend linearly on d). Employing the obvious subdivision for matrices l′ ∈ L′, l′ =(cid:18)x1 x2 x3 x4(cid:19) , x1 ∈ glℓ(A), x3g−1 whence for d ∈ D the left hand side of (3.6.1) becomes gx2 fL′(l′) = Int(g′)(l′) =(cid:18)gx1g−1 x4 (cid:19) , d · fL′(l′) − fL′(d · l′) =(cid:18)(dg)x1g−1 + gx1d(g−1) [fL′(l′), η′(d)] =(cid:18) [gx1g−1, α] x3g−1α − βx3g−1 x3d(g−1) (dg)x2 0 (cid:19) , [x4, β] (cid:19) . gx2β − αgx2 while the right hand side of (3.6.1) is we get Thus (3.6.2) (dg)x2 = gx2β − αgx2 (3.6.3) Since every x4 ∈ glm(A) is part of some matrix l′ ∈ L′, it follows that (3.6.3) holds for all x4 ∈ glm(A). Therefore, by (3.1.3), 0 = [x4, β] β = z Em for some x ∈ Z(A). We substitute this expression for β into (3.6.2) and obtain (dg)x2 = (zg − αg)x2. Since this holds for all x2 ∈ Mmn(A) we get dg = zg − αg or α = z Eℓ −(dg)g−1. Because g′ ∈ ELℓ+m(A) it follows from (3.2.2) and Lemma 3.3 that (dg′)(g′)−1 ∈ slℓ+m(A) for all d ∈ D. But so that (dg)g−1 ∈ slℓ(A) follows. Since η′(d) ∈ slℓ+m(A) we now get (dg′)(g′)−1 =(cid:18)(dg)g−1 0 0(cid:19) , 0 15 Tr(cid:0)η′(d)(cid:1) = Tr(α) + Tr(β) = ℓz − Tr(cid:0)(dg)g−1(cid:1) + mz = (ℓ + m)z − Tr(cid:0)(dg)g−1(cid:1) ∈ [A, A]. As A = Z(A) ⊕ [A, A] by assumption and Tr(cid:0)(dg)g−1(cid:1) ∈ [A, A], this forces (ℓ + m)z = 0 so that z = 0 and finally β = 0, i.e., η′(d) ∈ L follows. (cid:3) 3.7. Quantum tori (review). We will later specialize A = Q to be a quantum torus. Why we do so, is explained in 3.8: slℓ(Q) is then a centreless Lie torus. In this subsection we review some properties of quantum tori that we will use. Contrary to the standing as- sumption for this paper, in this subsection our base field k can have arbitrary characteristic. We let Λ be a free abelian group of rank n. (a) (Definition) By definition, a quantum torus (with grading group Λ) is an associative unital Λ-graded k-algebra Q =Lλ∈Λ Qλ such that (QT1) dim Qλ ≤ 1 for all λ ∈ Λ, (QT2) every 0 6= a ∈ Qλ is invertible, and (QT3) Λ is generated as abelian group by {λ ∈ Λ : Qλ 6= 0}. Since the invertible elements of an associative algebra form a group, {λ ∈ Λ : Qλ 6= 0} is a subgroup of Λ, whence equals Λ by (QT3). (b) After fixing a basis ε = (εi) of Λ, we can choose 0 6= xi ∈ Qεi and then get a quantum matrix q = (qij) ∈ Mn(k) defined by xixj = qijxjxi. We recall that q = (qij) ∈ Mn(k) is called a quantum matrix if qij = q−1 ji and qii = 1 for all 1 ≤ i, j ≤ n. i = the inverse of xi, we define xλ = xℓ1 1 · · · xℓn n for λ = ℓ1ε1+· · ·+ℓnεn ∈ Λ: Then, using x−1 (3.7.1) Q =Lλ∈Λ kxλ. One can then also realize a quantum torus as the unital associative k-algebra presented by generators x1, . . . , xn, x−1 n and relations 1 , . . . , x−1 xix−1 i = 1Q = x−1 i xi, xixj = qijxjxi. We will refer to this view of Q as a coordinatization. (c) The centre of Q is a Λ-graded subalgebra, where Ξ is the so-called central grading group: Z(Q) =Lξ∈Ξ Qξ Ξ = {λ ∈ Λ : Qλ ⊂ Z(Q)}. This is a free abelian group of rank z ≤ n. Hence Z(Q) is a Laurent polynomial ring in z variables, which we may take as t1, . . . , tz (these can be taken to be of the form xλ for suitable λ's). (d) The grading properties of a quantum torus Q show that Q is fgc in the sense that Q is finitely generated as a module over Z(Q) if and only if Ξ has finite index in Λ. Equivalently, for some (hence all) coordinatization all entries qij of the quantum matrix q have finite order. If this holds, then for every coordinatization the qij have finite order. 16 V. CHERNOUSOV, E. NEHER, AND A. PIANZOLA (e) We define [Q, Q] = spank{[a, b] : a, b ∈ Q}, a graded subspace of Q. One knows (see e.g. [BGK, Prop. 2.44(iii)] for k = C or [NY, (3.3.2)] in general) (3.7.2) Q = Z(Q) ⊕ [Q, Q]. (f) An element u of Q is invertible if and only if 0 6= u ∈ Qλ for some λ ∈ Λ. Derk(Q)λ consists of those derivations d satisfying d(Qµ) ⊂ Qλ+µ for all µ ∈ Λ. The inner derivations of Q are the maps ad q, given by ad(q)(q′) = qq′ − q′q for q, q′ ∈ Q. They form a (g) The derivation Lie algebra Derk(Q) is graded: Derk(Q) = Lλ∈Λ Derk(Q)λ where graded ideal IDer Q = {ad q : q ∈ Q} of Derk(Q). As in 2.2(e), the grading Q =Lλ∈Λ Qλ gives rise to degree derivations ∂θ of Q, defined by ∂θ(q) = θ(λ)q for θ ∈ HomZ(Λ, k) and q ∈ Qλ. We put DQ = {∂θ : θ ∈ HomZ(Λ, k)} and define the graded subalgebra of centroidal derivations. Then ([OP, Cor. 2.3]) CDer(Q) = Z(Q) DQ =Lξ∈Ξ Qξ DQ, Derk(Q) = IDer(Q) ⋊ CDer(Q), so IDer(Q) =Lλ6∈Ξ Derk(Q)λ. Let ε : Q → k be the linear form defined by ε(1Q) = 1 and ε(Qλ) = 0 for λ 6= 0. The skew- symmetric derivations with respect to the bilinear form (q, q′) 7→ ε(qq′) have the following description: SDer(Q) = SCDer(Q) ⊕ IDer(Q) (3.7.3) SCDer(Q) = SDer(Q) ∩ CDer(Q) =Lξ∈Ξ SCDer(Q)ξ where SCDer(Q)ξ = Qξ {∂θ : θ ∈ HomZ(Λ, k), θ(ξ) = 0}. 3.8. slℓ(A) as Lie torus. In this subsection we describe for which algebras A the Lie algebra slℓ(A) is a Lie torus as defined in 2.1 and identify the data 2.5 necessary to construct an EALA with centreless core slℓ(A). All un-attributed result can be found in [Ne3, §7] or are easily verified by the reader. We assume A 6= 0 throughout. (a) Let ∆ be the root system of type Aℓ−1, realized as ∆ = {εi − εj, 1 ≤ i, j ≤ ℓ} in standard notation. Then the Lie algebra slℓ(A) has a canonical grading by the root lattice Q(∆), (3.8.1) slℓ(A) =Lα∈∆ slℓ(A)α, slℓ(A)α =(A Eij, {x ∈ slℓ(A) : x diagonal}, α = 0. for α = εi − εj 6= 0, (b) Let e = aEij ∈ slℓ(A) for i 6= j. Then e is part of an sl2-triple (e, h, f ) satisfying [h, xβ] = hβ, α∨ixβ for all β ∈ ∆ and xβ ∈ Lβ if and only if a is invertible in A. In this case f = a−1Eji and h = Eii − Ejj. (c) Let Λ be an abelian group, and let A = Lλ∈Λ Aλ be a Λ-graded unital associative k-algebra. Then the Q(∆) grading (3.8.1) of slℓ(A) extends to a (Q(∆) ⊕ Λ)-grading of slℓ(A), α consist of those matrices, for which all entries lie in Aλ. Conversely, a by letting slℓ(A)λ Q(∆) ⊕ Λ-grading of slℓ(A) extending the Q(∆)-grading (3.8.1) arises from a Λ-grading of the associative algebra A as described above. slℓ(A) =Lα∈∆, λ∈Λ slℓ(A)λ α 17 (d) Because of (c), for slℓ(A) to satisfy the axiom (LT1) of 2.1 with Q(∆)-grading (3.8.1) it is necessary and sufficient for the associative k-algebra A to be Λ-graded. Observe that then also (LT2.b) holds since 0 6= 1A ∈ A0 and therefore 0 6= 1AEij ∈ slℓ(A)0 α for α = εi − εj 6= 0. torus. Because of (b), the axiom (LT2.a) holds if and only if A = Q = Lλ∈Λ Qλ is a quantum Since (LT3) is clear, (LT4) says that slℓ(Q) is a Lie torus of type (∆, Λ) if and only if Q is a quantum torus of type Λ, as defined in 3.7(a). In this case, it follows from (g) that L is fgc as defined in 2.2 if and only if Q is an fgc quantum torus in the sense of 3.7(d) -- but we will not assume this in the following. (e) In the remainder of this subsection we let L = slℓ(Q) for Q a quantum torus with grading group Λ. Because of (3.7.2), the assumption of Lemma 3.3 is fulfilled. Then (3.3.2) and (3.1.3) imply that L is a centreless Lie torus of type (∆, Λ). Hence, by Theorem 2.6, L is centreless core of an EALA obtained by the construction 2.5. We describe the bilinear forms (··) and derivation algebras D allowed in this construction in the next two items. (f) Every Λ-graded invariant symmetric bilinear form β on L has the form (3.4.2) where which we can assume to be 1 ∈ k. ε : Q → k is a linear form vanishing on L06=λ Qλ and is therefore given by the scalar ε(1Q) (g) 9 For z ∈ Z(Q) define χz ∈ Endk(cid:0)Mℓ(Q)) by χz (x) = (zxij) for x = (xij) ∈ Mℓ(Q). The map Z(Q) → Ctd(cid:0)slℓ(Q)(cid:1), z 7→ χz, is an isomorphism of k-algebras. Then χz stabilizes slℓ(Q) and defines by restriction a centroidal transformation of slℓ(Q). (h) For d ∈ Der(Q) we denote by Mℓ(d) the derivation of slℓ(Q) defined in (3.1.4). The maps d 7→ Mℓ(d) is clearly a monomorphism of Lie algebras. Moreover, Derk(cid:0)slℓ(Q)(cid:1) = IDer(cid:0)slℓ(Q)(cid:1) + Mℓ(cid:0) Der(Q)(cid:1) Mℓ(cid:0) IDer(Q)) = IDer(cid:0)slℓ(Q)(cid:1) ∩ Mℓ(cid:0) Der(Q)(cid:1) ≃ IDer(Q) CDer(cid:0)slℓ(Q)(cid:1) = Mℓ(cid:0) CDer(Q)(cid:1) ≃ CDer(Q) SDer(cid:0)slℓ(Q)(cid:1) = IDer(cid:0)slℓ(Q)(cid:1) + Mℓ(cid:0) SDer(Q)(cid:1) SCDer(cid:0)slℓ(Q)(cid:1) = Mℓ(SCDer(Q)(cid:1) ≃ SCDer(Q) for Der(Q), IDer(Q), CDer(Q), SDer(Q) and SCDer(Q) described in 3.7(g). Note that the first three equations above together with 3.7(g) prove (2.2.3) for the case L = slℓ(Q). (i) The maximal possible choice for D in the construction 2.5 is SCDer(cid:0)slℓ(Q)(cid:1) which we identify with SCDer(Q) using the isomorphism Mℓ of (h). For Q = k[x± n ], n ≥ 2, a non-zero affine cocycle τ has been exhibited in [BGK, Rem. 3.71]. It can be described as follows. 1 , . . . , x±1 Modulo the isomorphism Mℓ of (h) we identify SCDer(cid:0)slℓ(Q)(cid:1) with SCDer(Q). Denoting by h·, ·i the standard inner product of kn and using the natural embedding Zn ⊂ kn we can further identify SCDer(Q) =Lλ∈Λ=Zn SCDer(Q)λ, where for λ = (λ1, . . . , λn) ∈ Zn SCDer(Q)λ ≡ {u = (ui) ∈ kn :Pn i=1 uiλi = 0} =: Dλ, cf. (3.7.3). uα ∈ Dα, vβ ∈ Dβ and wγ ∈ Dγ define τ (uα, vβ) (wγ) =(α(v) β(w) γ(u) 0 if α + β + γ = 0, otherwise, 9 The items (g) and (h) are true for any algebra A in place of Q 18 V. CHERNOUSOV, E. NEHER, AND A. PIANZOLA Then τ is an affine cocycle. It is non-trivial in the sense that the EALAs associated with L = slℓ(Q), ℓ ≥ 3, D = SCDer(L) and the two affine cocycles τ as above respectively τ = 0 are not isomorphic ([Kr, Thm. 5.76]. 4. Proof of the Main Theorem The proof of our main result will be based on the computation of K-Theory of non- commutative (twisted) Laurent polynomial rings due to D. Quillen. We first briefly recall functors K0 and K1. A nice introduction to the subject can be found in [Ro] and [Wb]. 4.1. K0(A) and K1(A) for a ring A. Let A be a ring (unital, but not necessarily commu- tative). If P is a (left) A-module, we denote its isomorphism class by [P ]. Consider the free abelian group F K0(A) generated by the set of isomorphism classes of projective A-modules of finite type. Then K0(A) is the quotient of the group F K0(A) by the normal subgroup generated by the relation [P ] = [P ′] + [P ′′] whenever there exists an exact sequence of A-modules 0 → P ′ → P → P ′′ → 0. As in 3.2 we denote by GLℓ(A), ℓ ∈ N+, the group of invertible ℓ × ℓ matrices with entries in A. For each m ∈ N+ we have a natural embedding GLℓ(A) ֒→ GLℓ+m(A) given by (4.1.1) X −→(cid:18)X 0 Eℓ+m(cid:19) , 0 cf. (3.5.1) for the corresponding embedding on the level of Lie algebras. We let GL∞(A) be the direct limit of GLℓ(A) with respect to the embeddings (4.1.1). Again as in 3.2, we let ELℓ(A) be the elementary linear subgroup of GLℓ(A) and let EL∞(A) be the direct limit of the ELℓ(A). Then K1(A) = GL∞(A)/[GL∞(A), GL∞(A)] = GL∞(A)/EL∞(A) (the first equality is the standard definition of K1(A), while the second equality is a classical theorem of Whitehead.) 4.2. Remark. The construction of K0 and K1 is functorial on k-algebras. Given a k- algebra homomorphism η : A → B we will denote by η∗ the induced group homomorphisms K0(A) → K0(B) and K1(A) → K1(B). Next we recall the definition of noncommutative Laurent polynomial ring Aφ[t±1]. Con- sider an automorphism φ of a (unital, associative and not necessarily commutative) k- algebra A. The multiplication in A will be denoted by juxtaposition. We define a new unital and associative k-algebra Aφ[t±1] as follows. The underlying k-vector space struc- ture is the free left A-module with basis {tm}m∈Z. The multiplication on Aφ[t±1], which we will denoted by ·, is given by Σi∈Z aiti · Σj∈Z a′ (4.2.1) It is known that if A is noetherian (resp. regular), so is Aφ[t±1] (see [Ar] Prop. 2.21). We also observe that φ induces a natural action φ∗ on K0(A) and K1(A). Namely, if P is a projective A-module then φ∗([P ]) := [P ⊗φ A]. It is obvious that if P is free or projective of finite type, so is φ∗(P ). Also, for every matrix X = (xij) in GLℓ(A) we let φ∗(X) = (φ(xij)). This of course induces an action φ∗ on GL∞(A) stabilizing EL∞(A), and hence also an action on K1(A). jtj = Σi,j∈Z aiφi(a′ for all ai, a′ j)ti+j j ∈ A. The following result is due to D. Quillen [Qu, §6, page 122]. 4.3. Theorem. Let φ ∈ Autk(A). Assume that A is noetherian and regular. Let η : A → Aφ[t±1] be the canonical embedding of k-algebras. Then the following sequence of abelian groups (4.3.1) 1−φ∗ K1(A) / K1(A) η∗ / K1(Aφ[t±1]) ∂ / K0(A) 1−φ∗ / K0(A) η∗ / K0(Aφ[t±1]) / 0 is exact.10 19 We will apply Theorem 4.3 to a quantum torus Q, Thus, as explained in 3.7, we can view n and i xi, xixj = qijxjxi, where the qij are non-zero elements of k, ji . For convenience in what follows we assume that the elements qij are Q as the unital associative k-algebra presented by generators x1, . . . , xn, x−1 relations xix−1 qii = 1 and qij = q−1 fixed throughout our discussion, and write i = 1Q = x−1 1 , . . . , x−1 Q = k[x±1 1 , · · · , x±1 n ]. It is immediate from the defining relations that the k-vector space Q is a direct sum Q = Li1,...,in∈Z kxi1 1 · · · xin n . The quantum torus Q contains a subring Qn−1 = k[x±1 1 , . . . , x±1 n−1] 1 , . . . , x±1 generated by x±1 n−1. Obviously, the conjugation by xn stabilizes Qn−1 and thus induces an automorphism φ on Qn−1 so that we may view Q as a noncommutative Laurent polynomial ring Q = Aφ[x±1 n ] where A = Qn−1. The advantage of realizing Q in this form is that it allows us to compute K0(Q) and K1(Q) by induction on n. We start with computing K0(Q). 4.4. Lemma. The group K0(Q) is isomorphic to Z. Its generator is the class of a free Q-module of rank 1. This is [Ar, Thm. 3.17]. We include a short proof for the sake of completeness. Proof. We reason by induction on n. If n = 1 then Q = k[x±1] is a commutative Laurent polynomial ring. Since Q is then a principal ideal domain, every projective Q-module is free. Our result is then clear. Assume n > 1. Consider the natural k-algebra inclusion η : Qn−1 → Qn. By induction we may assume that K0(Qn−1) ≃ Z. Since φ∗ acts trivially on its generator, it acts trivially on K0(Qn−1). From Quillen's exact sequence (4.3.1) we see that the base change map η : K0(Qn−1) → K0(Q) is an isomorphism and the result follows. (cid:3) We now pass to the computation of the group K1(Q) for a quantum torus Q. We first remark that for an arbitrary ring A and a unit u ∈ A× the 1 × 1 matrix (u) is an element of GL1(A). Taking the composition of A× → GL1(A) with GL1(A) → GL∞(A) → K1(A) we obtain a canonical group homomorphism λA : A× → K1(A). In general, λA is neither injective nor surjective, but we will show that λQ is surjective when Q is a quantum torus. 4.5. Proposition. Let Q be a quantum torus. Then λQ : Q× → K1(Q) is surjective. Proof. We argue by induction on n ∈ N. In case n = 0 and n = 1 it is well-known that λQ is actually an isomorphism: K1(k) ≃ k× and K1(k[t±1]) ≃ k[t±1]× for any field k by (for example) [Ro, Prop. 2.2.2] and [Ro, Thm. 2.3.2] respectively, where both isomorphisms are induced by the determinant. Thus we can assume n ≥ 2 in the following. 10The maps φ∗ and η∗ have been defined already. The nature of ∂ is explained in Quillen's paper. / / / / / / 20 V. CHERNOUSOV, E. NEHER, AND A. PIANZOLA Consider the sequence (4.3.1) with A = Qn−1. We already know, by Lemma 4.4, that φ∗ acts trivially on K0(Qn−1) so that we have a commutative diagram with an exact horizontal row at the bottom: Q× n−1 1−φ Q× n−1 λQn−1 λQn−1 K1(Qn−1) 1−φ∗ / K1(Qn−1) η η∗ Q× λQ / K1(Q) ∂ / / K0(Qn−1) / 0 By induction, λQn−1 is surjective. By Lemma 4.4, K0(Qn−1) ≃ Z. Clearly Q× is generated by k× and x1, . . . , xn, cf.3.7(f). It is shown in the proof of Lemma 5.16 in [Qu] that ∂(λQ(xi)) is a generator of K0(Q) ≃ Z. To prove surjectivity of λQ, let a ∈ K1(Q). Then ∂(a) = m ∈ Z and either the element a − λQ(xm i ) lies in the kernel of ∂. The claim now follows by a standard diagram chase. (cid:3) i ) or a + λQ(xm 4.6. Remark. A further diagram chase yields more than surjectivity. Q×/[Q×, Q×]. We do not need this more detailed result for our purposes. In fact K1(Q) = Interpreted in terms of matrices, Proposition 4.5 yields the following corollary. 4.7. Corollary. Let Q be a quantum torus. Let h ∈ GLℓ(Q). Then there exists a nonnegative integer m and a unit u ∈ Q× such that the matrix (4.7.1) is contained in ELℓ+m(Q). 0 (cid:18)h 0 Em(cid:19)(cid:18)u 0 Eℓ+m−1(cid:19) 0 4.8. Proof of the Main Theorem. To prove the Main Theorem as stated in the intro- i=1 siEii : duction, we can assume that the Cartan subalgebra H of E is such that Hcc = {Pℓ si ∈ k, Pi si = 0} =: hst in the notation of [CNP]. Let (E, H ′) be a second EALA structure, and set H ′ cc = h. We then know by the main theorem of [CNP] that there exists h ∈ GLℓ(Q) such that Int(h) maps hst to h. We now apply Corollary 4.7 and get u ∈ Q× such that the matrix of (4.7.1) is elementary. Put Then also Int(g) maps hst to h ([CNP, Lemma 2.10]). Moreover, g = h (cid:18)u 0 Eℓ−1(cid:19) ∈ GLℓ(Q). 0 is elementary. Because of (3.7.2) we can now apply Theorem 3.6 and obtain that Int(g) lifts to an automorphism of E. This finishes the proof. (cid:3) 0 (cid:18)g 0 Em(cid:19) = g′ References [Al] [AABGP] B. Allison, S. Azam, S. Berman, Y. Gao and A. Pianzola, Extended affine Lie algebras and their B. Allison, Some Isomorphism Invariants for Lie Tori, J. Lie Theory, 22 (2012), 163-204. root systems, Mem. Amer. Math. Soc. 126 (1997), no. 603, x+122. [ABFP] B. Allison, S. Berman, J. Faulkner and A. Pianzola, Multiloop realization of extended affine Lie algebras and Lie tori, Trans. Amer. Math. Soc. 361 (2009), 4807 -- 4842. [ABGP] B. Allison, S. Berman, Y. Gao and A. Pianzola, A characterization of affine Kac-Moody Lie [Ar] algebras, Comm. Math. Phys. 185 (1997), 671 -- 688. V. A. Artamonov, The quantum Serre problem (in Russian), translated from Uspekhi Mat. Nauk 53 (1998), no. 4(322), 3 -- 76 in Russian Math. Surveys 53 (1998), no. 4, 657 -- 730. / / / / / / / 21 [BN] [BGK] [BGKN] [Bo] [CGP] G. Benkart and E. Neher, The centroid of extended affine and root graded Lie algebras, J. Pure Appl. Algebra 205 (2006), 117 -- 145. S. Berman, Y. Gao and Y. Krylyuk, Quantum tori and the structure of elliptic quasi-simple Lie algebras, J. Funct. Anal. 135 (1996), 339 -- 389. S. Berman, Y. Gao, Y. Krylyuk, and E. Neher, The alternative torus and the structure of elliptic quasi-simple Lie algebras of type A2, Trans. Amer. Math. Soc. 347 (1995), 4315 -- 4363. N. Bourbaki, Groupes et Alg`ebres de Lie, Ch. 7 et 8, Hermann, Paris (1968). V. Chernousov, P. Gille and A. Pianzola, Conjugacy theorems for loop reductive group schemes and Lie algebras, Bull. Math. Sci. 4 (2014), 281-324. [CNPY] V. Chernousov, E. Neher, A. Pianzola and U. Yahorau. On conjugacy of Cartan subalgebras in [CNP] [K] [Kr] [Ne1] [Ne2] [Ne3] [Ne4] [NPPS] [NY] [OP] [PK] [Qu] [Ro] [Wb] [Yo1] [Yo2] extended affine Lie algebras, Adv. Math. 290 (2016), 260 -- 292. V. Chernousov, E. Neher, and A. Pianzola, On conjugacy of Cartan subalgebras in non-fgc Lie tori, Transformation Groups, 21 (2016), 1003 -- 1037. V. Kac, Infinite dimensional Lie algebras Cambridge University Press (1985). Y. Krylyuk, On automorphisms and isomorphisms of quasi-simple Lie algebras, J. Math. Sci. (New York) 100 (2000), 1944 -- 2002. E. Neher, Lie tori, C. R. Math. Acad. Sci. Soc. R. Can. 26 (2004), 84-89. E. Neher, Extended affine Lie algebras, C. R. Math. Acad. Sci. Soc. R. Can. 26 (2004), 90-96. E. Neher, Extended affine Lie algebras and other generalizations of affine Lie algebras -- a survey, in Developments and trends in infinite-dimensional Lie theory, Progr. Math. 288 (2011), 53 -- 126, Birkhauser Boston Inc., Boston, MA. E. Neher, Extended affine Lie Algebras -- An Introductrion to Their Structure Theory, in Geometric representation theory and extended affine Lie algebras, Fields Inst. Commun. 59 (2011), 107 -- 167, Amer. Math. Soc., Providence, RI. E. Neher, A. Pianzola, D. Prelat and C. Sepp, by faithfully flat descent, Communications 10.1142/S0219199714500096. E. Neher and Y. Yoshii, Derivations and invariant forms of Jordan and alternative tori, Trans. Amer. Math. Soc. 355 (2003), 1079 -- 1108. J. M. Osborn and D. S. Passman, Derivations of skew polynomial rings, J. Algebra 176 (1995), 417 -- 448. D.H. Peterson and V. Kac, Infinite flag varieties and conjugacy theorems, Proc. Natl. Acad. Sci. USA 80 (1983), 1778 -- 1782. D. Quillen, Higher algebraic K-theory. I, in Algebraic K-theory, I: Higher K-theories, Proc. Conf., Battelle Memorial Inst., Seattle, Wash., Lecture Notes in Math., Vol. 341 (1973), pages 85 -- 147, Springer-Verlag, Berlin. J. Rosenberg, Algebraic K-Theory and Its Applications, Graduate Texts in Mathematics 147, Springer-Verlag New York (1994). C. Weibel, The K-book: Introduction to Algebraic K-Theory Graduate Studies in Mathematics 145, American Mathematical Society (2013). Y. Yoshii, Coordinate Algebras of Extended Affine Lie Algebras of Type A1, J. Algebra 234 (2000), 128 -- 168. forms of algebras given in Contemporary Mathematics 2014, DOI: Invariant , Lie tori -- a simple characterization of extended affine Lie algebras, Publ. Res. Inst. Math. Sci. 42 (2006), 739 -- 762. Department of Mathematics, University of Alberta, Edmonton, Alberta T6G 2G1, Canada E-mail address: [email protected] Department of Mathematics and Statistics, University of Ottawa, Ottawa, Ontario K1N 6N5, Canada E-mail address: [email protected] Department of Mathematics, University of Alberta, Edmonton, Alberta T6G 2G1, Canada. Centro de Altos Estudios en Ciencia Exactas, Avenida de Mayo 866, (1084) Buenos Aires, Argentina. E-mail address: [email protected]

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.